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>
61 * The SPA supports block sizes up to 16MB. However, very large blocks
62 * can have an impact on i/o latency (e.g. tying up a spinning disk for
63 * ~300ms), and also potentially on the memory allocator. Therefore,
64 * we do not allow the recordsize to be set larger than zfs_max_recordsize
65 * (default 1MB). Larger blocks can be created by changing this tunable,
66 * and pools with larger blocks can always be imported and used, regardless
69 int zfs_max_recordsize
= 1 * 1024 * 1024;
71 #define SWITCH64(x, y) \
73 uint64_t __tmp = (x); \
78 #define DS_REF_MAX (1ULL << 62)
80 extern inline dsl_dataset_phys_t
*dsl_dataset_phys(dsl_dataset_t
*ds
);
82 extern int spa_asize_inflation
;
85 * Figure out how much of this delta should be propogated to the dsl_dir
86 * layer. If there's a refreservation, that space has already been
87 * partially accounted for in our ancestors.
90 parent_delta(dsl_dataset_t
*ds
, int64_t delta
)
92 dsl_dataset_phys_t
*ds_phys
;
93 uint64_t old_bytes
, new_bytes
;
95 if (ds
->ds_reserved
== 0)
98 ds_phys
= dsl_dataset_phys(ds
);
99 old_bytes
= MAX(ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
100 new_bytes
= MAX(ds_phys
->ds_unique_bytes
+ delta
, ds
->ds_reserved
);
102 ASSERT3U(ABS((int64_t)(new_bytes
- old_bytes
)), <=, ABS(delta
));
103 return (new_bytes
- old_bytes
);
107 dsl_dataset_block_born(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
109 int used
, compressed
, uncompressed
;
112 used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
113 compressed
= BP_GET_PSIZE(bp
);
114 uncompressed
= BP_GET_UCSIZE(bp
);
116 dprintf_bp(bp
, "ds=%p", ds
);
118 ASSERT(dmu_tx_is_syncing(tx
));
119 /* It could have been compressed away to nothing */
122 ASSERT(BP_GET_TYPE(bp
) != DMU_OT_NONE
);
123 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp
)));
125 dsl_pool_mos_diduse_space(tx
->tx_pool
,
126 used
, compressed
, uncompressed
);
130 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
131 mutex_enter(&ds
->ds_lock
);
132 delta
= parent_delta(ds
, used
);
133 dsl_dataset_phys(ds
)->ds_referenced_bytes
+= used
;
134 dsl_dataset_phys(ds
)->ds_compressed_bytes
+= compressed
;
135 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
+= uncompressed
;
136 dsl_dataset_phys(ds
)->ds_unique_bytes
+= used
;
137 if (BP_GET_LSIZE(bp
) > SPA_OLD_MAXBLOCKSIZE
) {
138 ds
->ds_feature_activation_needed
[SPA_FEATURE_LARGE_BLOCKS
] =
141 mutex_exit(&ds
->ds_lock
);
142 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
, delta
,
143 compressed
, uncompressed
, tx
);
144 dsl_dir_transfer_space(ds
->ds_dir
, used
- delta
,
145 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
149 dsl_dataset_block_kill(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
,
152 int used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
153 int compressed
= BP_GET_PSIZE(bp
);
154 int uncompressed
= BP_GET_UCSIZE(bp
);
159 ASSERT(dmu_tx_is_syncing(tx
));
160 ASSERT(bp
->blk_birth
<= tx
->tx_txg
);
163 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
164 dsl_pool_mos_diduse_space(tx
->tx_pool
,
165 -used
, -compressed
, -uncompressed
);
168 ASSERT3P(tx
->tx_pool
, ==, ds
->ds_dir
->dd_pool
);
170 ASSERT(!ds
->ds_is_snapshot
);
171 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
173 if (bp
->blk_birth
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
176 dprintf_bp(bp
, "freeing ds=%llu", ds
->ds_object
);
177 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
179 mutex_enter(&ds
->ds_lock
);
180 ASSERT(dsl_dataset_phys(ds
)->ds_unique_bytes
>= used
||
181 !DS_UNIQUE_IS_ACCURATE(ds
));
182 delta
= parent_delta(ds
, -used
);
183 dsl_dataset_phys(ds
)->ds_unique_bytes
-= used
;
184 mutex_exit(&ds
->ds_lock
);
185 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
186 delta
, -compressed
, -uncompressed
, tx
);
187 dsl_dir_transfer_space(ds
->ds_dir
, -used
- delta
,
188 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
190 dprintf_bp(bp
, "putting on dead list: %s", "");
193 * We are here as part of zio's write done callback,
194 * which means we're a zio interrupt thread. We can't
195 * call dsl_deadlist_insert() now because it may block
196 * waiting for I/O. Instead, put bp on the deferred
197 * queue and let dsl_pool_sync() finish the job.
199 bplist_append(&ds
->ds_pending_deadlist
, bp
);
201 dsl_deadlist_insert(&ds
->ds_deadlist
, bp
, tx
);
203 ASSERT3U(ds
->ds_prev
->ds_object
, ==,
204 dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
205 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 0);
206 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
207 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
208 ds
->ds_object
&& bp
->blk_birth
>
209 dsl_dataset_phys(ds
->ds_prev
)->ds_prev_snap_txg
) {
210 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
211 mutex_enter(&ds
->ds_prev
->ds_lock
);
212 dsl_dataset_phys(ds
->ds_prev
)->ds_unique_bytes
+= used
;
213 mutex_exit(&ds
->ds_prev
->ds_lock
);
215 if (bp
->blk_birth
> ds
->ds_dir
->dd_origin_txg
) {
216 dsl_dir_transfer_space(ds
->ds_dir
, used
,
217 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
220 mutex_enter(&ds
->ds_lock
);
221 ASSERT3U(dsl_dataset_phys(ds
)->ds_referenced_bytes
, >=, used
);
222 dsl_dataset_phys(ds
)->ds_referenced_bytes
-= used
;
223 ASSERT3U(dsl_dataset_phys(ds
)->ds_compressed_bytes
, >=, compressed
);
224 dsl_dataset_phys(ds
)->ds_compressed_bytes
-= compressed
;
225 ASSERT3U(dsl_dataset_phys(ds
)->ds_uncompressed_bytes
, >=, uncompressed
);
226 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
-= uncompressed
;
227 mutex_exit(&ds
->ds_lock
);
233 dsl_dataset_prev_snap_txg(dsl_dataset_t
*ds
)
235 uint64_t trysnap
= 0;
240 * The snapshot creation could fail, but that would cause an
241 * incorrect FALSE return, which would only result in an
242 * overestimation of the amount of space that an operation would
243 * consume, which is OK.
245 * There's also a small window where we could miss a pending
246 * snapshot, because we could set the sync task in the quiescing
247 * phase. So this should only be used as a guess.
249 if (ds
->ds_trysnap_txg
>
250 spa_last_synced_txg(ds
->ds_dir
->dd_pool
->dp_spa
))
251 trysnap
= ds
->ds_trysnap_txg
;
252 return (MAX(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, trysnap
));
256 dsl_dataset_block_freeable(dsl_dataset_t
*ds
, const blkptr_t
*bp
,
259 if (blk_birth
<= dsl_dataset_prev_snap_txg(ds
) ||
260 (bp
!= NULL
&& BP_IS_HOLE(bp
)))
263 ddt_prefetch(dsl_dataset_get_spa(ds
), bp
);
269 dsl_dataset_evict(void *dbu
)
271 dsl_dataset_t
*ds
= dbu
;
273 ASSERT(ds
->ds_owner
== NULL
);
277 unique_remove(ds
->ds_fsid_guid
);
279 if (ds
->ds_objset
!= NULL
)
280 dmu_objset_evict(ds
->ds_objset
);
283 dsl_dataset_rele(ds
->ds_prev
, ds
);
287 bplist_destroy(&ds
->ds_pending_deadlist
);
288 if (ds
->ds_deadlist
.dl_os
!= NULL
)
289 dsl_deadlist_close(&ds
->ds_deadlist
);
291 dsl_dir_async_rele(ds
->ds_dir
, ds
);
293 ASSERT(!list_link_active(&ds
->ds_synced_link
));
295 list_destroy(&ds
->ds_prop_cbs
);
296 mutex_destroy(&ds
->ds_lock
);
297 mutex_destroy(&ds
->ds_opening_lock
);
298 mutex_destroy(&ds
->ds_sendstream_lock
);
299 refcount_destroy(&ds
->ds_longholds
);
301 kmem_free(ds
, sizeof (dsl_dataset_t
));
305 dsl_dataset_get_snapname(dsl_dataset_t
*ds
)
307 dsl_dataset_phys_t
*headphys
;
310 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
311 objset_t
*mos
= dp
->dp_meta_objset
;
313 if (ds
->ds_snapname
[0])
315 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0)
318 err
= dmu_bonus_hold(mos
, dsl_dir_phys(ds
->ds_dir
)->dd_head_dataset_obj
,
322 headphys
= headdbuf
->db_data
;
323 err
= zap_value_search(dp
->dp_meta_objset
,
324 headphys
->ds_snapnames_zapobj
, ds
->ds_object
, 0, ds
->ds_snapname
);
325 if (err
!= 0 && zfs_recover
== B_TRUE
) {
327 (void) snprintf(ds
->ds_snapname
, sizeof (ds
->ds_snapname
),
328 "SNAPOBJ=%llu-ERR=%d",
329 (unsigned long long)ds
->ds_object
, err
);
331 dmu_buf_rele(headdbuf
, FTAG
);
336 dsl_dataset_snap_lookup(dsl_dataset_t
*ds
, const char *name
, uint64_t *value
)
338 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
339 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
343 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
348 err
= zap_lookup_norm(mos
, snapobj
, name
, 8, 1,
349 value
, mt
, NULL
, 0, NULL
);
350 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
351 err
= zap_lookup(mos
, snapobj
, name
, 8, 1, value
);
356 dsl_dataset_snap_remove(dsl_dataset_t
*ds
, const char *name
, dmu_tx_t
*tx
,
359 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
360 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
364 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
366 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
371 err
= zap_remove_norm(mos
, snapobj
, name
, mt
, tx
);
372 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
373 err
= zap_remove(mos
, snapobj
, name
, tx
);
375 if (err
== 0 && adj_cnt
)
376 dsl_fs_ss_count_adjust(ds
->ds_dir
, -1,
377 DD_FIELD_SNAPSHOT_COUNT
, tx
);
383 dsl_dataset_try_add_ref(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *tag
)
385 dmu_buf_t
*dbuf
= ds
->ds_dbuf
;
386 boolean_t result
= B_FALSE
;
388 if (dbuf
!= NULL
&& dmu_buf_try_add_ref(dbuf
, dp
->dp_meta_objset
,
389 ds
->ds_object
, DMU_BONUS_BLKID
, tag
)) {
391 if (ds
== dmu_buf_get_user(dbuf
))
394 dmu_buf_rele(dbuf
, tag
);
401 dsl_dataset_hold_obj(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
404 objset_t
*mos
= dp
->dp_meta_objset
;
408 dmu_object_info_t doi
;
410 ASSERT(dsl_pool_config_held(dp
));
412 err
= dmu_bonus_hold(mos
, dsobj
, tag
, &dbuf
);
416 /* Make sure dsobj has the correct object type. */
417 dmu_object_info_from_db(dbuf
, &doi
);
418 if (doi
.doi_bonus_type
!= DMU_OT_DSL_DATASET
) {
419 dmu_buf_rele(dbuf
, tag
);
420 return (SET_ERROR(EINVAL
));
423 ds
= dmu_buf_get_user(dbuf
);
425 dsl_dataset_t
*winner
= NULL
;
427 ds
= kmem_zalloc(sizeof (dsl_dataset_t
), KM_SLEEP
);
429 ds
->ds_object
= dsobj
;
430 ds
->ds_is_snapshot
= dsl_dataset_phys(ds
)->ds_num_children
!= 0;
431 list_link_init(&ds
->ds_synced_link
);
433 mutex_init(&ds
->ds_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
434 mutex_init(&ds
->ds_opening_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
435 mutex_init(&ds
->ds_sendstream_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
436 refcount_create(&ds
->ds_longholds
);
438 bplist_create(&ds
->ds_pending_deadlist
);
439 dsl_deadlist_open(&ds
->ds_deadlist
,
440 mos
, dsl_dataset_phys(ds
)->ds_deadlist_obj
);
442 list_create(&ds
->ds_sendstreams
, sizeof (dmu_sendarg_t
),
443 offsetof(dmu_sendarg_t
, dsa_link
));
445 list_create(&ds
->ds_prop_cbs
, sizeof (dsl_prop_cb_record_t
),
446 offsetof(dsl_prop_cb_record_t
, cbr_ds_node
));
448 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
451 for (f
= 0; f
< SPA_FEATURES
; f
++) {
452 if (!(spa_feature_table
[f
].fi_flags
&
453 ZFEATURE_FLAG_PER_DATASET
))
455 err
= zap_contains(mos
, dsobj
,
456 spa_feature_table
[f
].fi_guid
);
458 ds
->ds_feature_inuse
[f
] = B_TRUE
;
460 ASSERT3U(err
, ==, ENOENT
);
466 err
= dsl_dir_hold_obj(dp
,
467 dsl_dataset_phys(ds
)->ds_dir_obj
, NULL
, ds
, &ds
->ds_dir
);
469 mutex_destroy(&ds
->ds_lock
);
470 mutex_destroy(&ds
->ds_opening_lock
);
471 mutex_destroy(&ds
->ds_sendstream_lock
);
472 refcount_destroy(&ds
->ds_longholds
);
473 bplist_destroy(&ds
->ds_pending_deadlist
);
474 dsl_deadlist_close(&ds
->ds_deadlist
);
475 kmem_free(ds
, sizeof (dsl_dataset_t
));
476 dmu_buf_rele(dbuf
, tag
);
480 if (!ds
->ds_is_snapshot
) {
481 ds
->ds_snapname
[0] = '\0';
482 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
483 err
= dsl_dataset_hold_obj(dp
,
