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) 2014 Integros [integros.com]
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/dmu_send.h>
60 #include <sys/zio_checksum.h>
61 #include <sys/zio_compress.h>
62 #include <zfs_fletcher.h>
65 * The SPA supports block sizes up to 16MB. However, very large blocks
66 * can have an impact on i/o latency (e.g. tying up a spinning disk for
67 * ~300ms), and also potentially on the memory allocator. Therefore,
68 * we do not allow the recordsize to be set larger than zfs_max_recordsize
69 * (default 1MB). Larger blocks can be created by changing this tunable,
70 * and pools with larger blocks can always be imported and used, regardless
73 int zfs_max_recordsize
= 1 * 1024 * 1024;
75 #define SWITCH64(x, y) \
77 uint64_t __tmp = (x); \
82 #define DS_REF_MAX (1ULL << 62)
84 extern inline dsl_dataset_phys_t
*dsl_dataset_phys(dsl_dataset_t
*ds
);
86 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t
*ds
,
87 uint64_t obj
, dmu_tx_t
*tx
);
88 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t
*ds
,
91 extern int spa_asize_inflation
;
93 static zil_header_t zero_zil
;
96 * Figure out how much of this delta should be propogated to the dsl_dir
97 * layer. If there's a refreservation, that space has already been
98 * partially accounted for in our ancestors.
101 parent_delta(dsl_dataset_t
*ds
, int64_t delta
)
103 dsl_dataset_phys_t
*ds_phys
;
104 uint64_t old_bytes
, new_bytes
;
106 if (ds
->ds_reserved
== 0)
109 ds_phys
= dsl_dataset_phys(ds
);
110 old_bytes
= MAX(ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
111 new_bytes
= MAX(ds_phys
->ds_unique_bytes
+ delta
, ds
->ds_reserved
);
113 ASSERT3U(ABS((int64_t)(new_bytes
- old_bytes
)), <=, ABS(delta
));
114 return (new_bytes
- old_bytes
);
118 dsl_dataset_block_born(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
120 int used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
121 int compressed
= BP_GET_PSIZE(bp
);
122 int uncompressed
= BP_GET_UCSIZE(bp
);
125 dprintf_bp(bp
, "ds=%p", ds
);
127 ASSERT(dmu_tx_is_syncing(tx
));
128 /* It could have been compressed away to nothing */
131 ASSERT(BP_GET_TYPE(bp
) != DMU_OT_NONE
);
132 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp
)));
134 dsl_pool_mos_diduse_space(tx
->tx_pool
,
135 used
, compressed
, uncompressed
);
139 ASSERT3U(bp
->blk_birth
, >, dsl_dataset_phys(ds
)->ds_prev_snap_txg
);
140 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
141 mutex_enter(&ds
->ds_lock
);
142 delta
= parent_delta(ds
, used
);
143 dsl_dataset_phys(ds
)->ds_referenced_bytes
+= used
;
144 dsl_dataset_phys(ds
)->ds_compressed_bytes
+= compressed
;
145 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
+= uncompressed
;
146 dsl_dataset_phys(ds
)->ds_unique_bytes
+= used
;
148 if (BP_GET_LSIZE(bp
) > SPA_OLD_MAXBLOCKSIZE
) {
149 ds
->ds_feature_activation_needed
[SPA_FEATURE_LARGE_BLOCKS
] =
153 spa_feature_t f
= zio_checksum_to_feature(BP_GET_CHECKSUM(bp
));
154 if (f
!= SPA_FEATURE_NONE
)
155 ds
->ds_feature_activation_needed
[f
] = B_TRUE
;
157 mutex_exit(&ds
->ds_lock
);
158 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
, delta
,
159 compressed
, uncompressed
, tx
);
160 dsl_dir_transfer_space(ds
->ds_dir
, used
- delta
,
161 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
165 * Called when the specified segment has been remapped, and is thus no
166 * longer referenced in the head dataset. The vdev must be indirect.
168 * If the segment is referenced by a snapshot, put it on the remap deadlist.
169 * Otherwise, add this segment to the obsolete spacemap.
