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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 RackTop Systems.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
29 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
30 * Copyright 2017 Nexenta Systems, Inc.
33 #include <sys/dmu_objset.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_prop.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dmu_traverse.h>
39 #include <sys/dmu_impl.h>
40 #include <sys/dmu_tx.h>
44 #include <sys/zfeature.h>
45 #include <sys/unique.h>
46 #include <sys/zfs_context.h>
47 #include <sys/zfs_ioctl.h>
49 #include <sys/spa_impl.h>
51 #include <sys/zfs_znode.h>
52 #include <sys/zfs_onexit.h>
54 #include <sys/dsl_scan.h>
55 #include <sys/dsl_deadlist.h>
56 #include <sys/dsl_destroy.h>
57 #include <sys/dsl_userhold.h>
58 #include <sys/dsl_bookmark.h>
59 #include <sys/policy.h>
60 #include <sys/dmu_send.h>
61 #include <sys/zio_compress.h>
62 #include <zfs_fletcher.h>
63 #include <sys/zio_checksum.h>
66 * The SPA supports block sizes up to 16MB. However, very large blocks
67 * can have an impact on i/o latency (e.g. tying up a spinning disk for
68 * ~300ms), and also potentially on the memory allocator. Therefore,
69 * we do not allow the recordsize to be set larger than zfs_max_recordsize
70 * (default 1MB). Larger blocks can be created by changing this tunable,
71 * and pools with larger blocks can always be imported and used, regardless
74 int zfs_max_recordsize
= 1 * 1024 * 1024;
76 #define SWITCH64(x, y) \
78 uint64_t __tmp = (x); \
83 #define DS_REF_MAX (1ULL << 62)
85 extern inline dsl_dataset_phys_t
*dsl_dataset_phys(dsl_dataset_t
*ds
);
87 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t
*ds
,
88 uint64_t obj
, dmu_tx_t
*tx
);
89 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t
*ds
,
92 extern int spa_asize_inflation
;
94 static zil_header_t zero_zil
;
97 * Figure out how much of this delta should be propagated to the dsl_dir
98 * layer. If there's a refreservation, that space has already been
99 * partially accounted for in our ancestors.
102 parent_delta(dsl_dataset_t
*ds
, int64_t delta
)
104 dsl_dataset_phys_t
*ds_phys
;
105 uint64_t old_bytes
, new_bytes
;
107 if (ds
->ds_reserved
== 0)
110 ds_phys
= dsl_dataset_phys(ds
);
111 old_bytes
= MAX(ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
112 new_bytes
= MAX(ds_phys
->ds_unique_bytes
+ delta
, ds
->ds_reserved
);
114 ASSERT3U(ABS((int64_t)(new_bytes
- old_bytes
)), <=, ABS(delta
));
115 return (new_bytes
- old_bytes
);
119 dsl_dataset_block_born(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
121 int used
, compressed
, uncompressed
;
124 used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
125 compressed
= BP_GET_PSIZE(bp
);
126 uncompressed
= BP_GET_UCSIZE(bp
);
128 dprintf_bp(bp
, "ds=%p", ds
);
130 ASSERT(dmu_tx_is_syncing(tx
));
131 /* It could have been compressed away to nothing */
134 ASSERT(BP_GET_TYPE(bp
) != DMU_OT_NONE
);
135 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp
)));
137 dsl_pool_mos_diduse_space(tx
->tx_pool
,
138 used
, compressed
, uncompressed
);
142 ASSERT3U(bp
->blk_birth
, >, dsl_dataset_phys(ds
)->ds_prev_snap_txg
);
143 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
144 mutex_enter(&ds
->ds_lock
);
145 delta
= parent_delta(ds
, used
);
146 dsl_dataset_phys(ds
)->ds_referenced_bytes
+= used
;
147 dsl_dataset_phys(ds
)->ds_compressed_bytes
+= compressed
;
148 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
+= uncompressed
;
149 dsl_dataset_phys(ds
)->ds_unique_bytes
+= used
;
151 if (BP_GET_LSIZE(bp
) > SPA_OLD_MAXBLOCKSIZE
) {
152 ds
->ds_feature_activation_needed
[SPA_FEATURE_LARGE_BLOCKS
] =
156 spa_feature_t f
= zio_checksum_to_feature(BP_GET_CHECKSUM(bp
));
157 if (f
!= SPA_FEATURE_NONE
)
158 ds
->ds_feature_activation_needed
[f
] = B_TRUE
;
160 mutex_exit(&ds
->ds_lock
);
161 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
, delta
,
162 compressed
, uncompressed
, tx
);
163 dsl_dir_transfer_space(ds
->ds_dir
, used
- delta
,
164 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
168 * Called when the specified segment has been remapped, and is thus no
169 * longer referenced in the head dataset. The vdev must be indirect.
171 * If the segment is referenced by a snapshot, put it on the remap deadlist.
172 * Otherwise, add this segment to the obsolete spacemap.
175 dsl_dataset_block_remapped(dsl_dataset_t
*ds
, uint64_t vdev
, uint64_t offset
,
176 uint64_t size
, uint64_t birth
, dmu_tx_t
*tx
)
178 spa_t
*spa
= ds
->ds_dir
->dd_pool
->dp_spa
;
180 ASSERT(dmu_tx_is_syncing(tx
));
181 ASSERT(birth
<= tx
->tx_txg
);
182 ASSERT(!ds
->ds_is_snapshot
);
184 if (birth
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
185 spa_vdev_indirect_mark_obsolete(spa
, vdev
, offset
, size
, tx
);
188 dva_t
*dva
= &fakebp
.blk_dva
[0];
192 mutex_enter(&ds
->ds_remap_deadlist_lock
);
193 if (!dsl_dataset_remap_deadlist_exists(ds
)) {
194 dsl_dataset_create_remap_deadlist(ds
, tx
);
196 mutex_exit(&ds
->ds_remap_deadlist_lock
);
199 fakebp
.blk_birth
= birth
;
200 DVA_SET_VDEV(dva
, vdev
);
201 DVA_SET_OFFSET(dva
, offset
);
202 DVA_SET_ASIZE(dva
, size
);
204 dsl_deadlist_insert(&ds
->ds_remap_deadlist
, &fakebp
, tx
);
209 dsl_dataset_block_kill(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
,
212 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
214 int used
= bp_get_dsize_sync(spa
, bp
);
215 int compressed
= BP_GET_PSIZE(bp
);
216 int uncompressed
= BP_GET_UCSIZE(bp
);
221 ASSERT(dmu_tx_is_syncing(tx
));
222 ASSERT(bp
->blk_birth
<= tx
->tx_txg
);
225 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
226 dsl_pool_mos_diduse_space(tx
->tx_pool
,
227 -used
, -compressed
, -uncompressed
);
230 ASSERT3P(tx
->tx_pool
, ==, ds
->ds_dir
->dd_pool
);
232 ASSERT(!ds
->ds_is_snapshot
);
233 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
235 if (bp
->blk_birth
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
238 dprintf_bp(bp
, "freeing ds=%llu", ds
->ds_object
);
239 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
241 mutex_enter(&ds
->ds_lock
);
242 ASSERT(dsl_dataset_phys(ds
)->ds_unique_bytes
>= used
||
243 !DS_UNIQUE_IS_ACCURATE(ds
));
244 delta
= parent_delta(ds
, -used
);
245 dsl_dataset_phys(ds
)->ds_unique_bytes
-= used
;
246 mutex_exit(&ds
->ds_lock
);
247 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
248 delta
, -compressed
, -uncompressed
, tx
);
249 dsl_dir_transfer_space(ds
->ds_dir
, -used
- delta
,
250 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
252 dprintf_bp(bp
, "putting on dead list: %s", "");
255 * We are here as part of zio's write done callback,
256 * which means we're a zio interrupt thread. We can't
257 * call dsl_deadlist_insert() now because it may block
258 * waiting for I/O. Instead, put bp on the deferred
259 * queue and let dsl_pool_sync() finish the job.
261 bplist_append(&ds
->ds_pending_deadlist
, bp
);
263 dsl_deadlist_insert(&ds
->ds_deadlist
, bp
, tx
);
265 ASSERT3U(ds
->ds_prev
->ds_object
, ==,
266 dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
267 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 0);
268 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
269 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
270 ds
->ds_object
&& bp
->blk_birth
>
271 dsl_dataset_phys(ds
->ds_prev
)->ds_prev_snap_txg
) {
272 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
273 mutex_enter(&ds
->ds_prev
->ds_lock
);
274 dsl_dataset_phys(ds
->ds_prev
)->ds_unique_bytes
+= used
;
275 mutex_exit(&ds
->ds_prev
->ds_lock
);
277 if (bp
->blk_birth
> ds
->ds_dir
->dd_origin_txg
) {
278 dsl_dir_transfer_space(ds
->ds_dir
, used
,
279 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
282 mutex_enter(&ds
->ds_lock
);
283 ASSERT3U(dsl_dataset_phys(ds
)->ds_referenced_bytes
, >=, used
);
284 dsl_dataset_phys(ds
)->ds_referenced_bytes
-= used
;
285 ASSERT3U(dsl_dataset_phys(ds
)->ds_compressed_bytes
, >=, compressed
);
286 dsl_dataset_phys(ds
)->ds_compressed_bytes
-= compressed
;
287 ASSERT3U(dsl_dataset_phys(ds
)->ds_uncompressed_bytes
, >=, uncompressed
);
288 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
-= uncompressed
;
289 mutex_exit(&ds
->ds_lock
);
295 * We have to release the fsid syncronously or we risk that a subsequent
296 * mount of the same dataset will fail to unique_insert the fsid. This
297 * failure would manifest itself as the fsid of this dataset changing
298 * between mounts which makes NFS clients quite unhappy.
301 dsl_dataset_evict_sync(void *dbu
)
303 dsl_dataset_t
*ds
= dbu
;
305 ASSERT(ds
->ds_owner
== NULL
);
307 unique_remove(ds
->ds_fsid_guid
);
311 dsl_dataset_evict_async(void *dbu
)
313 dsl_dataset_t
*ds
= dbu
;
315 ASSERT(ds
->ds_owner
== NULL
);
319 if (ds
->ds_objset
!= NULL
)
320 dmu_objset_evict(ds
->ds_objset
);
323 dsl_dataset_rele(ds
->ds_prev
, ds
);
327 bplist_destroy(&ds
->ds_pending_deadlist
);
328 if (dsl_deadlist_is_open(&ds
->ds_deadlist
))
329 dsl_deadlist_close(&ds
->ds_deadlist
);
330 if (dsl_deadlist_is_open(&ds
->ds_remap_deadlist
))
331 dsl_deadlist_close(&ds
->ds_remap_deadlist
);
333 dsl_dir_async_rele(ds
->ds_dir
, ds
);
335 ASSERT(!list_link_active(&ds
->ds_synced_link
));
337 list_destroy(&ds
->ds_prop_cbs
);
338 mutex_destroy(&ds
->ds_lock
);
339 mutex_destroy(&ds
->ds_opening_lock
);
340 mutex_destroy(&ds
->ds_sendstream_lock
);
341 mutex_destroy(&ds
->ds_remap_deadlist_lock
);
342 refcount_destroy(&ds
->ds_longholds
);
343 rrw_destroy(&ds
->ds_bp_rwlock
);
345 kmem_free(ds
, sizeof (dsl_dataset_t
));
349 dsl_dataset_get_snapname(dsl_dataset_t
*ds
)
351 dsl_dataset_phys_t
*headphys
;
354 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
355 objset_t
*mos
= dp
->dp_meta_objset
;
357 if (ds
->ds_snapname
[0])
359 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0)
362 err
= dmu_bonus_hold(mos
, dsl_dir_phys(ds
->ds_dir
)->dd_head_dataset_obj
,
366 headphys
= headdbuf
->db_data
;
367 err
= zap_value_search(dp
->dp_meta_objset
,
368 headphys
->ds_snapnames_zapobj
, ds
->ds_object
, 0, ds
->ds_snapname
);
369 if (err
!= 0 && zfs_recover
== B_TRUE
) {
371 (void) snprintf(ds
->ds_snapname
, sizeof (ds
->ds_snapname
),
372 "SNAPOBJ=%llu-ERR=%d",
373 (unsigned long long)ds
->ds_object
, err
);
375 dmu_buf_rele(headdbuf
, FTAG
);
380 dsl_dataset_snap_lookup(dsl_dataset_t
*ds
, const char *name
, uint64_t *value
)
382 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
383 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
387 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
390 err
= zap_lookup_norm(mos
, snapobj
, name
, 8, 1,
391 value
, mt
, NULL
, 0, NULL
);
392 if (err
== ENOTSUP
&& (mt
& MT_NORMALIZE
))
393 err
= zap_lookup(mos
, snapobj
, name
, 8, 1, value
);
398 dsl_dataset_snap_remove(dsl_dataset_t
*ds
, const char *name
, dmu_tx_t
*tx
,
401 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
402 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
406 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
408 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
411 err
= zap_remove_norm(mos
, snapobj
, name
, mt
, tx
);
412 if (err
== ENOTSUP
&& (mt
& MT_NORMALIZE
))
413 err
= zap_remove(mos
, snapobj
, name
, tx
);
415 if (err
== 0 && adj_cnt
)
416 dsl_fs_ss_count_adjust(ds
->ds_dir
, -1,
417 DD_FIELD_SNAPSHOT_COUNT
, tx
);
423 dsl_dataset_try_add_ref(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *tag
)
425 dmu_buf_t
*dbuf
= ds
->ds_dbuf
;
426 boolean_t result
= B_FALSE
;
428 if (dbuf
!= NULL
&& dmu_buf_try_add_ref(dbuf
, dp
->dp_meta_objset
,
429 ds
->ds_object
, DMU_BONUS_BLKID
, tag
)) {
431 if (ds
== dmu_buf_get_user(dbuf
))
434 dmu_buf_rele(dbuf
, tag
);
441 dsl_dataset_hold_obj_flags(dsl_pool_t
*dp
, uint64_t dsobj
,
442 ds_hold_flags_t flags
, void *tag
, dsl_dataset_t
**dsp
)
444 objset_t
*mos
= dp
->dp_meta_objset
;
448 dmu_object_info_t doi
;
450 ASSERT(dsl_pool_config_held(dp
));
452 err
= dmu_bonus_hold(mos
, dsobj
, tag
, &dbuf
);
456 /* Make sure dsobj has the correct object type. */
457 dmu_object_info_from_db(dbuf
, &doi
);
458 if (doi
.doi_bonus_type
!= DMU_OT_DSL_DATASET
) {
459 dmu_buf_rele(dbuf
, tag
);
460 return (SET_ERROR(EINVAL
));
463 ds
= dmu_buf_get_user(dbuf
);
465 dsl_dataset_t
*winner
= NULL
;
467 ds
= kmem_zalloc(sizeof (dsl_dataset_t
), KM_SLEEP
);
469 ds
->ds_object
= dsobj
;
470 ds
->ds_is_snapshot
= dsl_dataset_phys(ds
)->ds_num_children
!= 0;
471 list_link_init(&ds
->ds_synced_link
);
473 err
= dsl_dir_hold_obj(dp
, dsl_dataset_phys(ds
)->ds_dir_obj
,
474 NULL
, ds
, &ds
->ds_dir
);
476 kmem_free(ds
, sizeof (dsl_dataset_t
));
477 dmu_buf_rele(dbuf
, tag
);
481 mutex_init(&ds
->ds_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
482 mutex_init(&ds
->ds_opening_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
483 mutex_init(&ds
->ds_sendstream_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
484 mutex_init(&ds
->ds_remap_deadlist_lock
,
485 NULL
, MUTEX_DEFAULT
, NULL
);
486 rrw_init(&ds
->ds_bp_rwlock
, B_FALSE
);
487 refcount_create(&ds
->ds_longholds
);
489 bplist_create(&ds
->ds_pending_deadlist
);
491 list_create(&ds
->ds_sendstreams
, sizeof (dmu_sendarg_t
),
492 offsetof(dmu_sendarg_t
, dsa_link
));
494 list_create(&ds
->ds_prop_cbs
, sizeof (dsl_prop_cb_record_t
),
495 offsetof(dsl_prop_cb_record_t
, cbr_ds_node
));
497 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
500 for (f
= 0; f
< SPA_FEATURES
; f
++) {
501 if (!(spa_feature_table
[f
].fi_flags
&
502 ZFEATURE_FLAG_PER_DATASET
))
504 err
= zap_contains(mos
, dsobj
,
505 spa_feature_table
[f
].fi_guid
);
507 ds
->ds_feature_inuse
[f
] = B_TRUE
;
509 ASSERT3U(err
, ==, ENOENT
);
515 if (!ds
->ds_is_snapshot
) {
516 ds
->ds_snapname
[0] = '\0';
517 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
518 err
= dsl_dataset_hold_obj(dp