484 dsl_dataset_phys(ds
)->ds_prev_snap_obj
,
487 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
488 int zaperr
= zap_lookup(mos
, ds
->ds_object
,
489 DS_FIELD_BOOKMARK_NAMES
,
490 sizeof (ds
->ds_bookmarks
), 1,
492 if (zaperr
!= ENOENT
)
496 if (zfs_flags
& ZFS_DEBUG_SNAPNAMES
)
497 err
= dsl_dataset_get_snapname(ds
);
499 dsl_dataset_phys(ds
)->ds_userrefs_obj
!= 0) {
501 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
502 dsl_dataset_phys(ds
)->ds_userrefs_obj
,
507 if (err
== 0 && !ds
->ds_is_snapshot
) {
508 err
= dsl_prop_get_int_ds(ds
,
509 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
512 err
= dsl_prop_get_int_ds(ds
,
513 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
517 ds
->ds_reserved
= ds
->ds_quota
= 0;
520 dmu_buf_init_user(&ds
->ds_dbu
, dsl_dataset_evict
, &ds
->ds_dbuf
);
522 winner
= dmu_buf_set_user_ie(dbuf
, &ds
->ds_dbu
);
524 if (err
!= 0 || winner
!= NULL
) {
525 bplist_destroy(&ds
->ds_pending_deadlist
);
526 dsl_deadlist_close(&ds
->ds_deadlist
);
528 dsl_dataset_rele(ds
->ds_prev
, ds
);
529 dsl_dir_rele(ds
->ds_dir
, ds
);
530 mutex_destroy(&ds
->ds_lock
);
531 mutex_destroy(&ds
->ds_opening_lock
);
532 mutex_destroy(&ds
->ds_sendstream_lock
);
533 refcount_destroy(&ds
->ds_longholds
);
534 kmem_free(ds
, sizeof (dsl_dataset_t
));
536 dmu_buf_rele(dbuf
, tag
);
542 unique_insert(dsl_dataset_phys(ds
)->ds_fsid_guid
);
545 ASSERT3P(ds
->ds_dbuf
, ==, dbuf
);
546 ASSERT3P(dsl_dataset_phys(ds
), ==, dbuf
->db_data
);
547 ASSERT(dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0 ||
548 spa_version(dp
->dp_spa
) < SPA_VERSION_ORIGIN
||
549 dp
->dp_origin_snap
== NULL
|| ds
== dp
->dp_origin_snap
);
555 dsl_dataset_hold(dsl_pool_t
*dp
, const char *name
,
556 void *tag
, dsl_dataset_t
**dsp
)
559 const char *snapname
;
564 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, &snapname
);
568 ASSERT(dsl_pool_config_held(dp
));
569 obj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
571 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
573 err
= SET_ERROR(ENOENT
);
575 /* we may be looking for a snapshot */
576 if (err
== 0 && snapname
!= NULL
) {
577 dsl_dataset_t
*snap_ds
;
579 if (*snapname
++ != '@') {
580 dsl_dataset_rele(ds
, tag
);
581 dsl_dir_rele(dd
, FTAG
);
582 return (SET_ERROR(ENOENT
));
585 dprintf("looking for snapshot '%s'\n", snapname
);
586 err
= dsl_dataset_snap_lookup(ds
, snapname
, &obj
);
588 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &snap_ds
);
589 dsl_dataset_rele(ds
, tag
);
592 mutex_enter(&snap_ds
->ds_lock
);
593 if (snap_ds
->ds_snapname
[0] == 0)
594 (void) strlcpy(snap_ds
->ds_snapname
, snapname
,
595 sizeof (snap_ds
->ds_snapname
));
596 mutex_exit(&snap_ds
->ds_lock
);
602 dsl_dir_rele(dd
, FTAG
);
607 dsl_dataset_own_obj(dsl_pool_t
*dp
, uint64_t dsobj
,
608 void *tag
, dsl_dataset_t
**dsp
)
610 int err
= dsl_dataset_hold_obj(dp
, dsobj
, tag
, dsp
);
613 if (!dsl_dataset_tryown(*dsp
, tag
)) {
614 dsl_dataset_rele(*dsp
, tag
);
616 return (SET_ERROR(EBUSY
));
622 dsl_dataset_own(dsl_pool_t
*dp
, const char *name
,
623 void *tag
, dsl_dataset_t
**dsp
)
625 int err
= dsl_dataset_hold(dp
, name
, tag
, dsp
);
628 if (!dsl_dataset_tryown(*dsp
, tag
)) {
629 dsl_dataset_rele(*dsp
, tag
);
630 return (SET_ERROR(EBUSY
));
636 * See the comment above dsl_pool_hold() for details. In summary, a long
637 * hold is used to prevent destruction of a dataset while the pool hold
638 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
640 * The dataset and pool must be held when this function is called. After it
641 * is called, the pool hold may be released while the dataset is still held
645 dsl_dataset_long_hold(dsl_dataset_t
*ds
, void *tag
)
647 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
648 (void) refcount_add(&ds
->ds_longholds
, tag
);
652 dsl_dataset_long_rele(dsl_dataset_t
*ds
, void *tag
)
654 (void) refcount_remove(&ds
->ds_longholds
, tag
);
657 /* Return B_TRUE if there are any long holds on this dataset. */
659 dsl_dataset_long_held(dsl_dataset_t
*ds
)
661 return (!refcount_is_zero(&ds
->ds_longholds
));
665 dsl_dataset_name(dsl_dataset_t
*ds
, char *name
)
668 (void) strcpy(name
, "mos");
670 dsl_dir_name(ds
->ds_dir
, name
);
671 VERIFY0(dsl_dataset_get_snapname(ds
));
672 if (ds
->ds_snapname
[0]) {
673 VERIFY3U(strlcat(name
, "@", ZFS_MAX_DATASET_NAME_LEN
),
674 <, ZFS_MAX_DATASET_NAME_LEN
);
676 * We use a "recursive" mutex so that we
677 * can call dprintf_ds() with ds_lock held.
679 if (!MUTEX_HELD(&ds
->ds_lock
)) {
680 mutex_enter(&ds
->ds_lock
);
681 VERIFY3U(strlcat(name
, ds
->ds_snapname
,
682 ZFS_MAX_DATASET_NAME_LEN
), <,
683 ZFS_MAX_DATASET_NAME_LEN
);
684 mutex_exit(&ds
->ds_lock
);
686 VERIFY3U(strlcat(name
, ds
->ds_snapname
,
687 ZFS_MAX_DATASET_NAME_LEN
), <,
688 ZFS_MAX_DATASET_NAME_LEN
);
695 dsl_dataset_namelen(dsl_dataset_t
*ds
)
698 VERIFY0(dsl_dataset_get_snapname(ds
));
699 mutex_enter(&ds
->ds_lock
);
700 len
= dsl_dir_namelen(ds
->ds_dir
) + 1 + strlen(ds
->ds_snapname
);
701 mutex_exit(&ds
->ds_lock
);
706 dsl_dataset_rele(dsl_dataset_t
*ds
, void *tag
)
708 dmu_buf_rele(ds
->ds_dbuf
, tag
);
712 dsl_dataset_disown(dsl_dataset_t
*ds
, void *tag
)
714 ASSERT3P(ds
->ds_owner
, ==, tag
);
715 ASSERT(ds
->ds_dbuf
!= NULL
);
717 mutex_enter(&ds
->ds_lock
);
719 mutex_exit(&ds
->ds_lock
);
720 dsl_dataset_long_rele(ds
, tag
);
721 dsl_dataset_rele(ds
, tag
);
725 dsl_dataset_tryown(dsl_dataset_t
*ds
, void *tag
)
727 boolean_t gotit
= FALSE
;
729 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
730 mutex_enter(&ds
->ds_lock
);
731 if (ds
->ds_owner
== NULL
&& !DS_IS_INCONSISTENT(ds
)) {
733 dsl_dataset_long_hold(ds
, tag
);
736 mutex_exit(&ds
->ds_lock
);
741 dsl_dataset_has_owner(dsl_dataset_t
*ds
)
744 mutex_enter(&ds
->ds_lock
);
745 rv
= (ds
->ds_owner
!= NULL
);
746 mutex_exit(&ds
->ds_lock
);
751 dsl_dataset_activate_feature(uint64_t dsobj
, spa_feature_t f
, dmu_tx_t
*tx
)
753 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
754 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
757 VERIFY(spa_feature_table
[f
].fi_flags
& ZFEATURE_FLAG_PER_DATASET
);
759 spa_feature_incr(spa
, f
, tx
);
760 dmu_object_zapify(mos
, dsobj
, DMU_OT_DSL_DATASET
, tx
);
762 VERIFY0(zap_add(mos
, dsobj
, spa_feature_table
[f
].fi_guid
,
763 sizeof (zero
), 1, &zero
, tx
));
767 dsl_dataset_deactivate_feature(uint64_t dsobj
, spa_feature_t f
, dmu_tx_t
*tx
)
769 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
770 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
772 VERIFY(spa_feature_table
[f
].fi_flags
& ZFEATURE_FLAG_PER_DATASET
);
774 VERIFY0(zap_remove(mos
, dsobj
, spa_feature_table
[f
].fi_guid
, tx
));
775 spa_feature_decr(spa
, f
, tx
);
779 dsl_dataset_create_sync_dd(dsl_dir_t
*dd
, dsl_dataset_t
*origin
,
780 uint64_t flags
, dmu_tx_t
*tx
)
782 dsl_pool_t
*dp
= dd
->dd_pool
;
784 dsl_dataset_phys_t
*dsphys
;
786 objset_t
*mos
= dp
->dp_meta_objset
;
789 origin
= dp
->dp_origin_snap
;
791 ASSERT(origin
== NULL
|| origin
->ds_dir
->dd_pool
== dp
);
792 ASSERT(origin
== NULL
|| dsl_dataset_phys(origin
)->ds_num_children
> 0);
793 ASSERT(dmu_tx_is_syncing(tx
));
794 ASSERT(dsl_dir_phys(dd
)->dd_head_dataset_obj
== 0);
796 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
797 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
798 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
799 dmu_buf_will_dirty(dbuf
, tx
);
800 dsphys
= dbuf
->db_data
;
801 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
802 dsphys
->ds_dir_obj
= dd
->dd_object
;
803 dsphys
->ds_flags
= flags
;
804 dsphys
->ds_fsid_guid
= unique_create();
805 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
806 sizeof (dsphys
->ds_guid
));
807 dsphys
->ds_snapnames_zapobj
=
808 zap_create_norm(mos
, U8_TEXTPREP_TOUPPER
, DMU_OT_DSL_DS_SNAP_MAP
,
810 dsphys
->ds_creation_time
= gethrestime_sec();
811 dsphys
->ds_creation_txg
= tx
->tx_txg
== TXG_INITIAL
? 1 : tx
->tx_txg
;
813 if (origin
== NULL
) {
814 dsphys
->ds_deadlist_obj
= dsl_deadlist_alloc(mos
, tx
);
817 dsl_dataset_t
*ohds
; /* head of the origin snapshot */
819 dsphys
->ds_prev_snap_obj
= origin
->ds_object
;
820 dsphys
->ds_prev_snap_txg
=
821 dsl_dataset_phys(origin
)->ds_creation_txg
;
822 dsphys
->ds_referenced_bytes
=
823 dsl_dataset_phys(origin
)->ds_referenced_bytes
;
824 dsphys
->ds_compressed_bytes
=
825 dsl_dataset_phys(origin
)->ds_compressed_bytes
;
826 dsphys
->ds_uncompressed_bytes
=
827 dsl_dataset_phys(origin
)->ds_uncompressed_bytes
;
828 dsphys
->ds_bp
= dsl_dataset_phys(origin
)->ds_bp
;
831 * Inherit flags that describe the dataset's contents
832 * (INCONSISTENT) or properties (Case Insensitive).
834 dsphys
->ds_flags
|= dsl_dataset_phys(origin
)->ds_flags
&
835 (DS_FLAG_INCONSISTENT
| DS_FLAG_CI_DATASET
);
837 for (f
= 0; f
< SPA_FEATURES
; f
++) {
838 if (origin
->ds_feature_inuse
[f
])
839 dsl_dataset_activate_feature(dsobj
, f
, tx
);
842 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
843 dsl_dataset_phys(origin
)->ds_num_children
++;
845 VERIFY0(dsl_dataset_hold_obj(dp
,
846 dsl_dir_phys(origin
->ds_dir
)->dd_head_dataset_obj
,
848 dsphys
->ds_deadlist_obj
= dsl_deadlist_clone(&ohds
->ds_deadlist
,
849 dsphys
->ds_prev_snap_txg
, dsphys
->ds_prev_snap_obj
, tx
);
850 dsl_dataset_rele(ohds
, FTAG
);
852 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_NEXT_CLONES
) {
853 if (dsl_dataset_phys(origin
)->ds_next_clones_obj
== 0) {
854 dsl_dataset_phys(origin
)->ds_next_clones_obj
=
856 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
858 VERIFY0(zap_add_int(mos
,
859 dsl_dataset_phys(origin
)->ds_next_clones_obj
,
863 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
864 dsl_dir_phys(dd
)->dd_origin_obj
= origin
->ds_object
;
865 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
866 if (dsl_dir_phys(origin
->ds_dir
)->dd_clones
== 0) {
867 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
868 dsl_dir_phys(origin
->ds_dir
)->dd_clones
=
870 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
872 VERIFY0(zap_add_int(mos
,
873 dsl_dir_phys(origin
->ds_dir
)->dd_clones
,
878 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
879 dsphys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
881 dmu_buf_rele(dbuf
, FTAG
);
883 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
884 dsl_dir_phys(dd
)->dd_head_dataset_obj
= dsobj
;
890 dsl_dataset_zero_zil(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
894 VERIFY0(dmu_objset_from_ds(ds
, &os
));
895 bzero(&os
->os_zil_header
, sizeof (os
->os_zil_header
));
896 dsl_dataset_dirty(ds
, tx
);
900 dsl_dataset_create_sync(dsl_dir_t
*pdd
, const char *lastname
,
901 dsl_dataset_t
*origin
, uint64_t flags
, cred_t
*cr
, dmu_tx_t
*tx
)
903 dsl_pool_t
*dp
= pdd
->dd_pool
;
904 uint64_t dsobj
, ddobj
;
907 ASSERT(dmu_tx_is_syncing(tx
));
908 ASSERT(lastname
[0] != '@');
910 ddobj
= dsl_dir_create_sync(dp
, pdd
, lastname
, tx
);
911 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, lastname
, FTAG
, &dd
));
913 dsobj
= dsl_dataset_create_sync_dd(dd
, origin
,
914 flags
& ~DS_CREATE_FLAG_NODIRTY
, tx
);
916 dsl_deleg_set_create_perms(dd
, tx
, cr
);
919 * Since we're creating a new node we know it's a leaf, so we can
920 * initialize the counts if the limit feature is active.
922 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
924 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
926 dsl_dir_zapify(dd
, tx
);
927 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
928 sizeof (cnt
), 1, &cnt
, tx
));
929 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
930 sizeof (cnt
), 1, &cnt
, tx
));
933 dsl_dir_rele(dd
, FTAG
);
936 * If we are creating a clone, make sure we zero out any stale
937 * data from the origin snapshots zil header.
