4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2015 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.
28 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
31 #include <sys/dmu_objset.h>
32 #include <sys/dsl_dataset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_prop.h>
35 #include <sys/dsl_synctask.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dmu_impl.h>
38 #include <sys/dmu_tx.h>
42 #include <sys/zfeature.h>
43 #include <sys/unique.h>
44 #include <sys/zfs_context.h>
45 #include <sys/zfs_ioctl.h>
47 #include <sys/zfs_znode.h>
48 #include <sys/zfs_onexit.h>
50 #include <sys/dsl_scan.h>
51 #include <sys/dsl_deadlist.h>
52 #include <sys/dsl_destroy.h>
53 #include <sys/dsl_userhold.h>
54 #include <sys/dsl_bookmark.h>
55 #include <sys/policy.h>
56 #include <sys/dmu_send.h>
57 #include <sys/zio_compress.h>
58 #include <zfs_fletcher.h>
59 #include <sys/zio_checksum.h>
62 * The SPA supports block sizes up to 16MB. However, very large blocks
63 * can have an impact on i/o latency (e.g. tying up a spinning disk for
64 * ~300ms), and also potentially on the memory allocator. Therefore,
65 * we do not allow the recordsize to be set larger than zfs_max_recordsize
66 * (default 1MB). Larger blocks can be created by changing this tunable,
67 * and pools with larger blocks can always be imported and used, regardless
70 int zfs_max_recordsize
= 1 * 1024 * 1024;
72 #define SWITCH64(x, y) \
74 uint64_t __tmp = (x); \
79 #define DS_REF_MAX (1ULL << 62)
81 extern inline dsl_dataset_phys_t
*dsl_dataset_phys(dsl_dataset_t
*ds
);
83 extern int spa_asize_inflation
;
86 * Figure out how much of this delta should be propogated to the dsl_dir
87 * layer. If there's a refreservation, that space has already been
88 * partially accounted for in our ancestors.
91 parent_delta(dsl_dataset_t
*ds
, int64_t delta
)
93 dsl_dataset_phys_t
*ds_phys
;
94 uint64_t old_bytes
, new_bytes
;
96 if (ds
->ds_reserved
== 0)
99 ds_phys
= dsl_dataset_phys(ds
);
100 old_bytes
= MAX(ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
101 new_bytes
= MAX(ds_phys
->ds_unique_bytes
+ delta
, ds
->ds_reserved
);
103 ASSERT3U(ABS((int64_t)(new_bytes
- old_bytes
)), <=, ABS(delta
));
104 return (new_bytes
- old_bytes
);
108 dsl_dataset_block_born(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
110 int used
, compressed
, uncompressed
;
114 used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
115 compressed
= BP_GET_PSIZE(bp
);
116 uncompressed
= BP_GET_UCSIZE(bp
);
118 dprintf_bp(bp
, "ds=%p", ds
);
120 ASSERT(dmu_tx_is_syncing(tx
));
121 /* It could have been compressed away to nothing */
124 ASSERT(BP_GET_TYPE(bp
) != DMU_OT_NONE
);
125 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp
)));
127 dsl_pool_mos_diduse_space(tx
->tx_pool
,
128 used
, compressed
, uncompressed
);
132 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
133 mutex_enter(&ds
->ds_lock
);
134 delta
= parent_delta(ds
, used
);
135 dsl_dataset_phys(ds
)->ds_referenced_bytes
+= used
;
136 dsl_dataset_phys(ds
)->ds_compressed_bytes
+= compressed
;
137 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
+= uncompressed
;
138 dsl_dataset_phys(ds
)->ds_unique_bytes
+= used
;
140 if (BP_GET_LSIZE(bp
) > SPA_OLD_MAXBLOCKSIZE
) {
141 ds
->ds_feature_activation_needed
[SPA_FEATURE_LARGE_BLOCKS
] =
145 f
= zio_checksum_to_feature(BP_GET_CHECKSUM(bp
));
146 if (f
!= SPA_FEATURE_NONE
)
147 ds
->ds_feature_activation_needed
[f
] = B_TRUE
;
149 mutex_exit(&ds
->ds_lock
);
150 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
, delta
,
151 compressed
, uncompressed
, tx
);
152 dsl_dir_transfer_space(ds
->ds_dir
, used
- delta
,
153 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
157 dsl_dataset_block_kill(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
,
160 int used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
161 int compressed
= BP_GET_PSIZE(bp
);
162 int uncompressed
= BP_GET_UCSIZE(bp
);
167 ASSERT(dmu_tx_is_syncing(tx
));
168 ASSERT(bp
->blk_birth
<= tx
->tx_txg
);
171 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
172 dsl_pool_mos_diduse_space(tx
->tx_pool
,
173 -used
, -compressed
, -uncompressed
);
176 ASSERT3P(tx
->tx_pool
, ==, ds
->ds_dir
->dd_pool
);
178 ASSERT(!ds
->ds_is_snapshot
);
179 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
181 if (bp
->blk_birth
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
184 dprintf_bp(bp
, "freeing ds=%llu", ds
->ds_object
);
185 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
187 mutex_enter(&ds
->ds_lock
);
188 ASSERT(dsl_dataset_phys(ds
)->ds_unique_bytes
>= used
||
189 !DS_UNIQUE_IS_ACCURATE(ds
));
190 delta
= parent_delta(ds
, -used
);
191 dsl_dataset_phys(ds
)->ds_unique_bytes
-= used
;
192 mutex_exit(&ds
->ds_lock
);
193 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
194 delta
, -compressed
, -uncompressed
, tx
);
195 dsl_dir_transfer_space(ds
->ds_dir
, -used
- delta
,
196 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
198 dprintf_bp(bp
, "putting on dead list: %s", "");
201 * We are here as part of zio's write done callback,
202 * which means we're a zio interrupt thread. We can't
203 * call dsl_deadlist_insert() now because it may block
204 * waiting for I/O. Instead, put bp on the deferred
205 * queue and let dsl_pool_sync() finish the job.
207 bplist_append(&ds
->ds_pending_deadlist
, bp
);
209 dsl_deadlist_insert(&ds
->ds_deadlist
, bp
, tx
);
211 ASSERT3U(ds
->ds_prev
->ds_object
, ==,
212 dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
213 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 0);
214 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
215 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
216 ds
->ds_object
&& bp
->blk_birth
>
217 dsl_dataset_phys(ds
->ds_prev
)->ds_prev_snap_txg
) {
218 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
219 mutex_enter(&ds
->ds_prev
->ds_lock
);
220 dsl_dataset_phys(ds
->ds_prev
)->ds_unique_bytes
+= used
;
221 mutex_exit(&ds
->ds_prev
->ds_lock
);
223 if (bp
->blk_birth
> ds
->ds_dir
->dd_origin_txg
) {
224 dsl_dir_transfer_space(ds
->ds_dir
, used
,
225 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
228 mutex_enter(&ds
->ds_lock
);
229 ASSERT3U(dsl_dataset_phys(ds
)->ds_referenced_bytes
, >=, used
);
230 dsl_dataset_phys(ds
)->ds_referenced_bytes
-= used
;
231 ASSERT3U(dsl_dataset_phys(ds
)->ds_compressed_bytes
, >=, compressed
);
232 dsl_dataset_phys(ds
)->ds_compressed_bytes
-= compressed
;
233 ASSERT3U(dsl_dataset_phys(ds
)->ds_uncompressed_bytes
, >=, uncompressed
);
234 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
-= uncompressed
;
235 mutex_exit(&ds
->ds_lock
);
241 dsl_dataset_prev_snap_txg(dsl_dataset_t
*ds
)
243 uint64_t trysnap
= 0;
248 * The snapshot creation could fail, but that would cause an
249 * incorrect FALSE return, which would only result in an
250 * overestimation of the amount of space that an operation would
251 * consume, which is OK.
253 * There's also a small window where we could miss a pending
254 * snapshot, because we could set the sync task in the quiescing
255 * phase. So this should only be used as a guess.
257 if (ds
->ds_trysnap_txg
>
258 spa_last_synced_txg(ds
->ds_dir
->dd_pool
->dp_spa
))
259 trysnap
= ds
->ds_trysnap_txg
;
260 return (MAX(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, trysnap
));
264 dsl_dataset_block_freeable(dsl_dataset_t
*ds
, const blkptr_t
*bp
,
267 if (blk_birth
<= dsl_dataset_prev_snap_txg(ds
) ||
268 (bp
!= NULL
&& BP_IS_HOLE(bp
)))
271 ddt_prefetch(dsl_dataset_get_spa(ds
), bp
);
277 dsl_dataset_evict(void *dbu
)
279 dsl_dataset_t
*ds
= dbu
;
281 ASSERT(ds
->ds_owner
== NULL
);
285 unique_remove(ds
->ds_fsid_guid
);
287 if (ds
->ds_objset
!= NULL
)
288 dmu_objset_evict(ds
->ds_objset
);
291 dsl_dataset_rele(ds
->ds_prev
, ds
);
295 bplist_destroy(&ds
->ds_pending_deadlist
);
296 if (ds
->ds_deadlist
.dl_os
!= NULL
)
297 dsl_deadlist_close(&ds
->ds_deadlist
);
299 dsl_dir_async_rele(ds
->ds_dir
, ds
);
301 ASSERT(!list_link_active(&ds
->ds_synced_link
));
303 list_destroy(&ds
->ds_prop_cbs
);
304 mutex_destroy(&ds
->ds_lock
);
305 mutex_destroy(&ds
->ds_opening_lock
);
306 mutex_destroy(&ds
->ds_sendstream_lock
);
307 refcount_destroy(&ds
->ds_longholds
);
309 kmem_free(ds
, sizeof (dsl_dataset_t
));
313 dsl_dataset_get_snapname(dsl_dataset_t
*ds
)
315 dsl_dataset_phys_t
*headphys
;
318 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
319 objset_t
*mos
= dp
->dp_meta_objset
;
321 if (ds
->ds_snapname
[0])
323 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0)
326 err
= dmu_bonus_hold(mos
, dsl_dir_phys(ds
->ds_dir
)->dd_head_dataset_obj
,
330 headphys
= headdbuf
->db_data
;
331 err
= zap_value_search(dp
->dp_meta_objset
,
332 headphys
->ds_snapnames_zapobj
, ds
->ds_object
, 0, ds
->ds_snapname
);
333 if (err
!= 0 && zfs_recover
== B_TRUE
) {
335 (void) snprintf(ds
->ds_snapname
, sizeof (ds
->ds_snapname
),
336 "SNAPOBJ=%llu-ERR=%d",
337 (unsigned long long)ds
->ds_object
, err
);
339 dmu_buf_rele(headdbuf
, FTAG
);
344 dsl_dataset_snap_lookup(dsl_dataset_t
*ds
, const char *name
, uint64_t *value
)
346 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
347 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
351 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
356 err
= zap_lookup_norm(mos
, snapobj
, name
, 8, 1,
357 value
, mt
, NULL
, 0, NULL
);
358 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
359 err
= zap_lookup(mos
, snapobj
, name
, 8, 1, value
);
364 dsl_dataset_snap_remove(dsl_dataset_t
*ds
, const char *name
, dmu_tx_t
*tx
,
367 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
368 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
372 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
374 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
379 err
= zap_remove_norm(mos
, snapobj
, name
, mt
, tx
);
380 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
381 err
= zap_remove(mos
, snapobj
, name
, tx
);
383 if (err
== 0 && adj_cnt
)
384 dsl_fs_ss_count_adjust(ds
->ds_dir
, -1,
385 DD_FIELD_SNAPSHOT_COUNT
, tx
);
391 dsl_dataset_try_add_ref(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *tag
)
393 dmu_buf_t
*dbuf
= ds
->ds_dbuf
;
394 boolean_t result
= B_FALSE
;
396 if (dbuf
!= NULL
&& dmu_buf_try_add_ref(dbuf
, dp
->dp_meta_objset
,
397 ds
->ds_object
, DMU_BONUS_BLKID
, tag
)) {
399 if (ds
== dmu_buf_get_user(dbuf
))
402 dmu_buf_rele(dbuf
, tag
);
409 dsl_dataset_hold_obj(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
412 objset_t
*mos
= dp
->dp_meta_objset
;
416 dmu_object_info_t doi
;
418 ASSERT(dsl_pool_config_held(dp
));
420 err
= dmu_bonus_hold(mos
, dsobj
, tag
, &dbuf
);
424 /* Make sure dsobj has the correct object type. */
425 dmu_object_info_from_db(dbuf
, &doi
);
426 if (doi
.doi_bonus_type
!= DMU_OT_DSL_DATASET
) {
427 dmu_buf_rele(dbuf
, tag
);
428 return (SET_ERROR(EINVAL
));
431 ds
= dmu_buf_get_user(dbuf
);
433 dsl_dataset_t
*winner
= NULL
;
435 ds
= kmem_zalloc(sizeof (dsl_dataset_t
), KM_SLEEP
);
437 ds
->ds_object
= dsobj
;
438 ds
->ds_is_snapshot
= dsl_dataset_phys(ds
)->ds_num_children
!= 0;
439 list_link_init(&ds
->ds_synced_link
);
441 mutex_init(&ds
->ds_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
442 mutex_init(&ds
->ds_opening_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
443 mutex_init(&ds
->ds_sendstream_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
444 refcount_create(&ds
->ds_longholds
);
446 bplist_create(&ds
->ds_pending_deadlist
);
447 dsl_deadlist_open(&ds
->ds_deadlist
,
448 mos
, dsl_dataset_phys(ds
)->ds_deadlist_obj
);
450 list_create(&ds
->ds_sendstreams
, sizeof (dmu_sendarg_t
),
451 offsetof(dmu_sendarg_t
, dsa_link
));
453 list_create(&ds
->ds_prop_cbs
, sizeof (dsl_prop_cb_record_t
),
454 offsetof(dsl_prop_cb_record_t
, cbr_ds_node
));
456 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
459 for (f
= 0; f
< SPA_FEATURES
; f
++) {
460 if (!(spa_feature_table
[f
].fi_flags
&
461 ZFEATURE_FLAG_PER_DATASET
))
463 err