,
519 dsl_dataset_phys(ds
)->ds_prev_snap_obj
,
522 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
523 int zaperr
= zap_lookup(mos
, ds
->ds_object
,
524 DS_FIELD_BOOKMARK_NAMES
,
525 sizeof (ds
->ds_bookmarks
), 1,
527 if (zaperr
!= ENOENT
)
531 if (zfs_flags
& ZFS_DEBUG_SNAPNAMES
)
532 err
= dsl_dataset_get_snapname(ds
);
534 dsl_dataset_phys(ds
)->ds_userrefs_obj
!= 0) {
536 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
537 dsl_dataset_phys(ds
)->ds_userrefs_obj
,
542 if (err
== 0 && !ds
->ds_is_snapshot
) {
543 err
= dsl_prop_get_int_ds(ds
,
544 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
547 err
= dsl_prop_get_int_ds(ds
,
548 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
552 ds
->ds_reserved
= ds
->ds_quota
= 0;
555 dsl_deadlist_open(&ds
->ds_deadlist
,
556 mos
, dsl_dataset_phys(ds
)->ds_deadlist_obj
);
557 uint64_t remap_deadlist_obj
=
558 dsl_dataset_get_remap_deadlist_object(ds
);
559 if (remap_deadlist_obj
!= 0) {
560 dsl_deadlist_open(&ds
->ds_remap_deadlist
, mos
,
564 dmu_buf_init_user(&ds
->ds_dbu
, dsl_dataset_evict_sync
,
565 dsl_dataset_evict_async
, &ds
->ds_dbuf
);
567 winner
= dmu_buf_set_user_ie(dbuf
, &ds
->ds_dbu
);
569 if (err
!= 0 || winner
!= NULL
) {
570 bplist_destroy(&ds
->ds_pending_deadlist
);
571 dsl_deadlist_close(&ds
->ds_deadlist
);
572 if (dsl_deadlist_is_open(&ds
->ds_remap_deadlist
))
573 dsl_deadlist_close(&ds
->ds_remap_deadlist
);
575 dsl_dataset_rele(ds
->ds_prev
, ds
);
576 dsl_dir_rele(ds
->ds_dir
, ds
);
577 mutex_destroy(&ds
->ds_lock
);
578 mutex_destroy(&ds
->ds_opening_lock
);
579 mutex_destroy(&ds
->ds_sendstream_lock
);
580 refcount_destroy(&ds
->ds_longholds
);
581 kmem_free(ds
, sizeof (dsl_dataset_t
));
583 dmu_buf_rele(dbuf
, tag
);
589 unique_insert(dsl_dataset_phys(ds
)->ds_fsid_guid
);
590 if (ds
->ds_fsid_guid
!=
591 dsl_dataset_phys(ds
)->ds_fsid_guid
) {
592 zfs_dbgmsg("ds_fsid_guid changed from "
593 "%llx to %llx for pool %s dataset id %llu",
595 dsl_dataset_phys(ds
)->ds_fsid_guid
,
596 (long long)ds
->ds_fsid_guid
,
597 spa_name(dp
->dp_spa
),
602 ASSERT3P(ds
->ds_dbuf
, ==, dbuf
);
603 ASSERT3P(dsl_dataset_phys(ds
), ==, dbuf
->db_data
);
604 ASSERT(dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0 ||
605 spa_version(dp
->dp_spa
) < SPA_VERSION_ORIGIN
||
606 dp
->dp_origin_snap
== NULL
|| ds
== dp
->dp_origin_snap
);
609 if ((flags
& DS_HOLD_FLAG_DECRYPT
) && ds
->ds_dir
->dd_crypto_obj
!= 0) {
610 err
= spa_keystore_create_mapping(dp
->dp_spa
, ds
, ds
);
612 dsl_dataset_rele(ds
, tag
);
613 return (SET_ERROR(EACCES
));
621 dsl_dataset_hold_obj(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
624 return (dsl_dataset_hold_obj_flags(dp
, dsobj
, 0, tag
, dsp
));
628 dsl_dataset_hold_flags(dsl_pool_t
*dp
, const char *name
, ds_hold_flags_t flags
,
629 void *tag
, dsl_dataset_t
**dsp
)
632 const char *snapname
;
637 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, &snapname
);
641 ASSERT(dsl_pool_config_held(dp
));
642 obj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
644 err
= dsl_dataset_hold_obj_flags(dp
, obj
, flags
, tag
, &ds
);
646 err
= SET_ERROR(ENOENT
);
648 /* we may be looking for a snapshot */
649 if (err
== 0 && snapname
!= NULL
) {
650 dsl_dataset_t
*snap_ds
;
652 if (*snapname
++ != '@') {
653 dsl_dataset_rele_flags(ds
, flags
, tag
);
654 dsl_dir_rele(dd
, FTAG
);
655 return (SET_ERROR(ENOENT
));
658 dprintf("looking for snapshot '%s'\n", snapname
);
659 err
= dsl_dataset_snap_lookup(ds
, snapname
, &obj
);
661 err
= dsl_dataset_hold_obj_flags(dp
, obj
, flags
, tag
,
664 dsl_dataset_rele_flags(ds
, flags
, tag
);
667 mutex_enter(&snap_ds
->ds_lock
);
668 if (snap_ds
->ds_snapname
[0] == 0)
669 (void) strlcpy(snap_ds
->ds_snapname
, snapname
,
670 sizeof (snap_ds
->ds_snapname
));
671 mutex_exit(&snap_ds
->ds_lock
);
677 dsl_dir_rele(dd
, FTAG
);
682 dsl_dataset_hold(dsl_pool_t
*dp
, const char *name
, void *tag
,
685 return (dsl_dataset_hold_flags(dp
, name
, 0, tag
, dsp
));
689 dsl_dataset_own_obj(dsl_pool_t
*dp
, uint64_t dsobj
, ds_hold_flags_t flags
,
690 void *tag
, dsl_dataset_t
**dsp
)
692 int err
= dsl_dataset_hold_obj_flags(dp
, dsobj
, flags
, tag
, dsp
);
695 if (!dsl_dataset_tryown(*dsp
, tag
)) {
696 dsl_dataset_rele_flags(*dsp
, flags
, tag
);
698 return (SET_ERROR(EBUSY
));
704 dsl_dataset_own(dsl_pool_t
*dp
, const char *name
, ds_hold_flags_t flags
,
705 void *tag
, dsl_dataset_t
**dsp
)
707 int err
= dsl_dataset_hold_flags(dp
, name
, flags
, tag
, dsp
);
710 if (!dsl_dataset_tryown(*dsp
, tag
)) {
711 dsl_dataset_rele_flags(*dsp
, flags
, tag
);
712 return (SET_ERROR(EBUSY
));
718 * See the comment above dsl_pool_hold() for details. In summary, a long
719 * hold is used to prevent destruction of a dataset while the pool hold
720 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
722 * The dataset and pool must be held when this function is called. After it
723 * is called, the pool hold may be released while the dataset is still held
727 dsl_dataset_long_hold(dsl_dataset_t
*ds
, void *tag
)
729 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
730 (void) zfs_refcount_add(&ds
->ds_longholds
, tag
);
734 dsl_dataset_long_rele(dsl_dataset_t
*ds
, void *tag
)
736 (void) refcount_remove(&ds
->ds_longholds
, tag
);
739 /* Return B_TRUE if there are any long holds on this dataset. */
741 dsl_dataset_long_held(dsl_dataset_t
*ds
)
743 return (!refcount_is_zero(&ds
->ds_longholds
));
747 dsl_dataset_name(dsl_dataset_t
*ds
, char *name
)
750 (void) strcpy(name
, "mos");
752 dsl_dir_name(ds
->ds_dir
, name
);
753 VERIFY0(dsl_dataset_get_snapname(ds
));
754 if (ds
->ds_snapname
[0]) {
755 VERIFY3U(strlcat(name
, "@", ZFS_MAX_DATASET_NAME_LEN
),
756 <, ZFS_MAX_DATASET_NAME_LEN
);
758 * We use a "recursive" mutex so that we
759 * can call dprintf_ds() with ds_lock held.
761 if (!MUTEX_HELD(&ds
->ds_lock
)) {
762 mutex_enter(&ds
->ds_lock
);
763 VERIFY3U(strlcat(name
, ds
->ds_snapname
,
764 ZFS_MAX_DATASET_NAME_LEN
), <,
765 ZFS_MAX_DATASET_NAME_LEN
);
766 mutex_exit(&ds
->ds_lock
);
768 VERIFY3U(strlcat(name
, ds
->ds_snapname
,
769 ZFS_MAX_DATASET_NAME_LEN
), <,
770 ZFS_MAX_DATASET_NAME_LEN
);
777 dsl_dataset_namelen(dsl_dataset_t
*ds
)
779 VERIFY0(dsl_dataset_get_snapname(ds
));
780 mutex_enter(&ds
->ds_lock
);
781 int len
= strlen(ds
->ds_snapname
);
782 /* add '@' if ds is a snap */
785 len
+= dsl_dir_namelen(ds
->ds_dir
);
786 mutex_exit(&ds
->ds_lock
);
791 dsl_dataset_rele_flags(dsl_dataset_t
*ds
, ds_hold_flags_t flags
, void *tag
)
793 if (ds
->ds_dir
!= NULL
&& ds
->ds_dir
->dd_crypto_obj
!= 0 &&
794 (flags
& DS_HOLD_FLAG_DECRYPT
)) {
795 (void) spa_keystore_remove_mapping(ds
->ds_dir
->dd_pool
->dp_spa
,
799 dmu_buf_rele(ds
->ds_dbuf
, tag
);
803 dsl_dataset_rele(dsl_dataset_t
*ds
, void *tag
)
805 dsl_dataset_rele_flags(ds
, 0, tag
);
809 dsl_dataset_disown(dsl_dataset_t
*ds
, ds_hold_flags_t flags
, void *tag
)
811 ASSERT3P(ds
->ds_owner
, ==, tag
);
812 ASSERT(ds
->ds_dbuf
!= NULL
);
814 mutex_enter(&ds
->ds_lock
);
816 mutex_exit(&ds
->ds_lock
);
817 dsl_dataset_long_rele(ds
, tag
);
818 dsl_dataset_rele_flags(ds
, flags
, tag
);
822 dsl_dataset_tryown(dsl_dataset_t
*ds
, void *tag
)
824 boolean_t gotit
= FALSE
;
826 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
827 mutex_enter(&ds
->ds_lock
);
828 if (ds
->ds_owner
== NULL
&& !DS_IS_INCONSISTENT(ds
)) {
830 dsl_dataset_long_hold(ds
, tag
);
833 mutex_exit(&ds
->ds_lock
);
838 dsl_dataset_has_owner(dsl_dataset_t
*ds
)
841 mutex_enter(&ds
->ds_lock
);
842 rv
= (ds
->ds_owner
!= NULL
);
843 mutex_exit(&ds
->ds_lock
);
848 dsl_dataset_activate_feature(uint64_t dsobj
, spa_feature_t f
, dmu_tx_t
*tx
)
850 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
851 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
854 VERIFY(spa_feature_table
[f
].fi_flags
& ZFEATURE_FLAG_PER_DATASET
);
856 spa_feature_incr(spa
, f
, tx
);
857 dmu_object_zapify(mos
, dsobj
, DMU_OT_DSL_DATASET
, tx
);
859 VERIFY0(zap_add(mos
, dsobj
, spa_feature_table
[f
].fi_guid
,
860 sizeof (zero
), 1, &zero
, tx
));
864 dsl_dataset_deactivate_feature(uint64_t dsobj
, spa_feature_t f
, dmu_tx_t
*tx
)
866 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
867 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
869 VERIFY(spa_feature_table
[f
].fi_flags
& ZFEATURE_FLAG_PER_DATASET
);
871 VERIFY0(zap_remove(mos
, dsobj
, spa_feature_table
[f
].fi_guid
, tx
));
872 spa_feature_decr(spa
, f
, tx
);
876 dsl_dataset_create_sync_dd(dsl_dir_t
*dd
, dsl_dataset_t
*origin
,
877 dsl_crypto_params_t
*dcp
, uint64_t flags
, dmu_tx_t
*tx
)
879 dsl_pool_t
*dp
= dd
->dd_pool
;
881 dsl_dataset_phys_t
*dsphys
;
883 objset_t
*mos
= dp
->dp_meta_objset
;
886 origin
= dp
->dp_origin_snap
;
888 ASSERT(origin
== NULL
|| origin
->ds_dir
->dd_pool
== dp
);
889 ASSERT(origin
== NULL
|| dsl_dataset_phys(origin
)->ds_num_children
> 0);
890 ASSERT(dmu_tx_is_syncing(tx
));
891 ASSERT(dsl_dir_phys(dd
)->dd_head_dataset_obj
== 0);
893 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
894 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
895 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
896 dmu_buf_will_dirty(dbuf
, tx
);
897 dsphys
= dbuf
->db_data
;
898 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
899 dsphys
->ds_dir_obj
= dd
->dd_object
;
900 dsphys
->ds_flags
= flags
;
901 dsphys
->ds_fsid_guid
= unique_create();
902 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
903 sizeof (dsphys
->ds_guid
));
904 dsphys
->ds_snapnames_zapobj
=
905 zap_create_norm(mos
, U8_TEXTPREP_TOUPPER
, DMU_OT_DSL_DS_SNAP_MAP
,
907 dsphys
->ds_creation_time
= gethrestime_sec();
908 dsphys
->ds_creation_txg
= tx
->tx_txg
== TXG_INITIAL
? 1 : tx
->tx_txg
;
910 if (origin
== NULL
) {
911 dsphys
->ds_deadlist_obj
= dsl_deadlist_alloc(mos
, tx
);
913 dsl_dataset_t
*ohds
; /* head of the origin snapshot */
915 dsphys
->ds_prev_snap_obj
= origin
->ds_object
;
916 dsphys
->ds_prev_snap_txg
=
917 dsl_dataset_phys(origin
)->ds_creation_txg
;
918 dsphys
->ds_referenced_bytes
=
919 dsl_dataset_phys(origin
)->ds_referenced_bytes
;
920 dsphys
->ds_compressed_bytes
=
921 dsl_dataset_phys(origin
)->ds_compressed_bytes
;
922 dsphys
->ds_uncompressed_bytes
=
923 dsl_dataset_phys(origin
)->ds_uncompressed_bytes
;
924 rrw_enter(&origin
->ds_bp_rwlock
, RW_READER
, FTAG
);
925 dsphys
->ds_bp
= dsl_dataset_phys(origin
)->ds_bp
;
926 rrw_exit(&origin
->ds_bp_rwlock
, FTAG
);
929 * Inherit flags that describe the dataset's contents
930 * (INCONSISTENT) or properties (Case Insensitive).
932 dsphys
->ds_flags
|= dsl_dataset_phys(origin
)->ds_flags
&
933 (DS_FLAG_INCONSISTENT
| DS_FLAG_CI_DATASET
);
935 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++) {
936 if (origin
->ds_feature_inuse
[f
])
937 dsl_dataset_activate_feature(dsobj
, f
, tx
);
940 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
941 dsl_dataset_phys(origin
)->ds_num_children
++;
943 VERIFY0(dsl_dataset_hold_obj(dp
,
944 dsl_dir_phys(origin
->ds_dir
)->dd_head_dataset_obj
,
946 dsphys
->ds_deadlist_obj
= dsl_deadlist_clone(&ohds
->ds_deadlist
,
947 dsphys
->ds_prev_snap_txg
, dsphys
->ds_prev_snap_obj
, tx
);
948 dsl_dataset_rele(ohds
, FTAG
);
950 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_NEXT_CLONES
) {
951 if (dsl_dataset_phys(origin
)->ds_next_clones_obj
== 0) {
952 dsl_dataset_phys(origin
)->ds_next_clones_obj
=
954 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
956 VERIFY0(zap_add_int(mos
,
957 dsl_dataset_phys(origin
)->ds_next_clones_obj
,
961 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
962 dsl_dir_phys(dd
)->dd_origin_obj
= origin
->ds_object
;
963 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
964 if (dsl_dir_phys(origin
->ds_dir
)->dd_clones
== 0) {
965 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
966 dsl_dir_phys(origin
->ds_dir
)->dd_clones
=
968 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
970 VERIFY0(zap_add_int(mos
,
971 dsl_dir_phys(origin
->ds_dir
)->dd_clones
,
976 /* handle encryption */
977 dsl_dataset_create_crypt_sync(dsobj
, dd
, origin
, dcp
, tx
);
979 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
980 dsphys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
982 dmu_buf_rele(dbuf
, FTAG
);
984 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
985 dsl_dir_phys(dd
)->dd_head_dataset_obj
= dsobj
;
991 dsl_dataset_zero_zil(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
995 VERIFY0(dmu_objset_from_ds(ds
, &os
));
996 if (bcmp(&os
->os_zil_header
, &zero_zil
, sizeof (zero_zil
)) != 0) {
997 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1000 bzero(&os
->os_zil_header
, sizeof (os
->os_zil_header
));
1001 if (os
->os_encrypted
)
1002 os
->os_next_write_raw
[tx
->tx_txg
& TXG_MASK
] = B_TRUE
;
1004 zio
= zio_root(dp
->dp_spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
1005 dsl_dataset_sync(ds
, zio
, tx
);
1006 VERIFY0(zio_wait(zio
));
1008 /* dsl_dataset_sync_done will drop this reference. */
1009 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
1010 dsl_dataset_sync_done(ds
, tx
);
1015 dsl_dataset_create_sync(dsl_dir_t
*pdd
, const char *lastname
,
1016 dsl_dataset_t
*origin
, uint64_t flags
, cred_t
*cr
,
1017 dsl_crypto_params_t
*dcp
, dmu_tx_t
*tx
)
1019 dsl_pool_t
*dp
= pdd
->dd_pool
;
1020 uint64_t dsobj
, ddobj
;
1023 ASSERT(dmu_tx_is_syncing(tx
));
1024 ASSERT(lastname
[0] != '@');
1026 ddobj
= dsl_dir_create_sync(dp
, pdd
, lastname
, tx
);
1027 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, lastname
, FTAG
, &dd
));
1029 dsobj
= dsl_dataset_create_sync_dd(dd
, origin
, dcp
,
1030 flags
& ~DS_CREATE_FLAG_NODIRTY
, tx
);
1032 dsl_deleg_set_create_perms(dd
, tx
, cr
);
1035 * Since we're creating a new node we know it's a leaf, so we can
1036 * initialize the counts if the limit feature is active.