939 if (origin
!= NULL
&& !(flags
& DS_CREATE_FLAG_NODIRTY
)) {
942 VERIFY0(dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
943 dsl_dataset_zero_zil(ds
, tx
);
944 dsl_dataset_rele(ds
, FTAG
);
951 * The unique space in the head dataset can be calculated by subtracting
952 * the space used in the most recent snapshot, that is still being used
953 * in this file system, from the space currently in use. To figure out
954 * the space in the most recent snapshot still in use, we need to take
955 * the total space used in the snapshot and subtract out the space that
956 * has been freed up since the snapshot was taken.
959 dsl_dataset_recalc_head_uniq(dsl_dataset_t
*ds
)
962 uint64_t dlused
, dlcomp
, dluncomp
;
964 ASSERT(!ds
->ds_is_snapshot
);
966 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0)
967 mrs_used
= dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
;
971 dsl_deadlist_space(&ds
->ds_deadlist
, &dlused
, &dlcomp
, &dluncomp
);
973 ASSERT3U(dlused
, <=, mrs_used
);
974 dsl_dataset_phys(ds
)->ds_unique_bytes
=
975 dsl_dataset_phys(ds
)->ds_referenced_bytes
- (mrs_used
- dlused
);
977 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) >=
978 SPA_VERSION_UNIQUE_ACCURATE
)
979 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
983 dsl_dataset_remove_from_next_clones(dsl_dataset_t
*ds
, uint64_t obj
,
986 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
988 ASSERTV(uint64_t count
);
990 ASSERT(dsl_dataset_phys(ds
)->ds_num_children
>= 2);
991 err
= zap_remove_int(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
994 * The err should not be ENOENT, but a bug in a previous version
995 * of the code could cause upgrade_clones_cb() to not set
996 * ds_next_snap_obj when it should, leading to a missing entry.
997 * If we knew that the pool was created after
998 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
999 * ENOENT. However, at least we can check that we don't have
1000 * too many entries in the next_clones_obj even after failing to
1005 ASSERT0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1007 ASSERT3U(count
, <=, dsl_dataset_phys(ds
)->ds_num_children
- 2);
1012 dsl_dataset_get_blkptr(dsl_dataset_t
*ds
)
1014 return (&dsl_dataset_phys(ds
)->ds_bp
);
1018 dsl_dataset_set_blkptr(dsl_dataset_t
*ds
, blkptr_t
*bp
, dmu_tx_t
*tx
)
1020 ASSERT(dmu_tx_is_syncing(tx
));
1021 /* If it's the meta-objset, set dp_meta_rootbp */
1023 tx
->tx_pool
->dp_meta_rootbp
= *bp
;
1025 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1026 dsl_dataset_phys(ds
)->ds_bp
= *bp
;
1031 dsl_dataset_get_spa(dsl_dataset_t
*ds
)
1033 return (ds
->ds_dir
->dd_pool
->dp_spa
);
1037 dsl_dataset_dirty(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1041 if (ds
== NULL
) /* this is the meta-objset */
1044 ASSERT(ds
->ds_objset
!= NULL
);
1046 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
!= 0)
1047 panic("dirtying snapshot!");
1049 dp
= ds
->ds_dir
->dd_pool
;
1051 if (txg_list_add(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
1052 /* up the hold count until we can be written out */
1053 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
1058 dsl_dataset_is_dirty(dsl_dataset_t
*ds
)
1062 for (t
= 0; t
< TXG_SIZE
; t
++) {
1063 if (txg_list_member(&ds
->ds_dir
->dd_pool
->dp_dirty_datasets
,
1071 dsl_dataset_snapshot_reserve_space(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1075 if (!dmu_tx_is_syncing(tx
))
1079 * If there's an fs-only reservation, any blocks that might become
1080 * owned by the snapshot dataset must be accommodated by space
1081 * outside of the reservation.
1083 ASSERT(ds
->ds_reserved
== 0 || DS_UNIQUE_IS_ACCURATE(ds
));
1084 asize
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
, ds
->ds_reserved
);
1085 if (asize
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
1086 return (SET_ERROR(ENOSPC
));
1089 * Propagate any reserved space for this snapshot to other
1090 * snapshot checks in this sync group.
1093 dsl_dir_willuse_space(ds
->ds_dir
, asize
, tx
);
1098 typedef struct dsl_dataset_snapshot_arg
{
1099 nvlist_t
*ddsa_snaps
;
1100 nvlist_t
*ddsa_props
;
1101 nvlist_t
*ddsa_errors
;
1103 } dsl_dataset_snapshot_arg_t
;
1106 dsl_dataset_snapshot_check_impl(dsl_dataset_t
*ds
, const char *snapname
,
1107 dmu_tx_t
*tx
, boolean_t recv
, uint64_t cnt
, cred_t
*cr
)
1112 ds
->ds_trysnap_txg
= tx
->tx_txg
;
1114 if (!dmu_tx_is_syncing(tx
))
1118 * We don't allow multiple snapshots of the same txg. If there
1119 * is already one, try again.
1121 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
>= tx
->tx_txg
)
1122 return (SET_ERROR(EAGAIN
));
1125 * Check for conflicting snapshot name.
1127 error
= dsl_dataset_snap_lookup(ds
, snapname
, &value
);
1129 return (SET_ERROR(EEXIST
));
1130 if (error
!= ENOENT
)
1134 * We don't allow taking snapshots of inconsistent datasets, such as
1135 * those into which we are currently receiving. However, if we are
1136 * creating this snapshot as part of a receive, this check will be
1137 * executed atomically with respect to the completion of the receive
1138 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1139 * case we ignore this, knowing it will be fixed up for us shortly in
1140 * dmu_recv_end_sync().
1142 if (!recv
&& DS_IS_INCONSISTENT(ds
))
1143 return (SET_ERROR(EBUSY
));
1146 * Skip the check for temporary snapshots or if we have already checked
1147 * the counts in dsl_dataset_snapshot_check. This means we really only
1148 * check the count here when we're receiving a stream.
1150 if (cnt
!= 0 && cr
!= NULL
) {
1151 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1152 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, cr
);
1157 error
= dsl_dataset_snapshot_reserve_space(ds
, tx
);
1165 dsl_dataset_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1167 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1168 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1173 * Pre-compute how many total new snapshots will be created for each
1174 * level in the tree and below. This is needed for validating the
1175 * snapshot limit when either taking a recursive snapshot or when
1176 * taking multiple snapshots.
1178 * The problem is that the counts are not actually adjusted when
1179 * we are checking, only when we finally sync. For a single snapshot,
1180 * this is easy, the count will increase by 1 at each node up the tree,
1181 * but its more complicated for the recursive/multiple snapshot case.
1183 * The dsl_fs_ss_limit_check function does recursively check the count
1184 * at each level up the tree but since it is validating each snapshot
1185 * independently we need to be sure that we are validating the complete
1186 * count for the entire set of snapshots. We do this by rolling up the
1187 * counts for each component of the name into an nvlist and then
1188 * checking each of those cases with the aggregated count.
1190 * This approach properly handles not only the recursive snapshot
1191 * case (where we get all of those on the ddsa_snaps list) but also
1192 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1193 * validate the limit on 'a' using a count of 2).
1195 * We validate the snapshot names in the third loop and only report
1198 if (dmu_tx_is_syncing(tx
)) {
1200 nvlist_t
*cnt_track
= NULL
;
1201 cnt_track
= fnvlist_alloc();
1203 nm
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1205 /* Rollup aggregated counts into the cnt_track list */
1206 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1208 pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1212 (void) strlcpy(nm
, nvpair_name(pair
), MAXPATHLEN
);
1213 pdelim
= strchr(nm
, '@');
1219 if (nvlist_lookup_uint64(cnt_track
, nm
,
1221 /* update existing entry */
1222 fnvlist_add_uint64(cnt_track
, nm
,
1226 fnvlist_add_uint64(cnt_track
, nm
, 1);
1229 pdelim
= strrchr(nm
, '/');
1232 } while (pdelim
!= NULL
);
1235 kmem_free(nm
, MAXPATHLEN
);
1237 /* Check aggregated counts at each level */
1238 for (pair
= nvlist_next_nvpair(cnt_track
, NULL
);
1239 pair
!= NULL
; pair
= nvlist_next_nvpair(cnt_track
, pair
)) {
1245 name
= nvpair_name(pair
);
1246 cnt
= fnvpair_value_uint64(pair
);
1249 error
= dsl_dataset_hold(dp
, name
, FTAG
, &ds
);
1251 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1252 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
,
1254 dsl_dataset_rele(ds
, FTAG
);
1258 if (ddsa
->ddsa_errors
!= NULL
)
1259 fnvlist_add_int32(ddsa
->ddsa_errors
,
1262 /* only report one error for this check */
1266 nvlist_free(cnt_track
);
1269 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1270 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1273 char *name
, *atp
= NULL
;
1274 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
1276 name
= nvpair_name(pair
);
1277 if (strlen(name
) >= ZFS_MAX_DATASET_NAME_LEN
)
1278 error
= SET_ERROR(ENAMETOOLONG
);
1280 atp
= strchr(name
, '@');
1282 error
= SET_ERROR(EINVAL
);
1284 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1287 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
1289 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1290 error
= dsl_dataset_snapshot_check_impl(ds
,
1291 atp
+ 1, tx
, B_FALSE
, 0, NULL
);
1292 dsl_dataset_rele(ds
, FTAG
);
1296 if (ddsa
->ddsa_errors
!= NULL
) {
1297 fnvlist_add_int32(ddsa
->ddsa_errors
,
1308 dsl_dataset_snapshot_sync_impl(dsl_dataset_t
*ds
, const char *snapname
,
1311 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1313 dsl_dataset_phys_t
*dsphys
;
1314 uint64_t dsobj
, crtxg
;
1315 objset_t
*mos
= dp
->dp_meta_objset
;
1317 ASSERTV(static zil_header_t zero_zil
);
1318 ASSERTV(objset_t
*os
);
1320 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
1323 * If we are on an old pool, the zil must not be active, in which
1324 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1326 ASSERT(spa_version(dmu_tx_pool(tx
)->dp_spa
) >= SPA_VERSION_FAST_SNAP
||
1327 dmu_objset_from_ds(ds
, &os
) != 0 ||
1328 bcmp(&os
->os_phys
->os_zil_header
, &zero_zil
,
1329 sizeof (zero_zil
)) == 0);
1331 dsl_fs_ss_count_adjust(ds
->ds_dir
, 1, DD_FIELD_SNAPSHOT_COUNT
, tx
);
1334 * The origin's ds_creation_txg has to be < TXG_INITIAL
1336 if (strcmp(snapname
, ORIGIN_DIR_NAME
) == 0)
1341 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
1342 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
1343 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
1344 dmu_buf_will_dirty(dbuf
, tx
);
1345 dsphys
= dbuf
->db_data
;
1346 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
1347 dsphys
->ds_dir_obj
= ds
->ds_dir
->dd_object
;
1348 dsphys
->ds_fsid_guid
= unique_create();
1349 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
1350 sizeof (dsphys
->ds_guid
));
1351 dsphys
->ds_prev_snap_obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1352 dsphys
->ds_prev_snap_txg
= dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
1353 dsphys
->ds_next_snap_obj
= ds
->ds_object
;
1354 dsphys
->ds_num_children
= 1;
1355 dsphys
->ds_creation_time
= gethrestime_sec();
1356 dsphys
->ds_creation_txg
= crtxg
;
1357 dsphys
->ds_deadlist_obj
= dsl_dataset_phys(ds
)->ds_deadlist_obj
;
1358 dsphys
->ds_referenced_bytes
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1359 dsphys
->ds_compressed_bytes
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1360 dsphys
->ds_uncompressed_bytes
=
1361 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1362 dsphys
->ds_flags
= dsl_dataset_phys(ds
)->ds_flags
;
1363 dsphys
->ds_bp
= dsl_dataset_phys(ds
)->ds_bp
;
1364 dmu_buf_rele(dbuf
, FTAG
);
1366 for (f
= 0; f
< SPA_FEATURES
; f
++) {
1367 if (ds
->ds_feature_inuse
[f
])
1368 dsl_dataset_activate_feature(dsobj
, f
, tx
);
1371 ASSERT3U(ds
->ds_prev
!= 0, ==,
1372 dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0);
1374 uint64_t next_clones_obj
=
1375 dsl_dataset_phys(ds
->ds_prev
)->ds_next_clones_obj
;
1376 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1378 dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 1);
1379 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1381 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1382 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, ==,
1383 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_txg
);
1384 dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
= dsobj
;
1385 } else if (next_clones_obj
!= 0) {
1386 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
1387 dsphys
->ds_next_snap_obj
, tx
);
1388 VERIFY0(zap_add_int(mos
,
1389 next_clones_obj
, dsobj
, tx
));
1394 * If we have a reference-reservation on this dataset, we will
1395 * need to increase the amount of refreservation being charged
1396 * since our unique space is going to zero.
1398 if (ds
->ds_reserved
) {
1400 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
1401 delta
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
,
1403 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
,
1407 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1408 dsl_dataset_phys(ds
)->ds_deadlist_obj
=
1409 dsl_deadlist_clone(&ds
->ds_deadlist
, UINT64_MAX
,
1410 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, tx
);
1411 dsl_deadlist_close(&ds
->ds_deadlist
);
1412 dsl_deadlist_open(&ds
->ds_deadlist
, mos
,
1413 dsl_dataset_phys(ds
)->ds_deadlist_obj
);
1414 dsl_deadlist_add_key(&ds
->ds_deadlist
,
1415 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
);
1417 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, <, tx
->tx_txg
);
1418 dsl_dataset_phys(ds
)->ds_prev_snap_obj
= dsobj
;
1419 dsl_dataset_phys(ds
)->ds_prev_snap_txg
= crtxg
;
1420 dsl_dataset_phys(ds
)->ds_unique_bytes
= 0;
1421 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
1422 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1424 VERIFY0(zap_add(mos
, dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1425 snapname
, 8, 1, &dsobj
, tx
));
1428 dsl_dataset_rele(ds
->ds_prev
, ds
);
1429 VERIFY0(dsl_dataset_hold_obj(dp
,
1430 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, ds
, &ds
->ds_prev
));
1432 dsl_scan_ds_snapshotted(ds
, tx
);
1434 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
1436 spa_history_log_internal_ds(ds
->ds_prev
, "snapshot", tx
, "");
1440 dsl_dataset_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1442 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1443 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1446 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1447 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1450 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
1452 name
= nvpair_name(pair
);
1453 atp
= strchr(name
, '@');
1454 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1455 VERIFY0(dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
));
1457 dsl_dataset_snapshot_sync_impl(ds
, atp
+ 1, tx
);
1458 if (ddsa
->ddsa_props
!= NULL
) {
1459 dsl_props_set_sync_impl(ds
->ds_prev
,
1460 ZPROP_SRC_LOCAL
, ddsa
->ddsa_props
, tx
);
1462 zvol_create_minors(dp
->dp_spa
, nvpair_name(pair
), B_TRUE
);
1463 dsl_dataset_rele(ds
, FTAG
);
1468 * The snapshots must all be in the same pool.