172 dsl_dataset_block_remapped(dsl_dataset_t
*ds
, uint64_t vdev
, uint64_t offset
,
173 uint64_t size
, uint64_t birth
, dmu_tx_t
*tx
)
175 spa_t
*spa
= ds
->ds_dir
->dd_pool
->dp_spa
;
177 ASSERT(dmu_tx_is_syncing(tx
));
178 ASSERT(birth
<= tx
->tx_txg
);
179 ASSERT(!ds
->ds_is_snapshot
);
181 if (birth
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
182 spa_vdev_indirect_mark_obsolete(spa
, vdev
, offset
, size
, tx
);
185 dva_t
*dva
= &fakebp
.blk_dva
[0];
189 mutex_enter(&ds
->ds_remap_deadlist_lock
);
190 if (!dsl_dataset_remap_deadlist_exists(ds
)) {
191 dsl_dataset_create_remap_deadlist(ds
, tx
);
193 mutex_exit(&ds
->ds_remap_deadlist_lock
);
196 fakebp
.blk_birth
= birth
;
197 DVA_SET_VDEV(dva
, vdev
);
198 DVA_SET_OFFSET(dva
, offset
);
199 DVA_SET_ASIZE(dva
, size
);
201 dsl_deadlist_insert(&ds
->ds_remap_deadlist
, &fakebp
, tx
);
206 dsl_dataset_block_kill(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
,
209 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
211 int used
= bp_get_dsize_sync(spa
, bp
);
212 int compressed
= BP_GET_PSIZE(bp
);
213 int uncompressed
= BP_GET_UCSIZE(bp
);
218 ASSERT(dmu_tx_is_syncing(tx
));
219 ASSERT(bp
->blk_birth
<= tx
->tx_txg
);
222 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
223 dsl_pool_mos_diduse_space(tx
->tx_pool
,
224 -used
, -compressed
, -uncompressed
);
227 ASSERT3P(tx
->tx_pool
, ==, ds
->ds_dir
->dd_pool
);
229 ASSERT(!ds
->ds_is_snapshot
);
230 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
232 if (bp
->blk_birth
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
235 dprintf_bp(bp
, "freeing ds=%llu", ds
->ds_object
);
236 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
238 mutex_enter(&ds
->ds_lock
);
239 ASSERT(dsl_dataset_phys(ds
)->ds_unique_bytes
>= used
||
240 !DS_UNIQUE_IS_ACCURATE(ds
));
241 delta
= parent_delta(ds
, -used
);
242 dsl_dataset_phys(ds
)->ds_unique_bytes
-= used
;
243 mutex_exit(&ds
->ds_lock
);
244 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
245 delta
, -compressed
, -uncompressed
, tx
);
246 dsl_dir_transfer_space(ds
->ds_dir
, -used
- delta
,
247 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
249 dprintf_bp(bp
, "putting on dead list: %s", "");
252 * We are here as part of zio's write done callback,
253 * which means we're a zio interrupt thread. We can't
254 * call dsl_deadlist_insert() now because it may block
255 * waiting for I/O. Instead, put bp on the deferred
256 * queue and let dsl_pool_sync() finish the job.
258 bplist_append(&ds
->ds_pending_deadlist
, bp
);
260 dsl_deadlist_insert(&ds
->ds_deadlist
, bp
, tx
);
262 ASSERT3U(ds
->ds_prev
->ds_object
, ==,
263 dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
264 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 0);
265 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
266 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
267 ds
->ds_object
&& bp
->blk_birth
>
268 dsl_dataset_phys(ds
->ds_prev
)->ds_prev_snap_txg
) {
269 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
270 mutex_enter(&ds
->ds_prev
->ds_lock
);
271 dsl_dataset_phys(ds
->ds_prev
)->ds_unique_bytes
+= used
;
272 mutex_exit(&ds
->ds_prev
->ds_lock
);
274 if (bp
->blk_birth
> ds
->ds_dir
->dd_origin_txg
) {
275 dsl_dir_transfer_space(ds
->ds_dir
, used
,
276 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
279 mutex_enter(&ds
->ds_lock
);
280 ASSERT3U(dsl_dataset_phys(ds
)->ds_referenced_bytes
, >=, used
);
281 dsl_dataset_phys(ds
)->ds_referenced_bytes
-= used
;
282 ASSERT3U(dsl_dataset_phys(ds
)->ds_compressed_bytes
, >=, compressed
);
283 dsl_dataset_phys(ds
)->ds_compressed_bytes
-= compressed
;
284 ASSERT3U(dsl_dataset_phys(ds
)->ds_uncompressed_bytes
, >=, uncompressed
);
285 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
-= uncompressed
;
286 mutex_exit(&ds
->ds_lock
);
292 * We have to release the fsid syncronously or we risk that a subsequent
293 * mount of the same dataset will fail to unique_insert the fsid. This
294 * failure would manifest itself as the fsid of this dataset changing
295 * between mounts which makes NFS clients quite unhappy.