1038 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
1040 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
1042 dsl_dir_zapify(dd
, tx
);
1043 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
1044 sizeof (cnt
), 1, &cnt
, tx
));
1045 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
1046 sizeof (cnt
), 1, &cnt
, tx
));
1049 dsl_dir_rele(dd
, FTAG
);
1052 * If we are creating a clone, make sure we zero out any stale
1053 * data from the origin snapshots zil header.
1055 if (origin
!= NULL
&& !(flags
& DS_CREATE_FLAG_NODIRTY
)) {
1058 VERIFY0(dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
1059 dsl_dataset_zero_zil(ds
, tx
);
1060 dsl_dataset_rele(ds
, FTAG
);
1067 * The unique space in the head dataset can be calculated by subtracting
1068 * the space used in the most recent snapshot, that is still being used
1069 * in this file system, from the space currently in use. To figure out
1070 * the space in the most recent snapshot still in use, we need to take
1071 * the total space used in the snapshot and subtract out the space that
1072 * has been freed up since the snapshot was taken.
1075 dsl_dataset_recalc_head_uniq(dsl_dataset_t
*ds
)
1078 uint64_t dlused
, dlcomp
, dluncomp
;
1080 ASSERT(!ds
->ds_is_snapshot
);
1082 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0)
1083 mrs_used
= dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
;
1087 dsl_deadlist_space(&ds
->ds_deadlist
, &dlused
, &dlcomp
, &dluncomp
);
1089 ASSERT3U(dlused
, <=, mrs_used
);
1090 dsl_dataset_phys(ds
)->ds_unique_bytes
=
1091 dsl_dataset_phys(ds
)->ds_referenced_bytes
- (mrs_used
- dlused
);
1093 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) >=
1094 SPA_VERSION_UNIQUE_ACCURATE
)
1095 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1099 dsl_dataset_remove_from_next_clones(dsl_dataset_t
*ds
, uint64_t obj
,
1102 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1103 ASSERTV(uint64_t count
);
1106 ASSERT(dsl_dataset_phys(ds
)->ds_num_children
>= 2);
1107 err
= zap_remove_int(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1110 * The err should not be ENOENT, but a bug in a previous version
1111 * of the code could cause upgrade_clones_cb() to not set
1112 * ds_next_snap_obj when it should, leading to a missing entry.
1113 * If we knew that the pool was created after
1114 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1115 * ENOENT. However, at least we can check that we don't have
1116 * too many entries in the next_clones_obj even after failing to
1121 ASSERT0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1123 ASSERT3U(count
, <=, dsl_dataset_phys(ds
)->ds_num_children
- 2);
1128 dsl_dataset_get_blkptr(dsl_dataset_t
*ds
)
1130 return (&dsl_dataset_phys(ds
)->ds_bp
);
1134 dsl_dataset_get_spa(dsl_dataset_t
*ds
)
1136 return (ds
->ds_dir
->dd_pool
->dp_spa
);
1140 dsl_dataset_dirty(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1144 if (ds
== NULL
) /* this is the meta-objset */
1147 ASSERT(ds
->ds_objset
!= NULL
);
1149 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
!= 0)
1150 panic("dirtying snapshot!");
1152 /* Must not dirty a dataset in the same txg where it got snapshotted. */
1153 ASSERT3U(tx
->tx_txg
, >, dsl_dataset_phys(ds
)->ds_prev_snap_txg
);
1155 dp
= ds
->ds_dir
->dd_pool
;
1156 if (txg_list_add(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
1157 /* up the hold count until we can be written out */
1158 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
1163 dsl_dataset_snapshot_reserve_space(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1167 if (!dmu_tx_is_syncing(tx
))
1171 * If there's an fs-only reservation, any blocks that might become
1172 * owned by the snapshot dataset must be accommodated by space
1173 * outside of the reservation.
1175 ASSERT(ds
->ds_reserved
== 0 || DS_UNIQUE_IS_ACCURATE(ds
));
1176 asize
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
, ds
->ds_reserved
);
1177 if (asize
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
1178 return (SET_ERROR(ENOSPC
));
1181 * Propagate any reserved space for this snapshot to other
1182 * snapshot checks in this sync group.
1185 dsl_dir_willuse_space(ds
->ds_dir
, asize
, tx
);
1191 dsl_dataset_snapshot_check_impl(dsl_dataset_t
*ds
, const char *snapname
,
1192 dmu_tx_t
*tx
, boolean_t recv
, uint64_t cnt
, cred_t
*cr
)
1197 ds
->ds_trysnap_txg
= tx
->tx_txg
;
1199 if (!dmu_tx_is_syncing(tx
))
1203 * We don't allow multiple snapshots of the same txg. If there
1204 * is already one, try again.
1206 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
>= tx
->tx_txg
)
1207 return (SET_ERROR(EAGAIN
));
1210 * Check for conflicting snapshot name.
1212 error
= dsl_dataset_snap_lookup(ds
, snapname
, &value
);
1214 return (SET_ERROR(EEXIST
));
1215 if (error
!= ENOENT
)
1219 * We don't allow taking snapshots of inconsistent datasets, such as
1220 * those into which we are currently receiving. However, if we are
1221 * creating this snapshot as part of a receive, this check will be
1222 * executed atomically with respect to the completion of the receive
1223 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1224 * case we ignore this, knowing it will be fixed up for us shortly in
1225 * dmu_recv_end_sync().
1227 if (!recv
&& DS_IS_INCONSISTENT(ds
))
1228 return (SET_ERROR(EBUSY
));
1231 * Skip the check for temporary snapshots or if we have already checked
1232 * the counts in dsl_dataset_snapshot_check. This means we really only
1233 * check the count here when we're receiving a stream.
1235 if (cnt
!= 0 && cr
!= NULL
) {
1236 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1237 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, cr
);
1242 error
= dsl_dataset_snapshot_reserve_space(ds
, tx
);
1250 dsl_dataset_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1252 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1253 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1258 * Pre-compute how many total new snapshots will be created for each
1259 * level in the tree and below. This is needed for validating the
1260 * snapshot limit when either taking a recursive snapshot or when
1261 * taking multiple snapshots.
1263 * The problem is that the counts are not actually adjusted when
1264 * we are checking, only when we finally sync. For a single snapshot,
1265 * this is easy, the count will increase by 1 at each node up the tree,
1266 * but its more complicated for the recursive/multiple snapshot case.
1268 * The dsl_fs_ss_limit_check function does recursively check the count
1269 * at each level up the tree but since it is validating each snapshot
1270 * independently we need to be sure that we are validating the complete
1271 * count for the entire set of snapshots. We do this by rolling up the
1272 * counts for each component of the name into an nvlist and then
1273 * checking each of those cases with the aggregated count.
1275 * This approach properly handles not only the recursive snapshot
1276 * case (where we get all of those on the ddsa_snaps list) but also
1277 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1278 * validate the limit on 'a' using a count of 2).
1280 * We validate the snapshot names in the third loop and only report
1283 if (dmu_tx_is_syncing(tx
)) {
1285 nvlist_t
*cnt_track
= NULL
;
1286 cnt_track
= fnvlist_alloc();
1288 nm
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1290 /* Rollup aggregated counts into the cnt_track list */
1291 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1293 pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1297 (void) strlcpy(nm
, nvpair_name(pair
), MAXPATHLEN
);
1298 pdelim
= strchr(nm
, '@');
1304 if (nvlist_lookup_uint64(cnt_track
, nm
,
1306 /* update existing entry */
1307 fnvlist_add_uint64(cnt_track
, nm
,
1311 fnvlist_add_uint64(cnt_track
, nm
, 1);
1314 pdelim
= strrchr(nm
, '/');
1317 } while (pdelim
!= NULL
);
1320 kmem_free(nm
, MAXPATHLEN
);
1322 /* Check aggregated counts at each level */
1323 for (pair
= nvlist_next_nvpair(cnt_track
, NULL
);
1324 pair
!= NULL
; pair
= nvlist_next_nvpair(cnt_track
, pair
)) {
1330 name
= nvpair_name(pair
);
1331 cnt
= fnvpair_value_uint64(pair
);
1334 error
= dsl_dataset_hold(dp
, name
, FTAG
, &ds
);
1336 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1337 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
,
1339 dsl_dataset_rele(ds
, FTAG
);
1343 if (ddsa
->ddsa_errors
!= NULL
)
1344 fnvlist_add_int32(ddsa
->ddsa_errors
,
1347 /* only report one error for this check */
1351 nvlist_free(cnt_track
);
1354 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1355 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1358 char *name
, *atp
= NULL
;
1359 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
1361 name
= nvpair_name(pair
);
1362 if (strlen(name
) >= ZFS_MAX_DATASET_NAME_LEN
)
1363 error
= SET_ERROR(ENAMETOOLONG
);
1365 atp
= strchr(name
, '@');
1367 error
= SET_ERROR(EINVAL
);
1369 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1372 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
1374 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1375 error
= dsl_dataset_snapshot_check_impl(ds
,
1376 atp
+ 1, tx
, B_FALSE
, 0, NULL
);
1377 dsl_dataset_rele(ds
, FTAG
);
1381 if (ddsa
->ddsa_errors
!= NULL
) {
1382 fnvlist_add_int32(ddsa
->ddsa_errors
,
1393 dsl_dataset_snapshot_sync_impl(dsl_dataset_t
*ds
, const char *snapname
,
1396 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1398 dsl_dataset_phys_t
*dsphys
;
1399 uint64_t dsobj
, crtxg
;
1400 objset_t
*mos
= dp
->dp_meta_objset
;
1401 ASSERTV(static zil_header_t zero_zil
);
1402 ASSERTV(objset_t
*os
);
1404 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
1407 * If we are on an old pool, the zil must not be active, in which
1408 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1410 ASSERT(spa_version(dmu_tx_pool(tx
)->dp_spa
) >= SPA_VERSION_FAST_SNAP
||
1411 dmu_objset_from_ds(ds
, &os
) != 0 ||
1412 bcmp(&os
->os_phys
->os_zil_header
, &zero_zil
,
1413 sizeof (zero_zil
)) == 0);
1415 /* Should not snapshot a dirty dataset. */
1416 ASSERT(!txg_list_member(&ds
->ds_dir
->dd_pool
->dp_dirty_datasets
,
1419 dsl_fs_ss_count_adjust(ds
->ds_dir
, 1, DD_FIELD_SNAPSHOT_COUNT
, tx
);
1422 * The origin's ds_creation_txg has to be < TXG_INITIAL
1424 if (strcmp(snapname
, ORIGIN_DIR_NAME
) == 0)
1429 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
1430 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
1431 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
1432 dmu_buf_will_dirty(dbuf
, tx
);
1433 dsphys
= dbuf
->db_data
;
1434 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
1435 dsphys
->ds_dir_obj
= ds
->ds_dir
->dd_object
;
1436 dsphys
->ds_fsid_guid
= unique_create();
1437 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
1438 sizeof (dsphys
->ds_guid
));
1439 dsphys
->ds_prev_snap_obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1440 dsphys
->ds_prev_snap_txg
= dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
1441 dsphys
->ds_next_snap_obj
= ds
->ds_object
;
1442 dsphys
->ds_num_children
= 1;
1443 dsphys
->ds_creation_time
= gethrestime_sec();
1444 dsphys
->ds_creation_txg
= crtxg
;
1445 dsphys
->ds_deadlist_obj
= dsl_dataset_phys(ds
)->ds_deadlist_obj
;
1446 dsphys
->ds_referenced_bytes
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1447 dsphys
->ds_compressed_bytes
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1448 dsphys
->ds_uncompressed_bytes
=
1449 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1450 dsphys
->ds_flags
= dsl_dataset_phys(ds
)->ds_flags
;
1451 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1452 dsphys
->ds_bp
= dsl_dataset_phys(ds
)->ds_bp
;
1453 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
1454 dmu_buf_rele(dbuf
, FTAG
);
1456 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++) {
1457 if (ds
->ds_feature_inuse
[f
])
1458 dsl_dataset_activate_feature(dsobj
, f
, tx
);
1461 ASSERT3U(ds
->ds_prev
!= 0, ==,
1462 dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0);
1464 uint64_t next_clones_obj
=
1465 dsl_dataset_phys(ds
->ds_prev
)->ds_next_clones_obj
;
1466 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1468 dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 1);
1469 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1471 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1472 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, ==,
1473 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_txg
);
1474 dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
= dsobj
;
1475 } else if (next_clones_obj
!= 0) {
1476 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
1477 dsphys
->ds_next_snap_obj
, tx
);
1478 VERIFY0(zap_add_int(mos
,
1479 next_clones_obj
, dsobj
, tx
));
1484 * If we have a reference-reservation on this dataset, we will
1485 * need to increase the amount of refreservation being charged
1486 * since our unique space is going to zero.
1488 if (ds
->ds_reserved
) {
1490 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
1491 delta
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
,
1493 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
,
1497 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1498 dsl_dataset_phys(ds
)->ds_deadlist_obj
=
1499 dsl_deadlist_clone(&ds
->ds_deadlist
, UINT64_MAX
,
1500 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, tx
);
1501 dsl_deadlist_close(&ds
->ds_deadlist
);
1502 dsl_deadlist_open(&ds
->ds_deadlist
, mos
,
1503 dsl_dataset_phys(ds
)->ds_deadlist_obj
);
1504 dsl_deadlist_add_key(&ds
->ds_deadlist
,
1505 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
);
1507 if (dsl_dataset_remap_deadlist_exists(ds
)) {
1508 uint64_t remap_deadlist_obj
=
1509 dsl_dataset_get_remap_deadlist_object(ds
);
1511 * Move the remap_deadlist to the snapshot. The head
1512 * will create a new remap deadlist on demand, from
1513 * dsl_dataset_block_remapped().
1515 dsl_dataset_unset_remap_deadlist_object(ds
, tx
);
1516 dsl_deadlist_close(&ds
->ds_remap_deadlist
);
1518 dmu_object_zapify(mos
, dsobj
, DMU_OT_DSL_DATASET
, tx
);
1519 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_REMAP_DEADLIST
,
1520 sizeof (remap_deadlist_obj
), 1, &remap_deadlist_obj
, tx
));
1523 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, <, tx
->tx_txg
);
1524 dsl_dataset_phys(ds
)->ds_prev_snap_obj
= dsobj
;
1525 dsl_dataset_phys(ds
)->ds_prev_snap_txg
= crtxg
;
1526 dsl_dataset_phys(ds
)->ds_unique_bytes
= 0;
1528 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
1529 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1531 VERIFY0(zap_add(mos
, dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1532 snapname
, 8, 1, &dsobj
, tx
));
1535 dsl_dataset_rele(ds
->ds_prev
, ds
);
1536 VERIFY0(dsl_dataset_hold_obj(dp
,
1537 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, ds
, &ds
->ds_prev
));
1539 dsl_scan_ds_snapshotted(ds
, tx
);
1541 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
1543 spa_history_log_internal_ds(ds
->ds_prev
, "snapshot", tx
, "");
1547 dsl_dataset_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1549 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1550 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1553 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1554 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1557 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
1559 name
= nvpair_name(pair
);
1560 atp
= strchr(name
, '@');
1561 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1562 VERIFY0(dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
));
1564 dsl_dataset_snapshot_sync_impl(ds
, atp
+ 1, tx
);
1565 if (ddsa
->ddsa_props
!= NULL
) {
1566 dsl_props_set_sync_impl(ds
->ds_prev
,
1567 ZPROP_SRC_LOCAL
, ddsa
->ddsa_props
, tx
);
1569 zvol_create_minors(dp
->dp_spa
, nvpair_name(pair
), B_TRUE
);
1570 dsl_dataset_rele(ds
, FTAG
);
1575 * The snapshots must all be in the same pool.
1576 * All-or-nothing: if there are any failures, nothing will be modified.
