4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
27 #include <sys/dmu_objset.h>
28 #include <sys/dsl_dataset.h>
29 #include <sys/dsl_dir.h>
30 #include <sys/dsl_prop.h>
31 #include <sys/dsl_synctask.h>
32 #include <sys/dmu_traverse.h>
33 #include <sys/dmu_impl.h>
34 #include <sys/dmu_tx.h>
38 #include <sys/zfeature.h>
39 #include <sys/unique.h>
40 #include <sys/zfs_context.h>
41 #include <sys/zfs_ioctl.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/zfs_onexit.h>
46 #include <sys/dsl_scan.h>
47 #include <sys/dsl_deadlist.h>
48 #include <sys/dsl_destroy.h>
49 #include <sys/dsl_userhold.h>
51 #define SWITCH64(x, y) \
53 uint64_t __tmp = (x); \
58 #define DS_REF_MAX (1ULL << 62)
60 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
63 * Figure out how much of this delta should be propogated to the dsl_dir
64 * layer. If there's a refreservation, that space has already been
65 * partially accounted for in our ancestors.
68 parent_delta(dsl_dataset_t
*ds
, int64_t delta
)
70 uint64_t old_bytes
, new_bytes
;
72 if (ds
->ds_reserved
== 0)
75 old_bytes
= MAX(ds
->ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
76 new_bytes
= MAX(ds
->ds_phys
->ds_unique_bytes
+ delta
, ds
->ds_reserved
);
78 ASSERT3U(ABS((int64_t)(new_bytes
- old_bytes
)), <=, ABS(delta
));
79 return (new_bytes
- old_bytes
);
83 dsl_dataset_block_born(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
85 int used
, compressed
, uncompressed
;
88 used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
89 compressed
= BP_GET_PSIZE(bp
);
90 uncompressed
= BP_GET_UCSIZE(bp
);
92 dprintf_bp(bp
, "ds=%p", ds
);
94 ASSERT(dmu_tx_is_syncing(tx
));
95 /* It could have been compressed away to nothing */
98 ASSERT(BP_GET_TYPE(bp
) != DMU_OT_NONE
);
99 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp
)));
101 dsl_pool_mos_diduse_space(tx
->tx_pool
,
102 used
, compressed
, uncompressed
);
106 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
107 mutex_enter(&ds
->ds_lock
);
108 delta
= parent_delta(ds
, used
);
109 ds
->ds_phys
->ds_referenced_bytes
+= used
;
110 ds
->ds_phys
->ds_compressed_bytes
+= compressed
;
111 ds
->ds_phys
->ds_uncompressed_bytes
+= uncompressed
;
112 ds
->ds_phys
->ds_unique_bytes
+= used
;
113 mutex_exit(&ds
->ds_lock
);
114 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
, delta
,
115 compressed
, uncompressed
, tx
);
116 dsl_dir_transfer_space(ds
->ds_dir
, used
- delta
,
117 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
121 dsl_dataset_block_kill(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
,
124 int used
, compressed
, uncompressed
;
129 ASSERT(dmu_tx_is_syncing(tx
));
130 ASSERT(bp
->blk_birth
<= tx
->tx_txg
);
132 used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
133 compressed
= BP_GET_PSIZE(bp
);
134 uncompressed
= BP_GET_UCSIZE(bp
);
138 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
139 dsl_pool_mos_diduse_space(tx
->tx_pool
,
140 -used
, -compressed
, -uncompressed
);
143 ASSERT3P(tx
->tx_pool
, ==, ds
->ds_dir
->dd_pool
);
145 ASSERT(!dsl_dataset_is_snapshot(ds
));
146 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
148 if (bp
->blk_birth
> ds
->ds_phys
->ds_prev_snap_txg
) {
151 dprintf_bp(bp
, "freeing ds=%llu", ds
->ds_object
);
152 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
154 mutex_enter(&ds
->ds_lock
);
155 ASSERT(ds
->ds_phys
->ds_unique_bytes
>= used
||
156 !DS_UNIQUE_IS_ACCURATE(ds
));
157 delta
= parent_delta(ds
, -used
);
158 ds
->ds_phys
->ds_unique_bytes
-= used
;
159 mutex_exit(&ds
->ds_lock
);
160 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
161 delta
, -compressed
, -uncompressed
, tx
);
162 dsl_dir_transfer_space(ds
->ds_dir
, -used
- delta
,
163 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
165 dprintf_bp(bp
, "putting on dead list: %s", "");
168 * We are here as part of zio's write done callback,
169 * which means we're a zio interrupt thread. We can't
170 * call dsl_deadlist_insert() now because it may block
171 * waiting for I/O. Instead, put bp on the deferred
172 * queue and let dsl_pool_sync() finish the job.
174 bplist_append(&ds
->ds_pending_deadlist
, bp
);
176 dsl_deadlist_insert(&ds
->ds_deadlist
, bp
, tx
);
178 ASSERT3U(ds
->ds_prev
->ds_object
, ==,
179 ds
->ds_phys
->ds_prev_snap_obj
);
180 ASSERT(ds
->ds_prev
->ds_phys
->ds_num_children
> 0);
181 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
182 if (ds
->ds_prev
->ds_phys
->ds_next_snap_obj
==
183 ds
->ds_object
&& bp
->blk_birth
>
184 ds
->ds_prev
->ds_phys
->ds_prev_snap_txg
) {
185 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
186 mutex_enter(&ds
->ds_prev
->ds_lock
);
187 ds
->ds_prev
->ds_phys
->ds_unique_bytes
+= used
;
188 mutex_exit(&ds
->ds_prev
->ds_lock
);
190 if (bp
->blk_birth
> ds
->ds_dir
->dd_origin_txg
) {
191 dsl_dir_transfer_space(ds
->ds_dir
, used
,
192 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
195 mutex_enter(&ds
->ds_lock
);
196 ASSERT3U(ds
->ds_phys
->ds_referenced_bytes
, >=, used
);
197 ds
->ds_phys
->ds_referenced_bytes
-= used
;
198 ASSERT3U(ds
->ds_phys
->ds_compressed_bytes
, >=, compressed
);
199 ds
->ds_phys
->ds_compressed_bytes
-= compressed
;
200 ASSERT3U(ds
->ds_phys
->ds_uncompressed_bytes
, >=, uncompressed
);
201 ds
->ds_phys
->ds_uncompressed_bytes
-= uncompressed
;
202 mutex_exit(&ds
->ds_lock
);
208 dsl_dataset_prev_snap_txg(dsl_dataset_t
*ds
)
210 uint64_t trysnap
= 0;
215 * The snapshot creation could fail, but that would cause an
216 * incorrect FALSE return, which would only result in an
217 * overestimation of the amount of space that an operation would
218 * consume, which is OK.
220 * There's also a small window where we could miss a pending
221 * snapshot, because we could set the sync task in the quiescing
222 * phase. So this should only be used as a guess.
224 if (ds
->ds_trysnap_txg
>
225 spa_last_synced_txg(ds
->ds_dir
->dd_pool
->dp_spa
))
226 trysnap
= ds
->ds_trysnap_txg
;
227 return (MAX(ds
->ds_phys
->ds_prev_snap_txg
, trysnap
));
231 dsl_dataset_block_freeable(dsl_dataset_t
*ds
, const blkptr_t
*bp
,
234 if (blk_birth
<= dsl_dataset_prev_snap_txg(ds
))
237 ddt_prefetch(dsl_dataset_get_spa(ds
), bp
);
244 dsl_dataset_evict(dmu_buf_t
*db
, void *dsv
)
246 dsl_dataset_t
*ds
= dsv
;
248 ASSERT(ds
->ds_owner
== NULL
);
250 unique_remove(ds
->ds_fsid_guid
);
252 if (ds
->ds_objset
!= NULL
)
253 dmu_objset_evict(ds
->ds_objset
);
256 dsl_dataset_rele(ds
->ds_prev
, ds
);
260 bplist_destroy(&ds
->ds_pending_deadlist
);
261 if (ds
->ds_phys
->ds_deadlist_obj
!= 0)
262 dsl_deadlist_close(&ds
->ds_deadlist
);
264 dsl_dir_rele(ds
->ds_dir
, ds
);
266 ASSERT(!list_link_active(&ds
->ds_synced_link
));
268 mutex_destroy(&ds
->ds_lock
);
269 mutex_destroy(&ds
->ds_opening_lock
);
270 refcount_destroy(&ds
->ds_longholds
);
272 kmem_free(ds
, sizeof (dsl_dataset_t
));
276 dsl_dataset_get_snapname(dsl_dataset_t
*ds
)
278 dsl_dataset_phys_t
*headphys
;
281 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
282 objset_t
*mos
= dp
->dp_meta_objset
;
284 if (ds
->ds_snapname
[0])
286 if (ds
->ds_phys
->ds_next_snap_obj
== 0)
289 err
= dmu_bonus_hold(mos
, ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
,
293 headphys
= headdbuf
->db_data
;
294 err
= zap_value_search(dp
->dp_meta_objset
,
295 headphys
->ds_snapnames_zapobj
, ds
->ds_object
, 0, ds
->ds_snapname
);
296 dmu_buf_rele(headdbuf
, FTAG
);
301 dsl_dataset_snap_lookup(dsl_dataset_t
*ds
, const char *name
, uint64_t *value
)
303 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
304 uint64_t snapobj
= ds
->ds_phys
->ds_snapnames_zapobj
;
308 if (ds
->ds_phys
->ds_flags
& DS_FLAG_CI_DATASET
)
313 err
= zap_lookup_norm(mos
, snapobj
, name
, 8, 1,
314 value
, mt
, NULL
, 0, NULL
);
315 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
316 err
= zap_lookup(mos
, snapobj
, name
, 8, 1, value
);
321 dsl_dataset_snap_remove(dsl_dataset_t
*ds
, const char *name
, dmu_tx_t
*tx
)
323 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
324 uint64_t snapobj
= ds
->ds_phys
->ds_snapnames_zapobj
;
328 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
330 if (ds
->ds_phys
->ds_flags
& DS_FLAG_CI_DATASET
)
335 err
= zap_remove_norm(mos
, snapobj
, name
, mt
, tx
);
336 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
337 err
= zap_remove(mos
, snapobj
, name
, tx
);
342 dsl_dataset_hold_obj(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
345 objset_t
*mos
= dp
->dp_meta_objset
;
349 dmu_object_info_t doi
;
351 ASSERT(dsl_pool_config_held(dp
));
353 err
= dmu_bonus_hold(mos
, dsobj
, tag
, &dbuf
);
357 /* Make sure dsobj has the correct object type. */
358 dmu_object_info_from_db(dbuf
, &doi
);
359 if (doi
.doi_type
!= DMU_OT_DSL_DATASET
) {
360 dmu_buf_rele(dbuf
, tag
);
361 return (SET_ERROR(EINVAL
));
364 ds
= dmu_buf_get_user(dbuf
);
366 dsl_dataset_t
*winner
= NULL
;
368 ds
= kmem_zalloc(sizeof (dsl_dataset_t
), KM_PUSHPAGE
);
370 ds
->ds_object
= dsobj
;
371 ds
->ds_phys
= dbuf
->db_data
;
372 list_link_init(&ds
->ds_synced_link
);
374 mutex_init(&ds
->ds_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
375 mutex_init(&ds
->ds_opening_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
376 mutex_init(&ds
->ds_sendstream_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
377 refcount_create(&ds
->ds_longholds
);
379 bplist_create(&ds
->ds_pending_deadlist
);
380 dsl_deadlist_open(&ds
->ds_deadlist
,
381 mos
, ds
->ds_phys
->ds_deadlist_obj
);
383 list_create(&ds
->ds_sendstreams
, sizeof (dmu_sendarg_t
),
384 offsetof(dmu_sendarg_t
, dsa_link
));
387 err
= dsl_dir_hold_obj(dp
,
388 ds
->ds_phys
->ds_dir_obj
, NULL
, ds
, &ds
->ds_dir
);
391 mutex_destroy(&ds
->ds_lock
);
392 mutex_destroy(&ds
->ds_opening_lock
);
393 refcount_destroy(&ds
->ds_longholds
);
394 bplist_destroy(&ds
->ds_pending_deadlist
);
395 dsl_deadlist_close(&ds
->ds_deadlist
);
396 kmem_free(ds
, sizeof (dsl_dataset_t
));