298 dsl_dataset_evict_sync(void *dbu
)
300 dsl_dataset_t
*ds
= dbu
;
302 ASSERT(ds
->ds_owner
== NULL
);
304 unique_remove(ds
->ds_fsid_guid
);
308 dsl_dataset_evict_async(void *dbu
)
310 dsl_dataset_t
*ds
= dbu
;
312 ASSERT(ds
->ds_owner
== NULL
);
316 if (ds
->ds_objset
!= NULL
)
317 dmu_objset_evict(ds
->ds_objset
);
320 dsl_dataset_rele(ds
->ds_prev
, ds
);
324 bplist_destroy(&ds
->ds_pending_deadlist
);
325 if (dsl_deadlist_is_open(&ds
->ds_deadlist
))
326 dsl_deadlist_close(&ds
->ds_deadlist
);
327 if (dsl_deadlist_is_open(&ds
->ds_remap_deadlist
))
328 dsl_deadlist_close(&ds
->ds_remap_deadlist
);
330 dsl_dir_async_rele(ds
->ds_dir
, ds
);
332 ASSERT(!list_link_active(&ds
->ds_synced_link
));
334 list_destroy(&ds
->ds_prop_cbs
);
335 mutex_destroy(&ds
->ds_lock
);
336 mutex_destroy(&ds
->ds_opening_lock
);
337 mutex_destroy(&ds
->ds_sendstream_lock
);
338 mutex_destroy(&ds
->ds_remap_deadlist_lock
);
339 refcount_destroy(&ds
->ds_longholds
);
340 rrw_destroy(&ds
->ds_bp_rwlock
);
342 kmem_free(ds
, sizeof (dsl_dataset_t
));
346 dsl_dataset_get_snapname(dsl_dataset_t
*ds
)
348 dsl_dataset_phys_t
*headphys
;
351 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
352 objset_t
*mos
= dp
->dp_meta_objset
;
354 if (ds
->ds_snapname
[0])
356 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0)
359 err
= dmu_bonus_hold(mos
, dsl_dir_phys(ds
->ds_dir
)->dd_head_dataset_obj
,
363 headphys
= headdbuf
->db_data
;
364 err
= zap_value_search(dp
->dp_meta_objset
,
365 headphys
->ds_snapnames_zapobj
, ds
->ds_object
, 0, ds
->ds_snapname
);
366 dmu_buf_rele(headdbuf
, FTAG
);
371 dsl_dataset_snap_lookup(dsl_dataset_t
*ds
, const char *name
, uint64_t *value
)
373 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
374 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
378 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
381 err
= zap_lookup_norm(mos
, snapobj
, name
, 8, 1,
382 value
, mt
, NULL
, 0, NULL
);
383 if (err
== ENOTSUP
&& (mt
& MT_NORMALIZE
))
384 err
= zap_lookup(mos
, snapobj
, name
, 8, 1, value
);
389 dsl_dataset_snap_remove(dsl_dataset_t
*ds
, const char *name
, dmu_tx_t
*tx
,
392 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
393 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
397 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
399 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
402 err
= zap_remove_norm(mos
, snapobj
, name
, mt
, tx
);
403 if (err
== ENOTSUP
&& (mt
& MT_NORMALIZE
))
404 err
= zap_remove(mos
, snapobj
, name
, tx
);
406 if (err
== 0 && adj_cnt
)
407 dsl_fs_ss_count_adjust(ds
->ds_dir
, -1,
408 DD_FIELD_SNAPSHOT_COUNT
, tx
);
414 dsl_dataset_try_add_ref(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *tag
)
416 dmu_buf_t
*dbuf
= ds
->ds_dbuf
;
417 boolean_t result
= B_FALSE
;
419 if (dbuf
!= NULL
&& dmu_buf_try_add_ref(dbuf
, dp
->dp_meta_objset
,
420 ds
->ds_object
, DMU_BONUS_BLKID
, tag
)) {
422 if (ds
== dmu_buf_get_user(dbuf
))
425 dmu_buf_rele(dbuf
, tag
);
432 dsl_dataset_hold_obj(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
435 objset_t
*mos
= dp
->dp_meta_objset
;
439 dmu_object_info_t doi
;
441 ASSERT(dsl_pool_config_held(dp
));
443 err
= dmu_bonus_hold(mos
, dsobj
, tag
, &dbuf
);
447 /* Make sure dsobj has the correct object type. */
448 dmu_object_info_from_db(dbuf
, &doi
);
449 if (doi
.doi_bonus_type