1579 dsl_dataset_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
*errors
)
1581 dsl_dataset_snapshot_arg_t ddsa
;
1583 boolean_t needsuspend
;
1587 nvlist_t
*suspended
= NULL
;
1589 pair
= nvlist_next_nvpair(snaps
, NULL
);
1592 firstname
= nvpair_name(pair
);
1594 error
= spa_open(firstname
, &spa
, FTAG
);
1597 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1598 spa_close(spa
, FTAG
);
1601 suspended
= fnvlist_alloc();
1602 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1603 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1604 char fsname
[ZFS_MAX_DATASET_NAME_LEN
];
1605 char *snapname
= nvpair_name(pair
);
1609 atp
= strchr(snapname
, '@');
1611 error
= SET_ERROR(EINVAL
);
1614 (void) strlcpy(fsname
, snapname
, atp
- snapname
+ 1);
1616 error
= zil_suspend(fsname
, &cookie
);
1619 fnvlist_add_uint64(suspended
, fsname
,
1624 ddsa
.ddsa_snaps
= snaps
;
1625 ddsa
.ddsa_props
= props
;
1626 ddsa
.ddsa_errors
= errors
;
1627 ddsa
.ddsa_cr
= CRED();
1630 error
= dsl_sync_task(firstname
, dsl_dataset_snapshot_check
,
1631 dsl_dataset_snapshot_sync
, &ddsa
,
1632 fnvlist_num_pairs(snaps
) * 3, ZFS_SPACE_CHECK_NORMAL
);
1635 if (suspended
!= NULL
) {
1636 for (pair
= nvlist_next_nvpair(suspended
, NULL
); pair
!= NULL
;
1637 pair
= nvlist_next_nvpair(suspended
, pair
)) {
1638 zil_resume((void *)(uintptr_t)
1639 fnvpair_value_uint64(pair
));
1641 fnvlist_free(suspended
);
1647 typedef struct dsl_dataset_snapshot_tmp_arg
{
1648 const char *ddsta_fsname
;
1649 const char *ddsta_snapname
;
1650 minor_t ddsta_cleanup_minor
;
1651 const char *ddsta_htag
;
1652 } dsl_dataset_snapshot_tmp_arg_t
;
1655 dsl_dataset_snapshot_tmp_check(void *arg
, dmu_tx_t
*tx
)
1657 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1658 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1662 error
= dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
);
1666 /* NULL cred means no limit check for tmp snapshot */
1667 error
= dsl_dataset_snapshot_check_impl(ds
, ddsta
->ddsta_snapname
,
1668 tx
, B_FALSE
, 0, NULL
);
1670 dsl_dataset_rele(ds
, FTAG
);
1674 if (spa_version(dp
->dp_spa
) < SPA_VERSION_USERREFS
) {
1675 dsl_dataset_rele(ds
, FTAG
);
1676 return (SET_ERROR(ENOTSUP
));
1678 error
= dsl_dataset_user_hold_check_one(NULL
, ddsta
->ddsta_htag
,
1681 dsl_dataset_rele(ds
, FTAG
);
1685 dsl_dataset_rele(ds
, FTAG
);
1690 dsl_dataset_snapshot_tmp_sync(void *arg
, dmu_tx_t
*tx
)
1692 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1693 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1694 dsl_dataset_t
*ds
= NULL
;
1696 VERIFY0(dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
));
1698 dsl_dataset_snapshot_sync_impl(ds
, ddsta
->ddsta_snapname
, tx
);
1699 dsl_dataset_user_hold_sync_one(ds
->ds_prev
, ddsta
->ddsta_htag
,
1700 ddsta
->ddsta_cleanup_minor
, gethrestime_sec(), tx
);
1701 dsl_destroy_snapshot_sync_impl(ds
->ds_prev
, B_TRUE
, tx
);
1703 dsl_dataset_rele(ds
, FTAG
);
1707 dsl_dataset_snapshot_tmp(const char *fsname
, const char *snapname
,
1708 minor_t cleanup_minor
, const char *htag
)
1710 dsl_dataset_snapshot_tmp_arg_t ddsta
;
1713 boolean_t needsuspend
;
1716 ddsta
.ddsta_fsname
= fsname
;
1717 ddsta
.ddsta_snapname
= snapname
;
1718 ddsta
.ddsta_cleanup_minor
= cleanup_minor
;
1719 ddsta
.ddsta_htag
= htag
;
1721 error
= spa_open(fsname
, &spa
, FTAG
);
1724 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1725 spa_close(spa
, FTAG
);
1728 error
= zil_suspend(fsname
, &cookie
);
1733 error
= dsl_sync_task(fsname
, dsl_dataset_snapshot_tmp_check
,
1734 dsl_dataset_snapshot_tmp_sync
, &ddsta
, 3, ZFS_SPACE_CHECK_RESERVED
);
1742 dsl_dataset_sync(dsl_dataset_t
*ds
, zio_t
*zio
, dmu_tx_t
*tx
)
1744 ASSERT(dmu_tx_is_syncing(tx
));
1745 ASSERT(ds
->ds_objset
!= NULL
);
1746 ASSERT(dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0);
1749 * in case we had to change ds_fsid_guid when we opened it,
1752 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1753 dsl_dataset_phys(ds
)->ds_fsid_guid
= ds
->ds_fsid_guid
;
1755 if (ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
] != 0) {
1756 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1757 ds
->ds_object
, DS_FIELD_RESUME_OBJECT
, 8, 1,
1758 &ds
->ds_resume_object
[tx
->tx_txg
& TXG_MASK
], tx
));
1759 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1760 ds
->ds_object
, DS_FIELD_RESUME_OFFSET
, 8, 1,
1761 &ds
->ds_resume_offset
[tx
->tx_txg
& TXG_MASK
], tx
));
1762 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1763 ds
->ds_object
, DS_FIELD_RESUME_BYTES
, 8, 1,
1764 &ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
], tx
));
1765 ds
->ds_resume_object
[tx
->tx_txg
& TXG_MASK
] = 0;
1766 ds
->ds_resume_offset
[tx
->tx_txg
& TXG_MASK
] = 0;
1767 ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
] = 0;
1770 dmu_objset_sync(ds
->ds_objset
, zio
, tx
);
1772 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++) {
1773 if (ds
->ds_feature_activation_needed
[f
]) {
1774 if (ds
->ds_feature_inuse
[f
])
1776 dsl_dataset_activate_feature(ds
->ds_object
, f
, tx
);
1777 ds
->ds_feature_inuse
[f
] = B_TRUE
;
1783 deadlist_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
1785 dsl_deadlist_t
*dl
= arg
;
1786 dsl_deadlist_insert(dl
, bp
, tx
);
1791 dsl_dataset_sync_done(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1793 objset_t
*os
= ds
->ds_objset
;
1795 bplist_iterate(&ds
->ds_pending_deadlist
,
1796 deadlist_enqueue_cb
, &ds
->ds_deadlist
, tx
);
1798 if (os
->os_synced_dnodes
!= NULL
) {
1799 multilist_destroy(os
->os_synced_dnodes
);
1800 os
->os_synced_dnodes
= NULL
;
1803 ASSERT(!dmu_objset_is_dirty(os
, dmu_tx_get_txg(tx
)));
1805 dmu_buf_rele(ds
->ds_dbuf
, ds
);
1809 get_clones_stat_impl(dsl_dataset_t
*ds
, nvlist_t
*val
)
1812 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1816 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1819 * There may be missing entries in ds_next_clones_obj
1820 * due to a bug in a previous version of the code.
1821 * Only trust it if it has the right number of entries.
1823 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
!= 0) {
1824 VERIFY0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1827 if (count
!= dsl_dataset_phys(ds
)->ds_num_children
- 1) {
1830 for (zap_cursor_init(&zc
, mos
,
1831 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
1832 zap_cursor_retrieve(&zc
, &za
) == 0;
1833 zap_cursor_advance(&zc
)) {
1834 dsl_dataset_t
*clone
;
1835 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1836 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
1837 za
.za_first_integer
, FTAG
, &clone
));
1838 dsl_dir_name(clone
->ds_dir
, buf
);
1839 fnvlist_add_boolean(val
, buf
);
1840 dsl_dataset_rele(clone
, FTAG
);
1842 zap_cursor_fini(&zc
);
1847 get_clones_stat(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1849 nvlist_t
*propval
= fnvlist_alloc();
1853 * We use nvlist_alloc() instead of fnvlist_alloc() because the
1854 * latter would allocate the list with NV_UNIQUE_NAME flag.
1855 * As a result, every time a clone name is appended to the list
1856 * it would be (linearly) searched for for a duplicate name.
1857 * We already know that all clone names must be unique and we
1858 * want avoid the quadratic complexity of double-checking that
1859 * because we can have a large number of clones.
1861 VERIFY0(nvlist_alloc(&val
, 0, KM_SLEEP
));
1863 if (get_clones_stat_impl(ds
, val
) == 0) {
1864 fnvlist_add_nvlist(propval
, ZPROP_VALUE
, val
);
1865 fnvlist_add_nvlist(nv
, zfs_prop_to_name(ZFS_PROP_CLONES
),
1870 nvlist_free(propval
);
1874 * Returns a string that represents the receive resume stats token. It should
1875 * be freed with strfree().
1878 get_receive_resume_stats_impl(dsl_dataset_t
*ds
)
1880 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1882 if (dsl_dataset_has_resume_receive_state(ds
)) {
1885 uint8_t *compressed
;
1887 nvlist_t
*token_nv
= fnvlist_alloc();
1888 size_t packed_size
, compressed_size
;
1890 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1891 DS_FIELD_RESUME_FROMGUID
, sizeof (val
), 1, &val
) == 0) {
1892 fnvlist_add_uint64(token_nv
, "fromguid", val
);
1894 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1895 DS_FIELD_RESUME_OBJECT
, sizeof (val
), 1, &val
) == 0) {
1896 fnvlist_add_uint64(token_nv
, "object", val
);
1898 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1899 DS_FIELD_RESUME_OFFSET
, sizeof (val
), 1, &val
) == 0) {
1900 fnvlist_add_uint64(token_nv
, "offset", val
);
1902 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1903 DS_FIELD_RESUME_BYTES
, sizeof (val
), 1, &val
) == 0) {
1904 fnvlist_add_uint64(token_nv
, "bytes", val
);
1906 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1907 DS_FIELD_RESUME_TOGUID
, sizeof (val
), 1, &val
) == 0) {
1908 fnvlist_add_uint64(token_nv
, "toguid", val
);
1910 char buf
[MAXNAMELEN
];
1911 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1912 DS_FIELD_RESUME_TONAME
, 1, sizeof (buf
), buf
) == 0) {
1913 fnvlist_add_string(token_nv
, "toname", buf
);
1915 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1916 DS_FIELD_RESUME_LARGEBLOCK
) == 0) {
1917 fnvlist_add_boolean(token_nv
, "largeblockok");
1919 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1920 DS_FIELD_RESUME_EMBEDOK
) == 0) {
1921 fnvlist_add_boolean(token_nv
, "embedok");
1923 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1924 DS_FIELD_RESUME_COMPRESSOK
) == 0) {
1925 fnvlist_add_boolean(token_nv
, "compressok");
1927 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1928 DS_FIELD_RESUME_RAWOK
) == 0) {
1929 fnvlist_add_boolean(token_nv
, "rawok");
1931 packed
= fnvlist_pack(token_nv
, &packed_size
);
1932 fnvlist_free(token_nv
);
1933 compressed
= kmem_alloc(packed_size
, KM_SLEEP
);
1935 compressed_size
= gzip_compress(packed
, compressed
,
1936 packed_size
, packed_size
, 6);
1939 fletcher_4_native_varsize(compressed
, compressed_size
, &cksum
);
1941 str
= kmem_alloc(compressed_size
* 2 + 1, KM_SLEEP
);
1942 for (int i
= 0; i
< compressed_size
; i
++) {
1943 (void) sprintf(str
+ i
* 2, "%02x", compressed
[i
]);
1945 str
[compressed_size
* 2] = '\0';
1946 char *propval
= kmem_asprintf("%u-%llx-%llx-%s",
1947 ZFS_SEND_RESUME_TOKEN_VERSION
,
1948 (longlong_t
)cksum
.zc_word
[0],
1949 (longlong_t
)packed_size
, str
);
1950 kmem_free(packed
, packed_size
);
1951 kmem_free(str
, compressed_size
* 2 + 1);
1952 kmem_free(compressed
, packed_size
);
1955 return (strdup(""));
1959 * Returns a string that represents the receive resume stats token of the
1960 * dataset's child. It should be freed with strfree().
1963 get_child_receive_stats(dsl_dataset_t
*ds
)
1965 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1966 dsl_dataset_t
*recv_ds
;
1967 dsl_dataset_name(ds
, recvname
);
1968 if (strlcat(recvname
, "/", sizeof (recvname
)) <
1969 sizeof (recvname
) &&
1970 strlcat(recvname
, recv_clone_name
, sizeof (recvname
)) <
1971 sizeof (recvname
) &&
1972 dsl_dataset_hold(ds
->ds_dir
->dd_pool
, recvname
, FTAG
,
1974 char *propval
= get_receive_resume_stats_impl(recv_ds
);
1975 dsl_dataset_rele(recv_ds
, FTAG
);
1978 return (strdup(""));
1982 get_receive_resume_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1984 char *propval
= get_receive_resume_stats_impl(ds
);
1985 if (strcmp(propval
, "") != 0) {
1986 dsl_prop_nvlist_add_string(nv
,
1987 ZFS_PROP_RECEIVE_RESUME_TOKEN
, propval
);
1989 char *childval
= get_child_receive_stats(ds
);
1990 if (strcmp(childval
, "") != 0) {
1991 dsl_prop_nvlist_add_string(nv
,
1992 ZFS_PROP_RECEIVE_RESUME_TOKEN
, childval
);
2000 dsl_get_refratio(dsl_dataset_t
*ds
)
2002 uint64_t ratio
= dsl_dataset_phys(ds
)->ds_compressed_bytes
== 0 ? 100 :
2003 (dsl_dataset_phys(ds
)->ds_uncompressed_bytes
* 100 /
2004 dsl_dataset_phys(ds
)->ds_compressed_bytes
);
2009 dsl_get_logicalreferenced(dsl_dataset_t
*ds
)
2011 return (dsl_dataset_phys(ds
)->ds_uncompressed_bytes
);
2015 dsl_get_compressratio(dsl_dataset_t
*ds
)
2017 if (ds
->ds_is_snapshot
) {
2018 return (dsl_get_refratio(ds
));
2020 dsl_dir_t
*dd
= ds
->ds_dir
;
2021 mutex_enter(&dd
->dd_lock
);
2022 uint64_t val
= dsl_dir_get_compressratio(dd
);
2023 mutex_exit(&dd
->dd_lock
);
2029 dsl_get_used(dsl_dataset_t
*ds
)
2031 if (ds
->ds_is_snapshot
) {
2032 return (dsl_dataset_phys(ds
)->ds_unique_bytes
);
2034 dsl_dir_t
*dd
= ds
->ds_dir
;
2035 mutex_enter(&dd
->dd_lock
);
2036 uint64_t val
= dsl_dir_get_used(dd
);
2037 mutex_exit(&dd
->dd_lock
);
2043 dsl_get_creation(dsl_dataset_t
*ds
)
2045 return (dsl_dataset_phys(ds
)->ds_creation_time
);
2049 dsl_get_creationtxg(dsl_dataset_t
*ds
)
2051 return (dsl_dataset_phys(ds
)->ds_creation_txg
);
2055 dsl_get_refquota(dsl_dataset_t
*ds
)
2057 return (ds
->ds_quota
);
2061 dsl_get_refreservation(dsl_dataset_t
*ds
)
2063 return (ds
->ds_reserved
);
2067 dsl_get_guid(dsl_dataset_t
*ds
)
2069 return (dsl_dataset_phys(ds
)->ds_guid
);
2073 dsl_get_unique(dsl_dataset_t
*ds
)
2075 return (dsl_dataset_phys(ds
)->ds_unique_bytes
);
2079 dsl_get_objsetid(dsl_dataset_t
*ds
)
2081 return (ds
->ds_object
);
2085 dsl_get_userrefs(dsl_dataset_t
*ds
)
2087 return (ds
->ds_userrefs
);
2091 dsl_get_defer_destroy(dsl_dataset_t
*ds
)
2093 return (DS_IS_DEFER_DESTROY(ds
) ? 1 : 0);
2097 dsl_get_referenced(dsl_dataset_t
*ds
)
2099 return (dsl_dataset_phys(ds
)->ds_referenced_bytes
);
2103 dsl_get_numclones(dsl_dataset_t
*ds
)
2105 ASSERT(ds
->ds_is_snapshot
);
2106 return (dsl_dataset_phys(ds
)->ds_num_children
- 1);
2110 dsl_get_inconsistent(dsl_dataset_t
*ds
)
2112 return ((dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_INCONSISTENT
) ?
2117 dsl_get_available(dsl_dataset_t
*ds
)
2119 uint64_t refdbytes
= dsl_get_referenced(ds
);
2120 uint64_t availbytes
= dsl_dir_space_available(ds
->ds_dir
,
2122 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
) {
2124 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
;
2126 if (ds
->ds_quota
!= 0) {
2128 * Adjust available bytes according to refquota
2130 if (refdbytes
< ds
->ds_quota
) {
2131 availbytes
= MIN(availbytes
,
2132 ds
->ds_quota
- refdbytes
);
2137 return (availbytes
);
2141 dsl_get_written(dsl_dataset_t
*ds
, uint64_t *written
)
2143 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2144 dsl_dataset_t
*prev
;
2145 int err
= dsl_dataset_hold_obj(dp
,
2146 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
2148 uint64_t comp
, uncomp
;
2149 err
= dsl_dataset_space_written(prev
, ds
, written
,
2151 dsl_dataset_rele(prev
, FTAG
);
2157 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN.
2160 dsl_get_prev_snap(dsl_dataset_t
*ds
, char *snap
)
2162 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2163 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
!= dp
->dp_origin_snap
) {
2164 dsl_dataset_name(ds
->ds_prev
, snap
);
2172 * Returns the mountpoint property and source for the given dataset in the value
2173 * and source buffers. The value buffer must be at least as large as MAXPATHLEN
2174 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN.
2175 * Returns 0 on success and an error on failure.
2178 dsl_get_mountpoint(dsl_dataset_t
*ds
, const char *dsname
, char *value
,
2182 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2184 /* Retrieve the mountpoint value stored in the zap opbject */
2185 error
= dsl_prop_get_ds(ds
, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT
), 1,
2186 ZAP_MAXVALUELEN
, value
, source
);
2192 * Process the dsname and source to find the full mountpoint string.
2193 * Can be skipped for 'legacy' or 'none'.