397 dmu_buf_rele(dbuf
, tag
);
401 if (!dsl_dataset_is_snapshot(ds
)) {
402 ds
->ds_snapname
[0] = '\0';
403 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0) {
404 err
= dsl_dataset_hold_obj(dp
,
405 ds
->ds_phys
->ds_prev_snap_obj
,
409 if (zfs_flags
& ZFS_DEBUG_SNAPNAMES
)
410 err
= dsl_dataset_get_snapname(ds
);
411 if (err
== 0 && ds
->ds_phys
->ds_userrefs_obj
!= 0) {
413 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
414 ds
->ds_phys
->ds_userrefs_obj
,
419 if (err
== 0 && !dsl_dataset_is_snapshot(ds
)) {
420 err
= dsl_prop_get_int_ds(ds
,
421 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
424 err
= dsl_prop_get_int_ds(ds
,
425 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
429 ds
->ds_reserved
= ds
->ds_quota
= 0;
432 if (err
!= 0 || (winner
= dmu_buf_set_user_ie(dbuf
, ds
,
433 &ds
->ds_phys
, dsl_dataset_evict
)) != NULL
) {
434 bplist_destroy(&ds
->ds_pending_deadlist
);
435 dsl_deadlist_close(&ds
->ds_deadlist
);
437 dsl_dataset_rele(ds
->ds_prev
, ds
);
438 dsl_dir_rele(ds
->ds_dir
, ds
);
439 mutex_destroy(&ds
->ds_lock
);
440 mutex_destroy(&ds
->ds_opening_lock
);
441 refcount_destroy(&ds
->ds_longholds
);
442 kmem_free(ds
, sizeof (dsl_dataset_t
));
444 dmu_buf_rele(dbuf
, tag
);
450 unique_insert(ds
->ds_phys
->ds_fsid_guid
);
453 ASSERT3P(ds
->ds_dbuf
, ==, dbuf
);
454 ASSERT3P(ds
->ds_phys
, ==, dbuf
->db_data
);
455 ASSERT(ds
->ds_phys
->ds_prev_snap_obj
!= 0 ||
456 spa_version(dp
->dp_spa
) < SPA_VERSION_ORIGIN
||
457 dp
->dp_origin_snap
== NULL
|| ds
== dp
->dp_origin_snap
);
463 dsl_dataset_hold(dsl_pool_t
*dp
, const char *name
,
464 void *tag
, dsl_dataset_t
**dsp
)
467 const char *snapname
;
471 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, &snapname
);
475 ASSERT(dsl_pool_config_held(dp
));
476 obj
= dd
->dd_phys
->dd_head_dataset_obj
;
478 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, dsp
);
480 err
= SET_ERROR(ENOENT
);
482 /* we may be looking for a snapshot */
483 if (err
== 0 && snapname
!= NULL
) {
486 if (*snapname
++ != '@') {
487 dsl_dataset_rele(*dsp
, tag
);
488 dsl_dir_rele(dd
, FTAG
);
489 return (SET_ERROR(ENOENT
));
492 dprintf("looking for snapshot '%s'\n", snapname
);
493 err
= dsl_dataset_snap_lookup(*dsp
, snapname
, &obj
);
495 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
496 dsl_dataset_rele(*dsp
, tag
);
499 mutex_enter(&ds
->ds_lock
);
500 if (ds
->ds_snapname
[0] == 0)
501 (void) strlcpy(ds
->ds_snapname
, snapname
,
502 sizeof (ds
->ds_snapname
));
503 mutex_exit(&ds
->ds_lock
);
508 dsl_dir_rele(dd
, FTAG
);
513 dsl_dataset_own_obj(dsl_pool_t
*dp
, uint64_t dsobj
,
514 void *tag
, dsl_dataset_t
**dsp
)
516 int err
= dsl_dataset_hold_obj(dp
, dsobj
, tag
, dsp
);
519 if (!dsl_dataset_tryown(*dsp
, tag
)) {
520 dsl_dataset_rele(*dsp
, tag
);
522 return (SET_ERROR(EBUSY
));
528 dsl_dataset_own(dsl_pool_t
*dp
, const char *name
,
529 void *tag
, dsl_dataset_t
**dsp
)
531 int err
= dsl_dataset_hold(dp
, name
, tag
, dsp
);
534 if (!dsl_dataset_tryown(*dsp
, tag
)) {
535 dsl_dataset_rele(*dsp
, tag
);
536 return (SET_ERROR(EBUSY
));
542 * See the comment above dsl_pool_hold() for details. In summary, a long
543 * hold is used to prevent destruction of a dataset while the pool hold
544 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
546 * The dataset and pool must be held when this function is called. After it
547 * is called, the pool hold may be released while the dataset is still held
551 dsl_dataset_long_hold(dsl_dataset_t
*ds
, void *tag
)
553 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
554 (void) refcount_add(&ds
->ds_longholds
, tag
);
558 dsl_dataset_long_rele(dsl_dataset_t
*ds
, void *tag
)
560 (void) refcount_remove(&ds
->ds_longholds
, tag
);
563 /* Return B_TRUE if there are any long holds on this dataset. */
565 dsl_dataset_long_held(dsl_dataset_t
*ds
)
567 return (!refcount_is_zero(&ds
->ds_longholds
));
571 dsl_dataset_name(dsl_dataset_t
*ds
, char *name
)
574 (void) strcpy(name
, "mos");
576 dsl_dir_name(ds
->ds_dir
, name
);
577 VERIFY0(dsl_dataset_get_snapname(ds
));
578 if (ds
->ds_snapname
[0]) {
579 (void) strcat(name
, "@");
581 * We use a "recursive" mutex so that we
582 * can call dprintf_ds() with ds_lock held.
584 if (!MUTEX_HELD(&ds
->ds_lock
)) {
585 mutex_enter(&ds
->ds_lock
);
586 (void) strcat(name
, ds
->ds_snapname
);
587 mutex_exit(&ds
->ds_lock
);
589 (void) strcat(name
, ds
->ds_snapname
);
596 dsl_dataset_rele(dsl_dataset_t
*ds
, void *tag
)
598 dmu_buf_rele(ds
->ds_dbuf
, tag
);
602 dsl_dataset_disown(dsl_dataset_t
*ds
, void *tag
)
604 ASSERT(ds
->ds_owner
== tag
&& ds
->ds_dbuf
!= NULL
);
606 mutex_enter(&ds
->ds_lock
);
608 mutex_exit(&ds
->ds_lock
);
609 dsl_dataset_long_rele(ds
, tag
);
610 if (ds
->ds_dbuf
!= NULL
)
611 dsl_dataset_rele(ds
, tag
);
613 dsl_dataset_evict(NULL
, ds
);
617 dsl_dataset_tryown(dsl_dataset_t
*ds
, void *tag
)
619 boolean_t gotit
= FALSE
;
621 mutex_enter(&ds
->ds_lock
);
622 if (ds
->ds_owner
== NULL
&& !DS_IS_INCONSISTENT(ds
)) {
624 dsl_dataset_long_hold(ds
, tag
);
627 mutex_exit(&ds
->ds_lock
);
632 dsl_dataset_create_sync_dd(dsl_dir_t
*dd
, dsl_dataset_t
*origin
,
633 uint64_t flags
, dmu_tx_t
*tx
)
635 dsl_pool_t
*dp
= dd
->dd_pool
;
637 dsl_dataset_phys_t
*dsphys
;
639 objset_t
*mos
= dp
->dp_meta_objset
;
642 origin
= dp
->dp_origin_snap
;
644 ASSERT(origin
== NULL
|| origin
->ds_dir
->dd_pool
== dp
);
645 ASSERT(origin
== NULL
|| origin
->ds_phys
->ds_num_children
> 0);
646 ASSERT(dmu_tx_is_syncing(tx
));
647 ASSERT(dd
->dd_phys
->dd_head_dataset_obj
== 0);
649 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
650 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
651 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
652 dmu_buf_will_dirty(dbuf
, tx
);
653 dsphys
= dbuf
->db_data
;
654 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
655 dsphys
->ds_dir_obj
= dd
->dd_object
;
656 dsphys
->ds_flags
= flags
;
657 dsphys
->ds_fsid_guid
= unique_create();
658 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
659 sizeof (dsphys
->ds_guid
));
660 dsphys
->ds_snapnames_zapobj
=
661 zap_create_norm(mos
, U8_TEXTPREP_TOUPPER
, DMU_OT_DSL_DS_SNAP_MAP
,
663 dsphys
->ds_creation_time
= gethrestime_sec();
664 dsphys
->ds_creation_txg
= tx
->tx_txg
== TXG_INITIAL
? 1 : tx
->tx_txg
;
666 if (origin
== NULL
) {
667 dsphys
->ds_deadlist_obj
= dsl_deadlist_alloc(mos
, tx
);
669 dsl_dataset_t
*ohds
; /* head of the origin snapshot */
671 dsphys
->ds_prev_snap_obj
= origin
->ds_object
;
672 dsphys
->ds_prev_snap_txg
=
673 origin
->ds_phys
->ds_creation_txg
;
674 dsphys
->ds_referenced_bytes
=
675 origin
->ds_phys
->ds_referenced_bytes
;
676 dsphys
->ds_compressed_bytes
=
677 origin
->ds_phys
->ds_compressed_bytes
;
678 dsphys
->ds_uncompressed_bytes
=
679 origin
->ds_phys
->ds_uncompressed_bytes
;
680 dsphys
->ds_bp
= origin
->ds_phys
->ds_bp
;
681 dsphys
->ds_flags
|= origin
->ds_phys
->ds_flags
;
683 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
684 origin
->ds_phys
->ds_num_children
++;
686 VERIFY0(dsl_dataset_hold_obj(dp
,
687 origin
->ds_dir
->dd_phys
->dd_head_dataset_obj
, FTAG
, &ohds
));
688 dsphys
->ds_deadlist_obj
= dsl_deadlist_clone(&ohds
->ds_deadlist
,
689 dsphys
->ds_prev_snap_txg
, dsphys
->ds_prev_snap_obj
, tx
);
690 dsl_dataset_rele(ohds
, FTAG
);
692 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_NEXT_CLONES
) {
693 if (origin
->ds_phys
->ds_next_clones_obj
== 0) {
694 origin
->ds_phys
->ds_next_clones_obj
=
696 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
698 VERIFY0(zap_add_int(mos
,
699 origin
->ds_phys
->ds_next_clones_obj
, dsobj
, tx
));
702 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
703 dd
->dd_phys
->dd_origin_obj
= origin
->ds_object
;
704 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
705 if (origin
->ds_dir
->dd_phys
->dd_clones
== 0) {
706 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
707 origin
->ds_dir
->dd_phys
->dd_clones
=
709 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
711 VERIFY0(zap_add_int(mos
,
712 origin
->ds_dir
->dd_phys
->dd_clones
, dsobj
, tx
));
716 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
717 dsphys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
719 dmu_buf_rele(dbuf
, FTAG
);
721 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
722 dd
->dd_phys
->dd_head_dataset_obj
= dsobj
;
728 dsl_dataset_zero_zil(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
732 VERIFY0(dmu_objset_from_ds(ds
, &os
));
733 bzero(&os
->os_zil_header
, sizeof (os
->os_zil_header
));
734 dsl_dataset_dirty(ds
, tx
);
738 dsl_dataset_create_sync(dsl_dir_t
*pdd
, const char *lastname
,
739 dsl_dataset_t
*origin
, uint64_t flags
, cred_t
*cr
, dmu_tx_t
*tx
)
741 dsl_pool_t
*dp
= pdd
->dd_pool
;
742 uint64_t dsobj
, ddobj
;
745 ASSERT(dmu_tx_is_syncing(tx
));
746 ASSERT(lastname
[0] != '@');
748 ddobj
= dsl_dir_create_sync(dp
, pdd
, lastname
, tx
);
749 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, lastname
, FTAG
, &dd
));
751 dsobj
= dsl_dataset_create_sync_dd(dd
, origin
,
752 flags
& ~DS_CREATE_FLAG_NODIRTY
, tx
);
754 dsl_deleg_set_create_perms(dd
, tx
, cr
);
756 dsl_dir_rele(dd
, FTAG
);
759 * If we are creating a clone, make sure we zero out any stale
760 * data from the origin snapshots zil header.
762 if (origin
!= NULL
&& !(flags
& DS_CREATE_FLAG_NODIRTY
)) {
765 VERIFY0(dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
766 dsl_dataset_zero_zil(ds
, tx
);
767 dsl_dataset_rele(ds
, FTAG
);
774 * The unique space in the head dataset can be calculated by subtracting
775 * the space used in the most recent snapshot, that is still being used
776 * in this file system, from the space currently in use. To figure out
777 * the space in the most recent snapshot still in use, we need to take
778 * the total space used in the snapshot and subtract out the space that
779 * has been freed up since the snapshot was taken.