= zap_contains(mos
, dsobj
,
464 spa_feature_table
[f
].fi_guid
);
466 ds
->ds_feature_inuse
[f
] = B_TRUE
;
468 ASSERT3U(err
, ==, ENOENT
);
474 err
= dsl_dir_hold_obj(dp
,
475 dsl_dataset_phys(ds
)->ds_dir_obj
, NULL
, ds
, &ds
->ds_dir
);
477 mutex_destroy(&ds
->ds_lock
);
478 mutex_destroy(&ds
->ds_opening_lock
);
479 mutex_destroy(&ds
->ds_sendstream_lock
);
480 refcount_destroy(&ds
->ds_longholds
);
481 bplist_destroy(&ds
->ds_pending_deadlist
);
482 dsl_deadlist_close(&ds
->ds_deadlist
);
483 kmem_free(ds
, sizeof (dsl_dataset_t
));
484 dmu_buf_rele(dbuf
, tag
);
488 if (!ds
->ds_is_snapshot
) {
489 ds
->ds_snapname
[0] = '\0';
490 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
491 err
= dsl_dataset_hold_obj(dp
,
492 dsl_dataset_phys(ds
)->ds_prev_snap_obj
,
495 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
496 int zaperr
= zap_lookup(mos
, ds
->ds_object
,
497 DS_FIELD_BOOKMARK_NAMES
,
498 sizeof (ds
->ds_bookmarks
), 1,
500 if (zaperr
!= ENOENT
)
504 if (zfs_flags
& ZFS_DEBUG_SNAPNAMES
)
505 err
= dsl_dataset_get_snapname(ds
);
507 dsl_dataset_phys(ds
)->ds_userrefs_obj
!= 0) {
509 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
510 dsl_dataset_phys(ds
)->ds_userrefs_obj
,
515 if (err
== 0 && !ds
->ds_is_snapshot
) {
516 err
= dsl_prop_get_int_ds(ds
,
517 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
520 err
= dsl_prop_get_int_ds(ds
,
521 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
525 ds
->ds_reserved
= ds
->ds_quota
= 0;
528 dmu_buf_init_user(&ds
->ds_dbu
, dsl_dataset_evict
, &ds
->ds_dbuf
);
530 winner
= dmu_buf_set_user_ie(dbuf
, &ds
->ds_dbu
);
532 if (err
!= 0 || winner
!= NULL
) {
533 bplist_destroy(&ds
->ds_pending_deadlist
);
534 dsl_deadlist_close(&ds
->ds_deadlist
);
536 dsl_dataset_rele(ds
->ds_prev
, ds
);
537 dsl_dir_rele(ds
->ds_dir
, ds
);
538 mutex_destroy(&ds
->ds_lock
);
539 mutex_destroy(&ds
->ds_opening_lock
);
540 mutex_destroy(&ds
->ds_sendstream_lock
);
541 refcount_destroy(&ds
->ds_longholds
);
542 kmem_free(ds
, sizeof (dsl_dataset_t
));
544 dmu_buf_rele(dbuf
, tag
);
550 unique_insert(dsl_dataset_phys(ds
)->ds_fsid_guid
);
553 ASSERT3P(ds
->ds_dbuf
, ==, dbuf
);
554 ASSERT3P(dsl_dataset_phys(ds
), ==, dbuf
->db_data
);
555 ASSERT(dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0 ||
556 spa_version(dp
->dp_spa
) < SPA_VERSION_ORIGIN
||
557 dp
->dp_origin_snap
== NULL
|| ds
== dp
->dp_origin_snap
);
563 dsl_dataset_hold(dsl_pool_t
*dp
, const char *name
,
564 void *tag
, dsl_dataset_t
**dsp
)
567 const char *snapname
;
572 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, &snapname
);
576 ASSERT(dsl_pool_config_held(dp
));
577 obj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
579 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
581 err
= SET_ERROR(ENOENT
);
583 /* we may be looking for a snapshot */
584 if (err
== 0 && snapname
!= NULL
) {
585 dsl_dataset_t
*snap_ds
;
587 if (*snapname
++ != '@') {
588 dsl_dataset_rele(ds
, tag
);
589 dsl_dir_rele(dd
, FTAG
);
590 return (SET_ERROR(ENOENT
));
593 dprintf("looking for snapshot '%s'\n", snapname
);
594 err
= dsl_dataset_snap_lookup(ds
, snapname
, &obj
);
596 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &snap_ds
);
597 dsl_dataset_rele(ds
, tag
);
600 mutex_enter(&snap_ds
->ds_lock
);
601 if (snap_ds
->ds_snapname
[0] == 0)
602 (void) strlcpy(snap_ds
->ds_snapname
, snapname
,
603 sizeof (snap_ds
->ds_snapname
));
604 mutex_exit(&snap_ds
->ds_lock
);
610 dsl_dir_rele(dd
, FTAG
);
615 dsl_dataset_own_obj(dsl_pool_t
*dp
, uint64_t dsobj
,
616 void *tag
, dsl_dataset_t
**dsp
)
618 int err
= dsl_dataset_hold_obj(dp
, dsobj
, tag
, dsp
);
621 if (!dsl_dataset_tryown(*dsp
, tag
)) {
622 dsl_dataset_rele(*dsp
, tag
);
624 return (SET_ERROR(EBUSY
));
630 dsl_dataset_own(dsl_pool_t
*dp
, const char *name
,
631 void *tag
, dsl_dataset_t
**dsp
)
633 int err
= dsl_dataset_hold(dp
, name
, tag
, dsp
);
636 if (!dsl_dataset_tryown(*dsp
, tag
)) {
637 dsl_dataset_rele(*dsp
, tag
);
638 return (SET_ERROR(EBUSY
));
644 * See the comment above dsl_pool_hold() for details. In summary, a long
645 * hold is used to prevent destruction of a dataset while the pool hold
646 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
648 * The dataset and pool must be held when this function is called. After it
649 * is called, the pool hold may be released while the dataset is still held
653 dsl_dataset_long_hold(dsl_dataset_t
*ds
, void *tag
)
655 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
656 (void) refcount_add(&ds
->ds_longholds
, tag
);
660 dsl_dataset_long_rele(dsl_dataset_t
*ds
, void *tag
)
662 (void) refcount_remove(&ds
->ds_longholds
, tag
);
665 /* Return B_TRUE if there are any long holds on this dataset. */
667 dsl_dataset_long_held(dsl_dataset_t
*ds
)
669 return (!refcount_is_zero(&ds
->ds_longholds
));
673 dsl_dataset_name(dsl_dataset_t
*ds
, char *name
)
676 (void) strcpy(name
, "mos");
678 dsl_dir_name(ds
->ds_dir
, name
);
679 VERIFY0(dsl_dataset_get_snapname(ds
));
680 if (ds
->ds_snapname
[0]) {
681 VERIFY3U(strlcat(name
, "@", ZFS_MAX_DATASET_NAME_LEN
),
682 <, ZFS_MAX_DATASET_NAME_LEN
);
684 * We use a "recursive" mutex so that we
685 * can call dprintf_ds() with ds_lock held.
687 if (!MUTEX_HELD(&ds
->ds_lock
)) {
688 mutex_enter(&ds
->ds_lock
);
689 VERIFY3U(strlcat(name
, ds
->ds_snapname
,
690 ZFS_MAX_DATASET_NAME_LEN
), <,
691 ZFS_MAX_DATASET_NAME_LEN
);
692 mutex_exit(&ds
->ds_lock
);
694 VERIFY3U(strlcat(name
, ds
->ds_snapname
,
695 ZFS_MAX_DATASET_NAME_LEN
), <,
696 ZFS_MAX_DATASET_NAME_LEN
);
703 dsl_dataset_namelen(dsl_dataset_t
*ds
)
706 VERIFY0(dsl_dataset_get_snapname(ds
));
707 mutex_enter(&ds
->ds_lock
);
708 len
= dsl_dir_namelen(ds
->ds_dir
) + 1 + strlen(ds
->ds_snapname
);
709 mutex_exit(&ds
->ds_lock
);
714 dsl_dataset_rele(dsl_dataset_t
*ds
, void *tag
)
716 dmu_buf_rele(ds
->ds_dbuf
, tag
);
720 dsl_dataset_disown(dsl_dataset_t
*ds
, void *tag
)
722 ASSERT3P(ds
->ds_owner
, ==, tag
);
723 ASSERT(ds
->ds_dbuf
!= NULL
);
725 mutex_enter(&ds
->ds_lock
);
727 mutex_exit(&ds
->ds_lock
);
728 dsl_dataset_long_rele(ds
, tag
);
729 dsl_dataset_rele(ds
, tag
);
733 dsl_dataset_tryown(dsl_dataset_t
*ds
, void *tag
)
735 boolean_t gotit
= FALSE
;
737 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
738 mutex_enter(&ds
->ds_lock
);
739 if (ds
->ds_owner
== NULL
&& !DS_IS_INCONSISTENT(ds
)) {
741 dsl_dataset_long_hold(ds
, tag
);
744 mutex_exit(&ds
->ds_lock
);
749 dsl_dataset_has_owner(dsl_dataset_t
*ds
)
752 mutex_enter(&ds
->ds_lock
);
753 rv
= (ds
->ds_owner
!= NULL
);
754 mutex_exit(&ds
->ds_lock
);
759 dsl_dataset_activate_feature(uint64_t dsobj
, spa_feature_t f
, dmu_tx_t
*tx
)
761 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
762 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
765 VERIFY(spa_feature_table
[f
].fi_flags
& ZFEATURE_FLAG_PER_DATASET
);
767 spa_feature_incr(spa
, f
, tx
);
768 dmu_object_zapify(mos
, dsobj
, DMU_OT_DSL_DATASET
, tx
);
770 VERIFY0(zap_add(mos
, dsobj
, spa_feature_table
[f
].fi_guid
,
771 sizeof (zero
), 1, &zero
, tx
));
775 dsl_dataset_deactivate_feature(uint64_t dsobj
, spa_feature_t f
, dmu_tx_t
*tx
)
777 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
778 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
780 VERIFY(spa_feature_table
[f
].fi_flags
& ZFEATURE_FLAG_PER_DATASET
);
782 VERIFY0(zap_remove(mos
, dsobj
, spa_feature_table
[f
].fi_guid
, tx
));
783 spa_feature_decr(spa
, f
, tx
);
787 dsl_dataset_create_sync_dd(dsl_dir_t
*dd
, dsl_dataset_t
*origin
,
788 uint64_t flags
, dmu_tx_t
*tx
)
790 dsl_pool_t
*dp
= dd
->dd_pool
;
792 dsl_dataset_phys_t
*dsphys
;
794 objset_t
*mos
= dp
->dp_meta_objset
;
797 origin
= dp
->dp_origin_snap
;
799 ASSERT(origin
== NULL
|| origin
->ds_dir
->dd_pool
== dp
);
800 ASSERT(origin
== NULL
|| dsl_dataset_phys(origin
)->ds_num_children
> 0);
801 ASSERT(dmu_tx_is_syncing(tx
));
802 ASSERT(dsl_dir_phys(dd
)->dd_head_dataset_obj
== 0);
804 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
805 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
806 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
807 dmu_buf_will_dirty(dbuf
, tx
);
808 dsphys
= dbuf
->db_data
;
809 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
810 dsphys
->ds_dir_obj
= dd
->dd_object
;
811 dsphys
->ds_flags
= flags
;
812 dsphys
->ds_fsid_guid
= unique_create();
813 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
814 sizeof (dsphys
->ds_guid
));
815 dsphys
->ds_snapnames_zapobj
=
816 zap_create_norm(mos
, U8_TEXTPREP_TOUPPER
, DMU_OT_DSL_DS_SNAP_MAP
,
818 dsphys
->ds_creation_time
= gethrestime_sec();
819 dsphys
->ds_creation_txg
= tx
->tx_txg
== TXG_INITIAL
? 1 : tx
->tx_txg
;
821 if (origin
== NULL
) {
822 dsphys
->ds_deadlist_obj
= dsl_deadlist_alloc(mos
, tx
);
825 dsl_dataset_t
*ohds
; /* head of the origin snapshot */
827 dsphys
->ds_prev_snap_obj
= origin
->ds_object
;
828 dsphys
->ds_prev_snap_txg
=
829 dsl_dataset_phys(origin
)->ds_creation_txg
;
830 dsphys
->ds_referenced_bytes
=
831 dsl_dataset_phys(origin
)->ds_referenced_bytes
;
832 dsphys
->ds_compressed_bytes
=
833 dsl_dataset_phys(origin
)->ds_compressed_bytes
;
834 dsphys
->ds_uncompressed_bytes
=
835 dsl_dataset_phys(origin
)->ds_uncompressed_bytes
;
836 dsphys
->ds_bp
= dsl_dataset_phys(origin
)->ds_bp
;
839 * Inherit flags that describe the dataset's contents
840 * (INCONSISTENT) or properties (Case Insensitive).
842 dsphys
->ds_flags
|= dsl_dataset_phys(origin
)->ds_flags
&
843 (DS_FLAG_INCONSISTENT
| DS_FLAG_CI_DATASET
);
845 for (f
= 0; f
< SPA_FEATURES
; f
++) {
846 if (origin
->ds_feature_inuse
[f
])
847 dsl_dataset_activate_feature(dsobj
, f
, tx
);
850 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
851 dsl_dataset_phys(origin
)->ds_num_children
++;
853 VERIFY0(dsl_dataset_hold_obj(dp
,
854 dsl_dir_phys(origin
->ds_dir
)->dd_head_dataset_obj
,
856 dsphys
->ds_deadlist_obj
= dsl_deadlist_clone(&ohds
->ds_deadlist
,
857 dsphys
->ds_prev_snap_txg
, dsphys
->ds_prev_snap_obj
, tx
);
858 dsl_dataset_rele(ohds
, FTAG
);
860 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_NEXT_CLONES
) {
861 if (dsl_dataset_phys(origin
)->ds_next_clones_obj
== 0) {
862 dsl_dataset_phys(origin
)->ds_next_clones_obj
=
864 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
866 VERIFY0(zap_add_int(mos
,
867 dsl_dataset_phys(origin
)->ds_next_clones_obj
,
871 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
872 dsl_dir_phys(dd
)->dd_origin_obj
= origin
->ds_object
;
873 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
874 if (dsl_dir_phys(origin
->ds_dir
)->dd_clones
== 0) {
875 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
876 dsl_dir_phys(origin
->ds_dir
)->dd_clones
=
878 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
880 VERIFY0(zap_add_int(mos
,
881 dsl_dir_phys(origin
->ds_dir
)->dd_clones
,
886 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
887 dsphys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
889 dmu_buf_rele(dbuf
, FTAG
);
891 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
892 dsl_dir_phys(dd
)->dd_head_dataset_obj
= dsobj
;
898 dsl_dataset_zero_zil(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
902 VERIFY0(dmu_objset_from_ds(ds
, &os
));
903 bzero(&os
->os_zil_header
, sizeof (os
->os_zil_header
));
904 dsl_dataset_dirty(ds
, tx
);
908 dsl_dataset_create_sync(dsl_dir_t
*pdd
, const char *lastname
,
909 dsl_dataset_t
*origin
, uint64_t flags
, cred_t
*cr
, dmu_tx_t
*tx
)
911 dsl_pool_t
*dp
= pdd
->dd_pool
;
912 uint64_t dsobj
, ddobj
;
915 ASSERT(dmu_tx_is_syncing(tx
));
916 ASSERT(lastname
[0] != '@');
918 ddobj
= dsl_dir_create_sync(dp
, pdd
, lastname
, tx
);
919 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, lastname
, FTAG
, &dd
));
921 dsobj
= dsl_dataset_create_sync_dd(dd
, origin
,
922 flags
& ~DS_CREATE_FLAG_NODIRTY
, tx
);
924 dsl_deleg_set_create_perms(dd
, tx
, cr
);
927 * Since we're creating a new node we know it's a leaf, so we can
928 * initialize the counts if the limit feature is active.