1469 * All-or-nothing: if there are any failures, nothing will be modified.
1472 dsl_dataset_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
*errors
)
1474 dsl_dataset_snapshot_arg_t ddsa
;
1476 boolean_t needsuspend
;
1480 nvlist_t
*suspended
= NULL
;
1482 pair
= nvlist_next_nvpair(snaps
, NULL
);
1485 firstname
= nvpair_name(pair
);
1487 error
= spa_open(firstname
, &spa
, FTAG
);
1490 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1491 spa_close(spa
, FTAG
);
1494 suspended
= fnvlist_alloc();
1495 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1496 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1497 char fsname
[ZFS_MAX_DATASET_NAME_LEN
];
1498 char *snapname
= nvpair_name(pair
);
1502 atp
= strchr(snapname
, '@');
1504 error
= SET_ERROR(EINVAL
);
1507 (void) strlcpy(fsname
, snapname
, atp
- snapname
+ 1);
1509 error
= zil_suspend(fsname
, &cookie
);
1512 fnvlist_add_uint64(suspended
, fsname
,
1517 ddsa
.ddsa_snaps
= snaps
;
1518 ddsa
.ddsa_props
= props
;
1519 ddsa
.ddsa_errors
= errors
;
1520 ddsa
.ddsa_cr
= CRED();
1523 error
= dsl_sync_task(firstname
, dsl_dataset_snapshot_check
,
1524 dsl_dataset_snapshot_sync
, &ddsa
,
1525 fnvlist_num_pairs(snaps
) * 3, ZFS_SPACE_CHECK_NORMAL
);
1528 if (suspended
!= NULL
) {
1529 for (pair
= nvlist_next_nvpair(suspended
, NULL
); pair
!= NULL
;
1530 pair
= nvlist_next_nvpair(suspended
, pair
)) {
1531 zil_resume((void *)(uintptr_t)
1532 fnvpair_value_uint64(pair
));
1534 fnvlist_free(suspended
);
1540 typedef struct dsl_dataset_snapshot_tmp_arg
{
1541 const char *ddsta_fsname
;
1542 const char *ddsta_snapname
;
1543 minor_t ddsta_cleanup_minor
;
1544 const char *ddsta_htag
;
1545 } dsl_dataset_snapshot_tmp_arg_t
;
1548 dsl_dataset_snapshot_tmp_check(void *arg
, dmu_tx_t
*tx
)
1550 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1551 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1555 error
= dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
);
1559 /* NULL cred means no limit check for tmp snapshot */
1560 error
= dsl_dataset_snapshot_check_impl(ds
, ddsta
->ddsta_snapname
,
1561 tx
, B_FALSE
, 0, NULL
);
1563 dsl_dataset_rele(ds
, FTAG
);
1567 if (spa_version(dp
->dp_spa
) < SPA_VERSION_USERREFS
) {
1568 dsl_dataset_rele(ds
, FTAG
);
1569 return (SET_ERROR(ENOTSUP
));
1571 error
= dsl_dataset_user_hold_check_one(NULL
, ddsta
->ddsta_htag
,
1574 dsl_dataset_rele(ds
, FTAG
);
1578 dsl_dataset_rele(ds
, FTAG
);
1583 dsl_dataset_snapshot_tmp_sync(void *arg
, dmu_tx_t
*tx
)
1585 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1586 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1587 dsl_dataset_t
*ds
= NULL
;
1589 VERIFY0(dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
));
1591 dsl_dataset_snapshot_sync_impl(ds
, ddsta
->ddsta_snapname
, tx
);
1592 dsl_dataset_user_hold_sync_one(ds
->ds_prev
, ddsta
->ddsta_htag
,
1593 ddsta
->ddsta_cleanup_minor
, gethrestime_sec(), tx
);
1594 dsl_destroy_snapshot_sync_impl(ds
->ds_prev
, B_TRUE
, tx
);
1596 dsl_dataset_rele(ds
, FTAG
);
1600 dsl_dataset_snapshot_tmp(const char *fsname
, const char *snapname
,
1601 minor_t cleanup_minor
, const char *htag
)
1603 dsl_dataset_snapshot_tmp_arg_t ddsta
;
1606 boolean_t needsuspend
;
1609 ddsta
.ddsta_fsname
= fsname
;
1610 ddsta
.ddsta_snapname
= snapname
;
1611 ddsta
.ddsta_cleanup_minor
= cleanup_minor
;
1612 ddsta
.ddsta_htag
= htag
;
1614 error
= spa_open(fsname
, &spa
, FTAG
);
1617 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1618 spa_close(spa
, FTAG
);
1621 error
= zil_suspend(fsname
, &cookie
);
1626 error
= dsl_sync_task(fsname
, dsl_dataset_snapshot_tmp_check
,
1627 dsl_dataset_snapshot_tmp_sync
, &ddsta
, 3, ZFS_SPACE_CHECK_RESERVED
);
1636 dsl_dataset_sync(dsl_dataset_t
*ds
, zio_t
*zio
, dmu_tx_t
*tx
)
1640 ASSERT(dmu_tx_is_syncing(tx
));
1641 ASSERT(ds
->ds_objset
!= NULL
);
1642 ASSERT(dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0);
1645 * in case we had to change ds_fsid_guid when we opened it,
1648 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1649 dsl_dataset_phys(ds
)->ds_fsid_guid
= ds
->ds_fsid_guid
;
1651 if (ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
] != 0) {
1652 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1653 ds
->ds_object
, DS_FIELD_RESUME_OBJECT
, 8, 1,
1654 &ds
->ds_resume_object
[tx
->tx_txg
& TXG_MASK
], tx
));
1655 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1656 ds
->ds_object
, DS_FIELD_RESUME_OFFSET
, 8, 1,
1657 &ds
->ds_resume_offset
[tx
->tx_txg
& TXG_MASK
], tx
));
1658 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1659 ds
->ds_object
, DS_FIELD_RESUME_BYTES
, 8, 1,
1660 &ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
], tx
));
1661 ds
->ds_resume_object
[tx
->tx_txg
& TXG_MASK
] = 0;
1662 ds
->ds_resume_offset
[tx
->tx_txg
& TXG_MASK
] = 0;
1663 ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
] = 0;
1666 dmu_objset_sync(ds
->ds_objset
, zio
, tx
);
1668 for (f
= 0; f
< SPA_FEATURES
; f
++) {
1669 if (ds
->ds_feature_activation_needed
[f
]) {
1670 if (ds
->ds_feature_inuse
[f
])
1672 dsl_dataset_activate_feature(ds
->ds_object
, f
, tx
);
1673 ds
->ds_feature_inuse
[f
] = B_TRUE
;
1679 get_clones_stat(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1682 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1685 nvlist_t
*propval
= fnvlist_alloc();
1686 nvlist_t
*val
= fnvlist_alloc();
1688 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1691 * There may be missing entries in ds_next_clones_obj
1692 * due to a bug in a previous version of the code.
1693 * Only trust it if it has the right number of entries.
1695 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
!= 0) {
1696 VERIFY0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1699 if (count
!= dsl_dataset_phys(ds
)->ds_num_children
- 1)
1701 for (zap_cursor_init(&zc
, mos
,
1702 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
1703 zap_cursor_retrieve(&zc
, &za
) == 0;
1704 zap_cursor_advance(&zc
)) {
1705 dsl_dataset_t
*clone
;
1706 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1707 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
1708 za
.za_first_integer
, FTAG
, &clone
));
1709 dsl_dir_name(clone
->ds_dir
, buf
);
1710 fnvlist_add_boolean(val
, buf
);
1711 dsl_dataset_rele(clone
, FTAG
);
1713 zap_cursor_fini(&zc
);
1714 fnvlist_add_nvlist(propval
, ZPROP_VALUE
, val
);
1715 fnvlist_add_nvlist(nv
, zfs_prop_to_name(ZFS_PROP_CLONES
), propval
);
1718 nvlist_free(propval
);
1722 get_receive_resume_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1724 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1726 if (dsl_dataset_has_resume_receive_state(ds
)) {
1729 uint8_t *compressed
;
1731 nvlist_t
*token_nv
= fnvlist_alloc();
1732 size_t packed_size
, compressed_size
;
1735 char buf
[MAXNAMELEN
];
1738 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1739 DS_FIELD_RESUME_FROMGUID
, sizeof (val
), 1, &val
) == 0) {
1740 fnvlist_add_uint64(token_nv
, "fromguid", val
);
1742 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1743 DS_FIELD_RESUME_OBJECT
, sizeof (val
), 1, &val
) == 0) {
1744 fnvlist_add_uint64(token_nv
, "object", val
);
1746 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1747 DS_FIELD_RESUME_OFFSET
, sizeof (val
), 1, &val
) == 0) {
1748 fnvlist_add_uint64(token_nv
, "offset", val
);
1750 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1751 DS_FIELD_RESUME_BYTES
, sizeof (val
), 1, &val
) == 0) {
1752 fnvlist_add_uint64(token_nv
, "bytes", val
);
1754 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1755 DS_FIELD_RESUME_TOGUID
, sizeof (val
), 1, &val
) == 0) {
1756 fnvlist_add_uint64(token_nv
, "toguid", val
);
1758 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1759 DS_FIELD_RESUME_TONAME
, 1, sizeof (buf
), buf
) == 0) {
1760 fnvlist_add_string(token_nv
, "toname", buf
);
1762 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1763 DS_FIELD_RESUME_LARGEBLOCK
) == 0) {
1764 fnvlist_add_boolean(token_nv
, "largeblockok");
1766 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1767 DS_FIELD_RESUME_EMBEDOK
) == 0) {
1768 fnvlist_add_boolean(token_nv
, "embedok");
1770 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1771 DS_FIELD_RESUME_COMPRESSOK
) == 0) {
1772 fnvlist_add_boolean(token_nv
, "compressok");
1774 packed
= fnvlist_pack(token_nv
, &packed_size
);
1775 fnvlist_free(token_nv
);
1776 compressed
= kmem_alloc(packed_size
, KM_SLEEP
);
1778 compressed_size
= gzip_compress(packed
, compressed
,
1779 packed_size
, packed_size
, 6);
1781 fletcher_4_native_varsize(compressed
, compressed_size
, &cksum
);
1783 str
= kmem_alloc(compressed_size
* 2 + 1, KM_SLEEP
);
1784 for (i
= 0; i
< compressed_size
; i
++) {
1785 (void) sprintf(str
+ i
* 2, "%02x", compressed
[i
]);
1787 str
[compressed_size
* 2] = '\0';
1788 propval
= kmem_asprintf("%u-%llx-%llx-%s",
1789 ZFS_SEND_RESUME_TOKEN_VERSION
,
1790 (longlong_t
)cksum
.zc_word
[0],
1791 (longlong_t
)packed_size
, str
);
1792 dsl_prop_nvlist_add_string(nv
,
1793 ZFS_PROP_RECEIVE_RESUME_TOKEN
, propval
);
1794 kmem_free(packed
, packed_size
);
1795 kmem_free(str
, compressed_size
* 2 + 1);
1796 kmem_free(compressed
, packed_size
);
1802 dsl_dataset_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1804 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1805 uint64_t refd
, avail
, uobjs
, aobjs
, ratio
;
1807 ASSERT(dsl_pool_config_held(dp
));
1809 ratio
= dsl_dataset_phys(ds
)->ds_compressed_bytes
== 0 ? 100 :
1810 (dsl_dataset_phys(ds
)->ds_uncompressed_bytes
* 100 /
1811 dsl_dataset_phys(ds
)->ds_compressed_bytes
);
1813 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRATIO
, ratio
);
1814 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALREFERENCED
,
1815 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
);
1817 if (ds
->ds_is_snapshot
) {
1818 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
, ratio
);
1819 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
1820 dsl_dataset_phys(ds
)->ds_unique_bytes
);
1821 get_clones_stat(ds
, nv
);
1823 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
!= dp
->dp_origin_snap
) {
1824 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1825 dsl_dataset_name(ds
->ds_prev
, buf
);
1826 dsl_prop_nvlist_add_string(nv
, ZFS_PROP_PREV_SNAP
, buf
);
1829 dsl_dir_stats(ds
->ds_dir
, nv
);
1832 dsl_dataset_space(ds
, &refd
, &avail
, &uobjs
, &aobjs
);
1833 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_AVAILABLE
, avail
);
1834 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFERENCED
, refd
);
1836 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATION
,
1837 dsl_dataset_phys(ds
)->ds_creation_time
);
1838 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATETXG
,
1839 dsl_dataset_phys(ds
)->ds_creation_txg
);
1840 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFQUOTA
,
1842 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRESERVATION
,
1844 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_GUID
,
1845 dsl_dataset_phys(ds
)->ds_guid
);
1846 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_UNIQUE
,
1847 dsl_dataset_phys(ds
)->ds_unique_bytes
);
1848 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_OBJSETID
,
1850 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERREFS
,
1852 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_DEFER_DESTROY
,
1853 DS_IS_DEFER_DESTROY(ds
) ? 1 : 0);
1855 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
1856 uint64_t written
, comp
, uncomp
;
1857 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1858 dsl_dataset_t
*prev
;
1861 err
= dsl_dataset_hold_obj(dp
,
1862 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
1864 err
= dsl_dataset_space_written(prev
, ds
, &written
,
1866 dsl_dataset_rele(prev
, FTAG
);
1868 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_WRITTEN
,
1874 if (!dsl_dataset_is_snapshot(ds
)) {
1875 /* 6 extra bytes for /%recv */
1876 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1877 dsl_dataset_t
*recv_ds
;
1880 * A failed "newfs" (e.g. full) resumable receive leaves
1881 * the stats set on this dataset. Check here for the prop.