782 dsl_dataset_recalc_head_uniq(dsl_dataset_t
*ds
)
785 uint64_t dlused
, dlcomp
, dluncomp
;
787 ASSERT(!dsl_dataset_is_snapshot(ds
));
789 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0)
790 mrs_used
= ds
->ds_prev
->ds_phys
->ds_referenced_bytes
;
794 dsl_deadlist_space(&ds
->ds_deadlist
, &dlused
, &dlcomp
, &dluncomp
);
796 ASSERT3U(dlused
, <=, mrs_used
);
797 ds
->ds_phys
->ds_unique_bytes
=
798 ds
->ds_phys
->ds_referenced_bytes
- (mrs_used
- dlused
);
800 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) >=
801 SPA_VERSION_UNIQUE_ACCURATE
)
802 ds
->ds_phys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
806 dsl_dataset_remove_from_next_clones(dsl_dataset_t
*ds
, uint64_t obj
,
809 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
811 ASSERTV(uint64_t count
);
813 ASSERT(ds
->ds_phys
->ds_num_children
>= 2);
814 err
= zap_remove_int(mos
, ds
->ds_phys
->ds_next_clones_obj
, obj
, tx
);
816 * The err should not be ENOENT, but a bug in a previous version
817 * of the code could cause upgrade_clones_cb() to not set
818 * ds_next_snap_obj when it should, leading to a missing entry.
819 * If we knew that the pool was created after
820 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
821 * ENOENT. However, at least we can check that we don't have
822 * too many entries in the next_clones_obj even after failing to
827 ASSERT0(zap_count(mos
, ds
->ds_phys
->ds_next_clones_obj
,
829 ASSERT3U(count
, <=, ds
->ds_phys
->ds_num_children
- 2);
834 dsl_dataset_get_blkptr(dsl_dataset_t
*ds
)
836 return (&ds
->ds_phys
->ds_bp
);
840 dsl_dataset_set_blkptr(dsl_dataset_t
*ds
, blkptr_t
*bp
, dmu_tx_t
*tx
)
842 ASSERT(dmu_tx_is_syncing(tx
));
843 /* If it's the meta-objset, set dp_meta_rootbp */
845 tx
->tx_pool
->dp_meta_rootbp
= *bp
;
847 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
848 ds
->ds_phys
->ds_bp
= *bp
;
853 dsl_dataset_get_spa(dsl_dataset_t
*ds
)
855 return (ds
->ds_dir
->dd_pool
->dp_spa
);
859 dsl_dataset_dirty(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
863 if (ds
== NULL
) /* this is the meta-objset */
866 ASSERT(ds
->ds_objset
!= NULL
);
868 if (ds
->ds_phys
->ds_next_snap_obj
!= 0)
869 panic("dirtying snapshot!");
871 dp
= ds
->ds_dir
->dd_pool
;
873 if (txg_list_add(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
874 /* up the hold count until we can be written out */
875 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
880 dsl_dataset_is_dirty(dsl_dataset_t
*ds
)
884 for (t
= 0; t
< TXG_SIZE
; t
++) {
885 if (txg_list_member(&ds
->ds_dir
->dd_pool
->dp_dirty_datasets
,
893 dsl_dataset_snapshot_reserve_space(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
897 if (!dmu_tx_is_syncing(tx
))
901 * If there's an fs-only reservation, any blocks that might become
902 * owned by the snapshot dataset must be accommodated by space
903 * outside of the reservation.
905 ASSERT(ds
->ds_reserved
== 0 || DS_UNIQUE_IS_ACCURATE(ds
));
906 asize
= MIN(ds
->ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
907 if (asize
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
908 return (SET_ERROR(ENOSPC
));
911 * Propagate any reserved space for this snapshot to other
912 * snapshot checks in this sync group.
915 dsl_dir_willuse_space(ds
->ds_dir
, asize
, tx
);
920 typedef struct dsl_dataset_snapshot_arg
{
921 nvlist_t
*ddsa_snaps
;
922 nvlist_t
*ddsa_props
;
923 nvlist_t
*ddsa_errors
;
924 } dsl_dataset_snapshot_arg_t
;
927 dsl_dataset_snapshot_check_impl(dsl_dataset_t
*ds
, const char *snapname
,
928 dmu_tx_t
*tx
, boolean_t recv
)
933 ds
->ds_trysnap_txg
= tx
->tx_txg
;
935 if (!dmu_tx_is_syncing(tx
))
939 * We don't allow multiple snapshots of the same txg. If there
940 * is already one, try again.
942 if (ds
->ds_phys
->ds_prev_snap_txg
>= tx
->tx_txg
)
943 return (SET_ERROR(EAGAIN
));
946 * Check for conflicting snapshot name.
948 error
= dsl_dataset_snap_lookup(ds
, snapname
, &value
);
950 return (SET_ERROR(EEXIST
));
955 * We don't allow taking snapshots of inconsistent datasets, such as
956 * those into which we are currently receiving. However, if we are
957 * creating this snapshot as part of a receive, this check will be
958 * executed atomically with respect to the completion of the receive
959 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
960 * case we ignore this, knowing it will be fixed up for us shortly in
961 * dmu_recv_end_sync().
963 if (!recv
&& DS_IS_INCONSISTENT(ds
))
964 return (SET_ERROR(EBUSY
));
966 error
= dsl_dataset_snapshot_reserve_space(ds
, tx
);
974 dsl_dataset_snapshot_check(void *arg
, dmu_tx_t
*tx
)
976 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
977 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
981 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
982 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
986 char dsname
[MAXNAMELEN
];
988 name
= nvpair_name(pair
);
989 if (strlen(name
) >= MAXNAMELEN
)
990 error
= SET_ERROR(ENAMETOOLONG
);
992 atp
= strchr(name
, '@');
994 error
= SET_ERROR(EINVAL
);
996 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
999 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
1001 error
= dsl_dataset_snapshot_check_impl(ds
,
1002 atp
+ 1, tx
, B_FALSE
);
1003 dsl_dataset_rele(ds
, FTAG
);
1007 if (ddsa
->ddsa_errors
!= NULL
) {
1008 fnvlist_add_int32(ddsa
->ddsa_errors
,
1018 dsl_dataset_snapshot_sync_impl(dsl_dataset_t
*ds
, const char *snapname
,
1021 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1023 dsl_dataset_phys_t
*dsphys
;
1024 uint64_t dsobj
, crtxg
;
1025 objset_t
*mos
= dp
->dp_meta_objset
;
1026 ASSERTV(static zil_header_t zero_zil
);
1027 ASSERTV(objset_t
*os
);
1029 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
1032 * If we are on an old pool, the zil must not be active, in which
1033 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1035 ASSERT(spa_version(dmu_tx_pool(tx
)->dp_spa
) >= SPA_VERSION_FAST_SNAP
||
1036 dmu_objset_from_ds(ds
, &os
) != 0 ||
1037 bcmp(&os
->os_phys
->os_zil_header
, &zero_zil
,
1038 sizeof (zero_zil
)) == 0);
1042 * The origin's ds_creation_txg has to be < TXG_INITIAL
1044 if (strcmp(snapname
, ORIGIN_DIR_NAME
) == 0)
1049 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
1050 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
1051 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
1052 dmu_buf_will_dirty(dbuf
, tx
);
1053 dsphys
= dbuf
->db_data
;
1054 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
1055 dsphys
->ds_dir_obj
= ds
->ds_dir
->dd_object
;
1056 dsphys
->ds_fsid_guid
= unique_create();
1057 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
1058 sizeof (dsphys
->ds_guid
));
1059 dsphys
->ds_prev_snap_obj
= ds
->ds_phys
->ds_prev_snap_obj
;
1060 dsphys
->ds_prev_snap_txg
= ds
->ds_phys
->ds_prev_snap_txg
;
1061 dsphys
->ds_next_snap_obj
= ds
->ds_object
;
1062 dsphys
->ds_num_children
= 1;
1063 dsphys
->ds_creation_time
= gethrestime_sec();
1064 dsphys
->ds_creation_txg
= crtxg
;
1065 dsphys
->ds_deadlist_obj
= ds
->ds_phys
->ds_deadlist_obj
;
1066 dsphys
->ds_referenced_bytes
= ds
->ds_phys
->ds_referenced_bytes
;
1067 dsphys
->ds_compressed_bytes
= ds
->ds_phys
->ds_compressed_bytes
;
1068 dsphys
->ds_uncompressed_bytes
= ds
->ds_phys
->ds_uncompressed_bytes
;
1069 dsphys
->ds_flags
= ds
->ds_phys
->ds_flags
;
1070 dsphys
->ds_bp
= ds
->ds_phys
->ds_bp
;
1071 dmu_buf_rele(dbuf
, FTAG
);
1073 ASSERT3U(ds
->ds_prev
!= 0, ==, ds
->ds_phys
->ds_prev_snap_obj
!= 0);
1075 uint64_t next_clones_obj
=
1076 ds
->ds_prev
->ds_phys
->ds_next_clones_obj
;
1077 ASSERT(ds
->ds_prev
->ds_phys
->ds_next_snap_obj
==
1079 ds
->ds_prev
->ds_phys
->ds_num_children
> 1);
1080 if (ds
->ds_prev
->ds_phys
->ds_next_snap_obj
== ds
->ds_object
) {
1081 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1082 ASSERT3U(ds
->ds_phys
->ds_prev_snap_txg
, ==,
1083 ds
->ds_prev
->ds_phys
->ds_creation_txg
);
1084 ds
->ds_prev
->ds_phys
->ds_next_snap_obj
= dsobj
;
1085 } else if (next_clones_obj
!= 0) {
1086 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
1087 dsphys
->ds_next_snap_obj
, tx
);
1088 VERIFY0(zap_add_int(mos
,
1089 next_clones_obj
, dsobj
, tx
));
1094 * If we have a reference-reservation on this dataset, we will
1095 * need to increase the amount of refreservation being charged
1096 * since our unique space is going to zero.
1098 if (ds
->ds_reserved
) {
1100 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
1101 delta
= MIN(ds
->ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
1102 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
,
1106 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1107 ds
->ds_phys
->ds_deadlist_obj
= dsl_deadlist_clone(&ds
->ds_deadlist
,
1108 UINT64_MAX
, ds
->ds_phys
->ds_prev_snap_obj
, tx
);
1109 dsl_deadlist_close(&ds
->ds_deadlist
);
1110 dsl_deadlist_open(&ds
->ds_deadlist
, mos
, ds
->ds_phys
->ds_deadlist_obj
);
1111 dsl_deadlist_add_key(&ds
->ds_deadlist
,
1112 ds
->ds_phys
->ds_prev_snap_txg
, tx
);
1114 ASSERT3U(ds
->ds_phys
->ds_prev_snap_txg
, <, tx
->tx_txg
);
1115 ds
->ds_phys
->ds_prev_snap_obj
= dsobj
;
1116 ds
->ds_phys
->ds_prev_snap_txg
= crtxg
;
1117 ds
->ds_phys
->ds_unique_bytes
= 0;
1118 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
1119 ds
->ds_phys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1121 VERIFY0(zap_add(mos
, ds
->ds_phys
->ds_snapnames_zapobj
,
1122 snapname
, 8, 1, &dsobj
, tx
));
1125 dsl_dataset_rele(ds
->ds_prev
, ds
);
1126 VERIFY0(dsl_dataset_hold_obj(dp
,
1127 ds
->ds_phys
->ds_prev_snap_obj
, ds
, &ds
->ds_prev
));
1129 dsl_scan_ds_snapshotted(ds
, tx
);
1131 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
1133 spa_history_log_internal_ds(ds
->ds_prev
, "snapshot", tx
, "");
1137 dsl_dataset_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1139 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1140 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1143 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1144 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1147 char dsname
[MAXNAMELEN
];
1149 name
= nvpair_name(pair
);
1150 atp
= strchr(name
, '@');
1151 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1152 VERIFY0(dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
));
1154 dsl_dataset_snapshot_sync_impl(ds
, atp
+ 1, tx
);
1155 if (ddsa
->ddsa_props
!= NULL
) {
1156 dsl_props_set_sync_impl(ds
->ds_prev
,
1157 ZPROP_SRC_LOCAL
, ddsa
->ddsa_props
, tx
);
1159 dsl_dataset_rele(ds
, FTAG
);
1164 * The snapshots must all be in the same pool.
1165 * All-or-nothing: if there are any failures, nothing will be modified.