2195 if (value
[0] == '/') {
2196 char *buf
= kmem_alloc(ZAP_MAXVALUELEN
, KM_SLEEP
);
2198 const char *relpath
;
2201 * If we inherit the mountpoint, even from a dataset
2202 * with a received value, the source will be the path of
2203 * the dataset we inherit from. If source is
2204 * ZPROP_SOURCE_VAL_RECVD, the received value is not
2207 if (strcmp(source
, ZPROP_SOURCE_VAL_RECVD
) == 0) {
2210 ASSERT0(strncmp(dsname
, source
, strlen(source
)));
2211 relpath
= dsname
+ strlen(source
);
2212 if (relpath
[0] == '/')
2216 spa_altroot(dp
->dp_spa
, root
, ZAP_MAXVALUELEN
);
2219 * Special case an alternate root of '/'. This will
2220 * avoid having multiple leading slashes in the
2223 if (strcmp(root
, "/") == 0)
2227 * If the mountpoint is '/' then skip over this
2228 * if we are obtaining either an alternate root or
2229 * an inherited mountpoint.
2232 if (value
[1] == '\0' && (root
[0] != '\0' ||
2233 relpath
[0] != '\0'))
2236 if (relpath
[0] == '\0') {
2237 (void) snprintf(value
, ZAP_MAXVALUELEN
, "%s%s",
2240 (void) snprintf(value
, ZAP_MAXVALUELEN
, "%s%s%s%s",
2241 root
, mnt
, relpath
[0] == '@' ? "" : "/",
2244 kmem_free(buf
, ZAP_MAXVALUELEN
);
2251 dsl_dataset_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
2253 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2255 ASSERT(dsl_pool_config_held(dp
));
2257 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRATIO
,
2258 dsl_get_refratio(ds
));
2259 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALREFERENCED
,
2260 dsl_get_logicalreferenced(ds
));
2261 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
,
2262 dsl_get_compressratio(ds
));
2263 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
2266 if (ds
->ds_is_snapshot
) {
2267 get_clones_stat(ds
, nv
);
2269 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
2270 if (dsl_get_prev_snap(ds
, buf
) == 0)
2271 dsl_prop_nvlist_add_string(nv
, ZFS_PROP_PREV_SNAP
,
2273 dsl_dir_stats(ds
->ds_dir
, nv
);
2276 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_AVAILABLE
,
2277 dsl_get_available(ds
));
2278 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFERENCED
,
2279 dsl_get_referenced(ds
));
2280 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATION
,
2281 dsl_get_creation(ds
));
2282 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATETXG
,
2283 dsl_get_creationtxg(ds
));
2284 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFQUOTA
,
2285 dsl_get_refquota(ds
));
2286 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRESERVATION
,
2287 dsl_get_refreservation(ds
));
2288 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_GUID
,
2290 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_UNIQUE
,
2291 dsl_get_unique(ds
));
2292 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_OBJSETID
,
2293 dsl_get_objsetid(ds
));
2294 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERREFS
,
2295 dsl_get_userrefs(ds
));
2296 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_DEFER_DESTROY
,
2297 dsl_get_defer_destroy(ds
));
2298 dsl_dataset_crypt_stats(ds
, nv
);
2300 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
2302 if (dsl_get_written(ds
, &written
) == 0) {
2303 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_WRITTEN
,
2308 if (!dsl_dataset_is_snapshot(ds
)) {
2310 * A failed "newfs" (e.g. full) resumable receive leaves
2311 * the stats set on this dataset. Check here for the prop.
2313 get_receive_resume_stats(ds
, nv
);
2316 * A failed incremental resumable receive leaves the
2317 * stats set on our child named "%recv". Check the child
2320 /* 6 extra bytes for /%recv */
2321 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
2322 dsl_dataset_t
*recv_ds
;
2323 dsl_dataset_name(ds
, recvname
);
2324 if (strlcat(recvname
, "/", sizeof (recvname
)) <
2325 sizeof (recvname
) &&
2326 strlcat(recvname
, recv_clone_name
, sizeof (recvname
)) <
2327 sizeof (recvname
) &&
2328 dsl_dataset_hold(dp
, recvname
, FTAG
, &recv_ds
) == 0) {
2329 get_receive_resume_stats(recv_ds
, nv
);
2330 dsl_dataset_rele(recv_ds
, FTAG
);
2336 dsl_dataset_fast_stat(dsl_dataset_t
*ds
, dmu_objset_stats_t
*stat
)
2338 ASSERTV(dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
);
2339 ASSERT(dsl_pool_config_held(dp
));
2341 stat
->dds_creation_txg
= dsl_get_creationtxg(ds
);
2342 stat
->dds_inconsistent
= dsl_get_inconsistent(ds
);
2343 stat
->dds_guid
= dsl_get_guid(ds
);
2344 stat
->dds_origin
[0] = '\0';
2345 if (ds
->ds_is_snapshot
) {
2346 stat
->dds_is_snapshot
= B_TRUE
;
2347 stat
->dds_num_clones
= dsl_get_numclones(ds
);
2349 stat
->dds_is_snapshot
= B_FALSE
;
2350 stat
->dds_num_clones
= 0;
2352 if (dsl_dir_is_clone(ds
->ds_dir
)) {
2353 dsl_dir_get_origin(ds
->ds_dir
, stat
->dds_origin
);
2359 dsl_dataset_fsid_guid(dsl_dataset_t
*ds
)
2361 return (ds
->ds_fsid_guid
);
2365 dsl_dataset_space(dsl_dataset_t
*ds
,
2366 uint64_t *refdbytesp
, uint64_t *availbytesp
,
2367 uint64_t *usedobjsp
, uint64_t *availobjsp
)
2369 *refdbytesp
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
2370 *availbytesp
= dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
);
2371 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
)
2373 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
;
2374 if (ds
->ds_quota
!= 0) {
2376 * Adjust available bytes according to refquota
2378 if (*refdbytesp
< ds
->ds_quota
)
2379 *availbytesp
= MIN(*availbytesp
,
2380 ds
->ds_quota
- *refdbytesp
);
2384 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
2385 *usedobjsp
= BP_GET_FILL(&dsl_dataset_phys(ds
)->ds_bp
);
2386 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2387 *availobjsp
= DN_MAX_OBJECT
- *usedobjsp
;
2391 dsl_dataset_modified_since_snap(dsl_dataset_t
*ds
, dsl_dataset_t
*snap
)
2393 ASSERTV(dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
);
2396 ASSERT(dsl_pool_config_held(dp
));
2399 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
2400 birth
= dsl_dataset_get_blkptr(ds
)->blk_birth
;
2401 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2402 if (birth
> dsl_dataset_phys(snap
)->ds_creation_txg
) {
2403 objset_t
*os
, *os_snap
;
2405 * It may be that only the ZIL differs, because it was
2406 * reset in the head. Don't count that as being
2409 if (dmu_objset_from_ds(ds
, &os
) != 0)
2411 if (dmu_objset_from_ds(snap
, &os_snap
) != 0)
2413 return (bcmp(&os
->os_phys
->os_meta_dnode
,
2414 &os_snap
->os_phys
->os_meta_dnode
,
2415 sizeof (os
->os_phys
->os_meta_dnode
)) != 0);
2420 typedef struct dsl_dataset_rename_snapshot_arg
{
2421 const char *ddrsa_fsname
;
2422 const char *ddrsa_oldsnapname
;
2423 const char *ddrsa_newsnapname
;
2424 boolean_t ddrsa_recursive
;
2426 } dsl_dataset_rename_snapshot_arg_t
;
2430 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t
*dp
,
2431 dsl_dataset_t
*hds
, void *arg
)
2433 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2437 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
2439 /* ignore nonexistent snapshots */
2440 return (error
== ENOENT
? 0 : error
);
2443 /* new name should not exist */
2444 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_newsnapname
, &val
);
2446 error
= SET_ERROR(EEXIST
);
2447 else if (error
== ENOENT
)
2450 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2451 if (dsl_dir_namelen(hds
->ds_dir
) + 1 +
2452 strlen(ddrsa
->ddrsa_newsnapname
) >= ZFS_MAX_DATASET_NAME_LEN
)
2453 error
= SET_ERROR(ENAMETOOLONG
);
2459 dsl_dataset_rename_snapshot_check(void *arg
, dmu_tx_t
*tx
)
2461 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2462 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2466 error
= dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
);
2470 if (ddrsa
->ddrsa_recursive
) {
2471 error
= dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
2472 dsl_dataset_rename_snapshot_check_impl
, ddrsa
,
2475 error
= dsl_dataset_rename_snapshot_check_impl(dp
, hds
, ddrsa
);
2477 dsl_dataset_rele(hds
, FTAG
);
2482 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t
*dp
,
2483 dsl_dataset_t
*hds
, void *arg
)
2485 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2488 dmu_tx_t
*tx
= ddrsa
->ddrsa_tx
;
2491 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
2492 ASSERT(error
== 0 || error
== ENOENT
);
2493 if (error
== ENOENT
) {
2494 /* ignore nonexistent snapshots */
2498 VERIFY0(dsl_dataset_hold_obj(dp
, val
, FTAG
, &ds
));
2500 /* log before we change the name */
2501 spa_history_log_internal_ds(ds
, "rename", tx
,
2502 "-> @%s", ddrsa
->ddrsa_newsnapname
);
2504 VERIFY0(dsl_dataset_snap_remove(hds
, ddrsa
->ddrsa_oldsnapname
, tx
,
2506 mutex_enter(&ds
->ds_lock
);
2507 (void) strlcpy(ds
->ds_snapname
, ddrsa
->ddrsa_newsnapname
,
2508 sizeof (ds
->ds_snapname
));
2509 mutex_exit(&ds
->ds_lock
);
2510 VERIFY0(zap_add(dp
->dp_meta_objset
,
2511 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
,
2512 ds
->ds_snapname
, 8, 1, &ds
->ds_object
, tx
));
2513 zvol_rename_minors(dp
->dp_spa
, ddrsa
->ddrsa_oldsnapname
,
2514 ddrsa
->ddrsa_newsnapname
, B_TRUE
);
2516 dsl_dataset_rele(ds
, FTAG
);
2521 dsl_dataset_rename_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
2523 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2524 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2525 dsl_dataset_t
*hds
= NULL
;
2527 VERIFY0(dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
));
2528 ddrsa
->ddrsa_tx
= tx
;
2529 if (ddrsa
->ddrsa_recursive
) {
2530 VERIFY0(dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
2531 dsl_dataset_rename_snapshot_sync_impl
, ddrsa
,
2534 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp
, hds
, ddrsa
));
2536 dsl_dataset_rele(hds
, FTAG
);
2540 dsl_dataset_rename_snapshot(const char *fsname
,
2541 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
)
2543 dsl_dataset_rename_snapshot_arg_t ddrsa
;
2545 ddrsa
.ddrsa_fsname
= fsname
;
2546 ddrsa
.ddrsa_oldsnapname
= oldsnapname
;
2547 ddrsa
.ddrsa_newsnapname
= newsnapname
;
2548 ddrsa
.ddrsa_recursive
= recursive
;
2550 return (dsl_sync_task(fsname
, dsl_dataset_rename_snapshot_check
,
2551 dsl_dataset_rename_snapshot_sync
, &ddrsa
,
2552 1, ZFS_SPACE_CHECK_RESERVED
));
2556 * If we're doing an ownership handoff, we need to make sure that there is
2557 * only one long hold on the dataset. We're not allowed to change anything here
2558 * so we don't permanently release the long hold or regular hold here. We want
2559 * to do this only when syncing to avoid the dataset unexpectedly going away
2560 * when we release the long hold.
2563 dsl_dataset_handoff_check(dsl_dataset_t
*ds
, void *owner
, dmu_tx_t
*tx
)
2567 if (!dmu_tx_is_syncing(tx
))
2570 if (owner
!= NULL
) {
2571 VERIFY3P(ds
->ds_owner
, ==, owner
);
2572 dsl_dataset_long_rele(ds
, owner
);
2575 held
= dsl_dataset_long_held(ds
);
2578 dsl_dataset_long_hold(ds
, owner
);
2581 return (SET_ERROR(EBUSY
));
2587 dsl_dataset_rollback_check(void *arg
, dmu_tx_t
*tx
)
2589 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2590 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2592 int64_t unused_refres_delta
;
2595 error
= dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
);
2599 /* must not be a snapshot */
2600 if (ds
->ds_is_snapshot
) {
2601 dsl_dataset_rele(ds
, FTAG
);
2602 return (SET_ERROR(EINVAL
));
2605 /* must have a most recent snapshot */
2606 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
< TXG_INITIAL
) {
2607 dsl_dataset_rele(ds
, FTAG
);
2608 return (SET_ERROR(ESRCH
));
2612 * No rollback to a snapshot created in the current txg, because
2613 * the rollback may dirty the dataset and create blocks that are
2614 * not reachable from the rootbp while having a birth txg that
2615 * falls into the snapshot's range.
2617 if (dmu_tx_is_syncing(tx
) &&
2618 dsl_dataset_phys(ds
)->ds_prev_snap_txg
>= tx
->tx_txg
) {
2619 dsl_dataset_rele(ds
, FTAG
);
2620 return (SET_ERROR(EAGAIN
));
2624 * If the expected target snapshot is specified, then check that
2625 * the latest snapshot is it.
2627 if (ddra
->ddra_tosnap
!= NULL
) {
2628 dsl_dataset_t
*snapds
;
2630 /* Check if the target snapshot exists at all. */
2631 error
= dsl_dataset_hold(dp
, ddra
->ddra_tosnap
, FTAG
, &snapds
);
2634 * ESRCH is used to signal that the target snapshot does
2635 * not exist, while ENOENT is used to report that
2636 * the rolled back dataset does not exist.
2637 * ESRCH is also used to cover other cases where the
2638 * target snapshot is not related to the dataset being
2639 * rolled back such as being in a different pool.
2641 if (error
== ENOENT
|| error
== EXDEV
)
2642 error
= SET_ERROR(ESRCH
);
2643 dsl_dataset_rele(ds
, FTAG
);
2646 ASSERT(snapds
->ds_is_snapshot
);
2648 /* Check if the snapshot is the latest snapshot indeed. */
2649 if (snapds
!= ds
->ds_prev
) {
2651 * Distinguish between the case where the only problem
2652 * is intervening snapshots (EEXIST) vs the snapshot
2653 * not being a valid target for rollback (ESRCH).
2655 if (snapds
->ds_dir
== ds
->ds_dir
||
2656 (dsl_dir_is_clone(ds
->ds_dir
) &&
2657 dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
==
2658 snapds
->ds_object
)) {
2659 error
= SET_ERROR(EEXIST
);
2661 error
= SET_ERROR(ESRCH
);
2663 dsl_dataset_rele(snapds
, FTAG
);
2664 dsl_dataset_rele(ds
, FTAG
);
2667 dsl_dataset_rele(snapds
, FTAG
);
2670 /* must not have any bookmarks after the most recent snapshot */
2671 nvlist_t
*proprequest
= fnvlist_alloc();
2672 fnvlist_add_boolean(proprequest
, zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2673 nvlist_t
*bookmarks
= fnvlist_alloc();
2674 error
= dsl_get_bookmarks_impl(ds
, proprequest
, bookmarks
);
2675 fnvlist_free(proprequest
);
2677 dsl_dataset_rele(ds
, FTAG
);
2680 for (nvpair_t
*pair
= nvlist_next_nvpair(bookmarks
, NULL
);
2681 pair
!= NULL
; pair
= nvlist_next_nvpair(bookmarks
, pair
)) {
2683 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair
),
2684 zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2685 uint64_t createtxg
= fnvlist_lookup_uint64(valuenv
, "value");
2686 if (createtxg
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
2687 fnvlist_free(bookmarks
);
2688 dsl_dataset_rele(ds
, FTAG
);
2689 return (SET_ERROR(EEXIST
));
2692 fnvlist_free(bookmarks
);
2694 error
= dsl_dataset_handoff_check(ds
, ddra
->ddra_owner
, tx
);
2696 dsl_dataset_rele(ds
, FTAG
);
2701 * Check if the snap we are rolling back to uses more than
2704 if (ds
->ds_quota
!= 0 &&
2705 dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
> ds
->ds_quota
) {
2706 dsl_dataset_rele(ds
, FTAG
);
2707 return (SET_ERROR(EDQUOT
));
2711 * When we do the clone swap, we will temporarily use more space
2712 * due to the refreservation (the head will no longer have any
2713 * unique space, so the entire amount of the refreservation will need
2714 * to be free). We will immediately destroy the clone, freeing
2715 * this space, but the freeing happens over many txg's.
2717 unused_refres_delta
= (int64_t)MIN(ds
->ds_reserved
,
2718 dsl_dataset_phys(ds
)->ds_unique_bytes
);
2720 if (unused_refres_delta
> 0 &&
2721 unused_refres_delta
>
2722 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
)) {
2723 dsl_dataset_rele(ds
, FTAG
);
2724 return (SET_ERROR(ENOSPC
));
2727 dsl_dataset_rele(ds
, FTAG
);
2732 dsl_dataset_rollback_sync(void *arg
, dmu_tx_t
*tx
)
2734 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2735 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2736 dsl_dataset_t
*ds
, *clone
;
2738 char namebuf
[ZFS_MAX_DATASET_NAME_LEN
];
2740 VERIFY0(dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
));
2742 dsl_dataset_name(ds
->ds_prev
, namebuf
);
2743 fnvlist_add_string(ddra
->ddra_result
, "target", namebuf
);
2745 cloneobj
= dsl_dataset_create_sync(ds
->ds_dir
, "%rollback",
2746 ds
->ds_prev
, DS_CREATE_FLAG_NODIRTY
, kcred
, NULL
, tx
);
2748 VERIFY0(dsl_dataset_hold_obj(dp
, cloneobj
, FTAG
, &clone
));
2750 dsl_dataset_clone_swap_sync_impl(clone
, ds
, tx
);
2751 dsl_dataset_zero_zil(ds
, tx
);
2753 dsl_destroy_head_sync_impl(clone
, tx
);
2755 dsl_dataset_rele(clone
, FTAG
);
2756 dsl_dataset_rele(ds
, FTAG
);
2760 * Rolls back the given filesystem or volume to the most recent snapshot.