930 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
932 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
934 dsl_dir_zapify(dd
, tx
);
935 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
936 sizeof (cnt
), 1, &cnt
, tx
));
937 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
938 sizeof (cnt
), 1, &cnt
, tx
));
941 dsl_dir_rele(dd
, FTAG
);
944 * If we are creating a clone, make sure we zero out any stale
945 * data from the origin snapshots zil header.
947 if (origin
!= NULL
&& !(flags
& DS_CREATE_FLAG_NODIRTY
)) {
950 VERIFY0(dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
951 dsl_dataset_zero_zil(ds
, tx
);
952 dsl_dataset_rele(ds
, FTAG
);
959 * The unique space in the head dataset can be calculated by subtracting
960 * the space used in the most recent snapshot, that is still being used
961 * in this file system, from the space currently in use. To figure out
962 * the space in the most recent snapshot still in use, we need to take
963 * the total space used in the snapshot and subtract out the space that
964 * has been freed up since the snapshot was taken.
967 dsl_dataset_recalc_head_uniq(dsl_dataset_t
*ds
)
970 uint64_t dlused
, dlcomp
, dluncomp
;
972 ASSERT(!ds
->ds_is_snapshot
);
974 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0)
975 mrs_used
= dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
;
979 dsl_deadlist_space(&ds
->ds_deadlist
, &dlused
, &dlcomp
, &dluncomp
);
981 ASSERT3U(dlused
, <=, mrs_used
);
982 dsl_dataset_phys(ds
)->ds_unique_bytes
=
983 dsl_dataset_phys(ds
)->ds_referenced_bytes
- (mrs_used
- dlused
);
985 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) >=
986 SPA_VERSION_UNIQUE_ACCURATE
)
987 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
991 dsl_dataset_remove_from_next_clones(dsl_dataset_t
*ds
, uint64_t obj
,
994 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
996 ASSERTV(uint64_t count
);
998 ASSERT(dsl_dataset_phys(ds
)->ds_num_children
>= 2);
999 err
= zap_remove_int(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1002 * The err should not be ENOENT, but a bug in a previous version
1003 * of the code could cause upgrade_clones_cb() to not set
1004 * ds_next_snap_obj when it should, leading to a missing entry.
1005 * If we knew that the pool was created after
1006 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1007 * ENOENT. However, at least we can check that we don't have
1008 * too many entries in the next_clones_obj even after failing to
1013 ASSERT0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1015 ASSERT3U(count
, <=, dsl_dataset_phys(ds
)->ds_num_children
- 2);
1020 dsl_dataset_get_blkptr(dsl_dataset_t
*ds
)
1022 return (&dsl_dataset_phys(ds
)->ds_bp
);
1026 dsl_dataset_set_blkptr(dsl_dataset_t
*ds
, blkptr_t
*bp
, dmu_tx_t
*tx
)
1028 ASSERT(dmu_tx_is_syncing(tx
));
1029 /* If it's the meta-objset, set dp_meta_rootbp */
1031 tx
->tx_pool
->dp_meta_rootbp
= *bp
;
1033 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1034 dsl_dataset_phys(ds
)->ds_bp
= *bp
;
1039 dsl_dataset_get_spa(dsl_dataset_t
*ds
)
1041 return (ds
->ds_dir
->dd_pool
->dp_spa
);
1045 dsl_dataset_dirty(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1049 if (ds
== NULL
) /* this is the meta-objset */
1052 ASSERT(ds
->ds_objset
!= NULL
);
1054 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
!= 0)
1055 panic("dirtying snapshot!");
1057 dp
= ds
->ds_dir
->dd_pool
;
1059 if (txg_list_add(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
1060 /* up the hold count until we can be written out */
1061 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
1066 dsl_dataset_is_dirty(dsl_dataset_t
*ds
)
1070 for (t
= 0; t
< TXG_SIZE
; t
++) {
1071 if (txg_list_member(&ds
->ds_dir
->dd_pool
->dp_dirty_datasets
,
1079 dsl_dataset_snapshot_reserve_space(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1083 if (!dmu_tx_is_syncing(tx
))
1087 * If there's an fs-only reservation, any blocks that might become
1088 * owned by the snapshot dataset must be accommodated by space
1089 * outside of the reservation.
1091 ASSERT(ds
->ds_reserved
== 0 || DS_UNIQUE_IS_ACCURATE(ds
));
1092 asize
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
, ds
->ds_reserved
);
1093 if (asize
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
1094 return (SET_ERROR(ENOSPC
));
1097 * Propagate any reserved space for this snapshot to other
1098 * snapshot checks in this sync group.
1101 dsl_dir_willuse_space(ds
->ds_dir
, asize
, tx
);
1106 typedef struct dsl_dataset_snapshot_arg
{
1107 nvlist_t
*ddsa_snaps
;
1108 nvlist_t
*ddsa_props
;
1109 nvlist_t
*ddsa_errors
;
1111 } dsl_dataset_snapshot_arg_t
;
1114 dsl_dataset_snapshot_check_impl(dsl_dataset_t
*ds
, const char *snapname
,
1115 dmu_tx_t
*tx
, boolean_t recv
, uint64_t cnt
, cred_t
*cr
)
1120 ds
->ds_trysnap_txg
= tx
->tx_txg
;
1122 if (!dmu_tx_is_syncing(tx
))
1126 * We don't allow multiple snapshots of the same txg. If there
1127 * is already one, try again.
1129 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
>= tx
->tx_txg
)
1130 return (SET_ERROR(EAGAIN
));
1133 * Check for conflicting snapshot name.
1135 error
= dsl_dataset_snap_lookup(ds
, snapname
, &value
);
1137 return (SET_ERROR(EEXIST
));
1138 if (error
!= ENOENT
)
1142 * We don't allow taking snapshots of inconsistent datasets, such as
1143 * those into which we are currently receiving. However, if we are
1144 * creating this snapshot as part of a receive, this check will be
1145 * executed atomically with respect to the completion of the receive
1146 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1147 * case we ignore this, knowing it will be fixed up for us shortly in
1148 * dmu_recv_end_sync().
1150 if (!recv
&& DS_IS_INCONSISTENT(ds
))
1151 return (SET_ERROR(EBUSY
));
1154 * Skip the check for temporary snapshots or if we have already checked
1155 * the counts in dsl_dataset_snapshot_check. This means we really only
1156 * check the count here when we're receiving a stream.
1158 if (cnt
!= 0 && cr
!= NULL
) {
1159 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1160 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, cr
);
1165 error
= dsl_dataset_snapshot_reserve_space(ds
, tx
);
1173 dsl_dataset_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1175 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1176 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1181 * Pre-compute how many total new snapshots will be created for each
1182 * level in the tree and below. This is needed for validating the
1183 * snapshot limit when either taking a recursive snapshot or when
1184 * taking multiple snapshots.
1186 * The problem is that the counts are not actually adjusted when
1187 * we are checking, only when we finally sync. For a single snapshot,
1188 * this is easy, the count will increase by 1 at each node up the tree,
1189 * but its more complicated for the recursive/multiple snapshot case.
1191 * The dsl_fs_ss_limit_check function does recursively check the count
1192 * at each level up the tree but since it is validating each snapshot
1193 * independently we need to be sure that we are validating the complete
1194 * count for the entire set of snapshots. We do this by rolling up the
1195 * counts for each component of the name into an nvlist and then
1196 * checking each of those cases with the aggregated count.
1198 * This approach properly handles not only the recursive snapshot
1199 * case (where we get all of those on the ddsa_snaps list) but also
1200 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1201 * validate the limit on 'a' using a count of 2).
1203 * We validate the snapshot names in the third loop and only report
1206 if (dmu_tx_is_syncing(tx
)) {
1208 nvlist_t
*cnt_track
= NULL
;
1209 cnt_track
= fnvlist_alloc();
1211 nm
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1213 /* Rollup aggregated counts into the cnt_track list */
1214 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1216 pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1220 (void) strlcpy(nm
, nvpair_name(pair
), MAXPATHLEN
);
1221 pdelim
= strchr(nm
, '@');
1227 if (nvlist_lookup_uint64(cnt_track
, nm
,
1229 /* update existing entry */
1230 fnvlist_add_uint64(cnt_track
, nm
,
1234 fnvlist_add_uint64(cnt_track
, nm
, 1);
1237 pdelim
= strrchr(nm
, '/');
1240 } while (pdelim
!= NULL
);
1243 kmem_free(nm
, MAXPATHLEN
);
1245 /* Check aggregated counts at each level */
1246 for (pair
= nvlist_next_nvpair(cnt_track
, NULL
);
1247 pair
!= NULL
; pair
= nvlist_next_nvpair(cnt_track
, pair
)) {
1253 name
= nvpair_name(pair
);
1254 cnt
= fnvpair_value_uint64(pair
);
1257 error
= dsl_dataset_hold(dp
, name
, FTAG
, &ds
);
1259 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1260 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
,
1262 dsl_dataset_rele(ds
, FTAG
);
1266 if (ddsa
->ddsa_errors
!= NULL
)
1267 fnvlist_add_int32(ddsa
->ddsa_errors
,
1270 /* only report one error for this check */
1274 nvlist_free(cnt_track
);
1277 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1278 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1281 char *name
, *atp
= NULL
;
1282 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
1284 name
= nvpair_name(pair
);
1285 if (strlen(name
) >= ZFS_MAX_DATASET_NAME_LEN
)
1286 error
= SET_ERROR(ENAMETOOLONG
);
1288 atp
= strchr(name
, '@');
1290 error
= SET_ERROR(EINVAL
);
1292 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1295 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
1297 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1298 error
= dsl_dataset_snapshot_check_impl(ds
,
1299 atp
+ 1, tx
, B_FALSE
, 0, NULL
);
1300 dsl_dataset_rele(ds
, FTAG
);
1304 if (ddsa
->ddsa_errors
!= NULL
) {
1305 fnvlist_add_int32(ddsa
->ddsa_errors
,
1316 dsl_dataset_snapshot_sync_impl(dsl_dataset_t
*ds
, const char *snapname
,
1319 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1321 dsl_dataset_phys_t
*dsphys
;
1322 uint64_t dsobj
, crtxg
;
1323 objset_t
*mos
= dp
->dp_meta_objset
;
1325 ASSERTV(static zil_header_t zero_zil
);
1326 ASSERTV(objset_t
*os
);
1328 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
1331 * If we are on an old pool, the zil must not be active, in which
1332 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1334 ASSERT(spa_version(dmu_tx_pool(tx
)->dp_spa
) >= SPA_VERSION_FAST_SNAP
||
1335 dmu_objset_from_ds(ds
, &os
) != 0 ||
1336 bcmp(&os
->os_phys
->os_zil_header
, &zero_zil
,
1337 sizeof (zero_zil
)) == 0);
1339 dsl_fs_ss_count_adjust(ds
->ds_dir
, 1, DD_FIELD_SNAPSHOT_COUNT
, tx
);
1342 * The origin's ds_creation_txg has to be < TXG_INITIAL
1344 if (strcmp(snapname
, ORIGIN_DIR_NAME
) == 0)
1349 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
1350 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
1351 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
1352 dmu_buf_will_dirty(dbuf
, tx
);
1353 dsphys
= dbuf
->db_data
;
1354 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
1355 dsphys
->ds_dir_obj
= ds
->ds_dir
->dd_object
;
1356 dsphys
->ds_fsid_guid
= unique_create();
1357 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
1358 sizeof (dsphys
->ds_guid
));
1359 dsphys
->ds_prev_snap_obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1360 dsphys
->ds_prev_snap_txg
= dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
1361 dsphys
->ds_next_snap_obj
= ds
->ds_object
;
1362 dsphys
->ds_num_children
= 1;
1363 dsphys
->ds_creation_time
= gethrestime_sec();
1364 dsphys
->ds_creation_txg
= crtxg
;
1365 dsphys
->ds_deadlist_obj
= dsl_dataset_phys(ds
)->ds_deadlist_obj
;
1366 dsphys
->ds_referenced_bytes
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1367 dsphys
->ds_compressed_bytes
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1368 dsphys
->ds_uncompressed_bytes
=
1369 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1370 dsphys
->ds_flags
= dsl_dataset_phys(ds
)->ds_flags
;
1371 dsphys
->ds_bp
= dsl_dataset_phys(ds
)->ds_bp
;
1372 dmu_buf_rele(dbuf
, FTAG
);
1374 for (f
= 0; f
< SPA_FEATURES
; f
++) {
1375 if (ds
->ds_feature_inuse
[f
])
1376 dsl_dataset_activate_feature(dsobj
, f
, tx
);
1379 ASSERT3U(ds
->ds_prev
!= 0, ==,
1380 dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0);
1382 uint64_t next_clones_obj
=
1383 dsl_dataset_phys(ds
->ds_prev
)->ds_next_clones_obj
;
1384 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1386 dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 1);
1387 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1389 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1390 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, ==,
1391 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_txg
);
1392 dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
= dsobj
;
1393 } else if (next_clones_obj
!= 0) {
1394 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
1395 dsphys
->ds_next_snap_obj
, tx
);
1396 VERIFY0(zap_add_int(mos
,
1397 next_clones_obj
, dsobj
, tx
));
1402 * If we have a reference-reservation on this dataset, we will
1403 * need to increase the amount of refreservation being charged
1404 * since our unique space is going to zero.