1168 dsl_dataset_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
*errors
)
1170 dsl_dataset_snapshot_arg_t ddsa
;
1172 boolean_t needsuspend
;
1176 nvlist_t
*suspended
= NULL
;
1178 pair
= nvlist_next_nvpair(snaps
, NULL
);
1181 firstname
= nvpair_name(pair
);
1183 error
= spa_open(firstname
, &spa
, FTAG
);
1186 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1187 spa_close(spa
, FTAG
);
1190 suspended
= fnvlist_alloc();
1191 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1192 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1193 char fsname
[MAXNAMELEN
];
1194 char *snapname
= nvpair_name(pair
);
1198 atp
= strchr(snapname
, '@');
1200 error
= SET_ERROR(EINVAL
);
1203 (void) strlcpy(fsname
, snapname
, atp
- snapname
+ 1);
1205 error
= zil_suspend(fsname
, &cookie
);
1208 fnvlist_add_uint64(suspended
, fsname
,
1213 ddsa
.ddsa_snaps
= snaps
;
1214 ddsa
.ddsa_props
= props
;
1215 ddsa
.ddsa_errors
= errors
;
1218 error
= dsl_sync_task(firstname
, dsl_dataset_snapshot_check
,
1219 dsl_dataset_snapshot_sync
, &ddsa
,
1220 fnvlist_num_pairs(snaps
) * 3);
1223 if (suspended
!= NULL
) {
1224 for (pair
= nvlist_next_nvpair(suspended
, NULL
); pair
!= NULL
;
1225 pair
= nvlist_next_nvpair(suspended
, pair
)) {
1226 zil_resume((void *)(uintptr_t)
1227 fnvpair_value_uint64(pair
));
1229 fnvlist_free(suspended
);
1234 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1235 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1236 char *snapname
= nvpair_name(pair
);
1237 zvol_create_minors(snapname
);
1245 typedef struct dsl_dataset_snapshot_tmp_arg
{
1246 const char *ddsta_fsname
;
1247 const char *ddsta_snapname
;
1248 minor_t ddsta_cleanup_minor
;
1249 const char *ddsta_htag
;
1250 } dsl_dataset_snapshot_tmp_arg_t
;
1253 dsl_dataset_snapshot_tmp_check(void *arg
, dmu_tx_t
*tx
)
1255 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1256 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1260 error
= dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
);
1264 error
= dsl_dataset_snapshot_check_impl(ds
, ddsta
->ddsta_snapname
,
1267 dsl_dataset_rele(ds
, FTAG
);
1271 if (spa_version(dp
->dp_spa
) < SPA_VERSION_USERREFS
) {
1272 dsl_dataset_rele(ds
, FTAG
);
1273 return (SET_ERROR(ENOTSUP
));
1275 error
= dsl_dataset_user_hold_check_one(NULL
, ddsta
->ddsta_htag
,
1278 dsl_dataset_rele(ds
, FTAG
);
1282 dsl_dataset_rele(ds
, FTAG
);
1287 dsl_dataset_snapshot_tmp_sync(void *arg
, dmu_tx_t
*tx
)
1289 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1290 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1293 VERIFY0(dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
));
1295 dsl_dataset_snapshot_sync_impl(ds
, ddsta
->ddsta_snapname
, tx
);
1296 dsl_dataset_user_hold_sync_one(ds
->ds_prev
, ddsta
->ddsta_htag
,
1297 ddsta
->ddsta_cleanup_minor
, gethrestime_sec(), tx
);
1298 dsl_destroy_snapshot_sync_impl(ds
->ds_prev
, B_TRUE
, tx
);
1300 dsl_dataset_rele(ds
, FTAG
);
1304 dsl_dataset_snapshot_tmp(const char *fsname
, const char *snapname
,
1305 minor_t cleanup_minor
, const char *htag
)
1307 dsl_dataset_snapshot_tmp_arg_t ddsta
;
1310 boolean_t needsuspend
;
1313 ddsta
.ddsta_fsname
= fsname
;
1314 ddsta
.ddsta_snapname
= snapname
;
1315 ddsta
.ddsta_cleanup_minor
= cleanup_minor
;
1316 ddsta
.ddsta_htag
= htag
;
1318 error
= spa_open(fsname
, &spa
, FTAG
);
1321 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1322 spa_close(spa
, FTAG
);
1325 error
= zil_suspend(fsname
, &cookie
);
1330 error
= dsl_sync_task(fsname
, dsl_dataset_snapshot_tmp_check
,
1331 dsl_dataset_snapshot_tmp_sync
, &ddsta
, 3);
1340 dsl_dataset_sync(dsl_dataset_t
*ds
, zio_t
*zio
, dmu_tx_t
*tx
)
1342 ASSERT(dmu_tx_is_syncing(tx
));
1343 ASSERT(ds
->ds_objset
!= NULL
);
1344 ASSERT(ds
->ds_phys
->ds_next_snap_obj
== 0);
1347 * in case we had to change ds_fsid_guid when we opened it,
1350 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1351 ds
->ds_phys
->ds_fsid_guid
= ds
->ds_fsid_guid
;
1353 dmu_objset_sync(ds
->ds_objset
, zio
, tx
);
1357 get_clones_stat(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1360 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1363 nvlist_t
*propval
= fnvlist_alloc();
1364 nvlist_t
*val
= fnvlist_alloc();
1366 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1369 * There may be missing entries in ds_next_clones_obj
1370 * due to a bug in a previous version of the code.
1371 * Only trust it if it has the right number of entries.
1373 if (ds
->ds_phys
->ds_next_clones_obj
!= 0) {
1374 ASSERT0(zap_count(mos
, ds
->ds_phys
->ds_next_clones_obj
,
1377 if (count
!= ds
->ds_phys
->ds_num_children
- 1)
1379 for (zap_cursor_init(&zc
, mos
, ds
->ds_phys
->ds_next_clones_obj
);
1380 zap_cursor_retrieve(&zc
, &za
) == 0;
1381 zap_cursor_advance(&zc
)) {
1382 dsl_dataset_t
*clone
;
1383 char buf
[ZFS_MAXNAMELEN
];
1384 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
1385 za
.za_first_integer
, FTAG
, &clone
));
1386 dsl_dir_name(clone
->ds_dir
, buf
);
1387 fnvlist_add_boolean(val
, buf
);
1388 dsl_dataset_rele(clone
, FTAG
);
1390 zap_cursor_fini(&zc
);
1391 fnvlist_add_nvlist(propval
, ZPROP_VALUE
, val
);
1392 fnvlist_add_nvlist(nv
, zfs_prop_to_name(ZFS_PROP_CLONES
), propval
);
1395 nvlist_free(propval
);
1399 dsl_dataset_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1401 uint64_t refd
, avail
, uobjs
, aobjs
, ratio
;
1402 ASSERTV(dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
);
1404 ASSERT(dsl_pool_config_held(dp
));
1406 ratio
= ds
->ds_phys
->ds_compressed_bytes
== 0 ? 100 :
1407 (ds
->ds_phys
->ds_uncompressed_bytes
* 100 /
1408 ds
->ds_phys
->ds_compressed_bytes
);
1410 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRATIO
, ratio
);
1411 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALREFERENCED
,
1412 ds
->ds_phys
->ds_uncompressed_bytes
);
1414 if (dsl_dataset_is_snapshot(ds
)) {
1415 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
, ratio
);
1416 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
1417 ds
->ds_phys
->ds_unique_bytes
);
1418 get_clones_stat(ds
, nv
);
1420 dsl_dir_stats(ds
->ds_dir
, nv
);
1423 dsl_dataset_space(ds
, &refd
, &avail
, &uobjs
, &aobjs
);
1424 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_AVAILABLE
, avail
);
1425 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFERENCED
, refd
);
1427 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATION
,
1428 ds
->ds_phys
->ds_creation_time
);
1429 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATETXG
,
1430 ds
->ds_phys
->ds_creation_txg
);
1431 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFQUOTA
,
1433 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRESERVATION
,
1435 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_GUID
,
1436 ds
->ds_phys
->ds_guid
);
1437 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_UNIQUE
,
1438 ds
->ds_phys
->ds_unique_bytes
);
1439 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_OBJSETID
,
1441 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERREFS
,
1443 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_DEFER_DESTROY
,
1444 DS_IS_DEFER_DESTROY(ds
) ? 1 : 0);
1446 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0) {
1447 uint64_t written
, comp
, uncomp
;
1448 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1449 dsl_dataset_t
*prev
;
1452 err
= dsl_dataset_hold_obj(dp
,
1453 ds
->ds_phys
->ds_prev_snap_obj
, FTAG
, &prev
);
1455 err
= dsl_dataset_space_written(prev
, ds
, &written
,
1457 dsl_dataset_rele(prev
, FTAG
);
1459 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_WRITTEN
,
1468 dsl_dataset_fast_stat(dsl_dataset_t
*ds
, dmu_objset_stats_t
*stat
)
1470 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1471 ASSERT(dsl_pool_config_held(dp
));
1473 stat
->dds_creation_txg
= ds
->ds_phys
->ds_creation_txg
;
1474 stat
->dds_inconsistent
= ds
->ds_phys
->ds_flags
& DS_FLAG_INCONSISTENT
;
1475 stat
->dds_guid
= ds
->ds_phys
->ds_guid
;
1476 stat
->dds_origin
[0] = '\0';
1477 if (dsl_dataset_is_snapshot(ds
)) {
1478 stat
->dds_is_snapshot
= B_TRUE
;
1479 stat
->dds_num_clones
= ds
->ds_phys
->ds_num_children
- 1;
1481 stat
->dds_is_snapshot
= B_FALSE
;
1482 stat
->dds_num_clones
= 0;
1484 if (dsl_dir_is_clone(ds
->ds_dir
)) {
1487 VERIFY0(dsl_dataset_hold_obj(dp
,
1488 ds
->ds_dir
->dd_phys
->dd_origin_obj
, FTAG
, &ods
));
1489 dsl_dataset_name(ods
, stat
->dds_origin
);
1490 dsl_dataset_rele(ods
, FTAG
);
1496 dsl_dataset_fsid_guid(dsl_dataset_t
*ds
)
1498 return (ds
->ds_fsid_guid
);
1502 dsl_dataset_space(dsl_dataset_t
*ds
,
1503 uint64_t *refdbytesp
, uint64_t *availbytesp
,
1504 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1506 *refdbytesp
= ds
->ds_phys
->ds_referenced_bytes
;
1507 *availbytesp
= dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
);
1508 if (ds
->ds_reserved
> ds
->ds_phys
->ds_unique_bytes
)
1509 *availbytesp
+= ds
->ds_reserved
- ds
->ds_phys
->ds_unique_bytes
;
1510 if (ds
->ds_quota
!= 0) {
1512 * Adjust available bytes according to refquota
1514 if (*refdbytesp
< ds
->ds_quota
)
1515 *availbytesp
= MIN(*availbytesp
,
1516 ds
->ds_quota
- *refdbytesp
);
1520 *usedobjsp
= ds
->ds_phys
->ds_bp
.blk_fill
;
1521 *availobjsp
= DN_MAX_OBJECT
- *usedobjsp
;
1525 dsl_dataset_modified_since_snap(dsl_dataset_t
*ds
, dsl_dataset_t
*snap
)
1527 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1530 if (ds
->ds_phys
->ds_bp
.blk_birth
>
1531 snap
->ds_phys
->ds_creation_txg
) {
1532 objset_t
*os
, *os_snap
;
1534 * It may be that only the ZIL differs, because it was
1535 * reset in the head. Don't count that as being
1538 if (dmu_objset_from_ds(ds
, &os
) != 0)
1540 if (dmu_objset_from_ds(snap
, &os_snap
) != 0)
1542 return (bcmp(&os
->os_phys
->os_meta_dnode
,
1543 &os_snap
->os_phys
->os_meta_dnode
,
1544 sizeof (os
->os_phys
->os_meta_dnode
)) != 0);
1549 typedef struct dsl_dataset_rename_snapshot_arg
{
1550 const char *ddrsa_fsname
;
1551 const char *ddrsa_oldsnapname
;
1552 const char *ddrsa_newsnapname
;
1553 boolean_t ddrsa_recursive
;
1555 } dsl_dataset_rename_snapshot_arg_t
;
1559 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t
*dp
,
1560 dsl_dataset_t
*hds
, void *arg
)
1562 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1566 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
1568 /* ignore nonexistent snapshots */
1569 return (error
== ENOENT
? 0 : error
);
1572 /* new name should not exist */
1573 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_newsnapname
, &val
);
1575 error
= SET_ERROR(EEXIST
);
1576 else if (error
== ENOENT
)
1579 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1580 if (dsl_dir_namelen(hds
->ds_dir
) + 1 +
1581 strlen(ddrsa
->ddrsa_newsnapname
) >= MAXNAMELEN
)
1582 error
= SET_ERROR(ENAMETOOLONG
);
1588 dsl_dataset_rename_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1590 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1591 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1595 error
= dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
);
1599 if (ddrsa
->ddrsa_recursive
) {
1600 error
= dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
1601 dsl_dataset_rename_snapshot_check_impl
, ddrsa
,
1604 error
= dsl_dataset_rename_snapshot_check_impl(dp
, hds
, ddrsa
);
1606 dsl_dataset_rele(hds
, FTAG
);
1611 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t
*dp
,
1612 dsl_dataset_t
*hds
, void *arg
)
1615 char *oldname
, *newname
;
1617 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1620 dmu_tx_t
*tx
= ddrsa
->ddrsa_tx
;
1623 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
1624 ASSERT(error
== 0 || error
== ENOENT
);
1625 if (error
== ENOENT
) {
1626 /* ignore nonexistent snapshots */
1630 VERIFY0(dsl_dataset_hold_obj(dp
, val
, FTAG
, &ds
));
1632 /* log before we change the name */
1633 spa_history_log_internal_ds(ds
, "rename", tx
,
1634 "-> @%s", ddrsa
->ddrsa_newsnapname
);
1636 VERIFY0(dsl_dataset_snap_remove(hds
, ddrsa
->ddrsa_oldsnapname
, tx
));
1637 mutex_enter(&ds
->ds_lock
);
1638 (void) strcpy(ds
->ds_snapname
, ddrsa
->ddrsa_newsnapname
);
1639 mutex_exit(&ds
->ds_lock
);
1640 VERIFY0(zap_add(dp
->dp_meta_objset
, hds
->ds_phys
->ds_snapnames_zapobj
,
1641 ds
->ds_snapname
, 8, 1, &ds
->ds_object
, tx
));
1644 oldname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
1645 newname
= kmem_alloc(MAXPATHLEN
, KM_PUSHPAGE
);
1646 snprintf(oldname
, MAXPATHLEN
, "%s@%s", ddrsa
->ddrsa_fsname
,
1647 ddrsa
->ddrsa_oldsnapname
);
1648 snprintf(newname
, MAXPATHLEN
, "%s@%s", ddrsa
->ddrsa_fsname
,
1649 ddrsa
->ddrsa_newsnapname
);
1650 zvol_rename_minors(oldname
, newname
);
1651 kmem_free(newname
, MAXPATHLEN
);
1652 kmem_free(oldname
, MAXPATHLEN
);
1655 dsl_dataset_rele(ds
, FTAG
);
1660 dsl_dataset_rename_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1662 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1663 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1666 VERIFY0(dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
));
1667 ddrsa
->ddrsa_tx
= tx
;
1668 if (ddrsa
->ddrsa_recursive
) {
1669 VERIFY0(dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
1670 dsl_dataset_rename_snapshot_sync_impl
, ddrsa
,
1673 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp
, hds
, ddrsa
));
1675 dsl_dataset_rele(hds
, FTAG
);
1679 dsl_dataset_rename_snapshot(const char *fsname
,
1680 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
)
1682 dsl_dataset_rename_snapshot_arg_t ddrsa
;
1684 ddrsa
.ddrsa_fsname
= fsname
;
1685 ddrsa
.ddrsa_oldsnapname
= oldsnapname
;
1686 ddrsa
.ddrsa_newsnapname
= newsnapname
;
1687 ddrsa
.ddrsa_recursive
= recursive
;
1689 return (dsl_sync_task(fsname
, dsl_dataset_rename_snapshot_check
,
1690 dsl_dataset_rename_snapshot_sync
, &ddrsa
, 1));
1694 * If we're doing an ownership handoff, we need to make sure that there is
1695 * only one long hold on the dataset. We're not allowed to change anything here
1696 * so we don't permanently release the long hold or regular hold here. We want
1697 * to do this only when syncing to avoid the dataset unexpectedly going away
1698 * when we release the long hold.