2761 * The name of the most recent snapshot will be returned under key "target"
2762 * in the result nvlist.
2765 * - The existing dataset MUST be owned by the specified owner at entry
2766 * - Upon return, dataset will still be held by the same owner, whether we
2769 * This mode is required any time the existing filesystem is mounted. See
2770 * notes above zfs_suspend_fs() for further details.
2773 dsl_dataset_rollback(const char *fsname
, const char *tosnap
, void *owner
,
2776 dsl_dataset_rollback_arg_t ddra
;
2778 ddra
.ddra_fsname
= fsname
;
2779 ddra
.ddra_tosnap
= tosnap
;
2780 ddra
.ddra_owner
= owner
;
2781 ddra
.ddra_result
= result
;
2783 return (dsl_sync_task(fsname
, dsl_dataset_rollback_check
,
2784 dsl_dataset_rollback_sync
, &ddra
,
2785 1, ZFS_SPACE_CHECK_RESERVED
));
2788 struct promotenode
{
2793 static int snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
);
2794 static int promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
,
2796 static void promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
);
2799 dsl_dataset_promote_check(void *arg
, dmu_tx_t
*tx
)
2801 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2802 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2804 struct promotenode
*snap
;
2805 dsl_dataset_t
*origin_ds
;
2809 size_t max_snap_len
;
2810 boolean_t conflicting_snaps
;
2812 err
= promote_hold(ddpa
, dp
, FTAG
);
2816 hds
= ddpa
->ddpa_clone
;
2817 max_snap_len
= MAXNAMELEN
- strlen(ddpa
->ddpa_clonename
) - 1;
2819 if (dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
) {
2820 promote_rele(ddpa
, FTAG
);
2821 return (SET_ERROR(EXDEV
));
2824 snap
= list_head(&ddpa
->shared_snaps
);
2826 err
= SET_ERROR(ENOENT
);
2829 origin_ds
= snap
->ds
;
2832 * Encrypted clones share a DSL Crypto Key with their origin's dsl dir.
2833 * When doing a promote we must make sure the encryption root for
2834 * both the target and the target's origin does not change to avoid
2835 * needing to rewrap encryption keys
2837 err
= dsl_dataset_promote_crypt_check(hds
->ds_dir
, origin_ds
->ds_dir
);
2842 * Compute and check the amount of space to transfer. Since this is
2843 * so expensive, don't do the preliminary check.
2845 if (!dmu_tx_is_syncing(tx
)) {
2846 promote_rele(ddpa
, FTAG
);
2850 /* compute origin's new unique space */
2851 snap
= list_tail(&ddpa
->clone_snaps
);
2852 ASSERT(snap
!= NULL
);
2853 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2854 origin_ds
->ds_object
);
2855 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2856 dsl_dataset_phys(origin_ds
)->ds_prev_snap_txg
, UINT64_MAX
,
2857 &ddpa
->unique
, &unused
, &unused
);
2860 * Walk the snapshots that we are moving
2862 * Compute space to transfer. Consider the incremental changes
2863 * to used by each snapshot:
2864 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2865 * So each snapshot gave birth to:
2866 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2867 * So a sequence would look like:
2868 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2869 * Which simplifies to:
2870 * uN + kN + kN-1 + ... + k1 + k0
2871 * Note however, if we stop before we reach the ORIGIN we get:
2872 * uN + kN + kN-1 + ... + kM - uM-1
2874 conflicting_snaps
= B_FALSE
;
2876 ddpa
->used
= dsl_dataset_phys(origin_ds
)->ds_referenced_bytes
;
2877 ddpa
->comp
= dsl_dataset_phys(origin_ds
)->ds_compressed_bytes
;
2878 ddpa
->uncomp
= dsl_dataset_phys(origin_ds
)->ds_uncompressed_bytes
;
2879 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2880 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2881 uint64_t val
, dlused
, dlcomp
, dluncomp
;
2882 dsl_dataset_t
*ds
= snap
->ds
;
2887 * If there are long holds, we won't be able to evict
2890 if (dsl_dataset_long_held(ds
)) {
2891 err
= SET_ERROR(EBUSY
);
2895 /* Check that the snapshot name does not conflict */
2896 VERIFY0(dsl_dataset_get_snapname(ds
));
2897 if (strlen(ds
->ds_snapname
) >= max_snap_len
) {
2898 err
= SET_ERROR(ENAMETOOLONG
);
2901 err
= dsl_dataset_snap_lookup(hds
, ds
->ds_snapname
, &val
);
2903 fnvlist_add_boolean(ddpa
->err_ds
,
2904 snap
->ds
->ds_snapname
);
2905 conflicting_snaps
= B_TRUE
;
2906 } else if (err
!= ENOENT
) {
2910 /* The very first snapshot does not have a deadlist */
2911 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
== 0)
2914 dsl_deadlist_space(&ds
->ds_deadlist
,
2915 &dlused
, &dlcomp
, &dluncomp
);
2916 ddpa
->used
+= dlused
;
2917 ddpa
->comp
+= dlcomp
;
2918 ddpa
->uncomp
+= dluncomp
;
2922 * In order to return the full list of conflicting snapshots, we check
2923 * whether there was a conflict after traversing all of them.
2925 if (conflicting_snaps
) {
2926 err
= SET_ERROR(EEXIST
);
2931 * If we are a clone of a clone then we never reached ORIGIN,
2932 * so we need to subtract out the clone origin's used space.
2934 if (ddpa
->origin_origin
) {
2936 dsl_dataset_phys(ddpa
->origin_origin
)->ds_referenced_bytes
;
2938 dsl_dataset_phys(ddpa
->origin_origin
)->ds_compressed_bytes
;
2940 dsl_dataset_phys(ddpa
->origin_origin
)->
2941 ds_uncompressed_bytes
;
2944 /* Check that there is enough space and limit headroom here */
2945 err
= dsl_dir_transfer_possible(origin_ds
->ds_dir
, hds
->ds_dir
,
2946 0, ss_mv_cnt
, ddpa
->used
, ddpa
->cr
);
2951 * Compute the amounts of space that will be used by snapshots
2952 * after the promotion (for both origin and clone). For each,
2953 * it is the amount of space that will be on all of their
2954 * deadlists (that was not born before their new origin).
2956 if (dsl_dir_phys(hds
->ds_dir
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2960 * Note, typically this will not be a clone of a clone,
2961 * so dd_origin_txg will be < TXG_INITIAL, so
2962 * these snaplist_space() -> dsl_deadlist_space_range()
2963 * calls will be fast because they do not have to
2964 * iterate over all bps.
2966 snap
= list_head(&ddpa
->origin_snaps
);
2968 err
= SET_ERROR(ENOENT
);
2971 err
= snaplist_space(&ddpa
->shared_snaps
,
2972 snap
->ds
->ds_dir
->dd_origin_txg
, &ddpa
->cloneusedsnap
);
2976 err
= snaplist_space(&ddpa
->clone_snaps
,
2977 snap
->ds
->ds_dir
->dd_origin_txg
, &space
);
2980 ddpa
->cloneusedsnap
+= space
;
2982 if (dsl_dir_phys(origin_ds
->ds_dir
)->dd_flags
&
2983 DD_FLAG_USED_BREAKDOWN
) {
2984 err
= snaplist_space(&ddpa
->origin_snaps
,
2985 dsl_dataset_phys(origin_ds
)->ds_creation_txg
,
2986 &ddpa
->originusedsnap
);
2992 promote_rele(ddpa
, FTAG
);
2997 dsl_dataset_promote_sync(void *arg
, dmu_tx_t
*tx
)
2999 dsl_dataset_promote_arg_t
*ddpa
= arg
;
3000 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3002 struct promotenode
*snap
;
3003 dsl_dataset_t
*origin_ds
;
3004 dsl_dataset_t
*origin_head
;
3006 dsl_dir_t
*odd
= NULL
;
3007 uint64_t oldnext_obj
;
3010 VERIFY0(promote_hold(ddpa
, dp
, FTAG
));
3011 hds
= ddpa
->ddpa_clone
;
3013 ASSERT0(dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
);
3015 snap
= list_head(&ddpa
->shared_snaps
);
3016 origin_ds
= snap
->ds
;
3019 snap
= list_head(&ddpa
->origin_snaps
);
3020 origin_head
= snap
->ds
;
3023 * We need to explicitly open odd, since origin_ds's dd will be
3026 VERIFY0(dsl_dir_hold_obj(dp
, origin_ds
->ds_dir
->dd_object
,
3029 dsl_dataset_promote_crypt_sync(hds
->ds_dir
, odd
, tx
);
3031 /* change origin's next snap */
3032 dmu_buf_will_dirty(origin_ds
->ds_dbuf
, tx
);
3033 oldnext_obj
= dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
;
3034 snap
= list_tail(&ddpa
->clone_snaps
);
3035 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
3036 origin_ds
->ds_object
);
3037 dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
= snap
->ds
->ds_object
;
3039 /* change the origin's next clone */
3040 if (dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
) {
3041 dsl_dataset_remove_from_next_clones(origin_ds
,
3042 snap
->ds
->ds_object
, tx
);
3043 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
3044 dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
,
3049 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
3050 ASSERT3U(dsl_dir_phys(dd
)->dd_origin_obj
, ==, origin_ds
->ds_object
);
3051 dsl_dir_phys(dd
)->dd_origin_obj
= dsl_dir_phys(odd
)->dd_origin_obj
;
3052 dd
->dd_origin_txg
= origin_head
->ds_dir
->dd_origin_txg
;
3053 dmu_buf_will_dirty(odd
->dd_dbuf
, tx
);
3054 dsl_dir_phys(odd
)->dd_origin_obj
= origin_ds
->ds_object
;
3055 origin_head
->ds_dir
->dd_origin_txg
=
3056 dsl_dataset_phys(origin_ds
)->ds_creation_txg
;
3058 /* change dd_clone entries */
3059 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
3060 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
3061 dsl_dir_phys(odd
)->dd_clones
, hds
->ds_object
, tx
));
3062 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
3063 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
3064 hds
->ds_object
, tx
));
3066 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
3067 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
3068 origin_head
->ds_object
, tx
));
3069 if (dsl_dir_phys(dd
)->dd_clones
== 0) {
3070 dsl_dir_phys(dd
)->dd_clones
=
3071 zap_create(dp
->dp_meta_objset
, DMU_OT_DSL_CLONES
,
3072 DMU_OT_NONE
, 0, tx
);
3074 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
3075 dsl_dir_phys(dd
)->dd_clones
, origin_head
->ds_object
, tx
));
3078 /* move snapshots to this dir */
3079 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
3080 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
3081 dsl_dataset_t
*ds
= snap
->ds
;
3084 * Property callbacks are registered to a particular
3085 * dsl_dir. Since ours is changing, evict the objset
3086 * so that they will be unregistered from the old dsl_dir.
3088 if (ds
->ds_objset
) {
3089 dmu_objset_evict(ds
->ds_objset
);
3090 ds
->ds_objset
= NULL
;
3093 /* move snap name entry */
3094 VERIFY0(dsl_dataset_get_snapname(ds
));
3095 VERIFY0(dsl_dataset_snap_remove(origin_head
,
3096 ds
->ds_snapname
, tx
, B_TRUE
));
3097 VERIFY0(zap_add(dp
->dp_meta_objset
,
3098 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
, ds
->ds_snapname
,
3099 8, 1, &ds
->ds_object
, tx
));
3100 dsl_fs_ss_count_adjust(hds
->ds_dir
, 1,
3101 DD_FIELD_SNAPSHOT_COUNT
, tx
);
3103 /* change containing dsl_dir */
3104 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3105 ASSERT3U(dsl_dataset_phys(ds
)->ds_dir_obj
, ==, odd
->dd_object
);
3106 dsl_dataset_phys(ds
)->ds_dir_obj
= dd
->dd_object
;
3107 ASSERT3P(ds
->ds_dir
, ==, odd
);
3108 dsl_dir_rele(ds
->ds_dir
, ds
);
3109 VERIFY0(dsl_dir_hold_obj(dp
, dd
->dd_object
,
3110 NULL
, ds
, &ds
->ds_dir
));
3112 /* move any clone references */
3113 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
&&
3114 spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
3118 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
3119 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
3120 zap_cursor_retrieve(&zc
, &za
) == 0;
3121 zap_cursor_advance(&zc
)) {
3122 dsl_dataset_t
*cnds
;
3125 if (za
.za_first_integer
== oldnext_obj
) {
3127 * We've already moved the
3128 * origin's reference.
3133 VERIFY0(dsl_dataset_hold_obj(dp
,
3134 za
.za_first_integer
, FTAG
, &cnds
));
3135 o
= dsl_dir_phys(cnds
->ds_dir
)->
3136 dd_head_dataset_obj
;
3138 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
3139 dsl_dir_phys(odd
)->dd_clones
, o
, tx
));
3140 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
3141 dsl_dir_phys(dd
)->dd_clones
, o
, tx
));
3142 dsl_dataset_rele(cnds
, FTAG
);
3144 zap_cursor_fini(&zc
);
3147 ASSERT(!dsl_prop_hascb(ds
));
3151 * Change space accounting.
3152 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
3153 * both be valid, or both be 0 (resulting in delta == 0). This
3154 * is true for each of {clone,origin} independently.
3157 delta
= ddpa
->cloneusedsnap
-
3158 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_SNAP
];
3159 ASSERT3S(delta
, >=, 0);
3160 ASSERT3U(ddpa
->used
, >=, delta
);
3161 dsl_dir_diduse_space(dd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
3162 dsl_dir_diduse_space(dd
, DD_USED_HEAD
,
3163 ddpa
->used
- delta
, ddpa
->comp
, ddpa
->uncomp
, tx
);
3165 delta
= ddpa
->originusedsnap
-
3166 dsl_dir_phys(odd
)->dd_used_breakdown
[DD_USED_SNAP
];
3167 ASSERT3S(delta
, <=, 0);
3168 ASSERT3U(ddpa
->used
, >=, -delta
);
3169 dsl_dir_diduse_space(odd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
3170 dsl_dir_diduse_space(odd
, DD_USED_HEAD
,
3171 -ddpa
->used
- delta
, -ddpa
->comp
, -ddpa
->uncomp
, tx
);
3173 dsl_dataset_phys(origin_ds
)->ds_unique_bytes
= ddpa
->unique
;
3175 /* log history record */
3176 spa_history_log_internal_ds(hds
, "promote", tx
, "");
3178 dsl_dir_rele(odd
, FTAG
);
3179 promote_rele(ddpa
, FTAG
);
3183 * Make a list of dsl_dataset_t's for the snapshots between first_obj
3184 * (exclusive) and last_obj (inclusive). The list will be in reverse
3185 * order (last_obj will be the list_head()). If first_obj == 0, do all
3186 * snapshots back to this dataset's origin.
3189 snaplist_make(dsl_pool_t
*dp
,
3190 uint64_t first_obj
, uint64_t last_obj
, list_t
*l
, void *tag
)
3192 uint64_t obj
= last_obj
;
3194 list_create(l
, sizeof (struct promotenode
),
3195 offsetof(struct promotenode
, link
));
3197 while (obj
!= first_obj
) {
3199 struct promotenode
*snap
;
3202 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
3203 ASSERT(err
!= ENOENT
);
3208 first_obj
= dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
;
3210 snap
= kmem_alloc(sizeof (*snap
), KM_SLEEP
);
3212 list_insert_tail(l
, snap
);
3213 obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
3220 snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
)
3222 struct promotenode
*snap
;
3225 for (snap
= list_head(l
); snap
; snap
= list_next(l
, snap
)) {
3226 uint64_t used
, comp
, uncomp
;
3227 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
3228 mintxg
, UINT64_MAX
, &used
, &comp
, &uncomp
);
3235 snaplist_destroy(list_t
*l
, void *tag
)
3237 struct promotenode
*snap
;
3239 if (l
== NULL
|| !list_link_active(&l
->list_head
))
3242 while ((snap
= list_tail(l
)) != NULL
) {
3243 list_remove(l
, snap
);
3244 dsl_dataset_rele(snap
->ds
, tag
);
3245 kmem_free(snap
, sizeof (*snap
));
3251 promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
, void *tag
)
3255 struct promotenode
*snap
;
3257 error
= dsl_dataset_hold(dp
, ddpa
->ddpa_clonename
, tag
,
3261 dd
= ddpa
->ddpa_clone
->ds_dir
;
3263 if (ddpa
->ddpa_clone
->ds_is_snapshot
||
3264 !dsl_dir_is_clone(dd
)) {
3265 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
3266 return (SET_ERROR(EINVAL
));
3269 error
= snaplist_make(dp
, 0, dsl_dir_phys(dd
)->dd_origin_obj
,
3270 &ddpa
->shared_snaps
, tag
);
3274 error
= snaplist_make(dp
, 0, ddpa
->ddpa_clone
->ds_object
,
3275 &ddpa
->clone_snaps
, tag
);
3279 snap
= list_head(&ddpa
->shared_snaps
);
3280 ASSERT3U(snap
->ds
->ds_object
, ==, dsl_dir_phys(dd
)->dd_origin_obj
);
3281 error
= snaplist_make(dp
, dsl_dir_phys(dd
)->dd_origin_obj
,
3282 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_head_dataset_obj
,
3283 &ddpa
->origin_snaps
, tag
);
3287 if (dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
!= 0) {
3288 error
= dsl_dataset_hold_obj(dp
,
3289 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
,
3290 tag
, &ddpa
->origin_origin
);
3296 promote_rele(ddpa
, tag
);
3301 promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
)
3303 snaplist_destroy(&ddpa
->shared_snaps
, tag
);
3304 snaplist_destroy(&ddpa
->clone_snaps
, tag
);
3305 snaplist_destroy(&ddpa
->origin_snaps
, tag
);
3306 if (ddpa
->origin_origin
!= NULL
)
3307 dsl_dataset_rele(ddpa
->origin_origin
, tag
);
3308 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
3314 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
3315 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
3318 dsl_dataset_promote(const char *name
, char *conflsnap
)
3320 dsl_dataset_promote_arg_t ddpa
= { 0 };
3323 nvpair_t
*snap_pair
;
3327 * We will modify space proportional to the number of
3328 * snapshots. Compute numsnaps.