1883 get_receive_resume_stats(ds
, nv
);
1886 * A failed incremental resumable receive leaves the
1887 * stats set on our child named "%recv". Check the child
1890 dsl_dataset_name(ds
, recvname
);
1891 if (strlcat(recvname
, "/", sizeof (recvname
)) <
1892 sizeof (recvname
) &&
1893 strlcat(recvname
, recv_clone_name
, sizeof (recvname
)) <
1894 sizeof (recvname
) &&
1895 dsl_dataset_hold(dp
, recvname
, FTAG
, &recv_ds
) == 0) {
1896 get_receive_resume_stats(recv_ds
, nv
);
1897 dsl_dataset_rele(recv_ds
, FTAG
);
1903 dsl_dataset_fast_stat(dsl_dataset_t
*ds
, dmu_objset_stats_t
*stat
)
1905 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1906 ASSERT(dsl_pool_config_held(dp
));
1908 stat
->dds_creation_txg
= dsl_dataset_phys(ds
)->ds_creation_txg
;
1909 stat
->dds_inconsistent
=
1910 dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_INCONSISTENT
;
1911 stat
->dds_guid
= dsl_dataset_phys(ds
)->ds_guid
;
1912 stat
->dds_origin
[0] = '\0';
1913 if (ds
->ds_is_snapshot
) {
1914 stat
->dds_is_snapshot
= B_TRUE
;
1915 stat
->dds_num_clones
=
1916 dsl_dataset_phys(ds
)->ds_num_children
- 1;
1918 stat
->dds_is_snapshot
= B_FALSE
;
1919 stat
->dds_num_clones
= 0;
1921 if (dsl_dir_is_clone(ds
->ds_dir
)) {
1924 VERIFY0(dsl_dataset_hold_obj(dp
,
1925 dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
,
1927 dsl_dataset_name(ods
, stat
->dds_origin
);
1928 dsl_dataset_rele(ods
, FTAG
);
1934 dsl_dataset_fsid_guid(dsl_dataset_t
*ds
)
1936 return (ds
->ds_fsid_guid
);
1940 dsl_dataset_space(dsl_dataset_t
*ds
,
1941 uint64_t *refdbytesp
, uint64_t *availbytesp
,
1942 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1944 *refdbytesp
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1945 *availbytesp
= dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
);
1946 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
)
1948 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
;
1949 if (ds
->ds_quota
!= 0) {
1951 * Adjust available bytes according to refquota
1953 if (*refdbytesp
< ds
->ds_quota
)
1954 *availbytesp
= MIN(*availbytesp
,
1955 ds
->ds_quota
- *refdbytesp
);
1959 *usedobjsp
= BP_GET_FILL(&dsl_dataset_phys(ds
)->ds_bp
);
1960 *availobjsp
= DN_MAX_OBJECT
- *usedobjsp
;
1964 dsl_dataset_modified_since_snap(dsl_dataset_t
*ds
, dsl_dataset_t
*snap
)
1966 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1969 if (dsl_dataset_phys(ds
)->ds_bp
.blk_birth
>
1970 dsl_dataset_phys(snap
)->ds_creation_txg
) {
1971 objset_t
*os
, *os_snap
;
1973 * It may be that only the ZIL differs, because it was
1974 * reset in the head. Don't count that as being
1977 if (dmu_objset_from_ds(ds
, &os
) != 0)
1979 if (dmu_objset_from_ds(snap
, &os_snap
) != 0)
1981 return (bcmp(&os
->os_phys
->os_meta_dnode
,
1982 &os_snap
->os_phys
->os_meta_dnode
,
1983 sizeof (os
->os_phys
->os_meta_dnode
)) != 0);
1988 typedef struct dsl_dataset_rename_snapshot_arg
{
1989 const char *ddrsa_fsname
;
1990 const char *ddrsa_oldsnapname
;
1991 const char *ddrsa_newsnapname
;
1992 boolean_t ddrsa_recursive
;
1994 } dsl_dataset_rename_snapshot_arg_t
;
1998 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t
*dp
,
1999 dsl_dataset_t
*hds
, void *arg
)
2001 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2005 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
2007 /* ignore nonexistent snapshots */
2008 return (error
== ENOENT
? 0 : error
);
2011 /* new name should not exist */
2012 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_newsnapname
, &val
);
2014 error
= SET_ERROR(EEXIST
);
2015 else if (error
== ENOENT
)
2018 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2019 if (dsl_dir_namelen(hds
->ds_dir
) + 1 +
2020 strlen(ddrsa
->ddrsa_newsnapname
) >= ZFS_MAX_DATASET_NAME_LEN
)
2021 error
= SET_ERROR(ENAMETOOLONG
);
2027 dsl_dataset_rename_snapshot_check(void *arg
, dmu_tx_t
*tx
)
2029 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2030 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2034 error
= dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
);
2038 if (ddrsa
->ddrsa_recursive
) {
2039 error
= dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
2040 dsl_dataset_rename_snapshot_check_impl
, ddrsa
,
2043 error
= dsl_dataset_rename_snapshot_check_impl(dp
, hds
, ddrsa
);
2045 dsl_dataset_rele(hds
, FTAG
);
2050 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t
*dp
,
2051 dsl_dataset_t
*hds
, void *arg
)
2053 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2056 dmu_tx_t
*tx
= ddrsa
->ddrsa_tx
;
2059 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
2060 ASSERT(error
== 0 || error
== ENOENT
);
2061 if (error
== ENOENT
) {
2062 /* ignore nonexistent snapshots */
2066 VERIFY0(dsl_dataset_hold_obj(dp
, val
, FTAG
, &ds
));
2068 /* log before we change the name */
2069 spa_history_log_internal_ds(ds
, "rename", tx
,
2070 "-> @%s", ddrsa
->ddrsa_newsnapname
);
2072 VERIFY0(dsl_dataset_snap_remove(hds
, ddrsa
->ddrsa_oldsnapname
, tx
,
2074 mutex_enter(&ds
->ds_lock
);
2075 (void) strcpy(ds
->ds_snapname
, ddrsa
->ddrsa_newsnapname
);
2076 mutex_exit(&ds
->ds_lock
);
2077 VERIFY0(zap_add(dp
->dp_meta_objset
,
2078 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
,
2079 ds
->ds_snapname
, 8, 1, &ds
->ds_object
, tx
));
2080 zvol_rename_minors(dp
->dp_spa
, ddrsa
->ddrsa_oldsnapname
,
2081 ddrsa
->ddrsa_newsnapname
, B_TRUE
);
2083 dsl_dataset_rele(ds
, FTAG
);
2088 dsl_dataset_rename_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
2090 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2091 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2092 dsl_dataset_t
*hds
= NULL
;
2094 VERIFY0(dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
));
2095 ddrsa
->ddrsa_tx
= tx
;
2096 if (ddrsa
->ddrsa_recursive
) {
2097 VERIFY0(dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
2098 dsl_dataset_rename_snapshot_sync_impl
, ddrsa
,
2101 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp
, hds
, ddrsa
));
2103 dsl_dataset_rele(hds
, FTAG
);
2107 dsl_dataset_rename_snapshot(const char *fsname
,
2108 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
)
2110 dsl_dataset_rename_snapshot_arg_t ddrsa
;
2112 ddrsa
.ddrsa_fsname
= fsname
;
2113 ddrsa
.ddrsa_oldsnapname
= oldsnapname
;
2114 ddrsa
.ddrsa_newsnapname
= newsnapname
;
2115 ddrsa
.ddrsa_recursive
= recursive
;
2117 return (dsl_sync_task(fsname
, dsl_dataset_rename_snapshot_check
,
2118 dsl_dataset_rename_snapshot_sync
, &ddrsa
,
2119 1, ZFS_SPACE_CHECK_RESERVED
));
2123 * If we're doing an ownership handoff, we need to make sure that there is
2124 * only one long hold on the dataset. We're not allowed to change anything here
2125 * so we don't permanently release the long hold or regular hold here. We want
2126 * to do this only when syncing to avoid the dataset unexpectedly going away
2127 * when we release the long hold. Allow a long hold to exist for volumes, this
2128 * may occur when asynchronously registering the minor with the kernel.
2131 dsl_dataset_handoff_check(dsl_dataset_t
*ds
, void *owner
, dmu_tx_t
*tx
)
2135 if (!dmu_tx_is_syncing(tx
))
2138 if (owner
!= NULL
) {
2139 VERIFY3P(ds
->ds_owner
, ==, owner
);
2140 dsl_dataset_long_rele(ds
, owner
);
2143 held
= (dsl_dataset_long_held(ds
) && (ds
->ds_owner
!= zvol_tag
));
2146 dsl_dataset_long_hold(ds
, owner
);
2149 return (SET_ERROR(EBUSY
));
2154 typedef struct dsl_dataset_rollback_arg
{
2155 const char *ddra_fsname
;
2157 nvlist_t
*ddra_result
;
2158 } dsl_dataset_rollback_arg_t
;
2161 dsl_dataset_rollback_check(void *arg
, dmu_tx_t
*tx
)
2163 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2164 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2166 int64_t unused_refres_delta
;
2169 nvlist_t
*proprequest
, *bookmarks
;
2171 error
= dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
);
2175 /* must not be a snapshot */
2176 if (ds
->ds_is_snapshot
) {
2177 dsl_dataset_rele(ds
, FTAG
);
2178 return (SET_ERROR(EINVAL
));
2181 /* must have a most recent snapshot */
2182 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
< TXG_INITIAL
) {
2183 dsl_dataset_rele(ds
, FTAG
);
2184 return (SET_ERROR(EINVAL
));
2187 /* must not have any bookmarks after the most recent snapshot */
2188 proprequest
= fnvlist_alloc();
2189 fnvlist_add_boolean(proprequest
, zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2190 bookmarks
= fnvlist_alloc();
2191 error
= dsl_get_bookmarks_impl(ds
, proprequest
, bookmarks
);
2192 fnvlist_free(proprequest
);
2195 for (pair
= nvlist_next_nvpair(bookmarks
, NULL
);
2196 pair
!= NULL
; pair
= nvlist_next_nvpair(bookmarks
, pair
)) {
2198 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair
),
2199 zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2200 uint64_t createtxg
= fnvlist_lookup_uint64(valuenv
, "value");
2201 if (createtxg
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
2202 fnvlist_free(bookmarks
);
2203 dsl_dataset_rele(ds
, FTAG
);
2204 return (SET_ERROR(EEXIST
));
2207 fnvlist_free(bookmarks
);
2209 error
= dsl_dataset_handoff_check(ds
, ddra
->ddra_owner
, tx
);
2211 dsl_dataset_rele(ds
, FTAG
);
2216 * Check if the snap we are rolling back to uses more than
2219 if (ds
->ds_quota
!= 0 &&
2220 dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
> ds
->ds_quota
) {
2221 dsl_dataset_rele(ds
, FTAG
);
2222 return (SET_ERROR(EDQUOT
));
2226 * When we do the clone swap, we will temporarily use more space
2227 * due to the refreservation (the head will no longer have any
2228 * unique space, so the entire amount of the refreservation will need
2229 * to be free). We will immediately destroy the clone, freeing
2230 * this space, but the freeing happens over many txg's.
2232 unused_refres_delta
= (int64_t)MIN(ds
->ds_reserved
,
2233 dsl_dataset_phys(ds
)->ds_unique_bytes
);
2235 if (unused_refres_delta
> 0 &&
2236 unused_refres_delta
>
2237 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
)) {
2238 dsl_dataset_rele(ds
, FTAG
);
2239 return (SET_ERROR(ENOSPC
));
2242 dsl_dataset_rele(ds
, FTAG
);
2247 dsl_dataset_rollback_sync(void *arg
, dmu_tx_t
*tx
)
2249 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2250 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2251 dsl_dataset_t
*ds
, *clone
;
2253 char namebuf
[ZFS_MAX_DATASET_NAME_LEN
];
2255 VERIFY0(dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
));
2257 dsl_dataset_name(ds
->ds_prev
, namebuf
);
2258 fnvlist_add_string(ddra
->ddra_result
, "target", namebuf
);
2260 cloneobj
= dsl_dataset_create_sync(ds
->ds_dir
, "%rollback",
2261 ds
->ds_prev
, DS_CREATE_FLAG_NODIRTY
, kcred
, tx
);
2263 VERIFY0(dsl_dataset_hold_obj(dp
, cloneobj
, FTAG
, &clone
));
2265 dsl_dataset_clone_swap_sync_impl(clone
, ds
, tx
);
2266 dsl_dataset_zero_zil(ds
, tx
);
2268 dsl_destroy_head_sync_impl(clone
, tx
);
2270 dsl_dataset_rele(clone
, FTAG
);
2271 dsl_dataset_rele(ds
, FTAG
);
2275 * Rolls back the given filesystem or volume to the most recent snapshot.
2276 * The name of the most recent snapshot will be returned under key "target"
2277 * in the result nvlist.
2280 * - The existing dataset MUST be owned by the specified owner at entry
2281 * - Upon return, dataset will still be held by the same owner, whether we
2284 * This mode is required any time the existing filesystem is mounted. See
2285 * notes above zfs_suspend_fs() for further details.
2288 dsl_dataset_rollback(const char *fsname
, void *owner
, nvlist_t
*result
)
2290 dsl_dataset_rollback_arg_t ddra
;
2292 ddra
.ddra_fsname
= fsname
;
2293 ddra
.ddra_owner
= owner
;
2294 ddra
.ddra_result
= result
;
2296 return (dsl_sync_task(fsname
, dsl_dataset_rollback_check
,
2297 dsl_dataset_rollback_sync
, &ddra
,
2298 1, ZFS_SPACE_CHECK_RESERVED
));
2301 struct promotenode
{
2306 typedef struct dsl_dataset_promote_arg
{
2307 const char *ddpa_clonename
;
2308 dsl_dataset_t
*ddpa_clone
;
2309 list_t shared_snaps
, origin_snaps
, clone_snaps
;
2310 dsl_dataset_t
*origin_origin
; /* origin of the origin */
2311 uint64_t used
, comp
, uncomp
, unique
, cloneusedsnap
, originusedsnap
;
2314 } dsl_dataset_promote_arg_t
;
2316 static int snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
);
2317 static int promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
,
2319 static void promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
);
2322 dsl_dataset_promote_check(void *arg
, dmu_tx_t
*tx
)
2324 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2325 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2327 struct promotenode
*snap
;
2328 dsl_dataset_t
*origin_ds
;
2332 size_t max_snap_len
;
2334 err
= promote_hold(ddpa
, dp
, FTAG
);
2338 hds
= ddpa
->ddpa_clone
;
2339 max_snap_len
= MAXNAMELEN
- strlen(ddpa
->ddpa_clonename
) - 1;
2341 if (dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
) {
2342 promote_rele(ddpa
, FTAG
);
2343 return (SET_ERROR(EXDEV
));
2347 * Compute and check the amount of space to transfer. Since this is
2348 * so expensive, don't do the preliminary check.