1701 dsl_dataset_handoff_check(dsl_dataset_t
*ds
, void *owner
, dmu_tx_t
*tx
)
1705 if (!dmu_tx_is_syncing(tx
))
1708 if (owner
!= NULL
) {
1709 VERIFY3P(ds
->ds_owner
, ==, owner
);
1710 dsl_dataset_long_rele(ds
, owner
);
1713 held
= dsl_dataset_long_held(ds
);
1716 dsl_dataset_long_hold(ds
, owner
);
1719 return (SET_ERROR(EBUSY
));
1724 typedef struct dsl_dataset_rollback_arg
{
1725 const char *ddra_fsname
;
1727 nvlist_t
*ddra_result
;
1728 } dsl_dataset_rollback_arg_t
;
1731 dsl_dataset_rollback_check(void *arg
, dmu_tx_t
*tx
)
1733 dsl_dataset_rollback_arg_t
*ddra
= arg
;
1734 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1736 int64_t unused_refres_delta
;
1739 error
= dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
);
1743 /* must not be a snapshot */
1744 if (dsl_dataset_is_snapshot(ds
)) {
1745 dsl_dataset_rele(ds
, FTAG
);
1746 return (SET_ERROR(EINVAL
));
1749 /* must have a most recent snapshot */
1750 if (ds
->ds_phys
->ds_prev_snap_txg
< TXG_INITIAL
) {
1751 dsl_dataset_rele(ds
, FTAG
);
1752 return (SET_ERROR(EINVAL
));
1755 error
= dsl_dataset_handoff_check(ds
, ddra
->ddra_owner
, tx
);
1757 dsl_dataset_rele(ds
, FTAG
);
1762 * Check if the snap we are rolling back to uses more than
1765 if (ds
->ds_quota
!= 0 &&
1766 ds
->ds_prev
->ds_phys
->ds_referenced_bytes
> ds
->ds_quota
) {
1767 dsl_dataset_rele(ds
, FTAG
);
1768 return (SET_ERROR(EDQUOT
));
1772 * When we do the clone swap, we will temporarily use more space
1773 * due to the refreservation (the head will no longer have any
1774 * unique space, so the entire amount of the refreservation will need
1775 * to be free). We will immediately destroy the clone, freeing
1776 * this space, but the freeing happens over many txg's.
1778 unused_refres_delta
= (int64_t)MIN(ds
->ds_reserved
,
1779 ds
->ds_phys
->ds_unique_bytes
);
1781 if (unused_refres_delta
> 0 &&
1782 unused_refres_delta
>
1783 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
)) {
1784 dsl_dataset_rele(ds
, FTAG
);
1785 return (SET_ERROR(ENOSPC
));
1788 dsl_dataset_rele(ds
, FTAG
);
1793 dsl_dataset_rollback_sync(void *arg
, dmu_tx_t
*tx
)
1795 dsl_dataset_rollback_arg_t
*ddra
= arg
;
1796 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1797 dsl_dataset_t
*ds
, *clone
;
1799 char namebuf
[ZFS_MAXNAMELEN
];
1801 VERIFY0(dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
));
1803 dsl_dataset_name(ds
->ds_prev
, namebuf
);
1804 fnvlist_add_string(ddra
->ddra_result
, "target", namebuf
);
1806 cloneobj
= dsl_dataset_create_sync(ds
->ds_dir
, "%rollback",
1807 ds
->ds_prev
, DS_CREATE_FLAG_NODIRTY
, kcred
, tx
);
1809 VERIFY0(dsl_dataset_hold_obj(dp
, cloneobj
, FTAG
, &clone
));
1811 dsl_dataset_clone_swap_sync_impl(clone
, ds
, tx
);
1812 dsl_dataset_zero_zil(ds
, tx
);
1814 dsl_destroy_head_sync_impl(clone
, tx
);
1816 dsl_dataset_rele(clone
, FTAG
);
1817 dsl_dataset_rele(ds
, FTAG
);
1821 * Rolls back the given filesystem or volume to the most recent snapshot.
1822 * The name of the most recent snapshot will be returned under key "target"
1823 * in the result nvlist.
1826 * - The existing dataset MUST be owned by the specified owner at entry
1827 * - Upon return, dataset will still be held by the same owner, whether we
1830 * This mode is required any time the existing filesystem is mounted. See
1831 * notes above zfs_suspend_fs() for further details.
1834 dsl_dataset_rollback(const char *fsname
, void *owner
, nvlist_t
*result
)
1836 dsl_dataset_rollback_arg_t ddra
;
1838 ddra
.ddra_fsname
= fsname
;
1839 ddra
.ddra_owner
= owner
;
1840 ddra
.ddra_result
= result
;
1842 return (dsl_sync_task(fsname
, dsl_dataset_rollback_check
,
1843 dsl_dataset_rollback_sync
, &ddra
, 1));
1846 struct promotenode
{
1851 typedef struct dsl_dataset_promote_arg
{
1852 const char *ddpa_clonename
;
1853 dsl_dataset_t
*ddpa_clone
;
1854 list_t shared_snaps
, origin_snaps
, clone_snaps
;
1855 dsl_dataset_t
*origin_origin
; /* origin of the origin */
1856 uint64_t used
, comp
, uncomp
, unique
, cloneusedsnap
, originusedsnap
;
1858 } dsl_dataset_promote_arg_t
;
1860 static int snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
);
1861 static int promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
,
1863 static void promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
);
1866 dsl_dataset_promote_check(void *arg
, dmu_tx_t
*tx
)
1868 dsl_dataset_promote_arg_t
*ddpa
= arg
;
1869 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1871 struct promotenode
*snap
;
1872 dsl_dataset_t
*origin_ds
;
1876 err
= promote_hold(ddpa
, dp
, FTAG
);
1880 hds
= ddpa
->ddpa_clone
;
1882 if (hds
->ds_phys
->ds_flags
& DS_FLAG_NOPROMOTE
) {
1883 promote_rele(ddpa
, FTAG
);
1884 return (SET_ERROR(EXDEV
));
1888 * Compute and check the amount of space to transfer. Since this is
1889 * so expensive, don't do the preliminary check.
1891 if (!dmu_tx_is_syncing(tx
)) {
1892 promote_rele(ddpa
, FTAG
);
1896 snap
= list_head(&ddpa
->shared_snaps
);
1897 origin_ds
= snap
->ds
;
1899 /* compute origin's new unique space */
1900 snap
= list_tail(&ddpa
->clone_snaps
);
1901 ASSERT3U(snap
->ds
->ds_phys
->ds_prev_snap_obj
, ==, origin_ds
->ds_object
);
1902 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
1903 origin_ds
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
1904 &ddpa
->unique
, &unused
, &unused
);
1907 * Walk the snapshots that we are moving
1909 * Compute space to transfer. Consider the incremental changes
1910 * to used by each snapshot:
1911 * (my used) = (prev's used) + (blocks born) - (blocks killed)
1912 * So each snapshot gave birth to:
1913 * (blocks born) = (my used) - (prev's used) + (blocks killed)
1914 * So a sequence would look like:
1915 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
1916 * Which simplifies to:
1917 * uN + kN + kN-1 + ... + k1 + k0
1918 * Note however, if we stop before we reach the ORIGIN we get:
1919 * uN + kN + kN-1 + ... + kM - uM-1
1921 ddpa
->used
= origin_ds
->ds_phys
->ds_referenced_bytes
;
1922 ddpa
->comp
= origin_ds
->ds_phys
->ds_compressed_bytes
;
1923 ddpa
->uncomp
= origin_ds
->ds_phys
->ds_uncompressed_bytes
;
1924 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
1925 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
1926 uint64_t val
, dlused
, dlcomp
, dluncomp
;
1927 dsl_dataset_t
*ds
= snap
->ds
;
1930 * If there are long holds, we won't be able to evict
1933 if (dsl_dataset_long_held(ds
)) {
1934 err
= SET_ERROR(EBUSY
);
1938 /* Check that the snapshot name does not conflict */
1939 VERIFY0(dsl_dataset_get_snapname(ds
));
1940 err
= dsl_dataset_snap_lookup(hds
, ds
->ds_snapname
, &val
);
1942 (void) strcpy(ddpa
->err_ds
, snap
->ds
->ds_snapname
);
1943 err
= SET_ERROR(EEXIST
);
1949 /* The very first snapshot does not have a deadlist */
1950 if (ds
->ds_phys
->ds_prev_snap_obj
== 0)
1953 dsl_deadlist_space(&ds
->ds_deadlist
,
1954 &dlused
, &dlcomp
, &dluncomp
);
1955 ddpa
->used
+= dlused
;
1956 ddpa
->comp
+= dlcomp
;
1957 ddpa
->uncomp
+= dluncomp
;
1961 * If we are a clone of a clone then we never reached ORIGIN,
1962 * so we need to subtract out the clone origin's used space.