3330 error
= dmu_objset_hold(name
, FTAG
, &os
);
3333 error
= zap_count(dmu_objset_pool(os
)->dp_meta_objset
,
3334 dsl_dataset_phys(dmu_objset_ds(os
))->ds_snapnames_zapobj
,
3336 dmu_objset_rele(os
, FTAG
);
3340 ddpa
.ddpa_clonename
= name
;
3341 ddpa
.err_ds
= fnvlist_alloc();
3344 error
= dsl_sync_task(name
, dsl_dataset_promote_check
,
3345 dsl_dataset_promote_sync
, &ddpa
,
3346 2 + numsnaps
, ZFS_SPACE_CHECK_RESERVED
);
3349 * Return the first conflicting snapshot found.
3351 snap_pair
= nvlist_next_nvpair(ddpa
.err_ds
, NULL
);
3352 if (snap_pair
!= NULL
&& conflsnap
!= NULL
)
3353 (void) strcpy(conflsnap
, nvpair_name(snap_pair
));
3355 fnvlist_free(ddpa
.err_ds
);
3360 dsl_dataset_clone_swap_check_impl(dsl_dataset_t
*clone
,
3361 dsl_dataset_t
*origin_head
, boolean_t force
, void *owner
, dmu_tx_t
*tx
)
3364 * "slack" factor for received datasets with refquota set on them.
3365 * See the bottom of this function for details on its use.
3367 uint64_t refquota_slack
= (uint64_t)DMU_MAX_ACCESS
*
3368 spa_asize_inflation
;
3369 int64_t unused_refres_delta
;
3371 /* they should both be heads */
3372 if (clone
->ds_is_snapshot
||
3373 origin_head
->ds_is_snapshot
)
3374 return (SET_ERROR(EINVAL
));
3376 /* if we are not forcing, the branch point should be just before them */
3377 if (!force
&& clone
->ds_prev
!= origin_head
->ds_prev
)
3378 return (SET_ERROR(EINVAL
));
3380 /* clone should be the clone (unless they are unrelated) */
3381 if (clone
->ds_prev
!= NULL
&&
3382 clone
->ds_prev
!= clone
->ds_dir
->dd_pool
->dp_origin_snap
&&
3383 origin_head
->ds_dir
!= clone
->ds_prev
->ds_dir
)
3384 return (SET_ERROR(EINVAL
));
3386 /* the clone should be a child of the origin */
3387 if (clone
->ds_dir
->dd_parent
!= origin_head
->ds_dir
)
3388 return (SET_ERROR(EINVAL
));
3390 /* origin_head shouldn't be modified unless 'force' */
3392 dsl_dataset_modified_since_snap(origin_head
, origin_head
->ds_prev
))
3393 return (SET_ERROR(ETXTBSY
));
3395 /* origin_head should have no long holds (e.g. is not mounted) */
3396 if (dsl_dataset_handoff_check(origin_head
, owner
, tx
))
3397 return (SET_ERROR(EBUSY
));
3399 /* check amount of any unconsumed refreservation */
3400 unused_refres_delta
=
3401 (int64_t)MIN(origin_head
->ds_reserved
,
3402 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
3403 (int64_t)MIN(origin_head
->ds_reserved
,
3404 dsl_dataset_phys(clone
)->ds_unique_bytes
);
3406 if (unused_refres_delta
> 0 &&
3407 unused_refres_delta
>
3408 dsl_dir_space_available(origin_head
->ds_dir
, NULL
, 0, TRUE
))
3409 return (SET_ERROR(ENOSPC
));
3412 * The clone can't be too much over the head's refquota.
3414 * To ensure that the entire refquota can be used, we allow one
3415 * transaction to exceed the the refquota. Therefore, this check
3416 * needs to also allow for the space referenced to be more than the
3417 * refquota. The maximum amount of space that one transaction can use
3418 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this
3419 * overage ensures that we are able to receive a filesystem that
3420 * exceeds the refquota on the source system.
3422 * So that overage is the refquota_slack we use below.
3424 if (origin_head
->ds_quota
!= 0 &&
3425 dsl_dataset_phys(clone
)->ds_referenced_bytes
>
3426 origin_head
->ds_quota
+ refquota_slack
)
3427 return (SET_ERROR(EDQUOT
));
3433 dsl_dataset_swap_remap_deadlists(dsl_dataset_t
*clone
,
3434 dsl_dataset_t
*origin
, dmu_tx_t
*tx
)
3436 uint64_t clone_remap_dl_obj
, origin_remap_dl_obj
;
3437 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3439 ASSERT(dsl_pool_sync_context(dp
));
3441 clone_remap_dl_obj
= dsl_dataset_get_remap_deadlist_object(clone
);
3442 origin_remap_dl_obj
= dsl_dataset_get_remap_deadlist_object(origin
);
3444 if (clone_remap_dl_obj
!= 0) {
3445 dsl_deadlist_close(&clone
->ds_remap_deadlist
);
3446 dsl_dataset_unset_remap_deadlist_object(clone
, tx
);
3448 if (origin_remap_dl_obj
!= 0) {
3449 dsl_deadlist_close(&origin
->ds_remap_deadlist
);
3450 dsl_dataset_unset_remap_deadlist_object(origin
, tx
);
3453 if (clone_remap_dl_obj
!= 0) {
3454 dsl_dataset_set_remap_deadlist_object(origin
,
3455 clone_remap_dl_obj
, tx
);
3456 dsl_deadlist_open(&origin
->ds_remap_deadlist
,
3457 dp
->dp_meta_objset
, clone_remap_dl_obj
);
3459 if (origin_remap_dl_obj
!= 0) {
3460 dsl_dataset_set_remap_deadlist_object(clone
,
3461 origin_remap_dl_obj
, tx
);
3462 dsl_deadlist_open(&clone
->ds_remap_deadlist
,
3463 dp
->dp_meta_objset
, origin_remap_dl_obj
);
3468 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t
*clone
,
3469 dsl_dataset_t
*origin_head
, dmu_tx_t
*tx
)
3471 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3472 int64_t unused_refres_delta
;
3474 ASSERT(clone
->ds_reserved
== 0);
3476 * NOTE: On DEBUG kernels there could be a race between this and
3477 * the check function if spa_asize_inflation is adjusted...
3479 ASSERT(origin_head
->ds_quota
== 0 ||
3480 dsl_dataset_phys(clone
)->ds_unique_bytes
<= origin_head
->ds_quota
+
3481 DMU_MAX_ACCESS
* spa_asize_inflation
);
3482 ASSERT3P(clone
->ds_prev
, ==, origin_head
->ds_prev
);
3485 * Swap per-dataset feature flags.
3487 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++) {
3488 if (!(spa_feature_table
[f
].fi_flags
&
3489 ZFEATURE_FLAG_PER_DATASET
)) {
3490 ASSERT(!clone
->ds_feature_inuse
[f
]);
3491 ASSERT(!origin_head
->ds_feature_inuse
[f
]);
3495 boolean_t clone_inuse
= clone
->ds_feature_inuse
[f
];
3496 boolean_t origin_head_inuse
= origin_head
->ds_feature_inuse
[f
];
3499 dsl_dataset_deactivate_feature(clone
->ds_object
, f
, tx
);
3500 clone
->ds_feature_inuse
[f
] = B_FALSE
;
3502 if (origin_head_inuse
) {
3503 dsl_dataset_deactivate_feature(origin_head
->ds_object
,
3505 origin_head
->ds_feature_inuse
[f
] = B_FALSE
;
3508 dsl_dataset_activate_feature(origin_head
->ds_object
,
3510 origin_head
->ds_feature_inuse
[f
] = B_TRUE
;
3512 if (origin_head_inuse
) {
3513 dsl_dataset_activate_feature(clone
->ds_object
, f
, tx
);
3514 clone
->ds_feature_inuse
[f
] = B_TRUE
;
3518 dmu_buf_will_dirty(clone
->ds_dbuf
, tx
);
3519 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
3521 if (clone
->ds_objset
!= NULL
) {
3522 dmu_objset_evict(clone
->ds_objset
);
3523 clone
->ds_objset
= NULL
;
3526 if (origin_head
->ds_objset
!= NULL
) {
3527 dmu_objset_evict(origin_head
->ds_objset
);
3528 origin_head
->ds_objset
= NULL
;
3531 unused_refres_delta
=
3532 (int64_t)MIN(origin_head
->ds_reserved
,
3533 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
3534 (int64_t)MIN(origin_head
->ds_reserved
,
3535 dsl_dataset_phys(clone
)->ds_unique_bytes
);
3538 * Reset origin's unique bytes, if it exists.
3540 if (clone
->ds_prev
) {
3541 dsl_dataset_t
*origin
= clone
->ds_prev
;
3542 uint64_t comp
, uncomp
;
3544 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
3545 dsl_deadlist_space_range(&clone
->ds_deadlist
,
3546 dsl_dataset_phys(origin
)->ds_prev_snap_txg
, UINT64_MAX
,
3547 &dsl_dataset_phys(origin
)->ds_unique_bytes
, &comp
, &uncomp
);
3552 rrw_enter(&clone
->ds_bp_rwlock
, RW_WRITER
, FTAG
);
3553 rrw_enter(&origin_head
->ds_bp_rwlock
, RW_WRITER
, FTAG
);
3555 tmp
= dsl_dataset_phys(origin_head
)->ds_bp
;
3556 dsl_dataset_phys(origin_head
)->ds_bp
=
3557 dsl_dataset_phys(clone
)->ds_bp
;
3558 dsl_dataset_phys(clone
)->ds_bp
= tmp
;
3559 rrw_exit(&origin_head
->ds_bp_rwlock
, FTAG
);
3560 rrw_exit(&clone
->ds_bp_rwlock
, FTAG
);
3563 /* set dd_*_bytes */
3565 int64_t dused
, dcomp
, duncomp
;
3566 uint64_t cdl_used
, cdl_comp
, cdl_uncomp
;
3567 uint64_t odl_used
, odl_comp
, odl_uncomp
;
3569 ASSERT3U(dsl_dir_phys(clone
->ds_dir
)->
3570 dd_used_breakdown
[DD_USED_SNAP
], ==, 0);
3572 dsl_deadlist_space(&clone
->ds_deadlist
,
3573 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
3574 dsl_deadlist_space(&origin_head
->ds_deadlist
,
3575 &odl_used
, &odl_comp
, &odl_uncomp
);
3577 dused
= dsl_dataset_phys(clone
)->ds_referenced_bytes
+
3579 (dsl_dataset_phys(origin_head
)->ds_referenced_bytes
+
3581 dcomp
= dsl_dataset_phys(clone
)->ds_compressed_bytes
+
3583 (dsl_dataset_phys(origin_head
)->ds_compressed_bytes
+
3585 duncomp
= dsl_dataset_phys(clone
)->ds_uncompressed_bytes
+
3587 (dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
+
3590 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_HEAD
,
3591 dused
, dcomp
, duncomp
, tx
);
3592 dsl_dir_diduse_space(clone
->ds_dir
, DD_USED_HEAD
,
3593 -dused
, -dcomp
, -duncomp
, tx
);
3596 * The difference in the space used by snapshots is the
3597 * difference in snapshot space due to the head's
3598 * deadlist (since that's the only thing that's
3599 * changing that affects the snapused).
3601 dsl_deadlist_space_range(&clone
->ds_deadlist
,
3602 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
3603 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
3604 dsl_deadlist_space_range(&origin_head
->ds_deadlist
,
3605 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
3606 &odl_used
, &odl_comp
, &odl_uncomp
);
3607 dsl_dir_transfer_space(origin_head
->ds_dir
, cdl_used
- odl_used
,
3608 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
3611 /* swap ds_*_bytes */
3612 SWITCH64(dsl_dataset_phys(origin_head
)->ds_referenced_bytes
,
3613 dsl_dataset_phys(clone
)->ds_referenced_bytes
);
3614 SWITCH64(dsl_dataset_phys(origin_head
)->ds_compressed_bytes
,
3615 dsl_dataset_phys(clone
)->ds_compressed_bytes
);
3616 SWITCH64(dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
,
3617 dsl_dataset_phys(clone
)->ds_uncompressed_bytes
);
3618 SWITCH64(dsl_dataset_phys(origin_head
)->ds_unique_bytes
,
3619 dsl_dataset_phys(clone
)->ds_unique_bytes
);
3621 /* apply any parent delta for change in unconsumed refreservation */
3622 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_REFRSRV
,
3623 unused_refres_delta
, 0, 0, tx
);
3628 dsl_deadlist_close(&clone
->ds_deadlist
);
3629 dsl_deadlist_close(&origin_head
->ds_deadlist
);
3630 SWITCH64(dsl_dataset_phys(origin_head
)->ds_deadlist_obj
,
3631 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
3632 dsl_deadlist_open(&clone
->ds_deadlist
, dp
->dp_meta_objset
,
3633 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
3634 dsl_deadlist_open(&origin_head
->ds_deadlist
, dp
->dp_meta_objset
,
3635 dsl_dataset_phys(origin_head
)->ds_deadlist_obj
);
3636 dsl_dataset_swap_remap_deadlists(clone
, origin_head
, tx
);
3638 dsl_scan_ds_clone_swapped(origin_head
, clone
, tx
);
3640 spa_history_log_internal_ds(clone
, "clone swap", tx
,
3641 "parent=%s", origin_head
->ds_dir
->dd_myname
);
3645 * Given a pool name and a dataset object number in that pool,
3646 * return the name of that dataset.
3649 dsl_dsobj_to_dsname(char *pname
, uint64_t obj
, char *buf
)
3655 error
= dsl_pool_hold(pname
, FTAG
, &dp
);
3659 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
3661 dsl_dataset_name(ds
, buf
);
3662 dsl_dataset_rele(ds
, FTAG
);
3664 dsl_pool_rele(dp
, FTAG
);
3670 dsl_dataset_check_quota(dsl_dataset_t
*ds
, boolean_t check_quota
,
3671 uint64_t asize
, uint64_t inflight
, uint64_t *used
, uint64_t *ref_rsrv
)
3675 ASSERT3S(asize
, >, 0);
3678 * *ref_rsrv is the portion of asize that will come from any
3679 * unconsumed refreservation space.
3683 mutex_enter(&ds
->ds_lock
);
3685 * Make a space adjustment for reserved bytes.
3687 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
) {
3689 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
3691 (ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
3693 asize
- MIN(asize
, parent_delta(ds
, asize
+ inflight
));
3696 if (!check_quota
|| ds
->ds_quota
== 0) {
3697 mutex_exit(&ds
->ds_lock
);
3701 * If they are requesting more space, and our current estimate
3702 * is over quota, they get to try again unless the actual
3703 * on-disk is over quota and there are no pending changes (which
3704 * may free up space for us).
3706 if (dsl_dataset_phys(ds
)->ds_referenced_bytes
+ inflight
>=
3709 dsl_dataset_phys(ds
)->ds_referenced_bytes
< ds
->ds_quota
)
3710 error
= SET_ERROR(ERESTART
);
3712 error
= SET_ERROR(EDQUOT
);
3714 mutex_exit(&ds
->ds_lock
);
3719 typedef struct dsl_dataset_set_qr_arg
{
3720 const char *ddsqra_name
;
3721 zprop_source_t ddsqra_source
;
3722 uint64_t ddsqra_value
;
3723 } dsl_dataset_set_qr_arg_t
;
3728 dsl_dataset_set_refquota_check(void *arg
, dmu_tx_t
*tx
)
3730 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3731 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3736 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFQUOTA
)
3737 return (SET_ERROR(ENOTSUP
));
3739 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3743 if (ds
->ds_is_snapshot
) {
3744 dsl_dataset_rele(ds
, FTAG
);
3745 return (SET_ERROR(EINVAL
));
3748 error
= dsl_prop_predict(ds
->ds_dir
,
3749 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
3750 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3752 dsl_dataset_rele(ds
, FTAG
);
3757 dsl_dataset_rele(ds
, FTAG
);
3761 if (newval
< dsl_dataset_phys(ds
)->ds_referenced_bytes
||
3762 newval
< ds
->ds_reserved
) {
3763 dsl_dataset_rele(ds
, FTAG
);
3764 return (SET_ERROR(ENOSPC
));
3767 dsl_dataset_rele(ds
, FTAG
);
3772 dsl_dataset_set_refquota_sync(void *arg
, dmu_tx_t
*tx
)
3774 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3775 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3776 dsl_dataset_t
*ds
= NULL
;
3779 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3781 dsl_prop_set_sync_impl(ds
,
3782 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
3783 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
3784 &ddsqra
->ddsqra_value
, tx
);
3786 VERIFY0(dsl_prop_get_int_ds(ds
,
3787 zfs_prop_to_name(ZFS_PROP_REFQUOTA
), &newval
));
3789 if (ds
->ds_quota
!= newval
) {
3790 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3791 ds
->ds_quota
= newval
;
3793 dsl_dataset_rele(ds
, FTAG
);
3797 dsl_dataset_set_refquota(const char *dsname
, zprop_source_t source
,
3800 dsl_dataset_set_qr_arg_t ddsqra
;
3802 ddsqra
.ddsqra_name
= dsname
;
3803 ddsqra
.ddsqra_source
= source
;
3804 ddsqra
.ddsqra_value
= refquota
;
3806 return (dsl_sync_task(dsname
, dsl_dataset_set_refquota_check
,
3807 dsl_dataset_set_refquota_sync
, &ddsqra
, 0,
3808 ZFS_SPACE_CHECK_EXTRA_RESERVED
));
3812 dsl_dataset_set_refreservation_check(void *arg
, dmu_tx_t
*tx
)
3814 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3815 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3818 uint64_t newval
, unique
;
3820 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFRESERVATION
)
3821 return (SET_ERROR(ENOTSUP
));
3823 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3827 if (ds
->ds_is_snapshot
) {
3828 dsl_dataset_rele(ds
, FTAG
);
3829 return (SET_ERROR(EINVAL
));
3832 error
= dsl_prop_predict(ds
->ds_dir
,
3833 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3834 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3836 dsl_dataset_rele(ds
, FTAG
);
3841 * If we are doing the preliminary check in open context, the
3842 * space estimates may be inaccurate.