2350 if (!dmu_tx_is_syncing(tx
)) {
2351 promote_rele(ddpa
, FTAG
);
2355 snap
= list_head(&ddpa
->shared_snaps
);
2356 origin_ds
= snap
->ds
;
2358 /* compute origin's new unique space */
2359 snap
= list_tail(&ddpa
->clone_snaps
);
2360 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2361 origin_ds
->ds_object
);
2362 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2363 dsl_dataset_phys(origin_ds
)->ds_prev_snap_txg
, UINT64_MAX
,
2364 &ddpa
->unique
, &unused
, &unused
);
2367 * Walk the snapshots that we are moving
2369 * Compute space to transfer. Consider the incremental changes
2370 * to used by each snapshot:
2371 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2372 * So each snapshot gave birth to:
2373 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2374 * So a sequence would look like:
2375 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2376 * Which simplifies to:
2377 * uN + kN + kN-1 + ... + k1 + k0
2378 * Note however, if we stop before we reach the ORIGIN we get:
2379 * uN + kN + kN-1 + ... + kM - uM-1
2382 ddpa
->used
= dsl_dataset_phys(origin_ds
)->ds_referenced_bytes
;
2383 ddpa
->comp
= dsl_dataset_phys(origin_ds
)->ds_compressed_bytes
;
2384 ddpa
->uncomp
= dsl_dataset_phys(origin_ds
)->ds_uncompressed_bytes
;
2385 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2386 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2387 uint64_t val
, dlused
, dlcomp
, dluncomp
;
2388 dsl_dataset_t
*ds
= snap
->ds
;
2393 * If there are long holds, we won't be able to evict
2396 if (dsl_dataset_long_held(ds
)) {
2397 err
= SET_ERROR(EBUSY
);
2401 /* Check that the snapshot name does not conflict */
2402 VERIFY0(dsl_dataset_get_snapname(ds
));
2403 if (strlen(ds
->ds_snapname
) >= max_snap_len
) {
2404 err
= SET_ERROR(ENAMETOOLONG
);
2407 err
= dsl_dataset_snap_lookup(hds
, ds
->ds_snapname
, &val
);
2409 (void) strcpy(ddpa
->err_ds
, snap
->ds
->ds_snapname
);
2410 err
= SET_ERROR(EEXIST
);
2416 /* The very first snapshot does not have a deadlist */
2417 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
== 0)
2420 dsl_deadlist_space(&ds
->ds_deadlist
,
2421 &dlused
, &dlcomp
, &dluncomp
);
2422 ddpa
->used
+= dlused
;
2423 ddpa
->comp
+= dlcomp
;
2424 ddpa
->uncomp
+= dluncomp
;
2428 * If we are a clone of a clone then we never reached ORIGIN,
2429 * so we need to subtract out the clone origin's used space.
2431 if (ddpa
->origin_origin
) {
2433 dsl_dataset_phys(ddpa
->origin_origin
)->ds_referenced_bytes
;
2435 dsl_dataset_phys(ddpa
->origin_origin
)->ds_compressed_bytes
;
2437 dsl_dataset_phys(ddpa
->origin_origin
)->
2438 ds_uncompressed_bytes
;
2441 /* Check that there is enough space and limit headroom here */
2442 err
= dsl_dir_transfer_possible(origin_ds
->ds_dir
, hds
->ds_dir
,
2443 0, ss_mv_cnt
, ddpa
->used
, ddpa
->cr
);
2448 * Compute the amounts of space that will be used by snapshots
2449 * after the promotion (for both origin and clone). For each,
2450 * it is the amount of space that will be on all of their
2451 * deadlists (that was not born before their new origin).
2453 if (dsl_dir_phys(hds
->ds_dir
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2457 * Note, typically this will not be a clone of a clone,
2458 * so dd_origin_txg will be < TXG_INITIAL, so
2459 * these snaplist_space() -> dsl_deadlist_space_range()
2460 * calls will be fast because they do not have to
2461 * iterate over all bps.
2463 snap
= list_head(&ddpa
->origin_snaps
);
2464 err
= snaplist_space(&ddpa
->shared_snaps
,
2465 snap
->ds
->ds_dir
->dd_origin_txg
, &ddpa
->cloneusedsnap
);
2469 err
= snaplist_space(&ddpa
->clone_snaps
,
2470 snap
->ds
->ds_dir
->dd_origin_txg
, &space
);
2473 ddpa
->cloneusedsnap
+= space
;
2475 if (dsl_dir_phys(origin_ds
->ds_dir
)->dd_flags
&
2476 DD_FLAG_USED_BREAKDOWN
) {
2477 err
= snaplist_space(&ddpa
->origin_snaps
,
2478 dsl_dataset_phys(origin_ds
)->ds_creation_txg
,
2479 &ddpa
->originusedsnap
);
2485 promote_rele(ddpa
, FTAG
);
2490 dsl_dataset_promote_sync(void *arg
, dmu_tx_t
*tx
)
2492 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2493 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2495 struct promotenode
*snap
;
2496 dsl_dataset_t
*origin_ds
;
2497 dsl_dataset_t
*origin_head
;
2499 dsl_dir_t
*odd
= NULL
;
2500 uint64_t oldnext_obj
;
2503 VERIFY0(promote_hold(ddpa
, dp
, FTAG
));
2504 hds
= ddpa
->ddpa_clone
;
2506 ASSERT0(dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
);
2508 snap
= list_head(&ddpa
->shared_snaps
);
2509 origin_ds
= snap
->ds
;
2512 snap
= list_head(&ddpa
->origin_snaps
);
2513 origin_head
= snap
->ds
;
2516 * We need to explicitly open odd, since origin_ds's dd will be
2519 VERIFY0(dsl_dir_hold_obj(dp
, origin_ds
->ds_dir
->dd_object
,
2522 /* change origin's next snap */
2523 dmu_buf_will_dirty(origin_ds
->ds_dbuf
, tx
);
2524 oldnext_obj
= dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
;
2525 snap
= list_tail(&ddpa
->clone_snaps
);
2526 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2527 origin_ds
->ds_object
);
2528 dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
= snap
->ds
->ds_object
;
2530 /* change the origin's next clone */
2531 if (dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
) {
2532 dsl_dataset_remove_from_next_clones(origin_ds
,
2533 snap
->ds
->ds_object
, tx
);
2534 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2535 dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
,
2540 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
2541 ASSERT3U(dsl_dir_phys(dd
)->dd_origin_obj
, ==, origin_ds
->ds_object
);
2542 dsl_dir_phys(dd
)->dd_origin_obj
= dsl_dir_phys(odd
)->dd_origin_obj
;
2543 dd
->dd_origin_txg
= origin_head
->ds_dir
->dd_origin_txg
;
2544 dmu_buf_will_dirty(odd
->dd_dbuf
, tx
);
2545 dsl_dir_phys(odd
)->dd_origin_obj
= origin_ds
->ds_object
;
2546 origin_head
->ds_dir
->dd_origin_txg
=
2547 dsl_dataset_phys(origin_ds
)->ds_creation_txg
;
2549 /* change dd_clone entries */
2550 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2551 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2552 dsl_dir_phys(odd
)->dd_clones
, hds
->ds_object
, tx
));
2553 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2554 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2555 hds
->ds_object
, tx
));
2557 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2558 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2559 origin_head
->ds_object
, tx
));
2560 if (dsl_dir_phys(dd
)->dd_clones
== 0) {
2561 dsl_dir_phys(dd
)->dd_clones
=
2562 zap_create(dp
->dp_meta_objset
, DMU_OT_DSL_CLONES
,
2563 DMU_OT_NONE
, 0, tx
);
2565 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2566 dsl_dir_phys(dd
)->dd_clones
, origin_head
->ds_object
, tx
));
2569 /* move snapshots to this dir */
2570 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2571 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2572 dsl_dataset_t
*ds
= snap
->ds
;
2575 * Property callbacks are registered to a particular
2576 * dsl_dir. Since ours is changing, evict the objset
2577 * so that they will be unregistered from the old dsl_dir.
2579 if (ds
->ds_objset
) {
2580 dmu_objset_evict(ds
->ds_objset
);
2581 ds
->ds_objset
= NULL
;
2584 /* move snap name entry */
2585 VERIFY0(dsl_dataset_get_snapname(ds
));
2586 VERIFY0(dsl_dataset_snap_remove(origin_head
,
2587 ds
->ds_snapname
, tx
, B_TRUE
));
2588 VERIFY0(zap_add(dp
->dp_meta_objset
,
2589 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
, ds
->ds_snapname
,
2590 8, 1, &ds
->ds_object
, tx
));
2591 dsl_fs_ss_count_adjust(hds
->ds_dir
, 1,
2592 DD_FIELD_SNAPSHOT_COUNT
, tx
);
2594 /* change containing dsl_dir */
2595 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2596 ASSERT3U(dsl_dataset_phys(ds
)->ds_dir_obj
, ==, odd
->dd_object
);
2597 dsl_dataset_phys(ds
)->ds_dir_obj
= dd
->dd_object
;
2598 ASSERT3P(ds
->ds_dir
, ==, odd
);
2599 dsl_dir_rele(ds
->ds_dir
, ds
);
2600 VERIFY0(dsl_dir_hold_obj(dp
, dd
->dd_object
,
2601 NULL
, ds
, &ds
->ds_dir
));
2603 /* move any clone references */
2604 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
&&
2605 spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2609 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2610 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
2611 zap_cursor_retrieve(&zc
, &za
) == 0;
2612 zap_cursor_advance(&zc
)) {
2613 dsl_dataset_t
*cnds
;
2616 if (za
.za_first_integer
== oldnext_obj
) {
2618 * We've already moved the
2619 * origin's reference.
2624 VERIFY0(dsl_dataset_hold_obj(dp
,
2625 za
.za_first_integer
, FTAG
, &cnds
));
2626 o
= dsl_dir_phys(cnds
->ds_dir
)->
2627 dd_head_dataset_obj
;
2629 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2630 dsl_dir_phys(odd
)->dd_clones
, o
, tx
));
2631 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2632 dsl_dir_phys(dd
)->dd_clones
, o
, tx
));
2633 dsl_dataset_rele(cnds
, FTAG
);
2635 zap_cursor_fini(&zc
);
2638 ASSERT(!dsl_prop_hascb(ds
));
2642 * Change space accounting.
2643 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2644 * both be valid, or both be 0 (resulting in delta == 0). This
2645 * is true for each of {clone,origin} independently.
2648 delta
= ddpa
->cloneusedsnap
-
2649 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_SNAP
];
2650 ASSERT3S(delta
, >=, 0);
2651 ASSERT3U(ddpa
->used
, >=, delta
);
2652 dsl_dir_diduse_space(dd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2653 dsl_dir_diduse_space(dd
, DD_USED_HEAD
,
2654 ddpa
->used
- delta
, ddpa
->comp
, ddpa
->uncomp
, tx
);
2656 delta
= ddpa
->originusedsnap
-
2657 dsl_dir_phys(odd
)->dd_used_breakdown
[DD_USED_SNAP
];
2658 ASSERT3S(delta
, <=, 0);
2659 ASSERT3U(ddpa
->used
, >=, -delta
);
2660 dsl_dir_diduse_space(odd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2661 dsl_dir_diduse_space(odd
, DD_USED_HEAD
,
2662 -ddpa
->used
- delta
, -ddpa
->comp
, -ddpa
->uncomp
, tx
);
2664 dsl_dataset_phys(origin_ds
)->ds_unique_bytes
= ddpa
->unique
;
2666 /* log history record */
2667 spa_history_log_internal_ds(hds
, "promote", tx
, "");
2669 dsl_dir_rele(odd
, FTAG
);
2670 promote_rele(ddpa
, FTAG
);
2674 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2675 * (exclusive) and last_obj (inclusive). The list will be in reverse
2676 * order (last_obj will be the list_head()). If first_obj == 0, do all
2677 * snapshots back to this dataset's origin.
2680 snaplist_make(dsl_pool_t
*dp
,
2681 uint64_t first_obj
, uint64_t last_obj
, list_t
*l
, void *tag
)
2683 uint64_t obj
= last_obj
;
2685 list_create(l
, sizeof (struct promotenode
),
2686 offsetof(struct promotenode
, link
));
2688 while (obj
!= first_obj
) {
2690 struct promotenode
*snap
;
2693 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
2694 ASSERT(err
!= ENOENT
);
2699 first_obj
= dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
;
2701 snap
= kmem_alloc(sizeof (*snap
), KM_SLEEP
);
2703 list_insert_tail(l
, snap
);
2704 obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
2711 snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
)
2713 struct promotenode
*snap
;
2716 for (snap
= list_head(l
); snap
; snap
= list_next(l
, snap
)) {
2717 uint64_t used
, comp
, uncomp
;
2718 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2719 mintxg
, UINT64_MAX
, &used
, &comp
, &uncomp
);
2726 snaplist_destroy(list_t
*l
, void *tag
)
2728 struct promotenode
*snap
;
2730 if (l
== NULL
|| !list_link_active(&l
->list_head
))
2733 while ((snap
= list_tail(l
)) != NULL
) {
2734 list_remove(l
, snap
);
2735 dsl_dataset_rele(snap
->ds
, tag
);
2736 kmem_free(snap
, sizeof (*snap
));
2742 promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
, void *tag
)
2746 struct promotenode
*snap
;
2748 error
= dsl_dataset_hold(dp
, ddpa
->ddpa_clonename
, tag
,
2752 dd
= ddpa
->ddpa_clone
->ds_dir
;
2754 if (ddpa
->ddpa_clone
->ds_is_snapshot
||
2755 !dsl_dir_is_clone(dd
)) {
2756 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2757 return (SET_ERROR(EINVAL
));
2760 error
= snaplist_make(dp
, 0, dsl_dir_phys(dd
)->dd_origin_obj
,
2761 &ddpa
->shared_snaps
, tag
);
2765 error
= snaplist_make(dp
, 0, ddpa
->ddpa_clone
->ds_object
,
2766 &ddpa
->clone_snaps
, tag
);
2770 snap
= list_head(&ddpa
->shared_snaps
);
2771 ASSERT3U(snap
->ds
->ds_object
, ==, dsl_dir_phys(dd
)->dd_origin_obj
);
2772 error
= snaplist_make(dp
, dsl_dir_phys(dd
)->dd_origin_obj
,
2773 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_head_dataset_obj
,
2774 &ddpa
->origin_snaps
, tag
);
2778 if (dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
!= 0) {
2779 error
= dsl_dataset_hold_obj(dp
,
2780 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
,
2781 tag
, &ddpa
->origin_origin
);
2787 promote_rele(ddpa
, tag
);
2792 promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
)
2794 snaplist_destroy(&ddpa
->shared_snaps
, tag
);
2795 snaplist_destroy(&ddpa
->clone_snaps
, tag
);
2796 snaplist_destroy(&ddpa
->origin_snaps
, tag
);
2797 if (ddpa
->origin_origin
!= NULL
)
2798 dsl_dataset_rele(ddpa
->origin_origin
, tag
);
2799 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2805 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2806 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2809 dsl_dataset_promote(const char *name
, char *conflsnap
)
2811 dsl_dataset_promote_arg_t ddpa
= { 0 };
2817 * We will modify space proportional to the number of
2818 * snapshots. Compute numsnaps.