1964 if (ddpa
->origin_origin
) {
1965 ddpa
->used
-= ddpa
->origin_origin
->ds_phys
->ds_referenced_bytes
;
1966 ddpa
->comp
-= ddpa
->origin_origin
->ds_phys
->ds_compressed_bytes
;
1968 ddpa
->origin_origin
->ds_phys
->ds_uncompressed_bytes
;
1971 /* Check that there is enough space here */
1972 err
= dsl_dir_transfer_possible(origin_ds
->ds_dir
, hds
->ds_dir
,
1978 * Compute the amounts of space that will be used by snapshots
1979 * after the promotion (for both origin and clone). For each,
1980 * it is the amount of space that will be on all of their
1981 * deadlists (that was not born before their new origin).
1983 if (hds
->ds_dir
->dd_phys
->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
1987 * Note, typically this will not be a clone of a clone,
1988 * so dd_origin_txg will be < TXG_INITIAL, so
1989 * these snaplist_space() -> dsl_deadlist_space_range()
1990 * calls will be fast because they do not have to
1991 * iterate over all bps.
1993 snap
= list_head(&ddpa
->origin_snaps
);
1994 err
= snaplist_space(&ddpa
->shared_snaps
,
1995 snap
->ds
->ds_dir
->dd_origin_txg
, &ddpa
->cloneusedsnap
);
1999 err
= snaplist_space(&ddpa
->clone_snaps
,
2000 snap
->ds
->ds_dir
->dd_origin_txg
, &space
);
2003 ddpa
->cloneusedsnap
+= space
;
2005 if (origin_ds
->ds_dir
->dd_phys
->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2006 err
= snaplist_space(&ddpa
->origin_snaps
,
2007 origin_ds
->ds_phys
->ds_creation_txg
, &ddpa
->originusedsnap
);
2013 promote_rele(ddpa
, FTAG
);
2018 dsl_dataset_promote_sync(void *arg
, dmu_tx_t
*tx
)
2020 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2021 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2023 struct promotenode
*snap
;
2024 dsl_dataset_t
*origin_ds
;
2025 dsl_dataset_t
*origin_head
;
2027 dsl_dir_t
*odd
= NULL
;
2028 uint64_t oldnext_obj
;
2031 VERIFY0(promote_hold(ddpa
, dp
, FTAG
));
2032 hds
= ddpa
->ddpa_clone
;
2034 ASSERT0(hds
->ds_phys
->ds_flags
& DS_FLAG_NOPROMOTE
);
2036 snap
= list_head(&ddpa
->shared_snaps
);
2037 origin_ds
= snap
->ds
;
2040 snap
= list_head(&ddpa
->origin_snaps
);
2041 origin_head
= snap
->ds
;
2044 * We need to explicitly open odd, since origin_ds's dd will be
2047 VERIFY0(dsl_dir_hold_obj(dp
, origin_ds
->ds_dir
->dd_object
,
2050 /* change origin's next snap */
2051 dmu_buf_will_dirty(origin_ds
->ds_dbuf
, tx
);
2052 oldnext_obj
= origin_ds
->ds_phys
->ds_next_snap_obj
;
2053 snap
= list_tail(&ddpa
->clone_snaps
);
2054 ASSERT3U(snap
->ds
->ds_phys
->ds_prev_snap_obj
, ==, origin_ds
->ds_object
);
2055 origin_ds
->ds_phys
->ds_next_snap_obj
= snap
->ds
->ds_object
;
2057 /* change the origin's next clone */
2058 if (origin_ds
->ds_phys
->ds_next_clones_obj
) {
2059 dsl_dataset_remove_from_next_clones(origin_ds
,
2060 snap
->ds
->ds_object
, tx
);
2061 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2062 origin_ds
->ds_phys
->ds_next_clones_obj
,
2067 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
2068 ASSERT3U(dd
->dd_phys
->dd_origin_obj
, ==, origin_ds
->ds_object
);
2069 dd
->dd_phys
->dd_origin_obj
= odd
->dd_phys
->dd_origin_obj
;
2070 dd
->dd_origin_txg
= origin_head
->ds_dir
->dd_origin_txg
;
2071 dmu_buf_will_dirty(odd
->dd_dbuf
, tx
);
2072 odd
->dd_phys
->dd_origin_obj
= origin_ds
->ds_object
;
2073 origin_head
->ds_dir
->dd_origin_txg
=
2074 origin_ds
->ds_phys
->ds_creation_txg
;
2076 /* change dd_clone entries */
2077 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2078 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2079 odd
->dd_phys
->dd_clones
, hds
->ds_object
, tx
));
2080 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2081 ddpa
->origin_origin
->ds_dir
->dd_phys
->dd_clones
,
2082 hds
->ds_object
, tx
));
2084 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2085 ddpa
->origin_origin
->ds_dir
->dd_phys
->dd_clones
,
2086 origin_head
->ds_object
, tx
));
2087 if (dd
->dd_phys
->dd_clones
== 0) {
2088 dd
->dd_phys
->dd_clones
= zap_create(dp
->dp_meta_objset
,
2089 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
2091 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2092 dd
->dd_phys
->dd_clones
, origin_head
->ds_object
, tx
));
2095 /* move snapshots to this dir */
2096 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2097 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2098 dsl_dataset_t
*ds
= snap
->ds
;
2101 * Property callbacks are registered to a particular
2102 * dsl_dir. Since ours is changing, evict the objset
2103 * so that they will be unregistered from the old dsl_dir.
2105 if (ds
->ds_objset
) {
2106 dmu_objset_evict(ds
->ds_objset
);
2107 ds
->ds_objset
= NULL
;
2110 /* move snap name entry */
2111 VERIFY0(dsl_dataset_get_snapname(ds
));
2112 VERIFY0(dsl_dataset_snap_remove(origin_head
,
2113 ds
->ds_snapname
, tx
));
2114 VERIFY0(zap_add(dp
->dp_meta_objset
,
2115 hds
->ds_phys
->ds_snapnames_zapobj
, ds
->ds_snapname
,
2116 8, 1, &ds
->ds_object
, tx
));
2118 /* change containing dsl_dir */
2119 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2120 ASSERT3U(ds
->ds_phys
->ds_dir_obj
, ==, odd
->dd_object
);
2121 ds
->ds_phys
->ds_dir_obj
= dd
->dd_object
;
2122 ASSERT3P(ds
->ds_dir
, ==, odd
);
2123 dsl_dir_rele(ds
->ds_dir
, ds
);
2124 VERIFY0(dsl_dir_hold_obj(dp
, dd
->dd_object
,
2125 NULL
, ds
, &ds
->ds_dir
));
2127 /* move any clone references */
2128 if (ds
->ds_phys
->ds_next_clones_obj
&&
2129 spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2133 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2134 ds
->ds_phys
->ds_next_clones_obj
);
2135 zap_cursor_retrieve(&zc
, &za
) == 0;
2136 zap_cursor_advance(&zc
)) {
2137 dsl_dataset_t
*cnds
;
2140 if (za
.za_first_integer
== oldnext_obj
) {
2142 * We've already moved the
2143 * origin's reference.
2148 VERIFY0(dsl_dataset_hold_obj(dp
,
2149 za
.za_first_integer
, FTAG
, &cnds
));
2150 o
= cnds
->ds_dir
->dd_phys
->dd_head_dataset_obj
;
2152 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2153 odd
->dd_phys
->dd_clones
, o
, tx
));
2154 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2155 dd
->dd_phys
->dd_clones
, o
, tx
));
2156 dsl_dataset_rele(cnds
, FTAG
);
2158 zap_cursor_fini(&zc
);
2161 ASSERT(!dsl_prop_hascb(ds
));
2165 * Change space accounting.
2166 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2167 * both be valid, or both be 0 (resulting in delta == 0). This
2168 * is true for each of {clone,origin} independently.
2171 delta
= ddpa
->cloneusedsnap
-
2172 dd
->dd_phys
->dd_used_breakdown
[DD_USED_SNAP
];
2173 ASSERT3S(delta
, >=, 0);
2174 ASSERT3U(ddpa
->used
, >=, delta
);
2175 dsl_dir_diduse_space(dd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2176 dsl_dir_diduse_space(dd
, DD_USED_HEAD
,
2177 ddpa
->used
- delta
, ddpa
->comp
, ddpa
->uncomp
, tx
);
2179 delta
= ddpa
->originusedsnap
-
2180 odd
->dd_phys
->dd_used_breakdown
[DD_USED_SNAP
];
2181 ASSERT3S(delta
, <=, 0);
2182 ASSERT3U(ddpa
->used
, >=, -delta
);
2183 dsl_dir_diduse_space(odd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2184 dsl_dir_diduse_space(odd
, DD_USED_HEAD
,
2185 -ddpa
->used
- delta
, -ddpa
->comp
, -ddpa
->uncomp
, tx
);
2187 origin_ds
->ds_phys
->ds_unique_bytes
= ddpa
->unique
;
2189 /* log history record */
2190 spa_history_log_internal_ds(hds
, "promote", tx
, "");
2192 dsl_dir_rele(odd
, FTAG
);
2193 promote_rele(ddpa
, FTAG
);
2197 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2198 * (exclusive) and last_obj (inclusive). The list will be in reverse
2199 * order (last_obj will be the list_head()). If first_obj == 0, do all
2200 * snapshots back to this dataset's origin.
2203 snaplist_make(dsl_pool_t
*dp
,
2204 uint64_t first_obj
, uint64_t last_obj
, list_t
*l
, void *tag
)
2206 uint64_t obj
= last_obj
;
2208 list_create(l
, sizeof (struct promotenode
),
2209 offsetof(struct promotenode
, link
));
2211 while (obj
!= first_obj
) {
2213 struct promotenode
*snap
;
2216 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
2217 ASSERT(err
!= ENOENT
);
2222 first_obj
= ds
->ds_dir
->dd_phys
->dd_origin_obj
;
2224 snap
= kmem_alloc(sizeof (*snap
), KM_PUSHPAGE
);
2226 list_insert_tail(l
, snap
);
2227 obj
= ds
->ds_phys
->ds_prev_snap_obj
;
2234 snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
)
2236 struct promotenode
*snap
;
2239 for (snap
= list_head(l
); snap
; snap
= list_next(l
, snap
)) {
2240 uint64_t used
, comp
, uncomp
;
2241 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2242 mintxg
, UINT64_MAX
, &used
, &comp
, &uncomp
);
2249 snaplist_destroy(list_t
*l
, void *tag
)
2251 struct promotenode
*snap
;
2253 if (l
== NULL
|| !list_link_active(&l
->list_head
))
2256 while ((snap
= list_tail(l
)) != NULL
) {
2257 list_remove(l
, snap
);
2258 dsl_dataset_rele(snap
->ds
, tag
);
2259 kmem_free(snap
, sizeof (*snap
));
2265 promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
, void *tag
)
2269 struct promotenode
*snap
;
2271 error
= dsl_dataset_hold(dp
, ddpa
->ddpa_clonename
, tag
,
2275 dd
= ddpa
->ddpa_clone
->ds_dir
;
2277 if (dsl_dataset_is_snapshot(ddpa
->ddpa_clone
) ||
2278 !dsl_dir_is_clone(dd
)) {
2279 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2280 return (SET_ERROR(EINVAL
));
2283 error
= snaplist_make(dp
, 0, dd
->dd_phys
->dd_origin_obj
,
2284 &ddpa
->shared_snaps
, tag
);
2288 error
= snaplist_make(dp
, 0, ddpa
->ddpa_clone
->ds_object
,
2289 &ddpa
->clone_snaps
, tag
);
2293 snap
= list_head(&ddpa
->shared_snaps
);
2294 ASSERT3U(snap
->ds
->ds_object
, ==, dd
->dd_phys
->dd_origin_obj
);
2295 error
= snaplist_make(dp
, dd
->dd_phys
->dd_origin_obj
,
2296 snap
->ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
,
2297 &ddpa
->origin_snaps
, tag
);
2301 if (snap
->ds
->ds_dir
->dd_phys
->dd_origin_obj
!= 0) {
2302 error
= dsl_dataset_hold_obj(dp
,
2303 snap
->ds
->ds_dir
->dd_phys
->dd_origin_obj
,
2304 tag
, &ddpa
->origin_origin
);
2310 promote_rele(ddpa
, tag
);
2315 promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
)
2317 snaplist_destroy(&ddpa
->shared_snaps
, tag
);
2318 snaplist_destroy(&ddpa
->clone_snaps
, tag
);
2319 snaplist_destroy(&ddpa
->origin_snaps
, tag
);
2320 if (ddpa
->origin_origin
!= NULL
)
2321 dsl_dataset_rele(ddpa
->origin_origin
, tag
);
2322 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2328 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2329 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2332 dsl_dataset_promote(const char *name
, char *conflsnap
)
2334 dsl_dataset_promote_arg_t ddpa
= { 0 };
2340 * We will modify space proportional to the number of
2341 * snapshots. Compute numsnaps.