1406 if (ds
->ds_reserved
) {
1408 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
1409 delta
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
,
1411 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
,
1415 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1416 dsl_dataset_phys(ds
)->ds_deadlist_obj
=
1417 dsl_deadlist_clone(&ds
->ds_deadlist
, UINT64_MAX
,
1418 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, tx
);
1419 dsl_deadlist_close(&ds
->ds_deadlist
);
1420 dsl_deadlist_open(&ds
->ds_deadlist
, mos
,
1421 dsl_dataset_phys(ds
)->ds_deadlist_obj
);
1422 dsl_deadlist_add_key(&ds
->ds_deadlist
,
1423 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
);
1425 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, <, tx
->tx_txg
);
1426 dsl_dataset_phys(ds
)->ds_prev_snap_obj
= dsobj
;
1427 dsl_dataset_phys(ds
)->ds_prev_snap_txg
= crtxg
;
1428 dsl_dataset_phys(ds
)->ds_unique_bytes
= 0;
1429 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
1430 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1432 VERIFY0(zap_add(mos
, dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1433 snapname
, 8, 1, &dsobj
, tx
));
1436 dsl_dataset_rele(ds
->ds_prev
, ds
);
1437 VERIFY0(dsl_dataset_hold_obj(dp
,
1438 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, ds
, &ds
->ds_prev
));
1440 dsl_scan_ds_snapshotted(ds
, tx
);
1442 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
1444 spa_history_log_internal_ds(ds
->ds_prev
, "snapshot", tx
, "");
1448 dsl_dataset_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1450 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1451 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1454 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1455 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1458 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
1460 name
= nvpair_name(pair
);
1461 atp
= strchr(name
, '@');
1462 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1463 VERIFY0(dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
));
1465 dsl_dataset_snapshot_sync_impl(ds
, atp
+ 1, tx
);
1466 if (ddsa
->ddsa_props
!= NULL
) {
1467 dsl_props_set_sync_impl(ds
->ds_prev
,
1468 ZPROP_SRC_LOCAL
, ddsa
->ddsa_props
, tx
);
1470 zvol_create_minors(dp
->dp_spa
, nvpair_name(pair
), B_TRUE
);
1471 dsl_dataset_rele(ds
, FTAG
);
1476 * The snapshots must all be in the same pool.
1477 * All-or-nothing: if there are any failures, nothing will be modified.
1480 dsl_dataset_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
*errors
)
1482 dsl_dataset_snapshot_arg_t ddsa
;
1484 boolean_t needsuspend
;
1488 nvlist_t
*suspended
= NULL
;
1490 pair
= nvlist_next_nvpair(snaps
, NULL
);
1493 firstname
= nvpair_name(pair
);
1495 error
= spa_open(firstname
, &spa
, FTAG
);
1498 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1499 spa_close(spa
, FTAG
);
1502 suspended
= fnvlist_alloc();
1503 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1504 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1505 char fsname
[ZFS_MAX_DATASET_NAME_LEN
];
1506 char *snapname
= nvpair_name(pair
);
1510 atp
= strchr(snapname
, '@');
1512 error
= SET_ERROR(EINVAL
);
1515 (void) strlcpy(fsname
, snapname
, atp
- snapname
+ 1);
1517 error
= zil_suspend(fsname
, &cookie
);
1520 fnvlist_add_uint64(suspended
, fsname
,
1525 ddsa
.ddsa_snaps
= snaps
;
1526 ddsa
.ddsa_props
= props
;
1527 ddsa
.ddsa_errors
= errors
;
1528 ddsa
.ddsa_cr
= CRED();
1531 error
= dsl_sync_task(firstname
, dsl_dataset_snapshot_check
,
1532 dsl_dataset_snapshot_sync
, &ddsa
,
1533 fnvlist_num_pairs(snaps
) * 3, ZFS_SPACE_CHECK_NORMAL
);
1536 if (suspended
!= NULL
) {
1537 for (pair
= nvlist_next_nvpair(suspended
, NULL
); pair
!= NULL
;
1538 pair
= nvlist_next_nvpair(suspended
, pair
)) {
1539 zil_resume((void *)(uintptr_t)
1540 fnvpair_value_uint64(pair
));
1542 fnvlist_free(suspended
);
1548 typedef struct dsl_dataset_snapshot_tmp_arg
{
1549 const char *ddsta_fsname
;
1550 const char *ddsta_snapname
;
1551 minor_t ddsta_cleanup_minor
;
1552 const char *ddsta_htag
;
1553 } dsl_dataset_snapshot_tmp_arg_t
;
1556 dsl_dataset_snapshot_tmp_check(void *arg
, dmu_tx_t
*tx
)
1558 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1559 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1563 error
= dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
);
1567 /* NULL cred means no limit check for tmp snapshot */
1568 error
= dsl_dataset_snapshot_check_impl(ds
, ddsta
->ddsta_snapname
,
1569 tx
, B_FALSE
, 0, NULL
);
1571 dsl_dataset_rele(ds
, FTAG
);
1575 if (spa_version(dp
->dp_spa
) < SPA_VERSION_USERREFS
) {
1576 dsl_dataset_rele(ds
, FTAG
);
1577 return (SET_ERROR(ENOTSUP
));
1579 error
= dsl_dataset_user_hold_check_one(NULL
, ddsta
->ddsta_htag
,
1582 dsl_dataset_rele(ds
, FTAG
);
1586 dsl_dataset_rele(ds
, FTAG
);
1591 dsl_dataset_snapshot_tmp_sync(void *arg
, dmu_tx_t
*tx
)
1593 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1594 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1595 dsl_dataset_t
*ds
= NULL
;
1597 VERIFY0(dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
));
1599 dsl_dataset_snapshot_sync_impl(ds
, ddsta
->ddsta_snapname
, tx
);
1600 dsl_dataset_user_hold_sync_one(ds
->ds_prev
, ddsta
->ddsta_htag
,
1601 ddsta
->ddsta_cleanup_minor
, gethrestime_sec(), tx
);
1602 dsl_destroy_snapshot_sync_impl(ds
->ds_prev
, B_TRUE
, tx
);
1604 dsl_dataset_rele(ds
, FTAG
);
1608 dsl_dataset_snapshot_tmp(const char *fsname
, const char *snapname
,
1609 minor_t cleanup_minor
, const char *htag
)
1611 dsl_dataset_snapshot_tmp_arg_t ddsta
;
1614 boolean_t needsuspend
;
1617 ddsta
.ddsta_fsname
= fsname
;
1618 ddsta
.ddsta_snapname
= snapname
;
1619 ddsta
.ddsta_cleanup_minor
= cleanup_minor
;
1620 ddsta
.ddsta_htag
= htag
;
1622 error
= spa_open(fsname
, &spa
, FTAG
);
1625 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1626 spa_close(spa
, FTAG
);
1629 error
= zil_suspend(fsname
, &cookie
);
1634 error
= dsl_sync_task(fsname
, dsl_dataset_snapshot_tmp_check
,
1635 dsl_dataset_snapshot_tmp_sync
, &ddsta
, 3, ZFS_SPACE_CHECK_RESERVED
);
1644 dsl_dataset_sync(dsl_dataset_t
*ds
, zio_t
*zio
, dmu_tx_t
*tx
)
1648 ASSERT(dmu_tx_is_syncing(tx
));
1649 ASSERT(ds
->ds_objset
!= NULL
);
1650 ASSERT(dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0);
1653 * in case we had to change ds_fsid_guid when we opened it,
1656 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1657 dsl_dataset_phys(ds
)->ds_fsid_guid
= ds
->ds_fsid_guid
;
1659 if (ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
] != 0) {
1660 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1661 ds
->ds_object
, DS_FIELD_RESUME_OBJECT
, 8, 1,
1662 &ds
->ds_resume_object
[tx
->tx_txg
& TXG_MASK
], tx
));
1663 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1664 ds
->ds_object
, DS_FIELD_RESUME_OFFSET
, 8, 1,
1665 &ds
->ds_resume_offset
[tx
->tx_txg
& TXG_MASK
], tx
));
1666 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1667 ds
->ds_object
, DS_FIELD_RESUME_BYTES
, 8, 1,
1668 &ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
], tx
));
1669 ds
->ds_resume_object
[tx
->tx_txg
& TXG_MASK
] = 0;
1670 ds
->ds_resume_offset
[tx
->tx_txg
& TXG_MASK
] = 0;
1671 ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
] = 0;
1674 dmu_objset_sync(ds
->ds_objset
, zio
, tx
);
1676 for (f
= 0; f
< SPA_FEATURES
; f
++) {
1677 if (ds
->ds_feature_activation_needed
[f
]) {
1678 if (ds
->ds_feature_inuse
[f
])
1680 dsl_dataset_activate_feature(ds
->ds_object
, f
, tx
);
1681 ds
->ds_feature_inuse
[f
] = B_TRUE
;
1687 get_clones_stat(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1690 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1693 nvlist_t
*propval
= fnvlist_alloc();
1694 nvlist_t
*val
= fnvlist_alloc();
1696 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1699 * There may be missing entries in ds_next_clones_obj
1700 * due to a bug in a previous version of the code.
1701 * Only trust it if it has the right number of entries.
1703 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
!= 0) {
1704 VERIFY0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1707 if (count
!= dsl_dataset_phys(ds
)->ds_num_children
- 1)
1709 for (zap_cursor_init(&zc
, mos
,
1710 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
1711 zap_cursor_retrieve(&zc
, &za
) == 0;
1712 zap_cursor_advance(&zc
)) {
1713 dsl_dataset_t
*clone
;
1714 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1715 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
1716 za
.za_first_integer
, FTAG
, &clone
));
1717 dsl_dir_name(clone
->ds_dir
, buf
);
1718 fnvlist_add_boolean(val
, buf
);
1719 dsl_dataset_rele(clone
, FTAG
);
1721 zap_cursor_fini(&zc
);
1722 fnvlist_add_nvlist(propval
, ZPROP_VALUE
, val
);
1723 fnvlist_add_nvlist(nv
, zfs_prop_to_name(ZFS_PROP_CLONES
), propval
);
1726 nvlist_free(propval
);
1730 get_receive_resume_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1732 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1734 if (dsl_dataset_has_resume_receive_state(ds
)) {
1737 uint8_t *compressed
;
1739 nvlist_t
*token_nv
= fnvlist_alloc();
1740 size_t packed_size
, compressed_size
;
1743 char buf
[MAXNAMELEN
];
1746 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1747 DS_FIELD_RESUME_FROMGUID
, sizeof (val
), 1, &val
) == 0) {
1748 fnvlist_add_uint64(token_nv
, "fromguid", val
);
1750 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1751 DS_FIELD_RESUME_OBJECT
, sizeof (val
), 1, &val
) == 0) {
1752 fnvlist_add_uint64(token_nv
, "object", val
);
1754 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1755 DS_FIELD_RESUME_OFFSET
, sizeof (val
), 1, &val
) == 0) {
1756 fnvlist_add_uint64(token_nv
, "offset", val
);
1758 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1759 DS_FIELD_RESUME_BYTES
, sizeof (val
), 1, &val
) == 0) {
1760 fnvlist_add_uint64(token_nv
, "bytes", val
);
1762 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1763 DS_FIELD_RESUME_TOGUID
, sizeof (val
), 1, &val
) == 0) {
1764 fnvlist_add_uint64(token_nv
, "toguid", val
);
1766 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1767 DS_FIELD_RESUME_TONAME
, 1, sizeof (buf
), buf
) == 0) {
1768 fnvlist_add_string(token_nv
, "toname", buf
);
1770 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1771 DS_FIELD_RESUME_LARGEBLOCK
) == 0) {
1772 fnvlist_add_boolean(token_nv
, "largeblockok");
1774 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1775 DS_FIELD_RESUME_EMBEDOK
) == 0) {
1776 fnvlist_add_boolean(token_nv
, "embedok");
1778 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1779 DS_FIELD_RESUME_COMPRESSOK
) == 0) {
1780 fnvlist_add_boolean(token_nv
, "compressok");
1782 packed
= fnvlist_pack(token_nv
, &packed_size
);
1783 fnvlist_free(token_nv
);
1784 compressed
= kmem_alloc(packed_size
, KM_SLEEP
);
1786 compressed_size
= gzip_compress(packed
, compressed
,
1787 packed_size
, packed_size
, 6);
1789 fletcher_4_native_varsize(compressed
, compressed_size
, &cksum
);
1791 str
= kmem_alloc(compressed_size
* 2 + 1, KM_SLEEP
);
1792 for (i
= 0; i
< compressed_size
; i
++) {
1793 (void) sprintf(str
+ i
* 2, "%02x", compressed
[i
]);
1795 str
[compressed_size
* 2] = '\0';
1796 propval
= kmem_asprintf("%u-%llx-%llx-%s",
1797 ZFS_SEND_RESUME_TOKEN_VERSION
,
1798 (longlong_t
)cksum
.zc_word
[0],
1799 (longlong_t
)packed_size
, str
);
1800 dsl_prop_nvlist_add_string(nv
,
1801 ZFS_PROP_RECEIVE_RESUME_TOKEN
, propval
);
1802 kmem_free(packed
, packed_size
);
1803 kmem_free(str
, compressed_size
* 2 + 1);
1804 kmem_free(compressed
, packed_size
);
1810 dsl_dataset_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1812 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1813 uint64_t refd
, avail
, uobjs
, aobjs
, ratio
;
1815 ASSERT(dsl_pool_config_held(dp
));
1817 ratio
= dsl_dataset_phys(ds
)->ds_compressed_bytes
== 0 ? 100 :
1818 (dsl_dataset_phys(ds
)->ds_uncompressed_bytes
* 100 /
1819 dsl_dataset_phys(ds
)->ds_compressed_bytes
);
1821 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRATIO
, ratio
);
1822 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALREFERENCED
,
1823 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
);
1825 if (ds
->ds_is_snapshot
) {
1826 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
, ratio
);
1827 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
1828 dsl_dataset_phys(ds
)->ds_unique_bytes
);
1829 get_clones_stat(ds
, nv
);
1831 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
!= dp
->dp_origin_snap
) {
1832 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1833 dsl_dataset_name(ds
->ds_prev
, buf
);
1834 dsl_prop_nvlist_add_string(nv
, ZFS_PROP_PREV_SNAP
, buf
);
1837 dsl_dir_stats(ds
->ds_dir
, nv
);
1840 dsl_dataset_space(ds
, &refd
, &avail
, &uobjs
, &aobjs
);
1841 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_AVAILABLE
, avail
);
1842 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFERENCED
, refd
);
1844 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATION
,
1845 dsl_dataset_phys(ds
)->ds_creation_time
);
1846 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATETXG
,
1847 dsl_dataset_phys(ds
)->ds_creation_txg
);
1848 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFQUOTA
,
1850 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRESERVATION
,
1852 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_GUID
,
1853 dsl_dataset_phys(ds
)->ds_guid
);
1854 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_UNIQUE
,
1855 dsl_dataset_phys(ds
)->ds_unique_bytes
);
1856 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_OBJSETID
,
1858 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERREFS
,
1860 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_DEFER_DESTROY
,
1861 DS_IS_DEFER_DESTROY(ds
) ? 1 : 0);
1863 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
1864 uint64_t written
, comp
, uncomp
;
1865 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1866 dsl_dataset_t
*prev
;
1869 err
= dsl_dataset_hold_obj(dp
,
1870 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
1872 err
= dsl_dataset_space_written(prev
, ds
, &written
,
1874 dsl_dataset_rele(prev
, FTAG
);
1876 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_WRITTEN
,
1882 if (!dsl_dataset_is_snapshot(ds
)) {
1883 /* 6 extra bytes for /%recv */
1884 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1885 dsl_dataset_t
*recv_ds
;
1888 * A failed "newfs" (e.g. full) resumable receive leaves
1889 * the stats set on this dataset. Check here for the prop.