2820 error
= dmu_objset_hold(name
, FTAG
, &os
);
2823 error
= zap_count(dmu_objset_pool(os
)->dp_meta_objset
,
2824 dsl_dataset_phys(dmu_objset_ds(os
))->ds_snapnames_zapobj
,
2826 dmu_objset_rele(os
, FTAG
);
2830 ddpa
.ddpa_clonename
= name
;
2831 ddpa
.err_ds
= conflsnap
;
2834 return (dsl_sync_task(name
, dsl_dataset_promote_check
,
2835 dsl_dataset_promote_sync
, &ddpa
,
2836 2 + numsnaps
, ZFS_SPACE_CHECK_RESERVED
));
2840 dsl_dataset_clone_swap_check_impl(dsl_dataset_t
*clone
,
2841 dsl_dataset_t
*origin_head
, boolean_t force
, void *owner
, dmu_tx_t
*tx
)
2844 * "slack" factor for received datasets with refquota set on them.
2845 * See the bottom of this function for details on its use.
2847 uint64_t refquota_slack
= (uint64_t)DMU_MAX_ACCESS
*
2848 spa_asize_inflation
;
2849 int64_t unused_refres_delta
;
2851 /* they should both be heads */
2852 if (clone
->ds_is_snapshot
||
2853 origin_head
->ds_is_snapshot
)
2854 return (SET_ERROR(EINVAL
));
2856 /* if we are not forcing, the branch point should be just before them */
2857 if (!force
&& clone
->ds_prev
!= origin_head
->ds_prev
)
2858 return (SET_ERROR(EINVAL
));
2860 /* clone should be the clone (unless they are unrelated) */
2861 if (clone
->ds_prev
!= NULL
&&
2862 clone
->ds_prev
!= clone
->ds_dir
->dd_pool
->dp_origin_snap
&&
2863 origin_head
->ds_dir
!= clone
->ds_prev
->ds_dir
)
2864 return (SET_ERROR(EINVAL
));
2866 /* the clone should be a child of the origin */
2867 if (clone
->ds_dir
->dd_parent
!= origin_head
->ds_dir
)
2868 return (SET_ERROR(EINVAL
));
2870 /* origin_head shouldn't be modified unless 'force' */
2872 dsl_dataset_modified_since_snap(origin_head
, origin_head
->ds_prev
))
2873 return (SET_ERROR(ETXTBSY
));
2875 /* origin_head should have no long holds (e.g. is not mounted) */
2876 if (dsl_dataset_handoff_check(origin_head
, owner
, tx
))
2877 return (SET_ERROR(EBUSY
));
2879 /* check amount of any unconsumed refreservation */
2880 unused_refres_delta
=
2881 (int64_t)MIN(origin_head
->ds_reserved
,
2882 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
2883 (int64_t)MIN(origin_head
->ds_reserved
,
2884 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2886 if (unused_refres_delta
> 0 &&
2887 unused_refres_delta
>
2888 dsl_dir_space_available(origin_head
->ds_dir
, NULL
, 0, TRUE
))
2889 return (SET_ERROR(ENOSPC
));
2892 * The clone can't be too much over the head's refquota.
2894 * To ensure that the entire refquota can be used, we allow one
2895 * transaction to exceed the the refquota. Therefore, this check
2896 * needs to also allow for the space referenced to be more than the
2897 * refquota. The maximum amount of space that one transaction can use
2898 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this
2899 * overage ensures that we are able to receive a filesystem that
2900 * exceeds the refquota on the source system.
2902 * So that overage is the refquota_slack we use below.
2904 if (origin_head
->ds_quota
!= 0 &&
2905 dsl_dataset_phys(clone
)->ds_referenced_bytes
>
2906 origin_head
->ds_quota
+ refquota_slack
)
2907 return (SET_ERROR(EDQUOT
));
2913 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t
*clone
,
2914 dsl_dataset_t
*origin_head
, dmu_tx_t
*tx
)
2917 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2918 int64_t unused_refres_delta
;
2920 ASSERT(clone
->ds_reserved
== 0);
2922 * NOTE: On DEBUG kernels there could be a race between this and
2923 * the check function if spa_asize_inflation is adjusted...
2925 ASSERT(origin_head
->ds_quota
== 0 ||
2926 dsl_dataset_phys(clone
)->ds_unique_bytes
<= origin_head
->ds_quota
+
2927 DMU_MAX_ACCESS
* spa_asize_inflation
);
2928 ASSERT3P(clone
->ds_prev
, ==, origin_head
->ds_prev
);
2931 * Swap per-dataset feature flags.
2933 for (f
= 0; f
< SPA_FEATURES
; f
++) {
2934 boolean_t clone_inuse
;
2935 boolean_t origin_head_inuse
;
2937 if (!(spa_feature_table
[f
].fi_flags
&
2938 ZFEATURE_FLAG_PER_DATASET
)) {
2939 ASSERT(!clone
->ds_feature_inuse
[f
]);
2940 ASSERT(!origin_head
->ds_feature_inuse
[f
]);
2944 clone_inuse
= clone
->ds_feature_inuse
[f
];
2945 origin_head_inuse
= origin_head
->ds_feature_inuse
[f
];
2948 dsl_dataset_deactivate_feature(clone
->ds_object
, f
, tx
);
2949 clone
->ds_feature_inuse
[f
] = B_FALSE
;
2951 if (origin_head_inuse
) {
2952 dsl_dataset_deactivate_feature(origin_head
->ds_object
,
2954 origin_head
->ds_feature_inuse
[f
] = B_FALSE
;
2957 dsl_dataset_activate_feature(origin_head
->ds_object
,
2959 origin_head
->ds_feature_inuse
[f
] = B_TRUE
;
2961 if (origin_head_inuse
) {
2962 dsl_dataset_activate_feature(clone
->ds_object
, f
, tx
);
2963 clone
->ds_feature_inuse
[f
] = B_TRUE
;
2967 dmu_buf_will_dirty(clone
->ds_dbuf
, tx
);
2968 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
2970 if (clone
->ds_objset
!= NULL
) {
2971 dmu_objset_evict(clone
->ds_objset
);
2972 clone
->ds_objset
= NULL
;
2975 if (origin_head
->ds_objset
!= NULL
) {
2976 dmu_objset_evict(origin_head
->ds_objset
);
2977 origin_head
->ds_objset
= NULL
;
2980 unused_refres_delta
=
2981 (int64_t)MIN(origin_head
->ds_reserved
,
2982 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
2983 (int64_t)MIN(origin_head
->ds_reserved
,
2984 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2987 * Reset origin's unique bytes, if it exists.
2989 if (clone
->ds_prev
) {
2990 dsl_dataset_t
*origin
= clone
->ds_prev
;
2991 uint64_t comp
, uncomp
;
2993 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
2994 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2995 dsl_dataset_phys(origin
)->ds_prev_snap_txg
, UINT64_MAX
,
2996 &dsl_dataset_phys(origin
)->ds_unique_bytes
, &comp
, &uncomp
);
3002 tmp
= dsl_dataset_phys(origin_head
)->ds_bp
;
3003 dsl_dataset_phys(origin_head
)->ds_bp
=
3004 dsl_dataset_phys(clone
)->ds_bp
;
3005 dsl_dataset_phys(clone
)->ds_bp
= tmp
;
3008 /* set dd_*_bytes */
3010 int64_t dused
, dcomp
, duncomp
;
3011 uint64_t cdl_used
, cdl_comp
, cdl_uncomp
;
3012 uint64_t odl_used
, odl_comp
, odl_uncomp
;
3014 ASSERT3U(dsl_dir_phys(clone
->ds_dir
)->
3015 dd_used_breakdown
[DD_USED_SNAP
], ==, 0);
3017 dsl_deadlist_space(&clone
->ds_deadlist
,
3018 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
3019 dsl_deadlist_space(&origin_head
->ds_deadlist
,
3020 &odl_used
, &odl_comp
, &odl_uncomp
);
3022 dused
= dsl_dataset_phys(clone
)->ds_referenced_bytes
+
3024 (dsl_dataset_phys(origin_head
)->ds_referenced_bytes
+
3026 dcomp
= dsl_dataset_phys(clone
)->ds_compressed_bytes
+
3028 (dsl_dataset_phys(origin_head
)->ds_compressed_bytes
+
3030 duncomp
= dsl_dataset_phys(clone
)->ds_uncompressed_bytes
+
3032 (dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
+
3035 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_HEAD
,
3036 dused
, dcomp
, duncomp
, tx
);
3037 dsl_dir_diduse_space(clone
->ds_dir
, DD_USED_HEAD
,
3038 -dused
, -dcomp
, -duncomp
, tx
);
3041 * The difference in the space used by snapshots is the
3042 * difference in snapshot space due to the head's
3043 * deadlist (since that's the only thing that's
3044 * changing that affects the snapused).
3046 dsl_deadlist_space_range(&clone
->ds_deadlist
,
3047 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
3048 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
3049 dsl_deadlist_space_range(&origin_head
->ds_deadlist
,
3050 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
3051 &odl_used
, &odl_comp
, &odl_uncomp
);
3052 dsl_dir_transfer_space(origin_head
->ds_dir
, cdl_used
- odl_used
,
3053 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
3056 /* swap ds_*_bytes */
3057 SWITCH64(dsl_dataset_phys(origin_head
)->ds_referenced_bytes
,
3058 dsl_dataset_phys(clone
)->ds_referenced_bytes
);
3059 SWITCH64(dsl_dataset_phys(origin_head
)->ds_compressed_bytes
,
3060 dsl_dataset_phys(clone
)->ds_compressed_bytes
);
3061 SWITCH64(dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
,
3062 dsl_dataset_phys(clone
)->ds_uncompressed_bytes
);
3063 SWITCH64(dsl_dataset_phys(origin_head
)->ds_unique_bytes
,
3064 dsl_dataset_phys(clone
)->ds_unique_bytes
);
3066 /* apply any parent delta for change in unconsumed refreservation */
3067 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_REFRSRV
,
3068 unused_refres_delta
, 0, 0, tx
);
3073 dsl_deadlist_close(&clone
->ds_deadlist
);
3074 dsl_deadlist_close(&origin_head
->ds_deadlist
);
3075 SWITCH64(dsl_dataset_phys(origin_head
)->ds_deadlist_obj
,
3076 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
3077 dsl_deadlist_open(&clone
->ds_deadlist
, dp
->dp_meta_objset
,
3078 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
3079 dsl_deadlist_open(&origin_head
->ds_deadlist
, dp
->dp_meta_objset
,
3080 dsl_dataset_phys(origin_head
)->ds_deadlist_obj
);
3082 dsl_scan_ds_clone_swapped(origin_head
, clone
, tx
);
3084 spa_history_log_internal_ds(clone
, "clone swap", tx
,
3085 "parent=%s", origin_head
->ds_dir
->dd_myname
);
3089 * Given a pool name and a dataset object number in that pool,
3090 * return the name of that dataset.
3093 dsl_dsobj_to_dsname(char *pname
, uint64_t obj
, char *buf
)
3099 error
= dsl_pool_hold(pname
, FTAG
, &dp
);
3103 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
3105 dsl_dataset_name(ds
, buf
);
3106 dsl_dataset_rele(ds
, FTAG
);
3108 dsl_pool_rele(dp
, FTAG
);
3114 dsl_dataset_check_quota(dsl_dataset_t
*ds
, boolean_t check_quota
,
3115 uint64_t asize
, uint64_t inflight
, uint64_t *used
, uint64_t *ref_rsrv
)
3119 ASSERT3S(asize
, >, 0);
3122 * *ref_rsrv is the portion of asize that will come from any
3123 * unconsumed refreservation space.
3127 mutex_enter(&ds
->ds_lock
);
3129 * Make a space adjustment for reserved bytes.
3131 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
) {
3133 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
3135 (ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
3137 asize
- MIN(asize
, parent_delta(ds
, asize
+ inflight
));
3140 if (!check_quota
|| ds
->ds_quota
== 0) {
3141 mutex_exit(&ds
->ds_lock
);
3145 * If they are requesting more space, and our current estimate
3146 * is over quota, they get to try again unless the actual
3147 * on-disk is over quota and there are no pending changes (which
3148 * may free up space for us).