2343 error
= dmu_objset_hold(name
, FTAG
, &os
);
2346 error
= zap_count(dmu_objset_pool(os
)->dp_meta_objset
,
2347 dmu_objset_ds(os
)->ds_phys
->ds_snapnames_zapobj
, &numsnaps
);
2348 dmu_objset_rele(os
, FTAG
);
2352 ddpa
.ddpa_clonename
= name
;
2353 ddpa
.err_ds
= conflsnap
;
2355 return (dsl_sync_task(name
, dsl_dataset_promote_check
,
2356 dsl_dataset_promote_sync
, &ddpa
, 2 + numsnaps
));
2360 dsl_dataset_clone_swap_check_impl(dsl_dataset_t
*clone
,
2361 dsl_dataset_t
*origin_head
, boolean_t force
, void *owner
, dmu_tx_t
*tx
)
2363 int64_t unused_refres_delta
;
2365 /* they should both be heads */
2366 if (dsl_dataset_is_snapshot(clone
) ||
2367 dsl_dataset_is_snapshot(origin_head
))
2368 return (SET_ERROR(EINVAL
));
2370 /* if we are not forcing, the branch point should be just before them */
2371 if (!force
&& clone
->ds_prev
!= origin_head
->ds_prev
)
2372 return (SET_ERROR(EINVAL
));
2374 /* clone should be the clone (unless they are unrelated) */
2375 if (clone
->ds_prev
!= NULL
&&
2376 clone
->ds_prev
!= clone
->ds_dir
->dd_pool
->dp_origin_snap
&&
2377 origin_head
->ds_dir
!= clone
->ds_prev
->ds_dir
)
2378 return (SET_ERROR(EINVAL
));
2380 /* the clone should be a child of the origin */
2381 if (clone
->ds_dir
->dd_parent
!= origin_head
->ds_dir
)
2382 return (SET_ERROR(EINVAL
));
2384 /* origin_head shouldn't be modified unless 'force' */
2386 dsl_dataset_modified_since_snap(origin_head
, origin_head
->ds_prev
))
2387 return (SET_ERROR(ETXTBSY
));
2389 /* origin_head should have no long holds (e.g. is not mounted) */
2390 if (dsl_dataset_handoff_check(origin_head
, owner
, tx
))
2391 return (SET_ERROR(EBUSY
));
2393 /* check amount of any unconsumed refreservation */
2394 unused_refres_delta
=
2395 (int64_t)MIN(origin_head
->ds_reserved
,
2396 origin_head
->ds_phys
->ds_unique_bytes
) -
2397 (int64_t)MIN(origin_head
->ds_reserved
,
2398 clone
->ds_phys
->ds_unique_bytes
);
2400 if (unused_refres_delta
> 0 &&
2401 unused_refres_delta
>
2402 dsl_dir_space_available(origin_head
->ds_dir
, NULL
, 0, TRUE
))
2403 return (SET_ERROR(ENOSPC
));
2405 /* clone can't be over the head's refquota */
2406 if (origin_head
->ds_quota
!= 0 &&
2407 clone
->ds_phys
->ds_referenced_bytes
> origin_head
->ds_quota
)
2408 return (SET_ERROR(EDQUOT
));
2414 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t
*clone
,
2415 dsl_dataset_t
*origin_head
, dmu_tx_t
*tx
)
2417 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2418 int64_t unused_refres_delta
;
2420 ASSERT(clone
->ds_reserved
== 0);
2421 ASSERT(origin_head
->ds_quota
== 0 ||
2422 clone
->ds_phys
->ds_unique_bytes
<= origin_head
->ds_quota
);
2423 ASSERT3P(clone
->ds_prev
, ==, origin_head
->ds_prev
);
2425 dmu_buf_will_dirty(clone
->ds_dbuf
, tx
);
2426 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
2428 if (clone
->ds_objset
!= NULL
) {
2429 dmu_objset_evict(clone
->ds_objset
);
2430 clone
->ds_objset
= NULL
;
2433 if (origin_head
->ds_objset
!= NULL
) {
2434 dmu_objset_evict(origin_head
->ds_objset
);
2435 origin_head
->ds_objset
= NULL
;
2438 unused_refres_delta
=
2439 (int64_t)MIN(origin_head
->ds_reserved
,
2440 origin_head
->ds_phys
->ds_unique_bytes
) -
2441 (int64_t)MIN(origin_head
->ds_reserved
,
2442 clone
->ds_phys
->ds_unique_bytes
);
2445 * Reset origin's unique bytes, if it exists.
2447 if (clone
->ds_prev
) {
2448 dsl_dataset_t
*origin
= clone
->ds_prev
;
2449 uint64_t comp
, uncomp
;
2451 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
2452 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2453 origin
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
2454 &origin
->ds_phys
->ds_unique_bytes
, &comp
, &uncomp
);
2460 tmp
= origin_head
->ds_phys
->ds_bp
;
2461 origin_head
->ds_phys
->ds_bp
= clone
->ds_phys
->ds_bp
;
2462 clone
->ds_phys
->ds_bp
= tmp
;
2465 /* set dd_*_bytes */
2467 int64_t dused
, dcomp
, duncomp
;
2468 uint64_t cdl_used
, cdl_comp
, cdl_uncomp
;
2469 uint64_t odl_used
, odl_comp
, odl_uncomp
;
2471 ASSERT3U(clone
->ds_dir
->dd_phys
->
2472 dd_used_breakdown
[DD_USED_SNAP
], ==, 0);
2474 dsl_deadlist_space(&clone
->ds_deadlist
,
2475 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
2476 dsl_deadlist_space(&origin_head
->ds_deadlist
,
2477 &odl_used
, &odl_comp
, &odl_uncomp
);
2479 dused
= clone
->ds_phys
->ds_referenced_bytes
+ cdl_used
-
2480 (origin_head
->ds_phys
->ds_referenced_bytes
+ odl_used
);
2481 dcomp
= clone
->ds_phys
->ds_compressed_bytes
+ cdl_comp
-
2482 (origin_head
->ds_phys
->ds_compressed_bytes
+ odl_comp
);
2483 duncomp
= clone
->ds_phys
->ds_uncompressed_bytes
+
2485 (origin_head
->ds_phys
->ds_uncompressed_bytes
+ odl_uncomp
);
2487 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_HEAD
,
2488 dused
, dcomp
, duncomp
, tx
);
2489 dsl_dir_diduse_space(clone
->ds_dir
, DD_USED_HEAD
,
2490 -dused
, -dcomp
, -duncomp
, tx
);
2493 * The difference in the space used by snapshots is the
2494 * difference in snapshot space due to the head's
2495 * deadlist (since that's the only thing that's
2496 * changing that affects the snapused).
2498 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2499 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
2500 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
2501 dsl_deadlist_space_range(&origin_head
->ds_deadlist
,
2502 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
2503 &odl_used
, &odl_comp
, &odl_uncomp
);
2504 dsl_dir_transfer_space(origin_head
->ds_dir
, cdl_used
- odl_used
,
2505 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
2508 /* swap ds_*_bytes */
2509 SWITCH64(origin_head
->ds_phys
->ds_referenced_bytes
,
2510 clone
->ds_phys
->ds_referenced_bytes
);
2511 SWITCH64(origin_head
->ds_phys
->ds_compressed_bytes
,
2512 clone
->ds_phys
->ds_compressed_bytes
);
2513 SWITCH64(origin_head
->ds_phys
->ds_uncompressed_bytes
,
2514 clone
->ds_phys
->ds_uncompressed_bytes
);
2515 SWITCH64(origin_head
->ds_phys
->ds_unique_bytes
,
2516 clone
->ds_phys
->ds_unique_bytes
);
2518 /* apply any parent delta for change in unconsumed refreservation */
2519 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_REFRSRV
,
2520 unused_refres_delta
, 0, 0, tx
);
2525 dsl_deadlist_close(&clone
->ds_deadlist
);
2526 dsl_deadlist_close(&origin_head
->ds_deadlist
);
2527 SWITCH64(origin_head
->ds_phys
->ds_deadlist_obj
,
2528 clone
->ds_phys
->ds_deadlist_obj
);
2529 dsl_deadlist_open(&clone
->ds_deadlist
, dp
->dp_meta_objset
,
2530 clone
->ds_phys
->ds_deadlist_obj
);
2531 dsl_deadlist_open(&origin_head
->ds_deadlist
, dp
->dp_meta_objset
,
2532 origin_head
->ds_phys
->ds_deadlist_obj
);
2534 dsl_scan_ds_clone_swapped(origin_head
, clone
, tx
);
2536 spa_history_log_internal_ds(clone
, "clone swap", tx
,
2537 "parent=%s", origin_head
->ds_dir
->dd_myname
);
2541 * Given a pool name and a dataset object number in that pool,
2542 * return the name of that dataset.
2545 dsl_dsobj_to_dsname(char *pname
, uint64_t obj
, char *buf
)
2551 error
= dsl_pool_hold(pname
, FTAG
, &dp
);
2555 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
2557 dsl_dataset_name(ds
, buf
);
2558 dsl_dataset_rele(ds
, FTAG
);
2560 dsl_pool_rele(dp
, FTAG
);
2566 dsl_dataset_check_quota(dsl_dataset_t
*ds
, boolean_t check_quota
,
2567 uint64_t asize
, uint64_t inflight
, uint64_t *used
, uint64_t *ref_rsrv
)
2571 ASSERT3S(asize
, >, 0);
2574 * *ref_rsrv is the portion of asize that will come from any
2575 * unconsumed refreservation space.
2579 mutex_enter(&ds
->ds_lock
);
2581 * Make a space adjustment for reserved bytes.
2583 if (ds
->ds_reserved
> ds
->ds_phys
->ds_unique_bytes
) {
2585 ds
->ds_reserved
- ds
->ds_phys
->ds_unique_bytes
);
2586 *used
-= (ds
->ds_reserved
- ds
->ds_phys
->ds_unique_bytes
);
2588 asize
- MIN(asize
, parent_delta(ds
, asize
+ inflight
));
2591 if (!check_quota
|| ds
->ds_quota
== 0) {
2592 mutex_exit(&ds
->ds_lock
);
2596 * If they are requesting more space, and our current estimate
2597 * is over quota, they get to try again unless the actual
2598 * on-disk is over quota and there are no pending changes (which
2599 * may free up space for us).
2601 if (ds
->ds_phys
->ds_referenced_bytes
+ inflight
>= ds
->ds_quota
) {
2603 ds
->ds_phys
->ds_referenced_bytes
< ds
->ds_quota
)
2604 error
= SET_ERROR(ERESTART
);
2606 error
= SET_ERROR(EDQUOT
);
2608 mutex_exit(&ds
->ds_lock
);
2613 typedef struct dsl_dataset_set_qr_arg
{
2614 const char *ddsqra_name
;
2615 zprop_source_t ddsqra_source
;
2616 uint64_t ddsqra_value
;
2617 } dsl_dataset_set_qr_arg_t
;
2622 dsl_dataset_set_refquota_check(void *arg
, dmu_tx_t
*tx
)
2624 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2625 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2630 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFQUOTA
)
2631 return (SET_ERROR(ENOTSUP
));
2633 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
2637 if (dsl_dataset_is_snapshot(ds
)) {
2638 dsl_dataset_rele(ds
, FTAG
);
2639 return (SET_ERROR(EINVAL
));
2642 error
= dsl_prop_predict(ds
->ds_dir
,
2643 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
2644 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
2646 dsl_dataset_rele(ds
, FTAG
);
2651 dsl_dataset_rele(ds
, FTAG
);
2655 if (newval
< ds
->ds_phys
->ds_referenced_bytes
||
2656 newval
< ds
->ds_reserved
) {
2657 dsl_dataset_rele(ds
, FTAG
);
2658 return (SET_ERROR(ENOSPC
));
2661 dsl_dataset_rele(ds
, FTAG
);
2666 dsl_dataset_set_refquota_sync(void *arg
, dmu_tx_t
*tx
)
2668 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2669 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2673 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
2675 dsl_prop_set_sync_impl(ds
,
2676 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
2677 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
2678 &ddsqra
->ddsqra_value
, tx
);
2680 VERIFY0(dsl_prop_get_int_ds(ds
,
2681 zfs_prop_to_name(ZFS_PROP_REFQUOTA
), &newval
));
2683 if (ds
->ds_quota
!= newval
) {
2684 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2685 ds
->ds_quota
= newval
;
2687 dsl_dataset_rele(ds
, FTAG
);
2691 dsl_dataset_set_refquota(const char *dsname
, zprop_source_t source
,
2694 dsl_dataset_set_qr_arg_t ddsqra
;
2696 ddsqra
.ddsqra_name
= dsname
;
2697 ddsqra
.ddsqra_source
= source
;
2698 ddsqra
.ddsqra_value
= refquota
;
2700 return (dsl_sync_task(dsname
, dsl_dataset_set_refquota_check
,
2701 dsl_dataset_set_refquota_sync
, &ddsqra
, 0));
2705 dsl_dataset_set_refreservation_check(void *arg
, dmu_tx_t
*tx
)
2707 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2708 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2711 uint64_t newval
, unique
;
2713 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFRESERVATION
)
2714 return (SET_ERROR(ENOTSUP
));
2716 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
2720 if (dsl_dataset_is_snapshot(ds
)) {
2721 dsl_dataset_rele(ds
, FTAG
);
2722 return (SET_ERROR(EINVAL
));
2725 error
= dsl_prop_predict(ds
->ds_dir
,
2726 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
2727 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
2729 dsl_dataset_rele(ds
, FTAG
);
2734 * If we are doing the preliminary check in open context, the
2735 * space estimates may be inaccurate.