1891 get_receive_resume_stats(ds
, nv
);
1894 * A failed incremental resumable receive leaves the
1895 * stats set on our child named "%recv". Check the child
1898 dsl_dataset_name(ds
, recvname
);
1899 if (strlcat(recvname
, "/", sizeof (recvname
)) <
1900 sizeof (recvname
) &&
1901 strlcat(recvname
, recv_clone_name
, sizeof (recvname
)) <
1902 sizeof (recvname
) &&
1903 dsl_dataset_hold(dp
, recvname
, FTAG
, &recv_ds
) == 0) {
1904 get_receive_resume_stats(recv_ds
, nv
);
1905 dsl_dataset_rele(recv_ds
, FTAG
);
1911 dsl_dataset_fast_stat(dsl_dataset_t
*ds
, dmu_objset_stats_t
*stat
)
1913 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1914 ASSERT(dsl_pool_config_held(dp
));
1916 stat
->dds_creation_txg
= dsl_dataset_phys(ds
)->ds_creation_txg
;
1917 stat
->dds_inconsistent
=
1918 dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_INCONSISTENT
;
1919 stat
->dds_guid
= dsl_dataset_phys(ds
)->ds_guid
;
1920 stat
->dds_origin
[0] = '\0';
1921 if (ds
->ds_is_snapshot
) {
1922 stat
->dds_is_snapshot
= B_TRUE
;
1923 stat
->dds_num_clones
=
1924 dsl_dataset_phys(ds
)->ds_num_children
- 1;
1926 stat
->dds_is_snapshot
= B_FALSE
;
1927 stat
->dds_num_clones
= 0;
1929 if (dsl_dir_is_clone(ds
->ds_dir
)) {
1932 VERIFY0(dsl_dataset_hold_obj(dp
,
1933 dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
,
1935 dsl_dataset_name(ods
, stat
->dds_origin
);
1936 dsl_dataset_rele(ods
, FTAG
);
1942 dsl_dataset_fsid_guid(dsl_dataset_t
*ds
)
1944 return (ds
->ds_fsid_guid
);
1948 dsl_dataset_space(dsl_dataset_t
*ds
,
1949 uint64_t *refdbytesp
, uint64_t *availbytesp
,
1950 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1952 *refdbytesp
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1953 *availbytesp
= dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
);
1954 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
)
1956 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
;
1957 if (ds
->ds_quota
!= 0) {
1959 * Adjust available bytes according to refquota
1961 if (*refdbytesp
< ds
->ds_quota
)
1962 *availbytesp
= MIN(*availbytesp
,
1963 ds
->ds_quota
- *refdbytesp
);
1967 *usedobjsp
= BP_GET_FILL(&dsl_dataset_phys(ds
)->ds_bp
);
1968 *availobjsp
= DN_MAX_OBJECT
- *usedobjsp
;
1972 dsl_dataset_modified_since_snap(dsl_dataset_t
*ds
, dsl_dataset_t
*snap
)
1974 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1977 if (dsl_dataset_phys(ds
)->ds_bp
.blk_birth
>
1978 dsl_dataset_phys(snap
)->ds_creation_txg
) {
1979 objset_t
*os
, *os_snap
;
1981 * It may be that only the ZIL differs, because it was
1982 * reset in the head. Don't count that as being
1985 if (dmu_objset_from_ds(ds
, &os
) != 0)
1987 if (dmu_objset_from_ds(snap
, &os_snap
) != 0)
1989 return (bcmp(&os
->os_phys
->os_meta_dnode
,
1990 &os_snap
->os_phys
->os_meta_dnode
,
1991 sizeof (os
->os_phys
->os_meta_dnode
)) != 0);
1996 typedef struct dsl_dataset_rename_snapshot_arg
{
1997 const char *ddrsa_fsname
;
1998 const char *ddrsa_oldsnapname
;
1999 const char *ddrsa_newsnapname
;
2000 boolean_t ddrsa_recursive
;
2002 } dsl_dataset_rename_snapshot_arg_t
;
2006 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t
*dp
,
2007 dsl_dataset_t
*hds
, void *arg
)
2009 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2013 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
2015 /* ignore nonexistent snapshots */
2016 return (error
== ENOENT
? 0 : error
);
2019 /* new name should not exist */
2020 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_newsnapname
, &val
);
2022 error
= SET_ERROR(EEXIST
);
2023 else if (error
== ENOENT
)
2026 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2027 if (dsl_dir_namelen(hds
->ds_dir
) + 1 +
2028 strlen(ddrsa
->ddrsa_newsnapname
) >= ZFS_MAX_DATASET_NAME_LEN
)
2029 error
= SET_ERROR(ENAMETOOLONG
);
2035 dsl_dataset_rename_snapshot_check(void *arg
, dmu_tx_t
*tx
)
2037 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2038 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2042 error
= dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
);
2046 if (ddrsa
->ddrsa_recursive
) {
2047 error
= dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
2048 dsl_dataset_rename_snapshot_check_impl
, ddrsa
,
2051 error
= dsl_dataset_rename_snapshot_check_impl(dp
, hds
, ddrsa
);
2053 dsl_dataset_rele(hds
, FTAG
);
2058 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t
*dp
,
2059 dsl_dataset_t
*hds
, void *arg
)
2061 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2064 dmu_tx_t
*tx
= ddrsa
->ddrsa_tx
;
2067 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
2068 ASSERT(error
== 0 || error
== ENOENT
);
2069 if (error
== ENOENT
) {
2070 /* ignore nonexistent snapshots */
2074 VERIFY0(dsl_dataset_hold_obj(dp
, val
, FTAG
, &ds
));
2076 /* log before we change the name */
2077 spa_history_log_internal_ds(ds
, "rename", tx
,
2078 "-> @%s", ddrsa
->ddrsa_newsnapname
);
2080 VERIFY0(dsl_dataset_snap_remove(hds
, ddrsa
->ddrsa_oldsnapname
, tx
,
2082 mutex_enter(&ds
->ds_lock
);
2083 (void) strlcpy(ds
->ds_snapname
, ddrsa
->ddrsa_newsnapname
,
2084 sizeof (ds
->ds_snapname
));
2085 mutex_exit(&ds
->ds_lock
);
2086 VERIFY0(zap_add(dp
->dp_meta_objset
,
2087 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
,
2088 ds
->ds_snapname
, 8, 1, &ds
->ds_object
, tx
));
2089 zvol_rename_minors(dp
->dp_spa
, ddrsa
->ddrsa_oldsnapname
,
2090 ddrsa
->ddrsa_newsnapname
, B_TRUE
);
2092 dsl_dataset_rele(ds
, FTAG
);
2097 dsl_dataset_rename_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
2099 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2100 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2101 dsl_dataset_t
*hds
= NULL
;
2103 VERIFY0(dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
));
2104 ddrsa
->ddrsa_tx
= tx
;
2105 if (ddrsa
->ddrsa_recursive
) {
2106 VERIFY0(dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
2107 dsl_dataset_rename_snapshot_sync_impl
, ddrsa
,
2110 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp
, hds
, ddrsa
));
2112 dsl_dataset_rele(hds
, FTAG
);
2116 dsl_dataset_rename_snapshot(const char *fsname
,
2117 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
)
2119 dsl_dataset_rename_snapshot_arg_t ddrsa
;
2121 ddrsa
.ddrsa_fsname
= fsname
;
2122 ddrsa
.ddrsa_oldsnapname
= oldsnapname
;
2123 ddrsa
.ddrsa_newsnapname
= newsnapname
;
2124 ddrsa
.ddrsa_recursive
= recursive
;
2126 return (dsl_sync_task(fsname
, dsl_dataset_rename_snapshot_check
,
2127 dsl_dataset_rename_snapshot_sync
, &ddrsa
,
2128 1, ZFS_SPACE_CHECK_RESERVED
));
2132 * If we're doing an ownership handoff, we need to make sure that there is
2133 * only one long hold on the dataset. We're not allowed to change anything here
2134 * so we don't permanently release the long hold or regular hold here. We want
2135 * to do this only when syncing to avoid the dataset unexpectedly going away
2136 * when we release the long hold. Allow a long hold to exist for volumes, this
2137 * may occur when asynchronously registering the minor with the kernel.
2140 dsl_dataset_handoff_check(dsl_dataset_t
*ds
, void *owner
, dmu_tx_t
*tx
)
2144 if (!dmu_tx_is_syncing(tx
))
2147 if (owner
!= NULL
) {
2148 VERIFY3P(ds
->ds_owner
, ==, owner
);
2149 dsl_dataset_long_rele(ds
, owner
);
2152 held
= (dsl_dataset_long_held(ds
) && (ds
->ds_owner
!= zvol_tag
));
2155 dsl_dataset_long_hold(ds
, owner
);
2158 return (SET_ERROR(EBUSY
));
2163 typedef struct dsl_dataset_rollback_arg
{
2164 const char *ddra_fsname
;
2166 nvlist_t
*ddra_result
;
2167 } dsl_dataset_rollback_arg_t
;
2170 dsl_dataset_rollback_check(void *arg
, dmu_tx_t
*tx
)
2172 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2173 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2175 int64_t unused_refres_delta
;
2178 nvlist_t
*proprequest
, *bookmarks
;
2180 error
= dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
);
2184 /* must not be a snapshot */
2185 if (ds
->ds_is_snapshot
) {
2186 dsl_dataset_rele(ds
, FTAG
);
2187 return (SET_ERROR(EINVAL
));
2190 /* must have a most recent snapshot */
2191 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
< TXG_INITIAL
) {
2192 dsl_dataset_rele(ds
, FTAG
);
2193 return (SET_ERROR(EINVAL
));
2196 /* must not have any bookmarks after the most recent snapshot */
2197 proprequest
= fnvlist_alloc();
2198 fnvlist_add_boolean(proprequest
, zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2199 bookmarks
= fnvlist_alloc();
2200 error
= dsl_get_bookmarks_impl(ds
, proprequest
, bookmarks
);
2201 fnvlist_free(proprequest
);
2204 for (pair
= nvlist_next_nvpair(bookmarks
, NULL
);
2205 pair
!= NULL
; pair
= nvlist_next_nvpair(bookmarks
, pair
)) {
2207 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair
),
2208 zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2209 uint64_t createtxg
= fnvlist_lookup_uint64(valuenv
, "value");
2210 if (createtxg
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
2211 fnvlist_free(bookmarks
);
2212 dsl_dataset_rele(ds
, FTAG
);
2213 return (SET_ERROR(EEXIST
));
2216 fnvlist_free(bookmarks
);
2218 error
= dsl_dataset_handoff_check(ds
, ddra
->ddra_owner
, tx
);
2220 dsl_dataset_rele(ds
, FTAG
);
2225 * Check if the snap we are rolling back to uses more than
2228 if (ds
->ds_quota
!= 0 &&
2229 dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
> ds
->ds_quota
) {
2230 dsl_dataset_rele(ds
, FTAG
);
2231 return (SET_ERROR(EDQUOT
));
2235 * When we do the clone swap, we will temporarily use more space
2236 * due to the refreservation (the head will no longer have any
2237 * unique space, so the entire amount of the refreservation will need
2238 * to be free). We will immediately destroy the clone, freeing
2239 * this space, but the freeing happens over many txg's.
2241 unused_refres_delta
= (int64_t)MIN(ds
->ds_reserved
,
2242 dsl_dataset_phys(ds
)->ds_unique_bytes
);
2244 if (unused_refres_delta
> 0 &&
2245 unused_refres_delta
>
2246 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
)) {
2247 dsl_dataset_rele(ds
, FTAG
);
2248 return (SET_ERROR(ENOSPC
));
2251 dsl_dataset_rele(ds
, FTAG
);
2256 dsl_dataset_rollback_sync(void *arg
, dmu_tx_t
*tx
)
2258 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2259 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2260 dsl_dataset_t
*ds
, *clone
;
2262 char namebuf
[ZFS_MAX_DATASET_NAME_LEN
];
2264 VERIFY0(dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
));
2266 dsl_dataset_name(ds
->ds_prev
, namebuf
);
2267 fnvlist_add_string(ddra
->ddra_result
, "target", namebuf
);
2269 cloneobj
= dsl_dataset_create_sync(ds
->ds_dir
, "%rollback",
2270 ds
->ds_prev
, DS_CREATE_FLAG_NODIRTY
, kcred
, tx
);
2272 VERIFY0(dsl_dataset_hold_obj(dp
, cloneobj
, FTAG
, &clone
));
2274 dsl_dataset_clone_swap_sync_impl(clone
, ds
, tx
);
2275 dsl_dataset_zero_zil(ds
, tx
);
2277 dsl_destroy_head_sync_impl(clone
, tx
);
2279 dsl_dataset_rele(clone
, FTAG
);
2280 dsl_dataset_rele(ds
, FTAG
);
2284 * Rolls back the given filesystem or volume to the most recent snapshot.
2285 * The name of the most recent snapshot will be returned under key "target"
2286 * in the result nvlist.
2289 * - The existing dataset MUST be owned by the specified owner at entry
2290 * - Upon return, dataset will still be held by the same owner, whether we
2293 * This mode is required any time the existing filesystem is mounted. See
2294 * notes above zfs_suspend_fs() for further details.
2297 dsl_dataset_rollback(const char *fsname
, void *owner
, nvlist_t
*result
)
2299 dsl_dataset_rollback_arg_t ddra
;
2301 ddra
.ddra_fsname
= fsname
;
2302 ddra
.ddra_owner
= owner
;
2303 ddra
.ddra_result
= result
;
2305 return (dsl_sync_task(fsname
, dsl_dataset_rollback_check
,
2306 dsl_dataset_rollback_sync
, &ddra
,
2307 1, ZFS_SPACE_CHECK_RESERVED
));
2310 struct promotenode
{
2315 typedef struct dsl_dataset_promote_arg
{
2316 const char *ddpa_clonename
;
2317 dsl_dataset_t
*ddpa_clone
;
2318 list_t shared_snaps
, origin_snaps
, clone_snaps
;
2319 dsl_dataset_t
*origin_origin
; /* origin of the origin */
2320 uint64_t used
, comp
, uncomp
, unique
, cloneusedsnap
, originusedsnap
;
2323 } dsl_dataset_promote_arg_t
;
2325 static int snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
);
2326 static int promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
,
2328 static void promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
);
2331 dsl_dataset_promote_check(void *arg
, dmu_tx_t
*tx
)
2333 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2334 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2336 struct promotenode
*snap
;
2337 dsl_dataset_t
*origin_ds
;
2341 size_t max_snap_len
;
2343 err
= promote_hold(ddpa
, dp
, FTAG
);
2347 hds
= ddpa
->ddpa_clone
;
2348 max_snap_len
= MAXNAMELEN
- strlen(ddpa
->ddpa_clonename
) - 1;
2350 if (dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
) {
2351 promote_rele(ddpa
, FTAG
);
2352 return (SET_ERROR(EXDEV
));
2356 * Compute and check the amount of space to transfer. Since this is
2357 * so expensive, don't do the preliminary check.