3150 if (dsl_dataset_phys(ds
)->ds_referenced_bytes
+ inflight
>=
3153 dsl_dataset_phys(ds
)->ds_referenced_bytes
< ds
->ds_quota
)
3154 error
= SET_ERROR(ERESTART
);
3156 error
= SET_ERROR(EDQUOT
);
3158 mutex_exit(&ds
->ds_lock
);
3163 typedef struct dsl_dataset_set_qr_arg
{
3164 const char *ddsqra_name
;
3165 zprop_source_t ddsqra_source
;
3166 uint64_t ddsqra_value
;
3167 } dsl_dataset_set_qr_arg_t
;
3172 dsl_dataset_set_refquota_check(void *arg
, dmu_tx_t
*tx
)
3174 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3175 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3180 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFQUOTA
)
3181 return (SET_ERROR(ENOTSUP
));
3183 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3187 if (ds
->ds_is_snapshot
) {
3188 dsl_dataset_rele(ds
, FTAG
);
3189 return (SET_ERROR(EINVAL
));
3192 error
= dsl_prop_predict(ds
->ds_dir
,
3193 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
3194 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3196 dsl_dataset_rele(ds
, FTAG
);
3201 dsl_dataset_rele(ds
, FTAG
);
3205 if (newval
< dsl_dataset_phys(ds
)->ds_referenced_bytes
||
3206 newval
< ds
->ds_reserved
) {
3207 dsl_dataset_rele(ds
, FTAG
);
3208 return (SET_ERROR(ENOSPC
));
3211 dsl_dataset_rele(ds
, FTAG
);
3216 dsl_dataset_set_refquota_sync(void *arg
, dmu_tx_t
*tx
)
3218 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3219 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3220 dsl_dataset_t
*ds
= NULL
;
3223 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3225 dsl_prop_set_sync_impl(ds
,
3226 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
3227 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
3228 &ddsqra
->ddsqra_value
, tx
);
3230 VERIFY0(dsl_prop_get_int_ds(ds
,
3231 zfs_prop_to_name(ZFS_PROP_REFQUOTA
), &newval
));
3233 if (ds
->ds_quota
!= newval
) {
3234 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3235 ds
->ds_quota
= newval
;
3237 dsl_dataset_rele(ds
, FTAG
);
3241 dsl_dataset_set_refquota(const char *dsname
, zprop_source_t source
,
3244 dsl_dataset_set_qr_arg_t ddsqra
;
3246 ddsqra
.ddsqra_name
= dsname
;
3247 ddsqra
.ddsqra_source
= source
;
3248 ddsqra
.ddsqra_value
= refquota
;
3250 return (dsl_sync_task(dsname
, dsl_dataset_set_refquota_check
,
3251 dsl_dataset_set_refquota_sync
, &ddsqra
, 0, ZFS_SPACE_CHECK_NONE
));
3255 dsl_dataset_set_refreservation_check(void *arg
, dmu_tx_t
*tx
)
3257 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3258 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3261 uint64_t newval
, unique
;
3263 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFRESERVATION
)
3264 return (SET_ERROR(ENOTSUP
));
3266 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3270 if (ds
->ds_is_snapshot
) {
3271 dsl_dataset_rele(ds
, FTAG
);
3272 return (SET_ERROR(EINVAL
));
3275 error
= dsl_prop_predict(ds
->ds_dir
,
3276 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3277 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3279 dsl_dataset_rele(ds
, FTAG
);
3284 * If we are doing the preliminary check in open context, the
3285 * space estimates may be inaccurate.
3287 if (!dmu_tx_is_syncing(tx
)) {
3288 dsl_dataset_rele(ds
, FTAG
);
3292 mutex_enter(&ds
->ds_lock
);
3293 if (!DS_UNIQUE_IS_ACCURATE(ds
))
3294 dsl_dataset_recalc_head_uniq(ds
);
3295 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3296 mutex_exit(&ds
->ds_lock
);
3298 if (MAX(unique
, newval
) > MAX(unique
, ds
->ds_reserved
)) {
3299 uint64_t delta
= MAX(unique
, newval
) -
3300 MAX(unique
, ds
->ds_reserved
);
3303 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, B_TRUE
) ||
3304 (ds
->ds_quota
> 0 && newval
> ds
->ds_quota
)) {
3305 dsl_dataset_rele(ds
, FTAG
);
3306 return (SET_ERROR(ENOSPC
));
3310 dsl_dataset_rele(ds
, FTAG
);
3315 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t
*ds
,
3316 zprop_source_t source
, uint64_t value
, dmu_tx_t
*tx
)
3322 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3323 source
, sizeof (value
), 1, &value
, tx
);
3325 VERIFY0(dsl_prop_get_int_ds(ds
,
3326 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
), &newval
));
3328 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3329 mutex_enter(&ds
->ds_dir
->dd_lock
);
3330 mutex_enter(&ds
->ds_lock
);
3331 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
3332 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3333 delta
= MAX(0, (int64_t)(newval
- unique
)) -
3334 MAX(0, (int64_t)(ds
->ds_reserved
- unique
));
3335 ds
->ds_reserved
= newval
;
3336 mutex_exit(&ds
->ds_lock
);
3338 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
, delta
, 0, 0, tx
);
3339 mutex_exit(&ds
->ds_dir
->dd_lock
);
3343 dsl_dataset_set_refreservation_sync(void *arg
, dmu_tx_t
*tx
)
3345 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3346 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3347 dsl_dataset_t
*ds
= NULL
;
3349 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3350 dsl_dataset_set_refreservation_sync_impl(ds
,
3351 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, tx
);
3352 dsl_dataset_rele(ds
, FTAG
);
3356 dsl_dataset_set_refreservation(const char *dsname
, zprop_source_t source
,
3357 uint64_t refreservation
)
3359 dsl_dataset_set_qr_arg_t ddsqra
;
3361 ddsqra
.ddsqra_name
= dsname
;
3362 ddsqra
.ddsqra_source
= source
;
3363 ddsqra
.ddsqra_value
= refreservation
;
3365 return (dsl_sync_task(dsname
, dsl_dataset_set_refreservation_check
,
3366 dsl_dataset_set_refreservation_sync
, &ddsqra
,
3367 0, ZFS_SPACE_CHECK_NONE
));
3371 * Return (in *usedp) the amount of space written in new that is not
3372 * present in oldsnap. New may be a snapshot or the head. Old must be
3373 * a snapshot before new, in new's filesystem (or its origin). If not then
3374 * fail and return EINVAL.
3376 * The written space is calculated by considering two components: First, we
3377 * ignore any freed space, and calculate the written as new's used space
3378 * minus old's used space. Next, we add in the amount of space that was freed
3379 * between the two snapshots, thus reducing new's used space relative to old's.
3380 * Specifically, this is the space that was born before old->ds_creation_txg,
3381 * and freed before new (ie. on new's deadlist or a previous deadlist).
3383 * space freed [---------------------]
3384 * snapshots ---O-------O--------O-------O------
3388 dsl_dataset_space_written(dsl_dataset_t
*oldsnap
, dsl_dataset_t
*new,
3389 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3393 dsl_pool_t
*dp
= new->ds_dir
->dd_pool
;
3395 ASSERT(dsl_pool_config_held(dp
));
3398 *usedp
+= dsl_dataset_phys(new)->ds_referenced_bytes
;
3399 *usedp
-= dsl_dataset_phys(oldsnap
)->ds_referenced_bytes
;
3402 *compp
+= dsl_dataset_phys(new)->ds_compressed_bytes
;
3403 *compp
-= dsl_dataset_phys(oldsnap
)->ds_compressed_bytes
;
3406 *uncompp
+= dsl_dataset_phys(new)->ds_uncompressed_bytes
;
3407 *uncompp
-= dsl_dataset_phys(oldsnap
)->ds_uncompressed_bytes
;
3409 snapobj
= new->ds_object
;
3410 while (snapobj
!= oldsnap
->ds_object
) {
3411 dsl_dataset_t
*snap
;
3412 uint64_t used
, comp
, uncomp
;
3414 if (snapobj
== new->ds_object
) {
3417 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &snap
);
3422 if (dsl_dataset_phys(snap
)->ds_prev_snap_txg
==
3423 dsl_dataset_phys(oldsnap
)->ds_creation_txg
) {
3425 * The blocks in the deadlist can not be born after
3426 * ds_prev_snap_txg, so get the whole deadlist space,
3427 * which is more efficient (especially for old-format
3428 * deadlists). Unfortunately the deadlist code
3429 * doesn't have enough information to make this
3430 * optimization itself.
3432 dsl_deadlist_space(&snap
->ds_deadlist
,
3433 &used
, &comp
, &uncomp
);
3435 dsl_deadlist_space_range(&snap
->ds_deadlist
,
3436 0, dsl_dataset_phys(oldsnap
)->ds_creation_txg
,
3437 &used
, &comp
, &uncomp
);
3444 * If we get to the beginning of the chain of snapshots
3445 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3446 * was not a snapshot of/before new.
3448 snapobj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3450 dsl_dataset_rele(snap
, FTAG
);
3452 err
= SET_ERROR(EINVAL
);
3461 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3462 * lastsnap, and all snapshots in between are deleted.
3464 * blocks that would be freed [---------------------------]
3465 * snapshots ---O-------O--------O-------O--------O
3466 * firstsnap lastsnap
3468 * This is the set of blocks that were born after the snap before firstsnap,
3469 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3470 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3471 * We calculate this by iterating over the relevant deadlists (from the snap
3472 * after lastsnap, backward to the snap after firstsnap), summing up the
3473 * space on the deadlist that was born after the snap before firstsnap.
3476 dsl_dataset_space_wouldfree(dsl_dataset_t
*firstsnap
,
3477 dsl_dataset_t
*lastsnap
,
3478 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3482 dsl_pool_t
*dp
= firstsnap
->ds_dir
->dd_pool
;
3484 ASSERT(firstsnap
->ds_is_snapshot
);
3485 ASSERT(lastsnap
->ds_is_snapshot
);
3488 * Check that the snapshots are in the same dsl_dir, and firstsnap
3489 * is before lastsnap.
3491 if (firstsnap
->ds_dir
!= lastsnap
->ds_dir
||
3492 dsl_dataset_phys(firstsnap
)->ds_creation_txg
>
3493 dsl_dataset_phys(lastsnap
)->ds_creation_txg
)
3494 return (SET_ERROR(EINVAL
));
3496 *usedp
= *compp
= *uncompp
= 0;
3498 snapobj
= dsl_dataset_phys(lastsnap
)->ds_next_snap_obj
;
3499 while (snapobj
!= firstsnap
->ds_object
) {
3501 uint64_t used
, comp
, uncomp
;
3503 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &ds
);
3507 dsl_deadlist_space_range(&ds
->ds_deadlist
,
3508 dsl_dataset_phys(firstsnap
)->ds_prev_snap_txg
, UINT64_MAX
,
3509 &used
, &comp
, &uncomp
);
3514 snapobj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
3515 ASSERT3U(snapobj
, !=, 0);
3516 dsl_dataset_rele(ds
, FTAG
);
3522 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3523 * For example, they could both be snapshots of the same filesystem, and
3524 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3525 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3526 * filesystem. Or 'earlier' could be the origin's origin.
3528 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3531 dsl_dataset_is_before(dsl_dataset_t
*later
, dsl_dataset_t
*earlier
,
3532 uint64_t earlier_txg
)
3534 dsl_pool_t
*dp
= later
->ds_dir
->dd_pool
;
3537 dsl_dataset_t
*origin
;
3539 ASSERT(dsl_pool_config_held(dp
));
3540 ASSERT(earlier
->ds_is_snapshot
|| earlier_txg
!= 0);
3542 if (earlier_txg
== 0)
3543 earlier_txg
= dsl_dataset_phys(earlier
)->ds_creation_txg
;
3545 if (later
->ds_is_snapshot
&&
3546 earlier_txg
>= dsl_dataset_phys(later
)->ds_creation_txg
)
3549 if (later
->ds_dir
== earlier
->ds_dir
)
3551 if (!dsl_dir_is_clone(later
->ds_dir
))
3554 if (dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
== earlier
->ds_object
)
3556 error
= dsl_dataset_hold_obj(dp
,
3557 dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
, FTAG
, &origin
);
3560 ret
= dsl_dataset_is_before(origin
, earlier
, earlier_txg
);
3561 dsl_dataset_rele(origin
, FTAG
);
3566 dsl_dataset_zapify(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
3568 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
3569 dmu_object_zapify(mos
, ds
->ds_object
, DMU_OT_DSL_DATASET
, tx
);
3573 dsl_dataset_is_zapified(dsl_dataset_t
*ds
)
3575 dmu_object_info_t doi
;
3577 dmu_object_info_from_db(ds
->ds_dbuf
, &doi
);
3578 return (doi
.doi_type
== DMU_OTN_ZAP_METADATA
);
3582 dsl_dataset_has_resume_receive_state(dsl_dataset_t
*ds
)
3584 return (dsl_dataset_is_zapified(ds
) &&
3585 zap_contains(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
3586 ds
->ds_object
, DS_FIELD_RESUME_TOGUID
) == 0);
3589 #if defined(_KERNEL) && defined(HAVE_SPL)
3591 module_param(zfs_max_recordsize
, int, 0644);
3592 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
3594 /* Limited to 1M on 32-bit platforms due to lack of virtual address space */
3595 module_param(zfs_max_recordsize
, int, 0444);
3596 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
3599 EXPORT_SYMBOL(dsl_dataset_hold
);
3600 EXPORT_SYMBOL(dsl_dataset_hold_obj
);
3601 EXPORT_SYMBOL(dsl_dataset_own
);
3602 EXPORT_SYMBOL(dsl_dataset_own_obj
);
3603 EXPORT_SYMBOL(dsl_dataset_name
);
3604 EXPORT_SYMBOL(dsl_dataset_rele
);
3605 EXPORT_SYMBOL(dsl_dataset_disown
);
3606 EXPORT_SYMBOL(dsl_dataset_tryown
);
3607 EXPORT_SYMBOL(dsl_dataset_create_sync
);
3608 EXPORT_SYMBOL(dsl_dataset_create_sync_dd
);
3609 EXPORT_SYMBOL(dsl_dataset_snapshot_check
);
3610 EXPORT_SYMBOL(dsl_dataset_snapshot_sync
);
3611 EXPORT_SYMBOL(dsl_dataset_promote
);
3612 EXPORT_SYMBOL(dsl_dataset_user_hold
);
3613 EXPORT_SYMBOL(dsl_dataset_user_release
);
3614 EXPORT_SYMBOL(dsl_dataset_get_holds
);
3615 EXPORT_SYMBOL(dsl_dataset_get_blkptr
);
3616 EXPORT_SYMBOL(dsl_dataset_set_blkptr
);
3617 EXPORT_SYMBOL(dsl_dataset_get_spa
);
3618 EXPORT_SYMBOL(dsl_dataset_modified_since_snap
);
3619 EXPORT_SYMBOL(dsl_dataset_space_written
);
3620 EXPORT_SYMBOL(dsl_dataset_space_wouldfree
);
3621 EXPORT_SYMBOL(dsl_dataset_sync
);
3622 EXPORT_SYMBOL(dsl_dataset_block_born
);
3623 EXPORT_SYMBOL(dsl_dataset_block_kill
);
3624 EXPORT_SYMBOL(dsl_dataset_block_freeable
);
3625 EXPORT_SYMBOL(dsl_dataset_prev_snap_txg
);
3626 EXPORT_SYMBOL(dsl_dataset_dirty
);
3627 EXPORT_SYMBOL(dsl_dataset_stats
);
3628 EXPORT_SYMBOL(dsl_dataset_fast_stat
);
3629 EXPORT_SYMBOL(dsl_dataset_space
);
3630 EXPORT_SYMBOL(dsl_dataset_fsid_guid
);
3631 EXPORT_SYMBOL(dsl_dsobj_to_dsname
);
3632 EXPORT_SYMBOL(dsl_dataset_check_quota
);
3633 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl
);
3634 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl
);