3844 if (!dmu_tx_is_syncing(tx
)) {
3845 dsl_dataset_rele(ds
, FTAG
);
3849 mutex_enter(&ds
->ds_lock
);
3850 if (!DS_UNIQUE_IS_ACCURATE(ds
))
3851 dsl_dataset_recalc_head_uniq(ds
);
3852 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3853 mutex_exit(&ds
->ds_lock
);
3855 if (MAX(unique
, newval
) > MAX(unique
, ds
->ds_reserved
)) {
3856 uint64_t delta
= MAX(unique
, newval
) -
3857 MAX(unique
, ds
->ds_reserved
);
3860 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, B_TRUE
) ||
3861 (ds
->ds_quota
> 0 && newval
> ds
->ds_quota
)) {
3862 dsl_dataset_rele(ds
, FTAG
);
3863 return (SET_ERROR(ENOSPC
));
3867 dsl_dataset_rele(ds
, FTAG
);
3872 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t
*ds
,
3873 zprop_source_t source
, uint64_t value
, dmu_tx_t
*tx
)
3879 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3880 source
, sizeof (value
), 1, &value
, tx
);
3882 VERIFY0(dsl_prop_get_int_ds(ds
,
3883 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
), &newval
));
3885 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3886 mutex_enter(&ds
->ds_dir
->dd_lock
);
3887 mutex_enter(&ds
->ds_lock
);
3888 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
3889 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3890 delta
= MAX(0, (int64_t)(newval
- unique
)) -
3891 MAX(0, (int64_t)(ds
->ds_reserved
- unique
));
3892 ds
->ds_reserved
= newval
;
3893 mutex_exit(&ds
->ds_lock
);
3895 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
, delta
, 0, 0, tx
);
3896 mutex_exit(&ds
->ds_dir
->dd_lock
);
3900 dsl_dataset_set_refreservation_sync(void *arg
, dmu_tx_t
*tx
)
3902 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3903 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3904 dsl_dataset_t
*ds
= NULL
;
3906 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3907 dsl_dataset_set_refreservation_sync_impl(ds
,
3908 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, tx
);
3909 dsl_dataset_rele(ds
, FTAG
);
3913 dsl_dataset_set_refreservation(const char *dsname
, zprop_source_t source
,
3914 uint64_t refreservation
)
3916 dsl_dataset_set_qr_arg_t ddsqra
;
3918 ddsqra
.ddsqra_name
= dsname
;
3919 ddsqra
.ddsqra_source
= source
;
3920 ddsqra
.ddsqra_value
= refreservation
;
3922 return (dsl_sync_task(dsname
, dsl_dataset_set_refreservation_check
,
3923 dsl_dataset_set_refreservation_sync
, &ddsqra
, 0,
3924 ZFS_SPACE_CHECK_EXTRA_RESERVED
));
3928 * Return (in *usedp) the amount of space written in new that is not
3929 * present in oldsnap. New may be a snapshot or the head. Old must be
3930 * a snapshot before new, in new's filesystem (or its origin). If not then
3931 * fail and return EINVAL.
3933 * The written space is calculated by considering two components: First, we
3934 * ignore any freed space, and calculate the written as new's used space
3935 * minus old's used space. Next, we add in the amount of space that was freed
3936 * between the two snapshots, thus reducing new's used space relative to old's.
3937 * Specifically, this is the space that was born before old->ds_creation_txg,
3938 * and freed before new (ie. on new's deadlist or a previous deadlist).
3940 * space freed [---------------------]
3941 * snapshots ---O-------O--------O-------O------
3945 dsl_dataset_space_written(dsl_dataset_t
*oldsnap
, dsl_dataset_t
*new,
3946 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3950 dsl_pool_t
*dp
= new->ds_dir
->dd_pool
;
3952 ASSERT(dsl_pool_config_held(dp
));
3955 *usedp
+= dsl_dataset_phys(new)->ds_referenced_bytes
;
3956 *usedp
-= dsl_dataset_phys(oldsnap
)->ds_referenced_bytes
;
3959 *compp
+= dsl_dataset_phys(new)->ds_compressed_bytes
;
3960 *compp
-= dsl_dataset_phys(oldsnap
)->ds_compressed_bytes
;
3963 *uncompp
+= dsl_dataset_phys(new)->ds_uncompressed_bytes
;
3964 *uncompp
-= dsl_dataset_phys(oldsnap
)->ds_uncompressed_bytes
;
3966 snapobj
= new->ds_object
;
3967 while (snapobj
!= oldsnap
->ds_object
) {
3968 dsl_dataset_t
*snap
;
3969 uint64_t used
, comp
, uncomp
;
3971 if (snapobj
== new->ds_object
) {
3974 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &snap
);
3979 if (dsl_dataset_phys(snap
)->ds_prev_snap_txg
==
3980 dsl_dataset_phys(oldsnap
)->ds_creation_txg
) {
3982 * The blocks in the deadlist can not be born after
3983 * ds_prev_snap_txg, so get the whole deadlist space,
3984 * which is more efficient (especially for old-format
3985 * deadlists). Unfortunately the deadlist code
3986 * doesn't have enough information to make this
3987 * optimization itself.
3989 dsl_deadlist_space(&snap
->ds_deadlist
,
3990 &used
, &comp
, &uncomp
);
3992 dsl_deadlist_space_range(&snap
->ds_deadlist
,
3993 0, dsl_dataset_phys(oldsnap
)->ds_creation_txg
,
3994 &used
, &comp
, &uncomp
);
4001 * If we get to the beginning of the chain of snapshots
4002 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4003 * was not a snapshot of/before new.
4005 snapobj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
4007 dsl_dataset_rele(snap
, FTAG
);
4009 err
= SET_ERROR(EINVAL
);
4018 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4019 * lastsnap, and all snapshots in between are deleted.
4021 * blocks that would be freed [---------------------------]
4022 * snapshots ---O-------O--------O-------O--------O
4023 * firstsnap lastsnap
4025 * This is the set of blocks that were born after the snap before firstsnap,
4026 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4027 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4028 * We calculate this by iterating over the relevant deadlists (from the snap
4029 * after lastsnap, backward to the snap after firstsnap), summing up the
4030 * space on the deadlist that was born after the snap before firstsnap.
4033 dsl_dataset_space_wouldfree(dsl_dataset_t
*firstsnap
,
4034 dsl_dataset_t
*lastsnap
,
4035 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
4039 dsl_pool_t
*dp
= firstsnap
->ds_dir
->dd_pool
;
4041 ASSERT(firstsnap
->ds_is_snapshot
);
4042 ASSERT(lastsnap
->ds_is_snapshot
);
4045 * Check that the snapshots are in the same dsl_dir, and firstsnap
4046 * is before lastsnap.
4048 if (firstsnap
->ds_dir
!= lastsnap
->ds_dir
||
4049 dsl_dataset_phys(firstsnap
)->ds_creation_txg
>
4050 dsl_dataset_phys(lastsnap
)->ds_creation_txg
)
4051 return (SET_ERROR(EINVAL
));
4053 *usedp
= *compp
= *uncompp
= 0;
4055 snapobj
= dsl_dataset_phys(lastsnap
)->ds_next_snap_obj
;
4056 while (snapobj
!= firstsnap
->ds_object
) {
4058 uint64_t used
, comp
, uncomp
;
4060 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &ds
);
4064 dsl_deadlist_space_range(&ds
->ds_deadlist
,
4065 dsl_dataset_phys(firstsnap
)->ds_prev_snap_txg
, UINT64_MAX
,
4066 &used
, &comp
, &uncomp
);
4071 snapobj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
4072 ASSERT3U(snapobj
, !=, 0);
4073 dsl_dataset_rele(ds
, FTAG
);
4079 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
4080 * For example, they could both be snapshots of the same filesystem, and
4081 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
4082 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
4083 * filesystem. Or 'earlier' could be the origin's origin.
4085 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
4088 dsl_dataset_is_before(dsl_dataset_t
*later
, dsl_dataset_t
*earlier
,
4089 uint64_t earlier_txg
)
4091 dsl_pool_t
*dp
= later
->ds_dir
->dd_pool
;
4095 ASSERT(dsl_pool_config_held(dp
));
4096 ASSERT(earlier
->ds_is_snapshot
|| earlier_txg
!= 0);
4098 if (earlier_txg
== 0)
4099 earlier_txg
= dsl_dataset_phys(earlier
)->ds_creation_txg
;
4101 if (later
->ds_is_snapshot
&&
4102 earlier_txg
>= dsl_dataset_phys(later
)->ds_creation_txg
)
4105 if (later
->ds_dir
== earlier
->ds_dir
)
4107 if (!dsl_dir_is_clone(later
->ds_dir
))
4110 if (dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
== earlier
->ds_object
)
4112 dsl_dataset_t
*origin
;
4113 error
= dsl_dataset_hold_obj(dp
,
4114 dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
, FTAG
, &origin
);
4117 ret
= dsl_dataset_is_before(origin
, earlier
, earlier_txg
);
4118 dsl_dataset_rele(origin
, FTAG
);
4123 dsl_dataset_zapify(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
4125 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
4126 dmu_object_zapify(mos
, ds
->ds_object
, DMU_OT_DSL_DATASET
, tx
);
4130 dsl_dataset_is_zapified(dsl_dataset_t
*ds
)
4132 dmu_object_info_t doi
;
4134 dmu_object_info_from_db(ds
->ds_dbuf
, &doi
);
4135 return (doi
.doi_type
== DMU_OTN_ZAP_METADATA
);
4139 dsl_dataset_has_resume_receive_state(dsl_dataset_t
*ds
)
4141 return (dsl_dataset_is_zapified(ds
) &&
4142 zap_contains(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
4143 ds
->ds_object
, DS_FIELD_RESUME_TOGUID
) == 0);
4147 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t
*ds
)
4149 uint64_t remap_deadlist_obj
;
4152 if (!dsl_dataset_is_zapified(ds
))
4155 err
= zap_lookup(ds
->ds_dir
->dd_pool
->dp_meta_objset
, ds
->ds_object
,
4156 DS_FIELD_REMAP_DEADLIST
, sizeof (remap_deadlist_obj
), 1,
4157 &remap_deadlist_obj
);
4160 VERIFY3S(err
, ==, ENOENT
);
4164 ASSERT(remap_deadlist_obj
!= 0);
4165 return (remap_deadlist_obj
);
4169 dsl_dataset_remap_deadlist_exists(dsl_dataset_t
*ds
)
4171 EQUIV(dsl_deadlist_is_open(&ds
->ds_remap_deadlist
),
4172 dsl_dataset_get_remap_deadlist_object(ds
) != 0);
4173 return (dsl_deadlist_is_open(&ds
->ds_remap_deadlist
));
4177 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t
*ds
, uint64_t obj
,
4181 dsl_dataset_zapify(ds
, tx
);
4182 VERIFY0(zap_add(ds
->ds_dir
->dd_pool
->dp_meta_objset
, ds
->ds_object
,
4183 DS_FIELD_REMAP_DEADLIST
, sizeof (obj
), 1, &obj
, tx
));
4187 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
4189 VERIFY0(zap_remove(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
4190 ds
->ds_object
, DS_FIELD_REMAP_DEADLIST
, tx
));
4194 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
4196 uint64_t remap_deadlist_object
;
4197 spa_t
*spa
= ds
->ds_dir
->dd_pool
->dp_spa
;
4199 ASSERT(dmu_tx_is_syncing(tx
));
4200 ASSERT(dsl_dataset_remap_deadlist_exists(ds
));
4202 remap_deadlist_object
= ds
->ds_remap_deadlist
.dl_object
;
4203 dsl_deadlist_close(&ds
->ds_remap_deadlist
);
4204 dsl_deadlist_free(spa_meta_objset(spa
), remap_deadlist_object
, tx
);
4205 dsl_dataset_unset_remap_deadlist_object(ds
, tx
);
4206 spa_feature_decr(spa
, SPA_FEATURE_OBSOLETE_COUNTS
, tx
);
4210 dsl_dataset_create_remap_deadlist(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
4212 uint64_t remap_deadlist_obj
;
4213 spa_t
*spa
= ds
->ds_dir
->dd_pool
->dp_spa
;
4215 ASSERT(dmu_tx_is_syncing(tx
));
4216 ASSERT(MUTEX_HELD(&ds
->ds_remap_deadlist_lock
));
4218 * Currently we only create remap deadlists when there are indirect
4219 * vdevs with referenced mappings.
4221 ASSERT(spa_feature_is_active(spa
, SPA_FEATURE_DEVICE_REMOVAL
));
4223 remap_deadlist_obj
= dsl_deadlist_clone(
4224 &ds
->ds_deadlist
, UINT64_MAX
,
4225 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, tx
);
4226 dsl_dataset_set_remap_deadlist_object(ds
,
4227 remap_deadlist_obj
, tx
);
4228 dsl_deadlist_open(&ds
->ds_remap_deadlist
, spa_meta_objset(spa
),
4229 remap_deadlist_obj
);
4230 spa_feature_incr(spa
, SPA_FEATURE_OBSOLETE_COUNTS
, tx
);
4233 #if defined(_KERNEL)
4235 module_param(zfs_max_recordsize
, int, 0644);
4236 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
4238 /* Limited to 1M on 32-bit platforms due to lack of virtual address space */
4239 module_param(zfs_max_recordsize
, int, 0444);
4240 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
4243 EXPORT_SYMBOL(dsl_dataset_hold
);
4244 EXPORT_SYMBOL(dsl_dataset_hold_flags
);
4245 EXPORT_SYMBOL(dsl_dataset_hold_obj
);
4246 EXPORT_SYMBOL(dsl_dataset_hold_obj_flags
);
4247 EXPORT_SYMBOL(dsl_dataset_own
);
4248 EXPORT_SYMBOL(dsl_dataset_own_obj
);
4249 EXPORT_SYMBOL(dsl_dataset_name
);
4250 EXPORT_SYMBOL(dsl_dataset_rele
);
4251 EXPORT_SYMBOL(dsl_dataset_rele_flags
);
4252 EXPORT_SYMBOL(dsl_dataset_disown
);
4253 EXPORT_SYMBOL(dsl_dataset_tryown
);
4254 EXPORT_SYMBOL(dsl_dataset_create_sync
);
4255 EXPORT_SYMBOL(dsl_dataset_create_sync_dd
);
4256 EXPORT_SYMBOL(dsl_dataset_snapshot_check
);
4257 EXPORT_SYMBOL(dsl_dataset_snapshot_sync
);
4258 EXPORT_SYMBOL(dsl_dataset_promote
);
4259 EXPORT_SYMBOL(dsl_dataset_user_hold
);
4260 EXPORT_SYMBOL(dsl_dataset_user_release
);
4261 EXPORT_SYMBOL(dsl_dataset_get_holds
);
4262 EXPORT_SYMBOL(dsl_dataset_get_blkptr
);
4263 EXPORT_SYMBOL(dsl_dataset_get_spa
);
4264 EXPORT_SYMBOL(dsl_dataset_modified_since_snap
);
4265 EXPORT_SYMBOL(dsl_dataset_space_written
);
4266 EXPORT_SYMBOL(dsl_dataset_space_wouldfree
);
4267 EXPORT_SYMBOL(dsl_dataset_sync
);
4268 EXPORT_SYMBOL(dsl_dataset_block_born
);
4269 EXPORT_SYMBOL(dsl_dataset_block_kill
);
4270 EXPORT_SYMBOL(dsl_dataset_dirty
);
4271 EXPORT_SYMBOL(dsl_dataset_stats
);
4272 EXPORT_SYMBOL(dsl_dataset_fast_stat
);
4273 EXPORT_SYMBOL(dsl_dataset_space
);
4274 EXPORT_SYMBOL(dsl_dataset_fsid_guid
);
4275 EXPORT_SYMBOL(dsl_dsobj_to_dsname
);
4276 EXPORT_SYMBOL(dsl_dataset_check_quota
);
4277 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl
);
4278 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl
);