2359 if (!dmu_tx_is_syncing(tx
)) {
2360 promote_rele(ddpa
, FTAG
);
2364 snap
= list_head(&ddpa
->shared_snaps
);
2366 err
= SET_ERROR(ENOENT
);
2369 origin_ds
= snap
->ds
;
2371 /* compute origin's new unique space */
2372 snap
= list_tail(&ddpa
->clone_snaps
);
2373 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2374 origin_ds
->ds_object
);
2375 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2376 dsl_dataset_phys(origin_ds
)->ds_prev_snap_txg
, UINT64_MAX
,
2377 &ddpa
->unique
, &unused
, &unused
);
2380 * Walk the snapshots that we are moving
2382 * Compute space to transfer. Consider the incremental changes
2383 * to used by each snapshot:
2384 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2385 * So each snapshot gave birth to:
2386 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2387 * So a sequence would look like:
2388 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2389 * Which simplifies to:
2390 * uN + kN + kN-1 + ... + k1 + k0
2391 * Note however, if we stop before we reach the ORIGIN we get:
2392 * uN + kN + kN-1 + ... + kM - uM-1
2395 ddpa
->used
= dsl_dataset_phys(origin_ds
)->ds_referenced_bytes
;
2396 ddpa
->comp
= dsl_dataset_phys(origin_ds
)->ds_compressed_bytes
;
2397 ddpa
->uncomp
= dsl_dataset_phys(origin_ds
)->ds_uncompressed_bytes
;
2398 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2399 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2400 uint64_t val
, dlused
, dlcomp
, dluncomp
;
2401 dsl_dataset_t
*ds
= snap
->ds
;
2406 * If there are long holds, we won't be able to evict
2409 if (dsl_dataset_long_held(ds
)) {
2410 err
= SET_ERROR(EBUSY
);
2414 /* Check that the snapshot name does not conflict */
2415 VERIFY0(dsl_dataset_get_snapname(ds
));
2416 if (strlen(ds
->ds_snapname
) >= max_snap_len
) {
2417 err
= SET_ERROR(ENAMETOOLONG
);
2420 err
= dsl_dataset_snap_lookup(hds
, ds
->ds_snapname
, &val
);
2422 (void) strcpy(ddpa
->err_ds
, snap
->ds
->ds_snapname
);
2423 err
= SET_ERROR(EEXIST
);
2429 /* The very first snapshot does not have a deadlist */
2430 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
== 0)
2433 dsl_deadlist_space(&ds
->ds_deadlist
,
2434 &dlused
, &dlcomp
, &dluncomp
);
2435 ddpa
->used
+= dlused
;
2436 ddpa
->comp
+= dlcomp
;
2437 ddpa
->uncomp
+= dluncomp
;
2441 * If we are a clone of a clone then we never reached ORIGIN,
2442 * so we need to subtract out the clone origin's used space.
2444 if (ddpa
->origin_origin
) {
2446 dsl_dataset_phys(ddpa
->origin_origin
)->ds_referenced_bytes
;
2448 dsl_dataset_phys(ddpa
->origin_origin
)->ds_compressed_bytes
;
2450 dsl_dataset_phys(ddpa
->origin_origin
)->
2451 ds_uncompressed_bytes
;
2454 /* Check that there is enough space and limit headroom here */
2455 err
= dsl_dir_transfer_possible(origin_ds
->ds_dir
, hds
->ds_dir
,
2456 0, ss_mv_cnt
, ddpa
->used
, ddpa
->cr
);
2461 * Compute the amounts of space that will be used by snapshots
2462 * after the promotion (for both origin and clone). For each,
2463 * it is the amount of space that will be on all of their
2464 * deadlists (that was not born before their new origin).
2466 if (dsl_dir_phys(hds
->ds_dir
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2470 * Note, typically this will not be a clone of a clone,
2471 * so dd_origin_txg will be < TXG_INITIAL, so
2472 * these snaplist_space() -> dsl_deadlist_space_range()
2473 * calls will be fast because they do not have to
2474 * iterate over all bps.
2476 snap
= list_head(&ddpa
->origin_snaps
);
2478 err
= SET_ERROR(ENOENT
);
2481 err
= snaplist_space(&ddpa
->shared_snaps
,
2482 snap
->ds
->ds_dir
->dd_origin_txg
, &ddpa
->cloneusedsnap
);
2486 err
= snaplist_space(&ddpa
->clone_snaps
,
2487 snap
->ds
->ds_dir
->dd_origin_txg
, &space
);
2490 ddpa
->cloneusedsnap
+= space
;
2492 if (dsl_dir_phys(origin_ds
->ds_dir
)->dd_flags
&
2493 DD_FLAG_USED_BREAKDOWN
) {
2494 err
= snaplist_space(&ddpa
->origin_snaps
,
2495 dsl_dataset_phys(origin_ds
)->ds_creation_txg
,
2496 &ddpa
->originusedsnap
);
2502 promote_rele(ddpa
, FTAG
);
2507 dsl_dataset_promote_sync(void *arg
, dmu_tx_t
*tx
)
2509 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2510 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2512 struct promotenode
*snap
;
2513 dsl_dataset_t
*origin_ds
;
2514 dsl_dataset_t
*origin_head
;
2516 dsl_dir_t
*odd
= NULL
;
2517 uint64_t oldnext_obj
;
2520 VERIFY0(promote_hold(ddpa
, dp
, FTAG
));
2521 hds
= ddpa
->ddpa_clone
;
2523 ASSERT0(dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
);
2525 snap
= list_head(&ddpa
->shared_snaps
);
2526 origin_ds
= snap
->ds
;
2529 snap
= list_head(&ddpa
->origin_snaps
);
2530 origin_head
= snap
->ds
;
2533 * We need to explicitly open odd, since origin_ds's dd will be
2536 VERIFY0(dsl_dir_hold_obj(dp
, origin_ds
->ds_dir
->dd_object
,
2539 /* change origin's next snap */
2540 dmu_buf_will_dirty(origin_ds
->ds_dbuf
, tx
);
2541 oldnext_obj
= dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
;
2542 snap
= list_tail(&ddpa
->clone_snaps
);
2543 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2544 origin_ds
->ds_object
);
2545 dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
= snap
->ds
->ds_object
;
2547 /* change the origin's next clone */
2548 if (dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
) {
2549 dsl_dataset_remove_from_next_clones(origin_ds
,
2550 snap
->ds
->ds_object
, tx
);
2551 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2552 dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
,
2557 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
2558 ASSERT3U(dsl_dir_phys(dd
)->dd_origin_obj
, ==, origin_ds
->ds_object
);
2559 dsl_dir_phys(dd
)->dd_origin_obj
= dsl_dir_phys(odd
)->dd_origin_obj
;
2560 dd
->dd_origin_txg
= origin_head
->ds_dir
->dd_origin_txg
;
2561 dmu_buf_will_dirty(odd
->dd_dbuf
, tx
);
2562 dsl_dir_phys(odd
)->dd_origin_obj
= origin_ds
->ds_object
;
2563 origin_head
->ds_dir
->dd_origin_txg
=
2564 dsl_dataset_phys(origin_ds
)->ds_creation_txg
;
2566 /* change dd_clone entries */
2567 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2568 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2569 dsl_dir_phys(odd
)->dd_clones
, hds
->ds_object
, tx
));
2570 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2571 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2572 hds
->ds_object
, tx
));
2574 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2575 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2576 origin_head
->ds_object
, tx
));
2577 if (dsl_dir_phys(dd
)->dd_clones
== 0) {
2578 dsl_dir_phys(dd
)->dd_clones
=
2579 zap_create(dp
->dp_meta_objset
, DMU_OT_DSL_CLONES
,
2580 DMU_OT_NONE
, 0, tx
);
2582 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2583 dsl_dir_phys(dd
)->dd_clones
, origin_head
->ds_object
, tx
));
2586 /* move snapshots to this dir */
2587 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2588 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2589 dsl_dataset_t
*ds
= snap
->ds
;
2592 * Property callbacks are registered to a particular
2593 * dsl_dir. Since ours is changing, evict the objset
2594 * so that they will be unregistered from the old dsl_dir.
2596 if (ds
->ds_objset
) {
2597 dmu_objset_evict(ds
->ds_objset
);
2598 ds
->ds_objset
= NULL
;
2601 /* move snap name entry */
2602 VERIFY0(dsl_dataset_get_snapname(ds
));
2603 VERIFY0(dsl_dataset_snap_remove(origin_head
,
2604 ds
->ds_snapname
, tx
, B_TRUE
));
2605 VERIFY0(zap_add(dp
->dp_meta_objset
,
2606 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
, ds
->ds_snapname
,
2607 8, 1, &ds
->ds_object
, tx
));
2608 dsl_fs_ss_count_adjust(hds
->ds_dir
, 1,
2609 DD_FIELD_SNAPSHOT_COUNT
, tx
);
2611 /* change containing dsl_dir */
2612 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2613 ASSERT3U(dsl_dataset_phys(ds
)->ds_dir_obj
, ==, odd
->dd_object
);
2614 dsl_dataset_phys(ds
)->ds_dir_obj
= dd
->dd_object
;
2615 ASSERT3P(ds
->ds_dir
, ==, odd
);
2616 dsl_dir_rele(ds
->ds_dir
, ds
);
2617 VERIFY0(dsl_dir_hold_obj(dp
, dd
->dd_object
,
2618 NULL
, ds
, &ds
->ds_dir
));
2620 /* move any clone references */
2621 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
&&
2622 spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2626 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2627 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
2628 zap_cursor_retrieve(&zc
, &za
) == 0;
2629 zap_cursor_advance(&zc
)) {
2630 dsl_dataset_t
*cnds
;
2633 if (za
.za_first_integer
== oldnext_obj
) {
2635 * We've already moved the
2636 * origin's reference.
2641 VERIFY0(dsl_dataset_hold_obj(dp
,
2642 za
.za_first_integer
, FTAG
, &cnds
));
2643 o
= dsl_dir_phys(cnds
->ds_dir
)->
2644 dd_head_dataset_obj
;
2646 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2647 dsl_dir_phys(odd
)->dd_clones
, o
, tx
));
2648 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2649 dsl_dir_phys(dd
)->dd_clones
, o
, tx
));
2650 dsl_dataset_rele(cnds
, FTAG
);
2652 zap_cursor_fini(&zc
);
2655 ASSERT(!dsl_prop_hascb(ds
));
2659 * Change space accounting.
2660 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2661 * both be valid, or both be 0 (resulting in delta == 0). This
2662 * is true for each of {clone,origin} independently.
2665 delta
= ddpa
->cloneusedsnap
-
2666 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_SNAP
];
2667 ASSERT3S(delta
, >=, 0);
2668 ASSERT3U(ddpa
->used
, >=, delta
);
2669 dsl_dir_diduse_space(dd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2670 dsl_dir_diduse_space(dd
, DD_USED_HEAD
,
2671 ddpa
->used
- delta
, ddpa
->comp
, ddpa
->uncomp
, tx
);
2673 delta
= ddpa
->originusedsnap
-
2674 dsl_dir_phys(odd
)->dd_used_breakdown
[DD_USED_SNAP
];
2675 ASSERT3S(delta
, <=, 0);
2676 ASSERT3U(ddpa
->used
, >=, -delta
);
2677 dsl_dir_diduse_space(odd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2678 dsl_dir_diduse_space(odd
, DD_USED_HEAD
,
2679 -ddpa
->used
- delta
, -ddpa
->comp
, -ddpa
->uncomp
, tx
);
2681 dsl_dataset_phys(origin_ds
)->ds_unique_bytes
= ddpa
->unique
;
2683 /* log history record */
2684 spa_history_log_internal_ds(hds
, "promote", tx
, "");
2686 dsl_dir_rele(odd
, FTAG
);
2687 promote_rele(ddpa
, FTAG
);
2691 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2692 * (exclusive) and last_obj (inclusive). The list will be in reverse
2693 * order (last_obj will be the list_head()). If first_obj == 0, do all
2694 * snapshots back to this dataset's origin.
2697 snaplist_make(dsl_pool_t
*dp
,
2698 uint64_t first_obj
, uint64_t last_obj
, list_t
*l
, void *tag
)
2700 uint64_t obj
= last_obj
;
2702 list_create(l
, sizeof (struct promotenode
),
2703 offsetof(struct promotenode
, link
));
2705 while (obj
!= first_obj
) {
2707 struct promotenode
*snap
;
2710 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
2711 ASSERT(err
!= ENOENT
);
2716 first_obj
= dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
;
2718 snap
= kmem_alloc(sizeof (*snap
), KM_SLEEP
);
2720 list_insert_tail(l
, snap
);
2721 obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
2728 snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
)
2730 struct promotenode
*snap
;
2733 for (snap
= list_head(l
); snap
; snap
= list_next(l
, snap
)) {
2734 uint64_t used
, comp
, uncomp
;
2735 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2736 mintxg
, UINT64_MAX
, &used
, &comp
, &uncomp
);
2743 snaplist_destroy(list_t
*l
, void *tag
)
2745 struct promotenode
*snap
;
2747 if (l
== NULL
|| !list_link_active(&l
->list_head
))
2750 while ((snap
= list_tail(l
)) != NULL
) {
2751 list_remove(l
, snap
);
2752 dsl_dataset_rele(snap
->ds
, tag
);
2753 kmem_free(snap
, sizeof (*snap
));
2759 promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
, void *tag
)
2763 struct promotenode
*snap
;
2765 error
= dsl_dataset_hold(dp
, ddpa
->ddpa_clonename
, tag
,
2769 dd
= ddpa
->ddpa_clone
->ds_dir
;
2771 if (ddpa
->ddpa_clone
->ds_is_snapshot
||
2772 !dsl_dir_is_clone(dd
)) {
2773 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2774 return (SET_ERROR(EINVAL
));
2777 error
= snaplist_make(dp
, 0, dsl_dir_phys(dd
)->dd_origin_obj
,
2778 &ddpa
->shared_snaps
, tag
);
2782 error
= snaplist_make(dp
, 0, ddpa
->ddpa_clone
->ds_object
,
2783 &ddpa
->clone_snaps
, tag
);
2787 snap
= list_head(&ddpa
->shared_snaps
);
2788 ASSERT3U(snap
->ds
->ds_object
, ==, dsl_dir_phys(dd
)->dd_origin_obj
);
2789 error
= snaplist_make(dp
, dsl_dir_phys(dd
)->dd_origin_obj
,
2790 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_head_dataset_obj
,
2791 &ddpa
->origin_snaps
, tag
);
2795 if (dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
!= 0) {
2796 error
= dsl_dataset_hold_obj(dp
,
2797 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
,
2798 tag
, &ddpa
->origin_origin
);
2804 promote_rele(ddpa
, tag
);
2809 promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
)
2811 snaplist_destroy(&ddpa
->shared_snaps
, tag
);
2812 snaplist_destroy(&ddpa
->clone_snaps
, tag
);
2813 snaplist_destroy(&ddpa
->origin_snaps
, tag
);
2814 if (ddpa
->origin_origin
!= NULL
)
2815 dsl_dataset_rele(ddpa
->origin_origin
, tag
);
2816 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2822 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2823 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2826 dsl_dataset_promote(const char *name
, char *conflsnap
)
2828 dsl_dataset_promote_arg_t ddpa
= { 0 };
2834 * We will modify space proportional to the number of
2835 * snapshots. Compute numsnaps.