2737 if (!dmu_tx_is_syncing(tx
)) {
2738 dsl_dataset_rele(ds
, FTAG
);
2742 mutex_enter(&ds
->ds_lock
);
2743 if (!DS_UNIQUE_IS_ACCURATE(ds
))
2744 dsl_dataset_recalc_head_uniq(ds
);
2745 unique
= ds
->ds_phys
->ds_unique_bytes
;
2746 mutex_exit(&ds
->ds_lock
);
2748 if (MAX(unique
, newval
) > MAX(unique
, ds
->ds_reserved
)) {
2749 uint64_t delta
= MAX(unique
, newval
) -
2750 MAX(unique
, ds
->ds_reserved
);
2753 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, B_TRUE
) ||
2754 (ds
->ds_quota
> 0 && newval
> ds
->ds_quota
)) {
2755 dsl_dataset_rele(ds
, FTAG
);
2756 return (SET_ERROR(ENOSPC
));
2760 dsl_dataset_rele(ds
, FTAG
);
2765 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t
*ds
,
2766 zprop_source_t source
, uint64_t value
, dmu_tx_t
*tx
)
2772 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
2773 source
, sizeof (value
), 1, &value
, tx
);
2775 VERIFY0(dsl_prop_get_int_ds(ds
,
2776 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
), &newval
));
2778 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2779 mutex_enter(&ds
->ds_dir
->dd_lock
);
2780 mutex_enter(&ds
->ds_lock
);
2781 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
2782 unique
= ds
->ds_phys
->ds_unique_bytes
;
2783 delta
= MAX(0, (int64_t)(newval
- unique
)) -
2784 MAX(0, (int64_t)(ds
->ds_reserved
- unique
));
2785 ds
->ds_reserved
= newval
;
2786 mutex_exit(&ds
->ds_lock
);
2788 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
, delta
, 0, 0, tx
);
2789 mutex_exit(&ds
->ds_dir
->dd_lock
);
2793 dsl_dataset_set_refreservation_sync(void *arg
, dmu_tx_t
*tx
)
2795 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2796 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2799 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
2800 dsl_dataset_set_refreservation_sync_impl(ds
,
2801 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, tx
);
2802 dsl_dataset_rele(ds
, FTAG
);
2806 dsl_dataset_set_refreservation(const char *dsname
, zprop_source_t source
,
2807 uint64_t refreservation
)
2809 dsl_dataset_set_qr_arg_t ddsqra
;
2811 ddsqra
.ddsqra_name
= dsname
;
2812 ddsqra
.ddsqra_source
= source
;
2813 ddsqra
.ddsqra_value
= refreservation
;
2815 return (dsl_sync_task(dsname
, dsl_dataset_set_refreservation_check
,
2816 dsl_dataset_set_refreservation_sync
, &ddsqra
, 0));
2820 * Return (in *usedp) the amount of space written in new that is not
2821 * present in oldsnap. New may be a snapshot or the head. Old must be
2822 * a snapshot before new, in new's filesystem (or its origin). If not then
2823 * fail and return EINVAL.
2825 * The written space is calculated by considering two components: First, we
2826 * ignore any freed space, and calculate the written as new's used space
2827 * minus old's used space. Next, we add in the amount of space that was freed
2828 * between the two snapshots, thus reducing new's used space relative to old's.
2829 * Specifically, this is the space that was born before old->ds_creation_txg,
2830 * and freed before new (ie. on new's deadlist or a previous deadlist).
2832 * space freed [---------------------]
2833 * snapshots ---O-------O--------O-------O------
2837 dsl_dataset_space_written(dsl_dataset_t
*oldsnap
, dsl_dataset_t
*new,
2838 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
2842 dsl_pool_t
*dp
= new->ds_dir
->dd_pool
;
2844 ASSERT(dsl_pool_config_held(dp
));
2847 *usedp
+= new->ds_phys
->ds_referenced_bytes
;
2848 *usedp
-= oldsnap
->ds_phys
->ds_referenced_bytes
;
2851 *compp
+= new->ds_phys
->ds_compressed_bytes
;
2852 *compp
-= oldsnap
->ds_phys
->ds_compressed_bytes
;
2855 *uncompp
+= new->ds_phys
->ds_uncompressed_bytes
;
2856 *uncompp
-= oldsnap
->ds_phys
->ds_uncompressed_bytes
;
2858 snapobj
= new->ds_object
;
2859 while (snapobj
!= oldsnap
->ds_object
) {
2860 dsl_dataset_t
*snap
;
2861 uint64_t used
, comp
, uncomp
;
2863 if (snapobj
== new->ds_object
) {
2866 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &snap
);
2871 if (snap
->ds_phys
->ds_prev_snap_txg
==
2872 oldsnap
->ds_phys
->ds_creation_txg
) {
2874 * The blocks in the deadlist can not be born after
2875 * ds_prev_snap_txg, so get the whole deadlist space,
2876 * which is more efficient (especially for old-format
2877 * deadlists). Unfortunately the deadlist code
2878 * doesn't have enough information to make this
2879 * optimization itself.
2881 dsl_deadlist_space(&snap
->ds_deadlist
,
2882 &used
, &comp
, &uncomp
);
2884 dsl_deadlist_space_range(&snap
->ds_deadlist
,
2885 0, oldsnap
->ds_phys
->ds_creation_txg
,
2886 &used
, &comp
, &uncomp
);
2893 * If we get to the beginning of the chain of snapshots
2894 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
2895 * was not a snapshot of/before new.
2897 snapobj
= snap
->ds_phys
->ds_prev_snap_obj
;
2899 dsl_dataset_rele(snap
, FTAG
);
2901 err
= SET_ERROR(EINVAL
);
2910 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
2911 * lastsnap, and all snapshots in between are deleted.
2913 * blocks that would be freed [---------------------------]
2914 * snapshots ---O-------O--------O-------O--------O
2915 * firstsnap lastsnap
2917 * This is the set of blocks that were born after the snap before firstsnap,
2918 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
2919 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
2920 * We calculate this by iterating over the relevant deadlists (from the snap
2921 * after lastsnap, backward to the snap after firstsnap), summing up the
2922 * space on the deadlist that was born after the snap before firstsnap.
2925 dsl_dataset_space_wouldfree(dsl_dataset_t
*firstsnap
,
2926 dsl_dataset_t
*lastsnap
,
2927 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
2931 dsl_pool_t
*dp
= firstsnap
->ds_dir
->dd_pool
;
2933 ASSERT(dsl_dataset_is_snapshot(firstsnap
));
2934 ASSERT(dsl_dataset_is_snapshot(lastsnap
));
2937 * Check that the snapshots are in the same dsl_dir, and firstsnap
2938 * is before lastsnap.
2940 if (firstsnap
->ds_dir
!= lastsnap
->ds_dir
||
2941 firstsnap
->ds_phys
->ds_creation_txg
>
2942 lastsnap
->ds_phys
->ds_creation_txg
)
2943 return (SET_ERROR(EINVAL
));
2945 *usedp
= *compp
= *uncompp
= 0;
2947 snapobj
= lastsnap
->ds_phys
->ds_next_snap_obj
;
2948 while (snapobj
!= firstsnap
->ds_object
) {
2950 uint64_t used
, comp
, uncomp
;
2952 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &ds
);
2956 dsl_deadlist_space_range(&ds
->ds_deadlist
,
2957 firstsnap
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
2958 &used
, &comp
, &uncomp
);
2963 snapobj
= ds
->ds_phys
->ds_prev_snap_obj
;
2964 ASSERT3U(snapobj
, !=, 0);
2965 dsl_dataset_rele(ds
, FTAG
);
2971 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
2972 * For example, they could both be snapshots of the same filesystem, and
2973 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
2974 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
2975 * filesystem. Or 'earlier' could be the origin's origin.
2978 dsl_dataset_is_before(dsl_dataset_t
*later
, dsl_dataset_t
*earlier
)
2980 dsl_pool_t
*dp
= later
->ds_dir
->dd_pool
;
2983 dsl_dataset_t
*origin
;
2985 ASSERT(dsl_pool_config_held(dp
));
2987 if (earlier
->ds_phys
->ds_creation_txg
>=
2988 later
->ds_phys
->ds_creation_txg
)
2991 if (later
->ds_dir
== earlier
->ds_dir
)
2993 if (!dsl_dir_is_clone(later
->ds_dir
))
2996 if (later
->ds_dir
->dd_phys
->dd_origin_obj
== earlier
->ds_object
)
2998 error
= dsl_dataset_hold_obj(dp
,
2999 later
->ds_dir
->dd_phys
->dd_origin_obj
, FTAG
, &origin
);
3002 ret
= dsl_dataset_is_before(origin
, earlier
);
3003 dsl_dataset_rele(origin
, FTAG
);
3007 #if defined(_KERNEL) && defined(HAVE_SPL)
3008 EXPORT_SYMBOL(dsl_dataset_hold
);
3009 EXPORT_SYMBOL(dsl_dataset_hold_obj
);
3010 EXPORT_SYMBOL(dsl_dataset_own
);
3011 EXPORT_SYMBOL(dsl_dataset_own_obj
);
3012 EXPORT_SYMBOL(dsl_dataset_name
);
3013 EXPORT_SYMBOL(dsl_dataset_rele
);
3014 EXPORT_SYMBOL(dsl_dataset_disown
);
3015 EXPORT_SYMBOL(dsl_dataset_tryown
);
3016 EXPORT_SYMBOL(dsl_dataset_create_sync
);
3017 EXPORT_SYMBOL(dsl_dataset_create_sync_dd
);
3018 EXPORT_SYMBOL(dsl_dataset_snapshot_check
);
3019 EXPORT_SYMBOL(dsl_dataset_snapshot_sync
);
3020 EXPORT_SYMBOL(dsl_dataset_promote
);
3021 EXPORT_SYMBOL(dsl_dataset_user_hold
);
3022 EXPORT_SYMBOL(dsl_dataset_user_release
);
3023 EXPORT_SYMBOL(dsl_dataset_get_holds
);
3024 EXPORT_SYMBOL(dsl_dataset_get_blkptr
);
3025 EXPORT_SYMBOL(dsl_dataset_set_blkptr
);
3026 EXPORT_SYMBOL(dsl_dataset_get_spa
);
3027 EXPORT_SYMBOL(dsl_dataset_modified_since_snap
);
3028 EXPORT_SYMBOL(dsl_dataset_space_written
);
3029 EXPORT_SYMBOL(dsl_dataset_space_wouldfree
);
3030 EXPORT_SYMBOL(dsl_dataset_sync
);
3031 EXPORT_SYMBOL(dsl_dataset_block_born
);
3032 EXPORT_SYMBOL(dsl_dataset_block_kill
);
3033 EXPORT_SYMBOL(dsl_dataset_block_freeable
);
3034 EXPORT_SYMBOL(dsl_dataset_prev_snap_txg
);
3035 EXPORT_SYMBOL(dsl_dataset_dirty
);
3036 EXPORT_SYMBOL(dsl_dataset_stats
);
3037 EXPORT_SYMBOL(dsl_dataset_fast_stat
);
3038 EXPORT_SYMBOL(dsl_dataset_space
);
3039 EXPORT_SYMBOL(dsl_dataset_fsid_guid
);
3040 EXPORT_SYMBOL(dsl_dsobj_to_dsname
);
3041 EXPORT_SYMBOL(dsl_dataset_check_quota
);
3042 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl
);
3043 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl
);