2837 error
= dmu_objset_hold(name
, FTAG
, &os
);
2840 error
= zap_count(dmu_objset_pool(os
)->dp_meta_objset
,
2841 dsl_dataset_phys(dmu_objset_ds(os
))->ds_snapnames_zapobj
,
2843 dmu_objset_rele(os
, FTAG
);
2847 ddpa
.ddpa_clonename
= name
;
2848 ddpa
.err_ds
= conflsnap
;
2851 return (dsl_sync_task(name
, dsl_dataset_promote_check
,
2852 dsl_dataset_promote_sync
, &ddpa
,
2853 2 + numsnaps
, ZFS_SPACE_CHECK_RESERVED
));
2857 dsl_dataset_clone_swap_check_impl(dsl_dataset_t
*clone
,
2858 dsl_dataset_t
*origin_head
, boolean_t force
, void *owner
, dmu_tx_t
*tx
)
2861 * "slack" factor for received datasets with refquota set on them.
2862 * See the bottom of this function for details on its use.
2864 uint64_t refquota_slack
= (uint64_t)DMU_MAX_ACCESS
*
2865 spa_asize_inflation
;
2866 int64_t unused_refres_delta
;
2868 /* they should both be heads */
2869 if (clone
->ds_is_snapshot
||
2870 origin_head
->ds_is_snapshot
)
2871 return (SET_ERROR(EINVAL
));
2873 /* if we are not forcing, the branch point should be just before them */
2874 if (!force
&& clone
->ds_prev
!= origin_head
->ds_prev
)
2875 return (SET_ERROR(EINVAL
));
2877 /* clone should be the clone (unless they are unrelated) */
2878 if (clone
->ds_prev
!= NULL
&&
2879 clone
->ds_prev
!= clone
->ds_dir
->dd_pool
->dp_origin_snap
&&
2880 origin_head
->ds_dir
!= clone
->ds_prev
->ds_dir
)
2881 return (SET_ERROR(EINVAL
));
2883 /* the clone should be a child of the origin */
2884 if (clone
->ds_dir
->dd_parent
!= origin_head
->ds_dir
)
2885 return (SET_ERROR(EINVAL
));
2887 /* origin_head shouldn't be modified unless 'force' */
2889 dsl_dataset_modified_since_snap(origin_head
, origin_head
->ds_prev
))
2890 return (SET_ERROR(ETXTBSY
));
2892 /* origin_head should have no long holds (e.g. is not mounted) */
2893 if (dsl_dataset_handoff_check(origin_head
, owner
, tx
))
2894 return (SET_ERROR(EBUSY
));
2896 /* check amount of any unconsumed refreservation */
2897 unused_refres_delta
=
2898 (int64_t)MIN(origin_head
->ds_reserved
,
2899 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
2900 (int64_t)MIN(origin_head
->ds_reserved
,
2901 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2903 if (unused_refres_delta
> 0 &&
2904 unused_refres_delta
>
2905 dsl_dir_space_available(origin_head
->ds_dir
, NULL
, 0, TRUE
))
2906 return (SET_ERROR(ENOSPC
));
2909 * The clone can't be too much over the head's refquota.
2911 * To ensure that the entire refquota can be used, we allow one
2912 * transaction to exceed the the refquota. Therefore, this check
2913 * needs to also allow for the space referenced to be more than the
2914 * refquota. The maximum amount of space that one transaction can use
2915 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this
2916 * overage ensures that we are able to receive a filesystem that
2917 * exceeds the refquota on the source system.
2919 * So that overage is the refquota_slack we use below.
2921 if (origin_head
->ds_quota
!= 0 &&
2922 dsl_dataset_phys(clone
)->ds_referenced_bytes
>
2923 origin_head
->ds_quota
+ refquota_slack
)
2924 return (SET_ERROR(EDQUOT
));
2930 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t
*clone
,
2931 dsl_dataset_t
*origin_head
, dmu_tx_t
*tx
)
2934 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2935 int64_t unused_refres_delta
;
2937 ASSERT(clone
->ds_reserved
== 0);
2939 * NOTE: On DEBUG kernels there could be a race between this and
2940 * the check function if spa_asize_inflation is adjusted...
2942 ASSERT(origin_head
->ds_quota
== 0 ||
2943 dsl_dataset_phys(clone
)->ds_unique_bytes
<= origin_head
->ds_quota
+
2944 DMU_MAX_ACCESS
* spa_asize_inflation
);
2945 ASSERT3P(clone
->ds_prev
, ==, origin_head
->ds_prev
);
2948 * Swap per-dataset feature flags.
2950 for (f
= 0; f
< SPA_FEATURES
; f
++) {
2951 boolean_t clone_inuse
;
2952 boolean_t origin_head_inuse
;
2954 if (!(spa_feature_table
[f
].fi_flags
&
2955 ZFEATURE_FLAG_PER_DATASET
)) {
2956 ASSERT(!clone
->ds_feature_inuse
[f
]);
2957 ASSERT(!origin_head
->ds_feature_inuse
[f
]);
2961 clone_inuse
= clone
->ds_feature_inuse
[f
];
2962 origin_head_inuse
= origin_head
->ds_feature_inuse
[f
];
2965 dsl_dataset_deactivate_feature(clone
->ds_object
, f
, tx
);
2966 clone
->ds_feature_inuse
[f
] = B_FALSE
;
2968 if (origin_head_inuse
) {
2969 dsl_dataset_deactivate_feature(origin_head
->ds_object
,
2971 origin_head
->ds_feature_inuse
[f
] = B_FALSE
;
2974 dsl_dataset_activate_feature(origin_head
->ds_object
,
2976 origin_head
->ds_feature_inuse
[f
] = B_TRUE
;
2978 if (origin_head_inuse
) {
2979 dsl_dataset_activate_feature(clone
->ds_object
, f
, tx
);
2980 clone
->ds_feature_inuse
[f
] = B_TRUE
;
2984 dmu_buf_will_dirty(clone
->ds_dbuf
, tx
);
2985 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
2987 if (clone
->ds_objset
!= NULL
) {
2988 dmu_objset_evict(clone
->ds_objset
);
2989 clone
->ds_objset
= NULL
;
2992 if (origin_head
->ds_objset
!= NULL
) {
2993 dmu_objset_evict(origin_head
->ds_objset
);
2994 origin_head
->ds_objset
= NULL
;
2997 unused_refres_delta
=
2998 (int64_t)MIN(origin_head
->ds_reserved
,
2999 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
3000 (int64_t)MIN(origin_head
->ds_reserved
,
3001 dsl_dataset_phys(clone
)->ds_unique_bytes
);
3004 * Reset origin's unique bytes, if it exists.
3006 if (clone
->ds_prev
) {
3007 dsl_dataset_t
*origin
= clone
->ds_prev
;
3008 uint64_t comp
, uncomp
;
3010 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
3011 dsl_deadlist_space_range(&clone
->ds_deadlist
,
3012 dsl_dataset_phys(origin
)->ds_prev_snap_txg
, UINT64_MAX
,
3013 &dsl_dataset_phys(origin
)->ds_unique_bytes
, &comp
, &uncomp
);
3019 tmp
= dsl_dataset_phys(origin_head
)->ds_bp
;
3020 dsl_dataset_phys(origin_head
)->ds_bp
=
3021 dsl_dataset_phys(clone
)->ds_bp
;
3022 dsl_dataset_phys(clone
)->ds_bp
= tmp
;
3025 /* set dd_*_bytes */
3027 int64_t dused
, dcomp
, duncomp
;
3028 uint64_t cdl_used
, cdl_comp
, cdl_uncomp
;
3029 uint64_t odl_used
, odl_comp
, odl_uncomp
;
3031 ASSERT3U(dsl_dir_phys(clone
->ds_dir
)->
3032 dd_used_breakdown
[DD_USED_SNAP
], ==, 0);
3034 dsl_deadlist_space(&clone
->ds_deadlist
,
3035 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
3036 dsl_deadlist_space(&origin_head
->ds_deadlist
,
3037 &odl_used
, &odl_comp
, &odl_uncomp
);
3039 dused
= dsl_dataset_phys(clone
)->ds_referenced_bytes
+
3041 (dsl_dataset_phys(origin_head
)->ds_referenced_bytes
+
3043 dcomp
= dsl_dataset_phys(clone
)->ds_compressed_bytes
+
3045 (dsl_dataset_phys(origin_head
)->ds_compressed_bytes
+
3047 duncomp
= dsl_dataset_phys(clone
)->ds_uncompressed_bytes
+
3049 (dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
+
3052 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_HEAD
,
3053 dused
, dcomp
, duncomp
, tx
);
3054 dsl_dir_diduse_space(clone
->ds_dir
, DD_USED_HEAD
,
3055 -dused
, -dcomp
, -duncomp
, tx
);
3058 * The difference in the space used by snapshots is the
3059 * difference in snapshot space due to the head's
3060 * deadlist (since that's the only thing that's
3061 * changing that affects the snapused).
3063 dsl_deadlist_space_range(&clone
->ds_deadlist
,
3064 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
3065 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
3066 dsl_deadlist_space_range(&origin_head
->ds_deadlist
,
3067 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
3068 &odl_used
, &odl_comp
, &odl_uncomp
);
3069 dsl_dir_transfer_space(origin_head
->ds_dir
, cdl_used
- odl_used
,
3070 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
3073 /* swap ds_*_bytes */
3074 SWITCH64(dsl_dataset_phys(origin_head
)->ds_referenced_bytes
,
3075 dsl_dataset_phys(clone
)->ds_referenced_bytes
);
3076 SWITCH64(dsl_dataset_phys(origin_head
)->ds_compressed_bytes
,
3077 dsl_dataset_phys(clone
)->ds_compressed_bytes
);
3078 SWITCH64(dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
,
3079 dsl_dataset_phys(clone
)->ds_uncompressed_bytes
);
3080 SWITCH64(dsl_dataset_phys(origin_head
)->ds_unique_bytes
,
3081 dsl_dataset_phys(clone
)->ds_unique_bytes
);
3083 /* apply any parent delta for change in unconsumed refreservation */
3084 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_REFRSRV
,
3085 unused_refres_delta
, 0, 0, tx
);
3090 dsl_deadlist_close(&clone
->ds_deadlist
);
3091 dsl_deadlist_close(&origin_head
->ds_deadlist
);
3092 SWITCH64(dsl_dataset_phys(origin_head
)->ds_deadlist_obj
,
3093 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
3094 dsl_deadlist_open(&clone
->ds_deadlist
, dp
->dp_meta_objset
,
3095 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
3096 dsl_deadlist_open(&origin_head
->ds_deadlist
, dp
->dp_meta_objset
,
3097 dsl_dataset_phys(origin_head
)->ds_deadlist_obj
);
3099 dsl_scan_ds_clone_swapped(origin_head
, clone
, tx
);
3101 spa_history_log_internal_ds(clone
, "clone swap", tx
,
3102 "parent=%s", origin_head
->ds_dir
->dd_myname
);
3106 * Given a pool name and a dataset object number in that pool,
3107 * return the name of that dataset.
3110 dsl_dsobj_to_dsname(char *pname
, uint64_t obj
, char *buf
)
3116 error
= dsl_pool_hold(pname
, FTAG
, &dp
);
3120 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
3122 dsl_dataset_name(ds
, buf
);
3123 dsl_dataset_rele(ds
, FTAG
);
3125 dsl_pool_rele(dp
, FTAG
);
3131 dsl_dataset_check_quota(dsl_dataset_t
*ds
, boolean_t check_quota
,
3132 uint64_t asize
, uint64_t inflight
, uint64_t *used
, uint64_t *ref_rsrv
)
3136 ASSERT3S(asize
, >, 0);
3139 * *ref_rsrv is the portion of asize that will come from any
3140 * unconsumed refreservation space.
3144 mutex_enter(&ds
->ds_lock
);
3146 * Make a space adjustment for reserved bytes.
3148 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
) {
3150 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
3152 (ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
3154 asize
- MIN(asize
, parent_delta(ds
, asize
+ inflight
));
3157 if (!check_quota
|| ds
->ds_quota
== 0) {
3158 mutex_exit(&ds
->ds_lock
);
3162 * If they are requesting more space, and our current estimate
3163 * is over quota, they get to try again unless the actual
3164 * on-disk is over quota and there are no pending changes (which
3165 * may free up space for us).
3167 if (dsl_dataset_phys(ds
)->ds_referenced_bytes
+ inflight
>=
3170 dsl_dataset_phys(ds
)->ds_referenced_bytes
< ds
->ds_quota
)
3171 error
= SET_ERROR(ERESTART
);
3173 error
= SET_ERROR(EDQUOT
);
3175 mutex_exit(&ds
->ds_lock
);
3180 typedef struct dsl_dataset_set_qr_arg
{
3181 const char *ddsqra_name
;
3182 zprop_source_t ddsqra_source
;
3183 uint64_t ddsqra_value
;
3184 } dsl_dataset_set_qr_arg_t
;
3189 dsl_dataset_set_refquota_check(void *arg
, dmu_tx_t
*tx
)
3191 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3192 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3197 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFQUOTA
)
3198 return (SET_ERROR(ENOTSUP
));
3200 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3204 if (ds
->ds_is_snapshot
) {
3205 dsl_dataset_rele(ds
, FTAG
);
3206 return (SET_ERROR(EINVAL
));
3209 error
= dsl_prop_predict(ds
->ds_dir
,
3210 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
3211 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3213 dsl_dataset_rele(ds
, FTAG
);
3218 dsl_dataset_rele(ds
, FTAG
);
3222 if (newval
< dsl_dataset_phys(ds
)->ds_referenced_bytes
||
3223 newval
< ds
->ds_reserved
) {
3224 dsl_dataset_rele(ds
, FTAG
);
3225 return (SET_ERROR(ENOSPC
));
3228 dsl_dataset_rele(ds
, FTAG
);
3233 dsl_dataset_set_refquota_sync(void *arg
, dmu_tx_t
*tx
)
3235 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3236 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3237 dsl_dataset_t
*ds
= NULL
;
3240 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3242 dsl_prop_set_sync_impl(ds
,
3243 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
3244 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
3245 &ddsqra
->ddsqra_value
, tx
);
3247 VERIFY0(dsl_prop_get_int_ds(ds
,
3248 zfs_prop_to_name(ZFS_PROP_REFQUOTA
), &newval
));
3250 if (ds
->ds_quota
!= newval
) {
3251 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3252 ds
->ds_quota
= newval
;
3254 dsl_dataset_rele(ds
, FTAG
);
3258 dsl_dataset_set_refquota(const char *dsname
, zprop_source_t source
,
3261 dsl_dataset_set_qr_arg_t ddsqra
;
3263 ddsqra
.ddsqra_name
= dsname
;
3264 ddsqra
.ddsqra_source
= source
;
3265 ddsqra
.ddsqra_value
= refquota
;
3267 return (dsl_sync_task(dsname
, dsl_dataset_set_refquota_check
,
3268 dsl_dataset_set_refquota_sync
, &ddsqra
, 0, ZFS_SPACE_CHECK_NONE
));
3272 dsl_dataset_set_refreservation_check(void *arg
, dmu_tx_t
*tx
)
3274 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3275 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3278 uint64_t newval
, unique
;
3280 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFRESERVATION
)
3281 return (SET_ERROR(ENOTSUP
));
3283 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3287 if (ds
->ds_is_snapshot
) {
3288 dsl_dataset_rele(ds
, FTAG
);
3289 return (SET_ERROR(EINVAL
));
3292 error
= dsl_prop_predict(ds
->ds_dir
,
3293 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3294 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3296 dsl_dataset_rele(ds
, FTAG
);
3301 * If we are doing the preliminary check in open context, the
3302 * space estimates may be inaccurate.
3304 if (!dmu_tx_is_syncing(tx
)) {
3305 dsl_dataset_rele(ds
, FTAG
);
3309 mutex_enter(&ds
->ds_lock
);
3310 if (!DS_UNIQUE_IS_ACCURATE(ds
))
3311 dsl_dataset_recalc_head_uniq(ds
);
3312 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3313 mutex_exit(&ds
->ds_lock
);
3315 if (MAX(unique
, newval
) > MAX(unique
, ds
->ds_reserved
)) {
3316 uint64_t delta
= MAX(unique
, newval
) -
3317 MAX(unique
, ds
->ds_reserved
);
3320 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, B_TRUE
) ||
3321 (ds
->ds_quota
> 0 && newval
> ds
->ds_quota
)) {
3322 dsl_dataset_rele(ds
, FTAG
);
3323 return (SET_ERROR(ENOSPC
));
3327 dsl_dataset_rele(ds
, FTAG
);
3332 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t
*ds
,
3333 zprop_source_t source
, uint64_t value
, dmu_tx_t
*tx
)
3339 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3340 source
, sizeof (value
), 1, &value
, tx
);
3342 VERIFY0(dsl_prop_get_int_ds(ds
,
3343 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
), &newval
));
3345 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3346 mutex_enter(&ds
->ds_dir
->dd_lock
);
3347 mutex_enter(&ds
->ds_lock
);
3348 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
3349 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3350 delta
= MAX(0, (int64_t)(newval
- unique
)) -
3351 MAX(0, (int64_t)(ds
->ds_reserved
- unique
));
3352 ds
->ds_reserved
= newval
;
3353 mutex_exit(&ds
->ds_lock
);
3355 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
, delta
, 0, 0, tx
);
3356 mutex_exit(&ds
->ds_dir
->dd_lock
);
3360 dsl_dataset_set_refreservation_sync(void *arg
, dmu_tx_t
*tx
)
3362 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3363 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3364 dsl_dataset_t
*ds
= NULL
;
3366 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3367 dsl_dataset_set_refreservation_sync_impl(ds
,
3368 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, tx
);
3369 dsl_dataset_rele(ds
, FTAG
);
3373 dsl_dataset_set_refreservation(const char *dsname
, zprop_source_t source
,
3374 uint64_t refreservation
)
3376 dsl_dataset_set_qr_arg_t ddsqra
;
3378 ddsqra
.ddsqra_name
= dsname
;
3379 ddsqra
.ddsqra_source
= source
;
3380 ddsqra
.ddsqra_value
= refreservation
;
3382 return (dsl_sync_task(dsname
, dsl_dataset_set_refreservation_check
,
3383 dsl_dataset_set_refreservation_sync
, &ddsqra
,
3384 0, ZFS_SPACE_CHECK_NONE
));
3388 * Return (in *usedp) the amount of space written in new that is not
3389 * present in oldsnap. New may be a snapshot or the head. Old must be
3390 * a snapshot before new, in new's filesystem (or its origin). If not then
3391 * fail and return EINVAL.
3393 * The written space is calculated by considering two components: First, we
3394 * ignore any freed space, and calculate the written as new's used space
3395 * minus old's used space. Next, we add in the amount of space that was freed
3396 * between the two snapshots, thus reducing new's used space relative to old's.
3397 * Specifically, this is the space that was born before old->ds_creation_txg,
3398 * and freed before new (ie. on new's deadlist or a previous deadlist).
3400 * space freed [---------------------]
3401 * snapshots ---O-------O--------O-------O------
3405 dsl_dataset_space_written(dsl_dataset_t
*oldsnap
, dsl_dataset_t
*new,
3406 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3410 dsl_pool_t
*dp
= new->ds_dir
->dd_pool
;
3412 ASSERT(dsl_pool_config_held(dp
));
3415 *usedp
+= dsl_dataset_phys(new)->ds_referenced_bytes
;
3416 *usedp
-= dsl_dataset_phys(oldsnap
)->ds_referenced_bytes
;
3419 *compp
+= dsl_dataset_phys(new)->ds_compressed_bytes
;
3420 *compp
-= dsl_dataset_phys(oldsnap
)->ds_compressed_bytes
;
3423 *uncompp
+= dsl_dataset_phys(new)->ds_uncompressed_bytes
;
3424 *uncompp
-= dsl_dataset_phys(oldsnap
)->ds_uncompressed_bytes
;
3426 snapobj
= new->ds_object
;
3427 while (snapobj
!= oldsnap
->ds_object
) {
3428 dsl_dataset_t
*snap
;
3429 uint64_t used
, comp
, uncomp
;
3431 if (snapobj
== new->ds_object
) {
3434 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &snap
);
3439 if (dsl_dataset_phys(snap
)->ds_prev_snap_txg
==
3440 dsl_dataset_phys(oldsnap
)->ds_creation_txg
) {
3442 * The blocks in the deadlist can not be born after
3443 * ds_prev_snap_txg, so get the whole deadlist space,
3444 * which is more efficient (especially for old-format
3445 * deadlists). Unfortunately the deadlist code
3446 * doesn't have enough information to make this
3447 * optimization itself.
3449 dsl_deadlist_space(&snap
->ds_deadlist
,
3450 &used
, &comp
, &uncomp
);
3452 dsl_deadlist_space_range(&snap
->ds_deadlist
,
3453 0, dsl_dataset_phys(oldsnap
)->ds_creation_txg
,
3454 &used
, &comp
, &uncomp
);
3461 * If we get to the beginning of the chain of snapshots
3462 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3463 * was not a snapshot of/before new.
3465 snapobj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3467 dsl_dataset_rele(snap
, FTAG
);
3469 err
= SET_ERROR(EINVAL
);
3478 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3479 * lastsnap, and all snapshots in between are deleted.
3481 * blocks that would be freed [---------------------------]
3482 * snapshots ---O-------O--------O-------O--------O
3483 * firstsnap lastsnap
3485 * This is the set of blocks that were born after the snap before firstsnap,
3486 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3487 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3488 * We calculate this by iterating over the relevant deadlists (from the snap
3489 * after lastsnap, backward to the snap after firstsnap), summing up the
3490 * space on the deadlist that was born after the snap before firstsnap.
3493 dsl_dataset_space_wouldfree(dsl_dataset_t
*firstsnap
,
3494 dsl_dataset_t
*lastsnap
,
3495 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3499 dsl_pool_t
*dp
= firstsnap
->ds_dir
->dd_pool
;
3501 ASSERT(firstsnap
->ds_is_snapshot
);
3502 ASSERT(lastsnap
->ds_is_snapshot
);
3505 * Check that the snapshots are in the same dsl_dir, and firstsnap
3506 * is before lastsnap.
3508 if (firstsnap
->ds_dir
!= lastsnap
->ds_dir
||
3509 dsl_dataset_phys(firstsnap
)->ds_creation_txg
>
3510 dsl_dataset_phys(lastsnap
)->ds_creation_txg
)
3511 return (SET_ERROR(EINVAL
));
3513 *usedp
= *compp
= *uncompp
= 0;
3515 snapobj
= dsl_dataset_phys(lastsnap
)->ds_next_snap_obj
;
3516 while (snapobj
!= firstsnap
->ds_object
) {
3518 uint64_t used
, comp
, uncomp
;
3520 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &ds
);
3524 dsl_deadlist_space_range(&ds
->ds_deadlist
,
3525 dsl_dataset_phys(firstsnap
)->ds_prev_snap_txg
, UINT64_MAX
,
3526 &used
, &comp
, &uncomp
);
3531 snapobj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
3532 ASSERT3U(snapobj
, !=, 0);
3533 dsl_dataset_rele(ds
, FTAG
);
3539 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3540 * For example, they could both be snapshots of the same filesystem, and
3541 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3542 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3543 * filesystem. Or 'earlier' could be the origin's origin.
3545 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3548 dsl_dataset_is_before(dsl_dataset_t
*later
, dsl_dataset_t
*earlier
,
3549 uint64_t earlier_txg
)
3551 dsl_pool_t
*dp
= later
->ds_dir
->dd_pool
;
3554 dsl_dataset_t
*origin
;
3556 ASSERT(dsl_pool_config_held(dp
));
3557 ASSERT(earlier
->ds_is_snapshot
|| earlier_txg
!= 0);
3559 if (earlier_txg
== 0)
3560 earlier_txg
= dsl_dataset_phys(earlier
)->ds_creation_txg
;
3562 if (later
->ds_is_snapshot
&&
3563 earlier_txg
>= dsl_dataset_phys(later
)->ds_creation_txg
)
3566 if (later
->ds_dir
== earlier
->ds_dir
)
3568 if (!dsl_dir_is_clone(later
->ds_dir
))
3571 if (dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
== earlier
->ds_object
)
3573 error
= dsl_dataset_hold_obj(dp
,
3574 dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
, FTAG
, &origin
);
3577 ret
= dsl_dataset_is_before(origin
, earlier
, earlier_txg
);
3578 dsl_dataset_rele(origin
, FTAG
);
3583 dsl_dataset_zapify(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
3585 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
3586 dmu_object_zapify(mos
, ds
->ds_object
, DMU_OT_DSL_DATASET
, tx
);
3590 dsl_dataset_is_zapified(dsl_dataset_t
*ds
)
3592 dmu_object_info_t doi
;
3594 dmu_object_info_from_db(ds
->ds_dbuf
, &doi
);
3595 return (doi
.doi_type
== DMU_OTN_ZAP_METADATA
);
3599 dsl_dataset_has_resume_receive_state(dsl_dataset_t
*ds
)
3601 return (dsl_dataset_is_zapified(ds
) &&
3602 zap_contains(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
3603 ds
->ds_object
, DS_FIELD_RESUME_TOGUID
) == 0);
3606 #if defined(_KERNEL) && defined(HAVE_SPL)
3608 module_param(zfs_max_recordsize
, int, 0644);
3609 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
3611 /* Limited to 1M on 32-bit platforms due to lack of virtual address space */
3612 module_param(zfs_max_recordsize
, int, 0444);
3613 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
3616 EXPORT_SYMBOL(dsl_dataset_hold
);
3617 EXPORT_SYMBOL(dsl_dataset_hold_obj
);
3618 EXPORT_SYMBOL(dsl_dataset_own
);
3619 EXPORT_SYMBOL(dsl_dataset_own_obj
);
3620 EXPORT_SYMBOL(dsl_dataset_name
);
3621 EXPORT_SYMBOL(dsl_dataset_rele
);
3622 EXPORT_SYMBOL(dsl_dataset_disown
);
3623 EXPORT_SYMBOL(dsl_dataset_tryown
);
3624 EXPORT_SYMBOL(dsl_dataset_create_sync
);
3625 EXPORT_SYMBOL(dsl_dataset_create_sync_dd
);
3626 EXPORT_SYMBOL(dsl_dataset_snapshot_check
);
3627 EXPORT_SYMBOL(dsl_dataset_snapshot_sync
);
3628 EXPORT_SYMBOL(dsl_dataset_promote
);
3629 EXPORT_SYMBOL(dsl_dataset_user_hold
);
3630 EXPORT_SYMBOL(dsl_dataset_user_release
);
3631 EXPORT_SYMBOL(dsl_dataset_get_holds
);
3632 EXPORT_SYMBOL(dsl_dataset_get_blkptr
);
3633 EXPORT_SYMBOL(dsl_dataset_set_blkptr
);
3634 EXPORT_SYMBOL(dsl_dataset_get_spa
);
3635 EXPORT_SYMBOL(dsl_dataset_modified_since_snap
);
3636 EXPORT_SYMBOL(dsl_dataset_space_written
);
3637 EXPORT_SYMBOL(dsl_dataset_space_wouldfree
);
3638 EXPORT_SYMBOL(dsl_dataset_sync
);
3639 EXPORT_SYMBOL(dsl_dataset_block_born
);
3640 EXPORT_SYMBOL(dsl_dataset_block_kill
);
3641 EXPORT_SYMBOL(dsl_dataset_block_freeable
);
3642 EXPORT_SYMBOL(dsl_dataset_prev_snap_txg
);
3643 EXPORT_SYMBOL(dsl_dataset_dirty
);
3644 EXPORT_SYMBOL(dsl_dataset_stats
);
3645 EXPORT_SYMBOL(dsl_dataset_fast_stat
);
3646 EXPORT_SYMBOL(dsl_dataset_space
);
3647 EXPORT_SYMBOL(dsl_dataset_fsid_guid
);
3648 EXPORT_SYMBOL(dsl_dsobj_to_dsname
);
3649 EXPORT_SYMBOL(dsl_dataset_check_quota
);
3650 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl
);
3651 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl
);