!= DMU_OT_DSL_DATASET
) {
450 dmu_buf_rele(dbuf
, tag
);
451 return (SET_ERROR(EINVAL
));
454 ds
= dmu_buf_get_user(dbuf
);
456 dsl_dataset_t
*winner
= NULL
;
458 ds
= kmem_zalloc(sizeof (dsl_dataset_t
), KM_SLEEP
);
460 ds
->ds_object
= dsobj
;
461 ds
->ds_is_snapshot
= dsl_dataset_phys(ds
)->ds_num_children
!= 0;
463 err
= dsl_dir_hold_obj(dp
, dsl_dataset_phys(ds
)->ds_dir_obj
,
464 NULL
, ds
, &ds
->ds_dir
);
466 kmem_free(ds
, sizeof (dsl_dataset_t
));
467 dmu_buf_rele(dbuf
, tag
);
471 mutex_init(&ds
->ds_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
472 mutex_init(&ds
->ds_opening_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
473 mutex_init(&ds
->ds_sendstream_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
474 mutex_init(&ds
->ds_remap_deadlist_lock
,
475 NULL
, MUTEX_DEFAULT
, NULL
);
476 rrw_init(&ds
->ds_bp_rwlock
, B_FALSE
);
477 refcount_create(&ds
->ds_longholds
);
479 bplist_create(&ds
->ds_pending_deadlist
);
481 list_create(&ds
->ds_sendstreams
, sizeof (dmu_sendarg_t
),
482 offsetof(dmu_sendarg_t
, dsa_link
));
484 list_create(&ds
->ds_prop_cbs
, sizeof (dsl_prop_cb_record_t
),
485 offsetof(dsl_prop_cb_record_t
, cbr_ds_node
));
487 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
488 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++) {
489 if (!(spa_feature_table
[f
].fi_flags
&
490 ZFEATURE_FLAG_PER_DATASET
))
492 err
= zap_contains(mos
, dsobj
,
493 spa_feature_table
[f
].fi_guid
);
495 ds
->ds_feature_inuse
[f
] = B_TRUE
;
497 ASSERT3U(err
, ==, ENOENT
);
503 if (!ds
->ds_is_snapshot
) {
504 ds
->ds_snapname
[0] = '\0';
505 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
506 err
= dsl_dataset_hold_obj(dp
,
507 dsl_dataset_phys(ds
)->ds_prev_snap_obj
,
510 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
511 int zaperr
= zap_lookup(mos
, ds
->ds_object
,
512 DS_FIELD_BOOKMARK_NAMES
,
513 sizeof (ds
->ds_bookmarks
), 1,
515 if (zaperr
!= ENOENT
)
519 if (zfs_flags
& ZFS_DEBUG_SNAPNAMES
)
520 err
= dsl_dataset_get_snapname(ds
);
522 dsl_dataset_phys(ds
)->ds_userrefs_obj
!= 0) {
524 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
525 dsl_dataset_phys(ds
)->ds_userrefs_obj
,
530 if (err
== 0 && !ds
->ds_is_snapshot
) {
531 err
= dsl_prop_get_int_ds(ds
,
532 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
535 err
= dsl_prop_get_int_ds(ds
,
536 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
540 ds
->ds_reserved
= ds
->ds_quota
= 0;
543 dsl_deadlist_open(&ds
->ds_deadlist
,
544 mos
, dsl_dataset_phys(ds
)->ds_deadlist_obj
);
545 uint64_t remap_deadlist_obj
=
546 dsl_dataset_get_remap_deadlist_object(ds
);
547 if (remap_deadlist_obj
!= 0) {
548 dsl_deadlist_open(&ds
->ds_remap_deadlist
, mos
,
552 dmu_buf_init_user(&ds
->ds_dbu
, dsl_dataset_evict_sync
,
553 dsl_dataset_evict_async
, &ds
->ds_dbuf
);
555 winner
= dmu_buf_set_user_ie(dbuf
, &ds
->ds_dbu
);
557 if (err
!= 0 || winner
!= NULL
) {
558 bplist_destroy(&ds
->ds_pending_deadlist
);
559 dsl_deadlist_close(&ds
->ds_deadlist
);
560 if (dsl_deadlist_is_open(&ds
->ds_remap_deadlist
))
561 dsl_deadlist_close(&ds
->ds_remap_deadlist
);
563 dsl_dataset_rele(ds
->ds_prev
, ds
);
564 dsl_dir_rele(ds
->ds_dir
, ds
);
565 mutex_destroy(&ds
->ds_lock
);
566 mutex_destroy(&ds
->ds_opening_lock
);
567 mutex_destroy(&ds
->ds_sendstream_lock
);
568 refcount_destroy(&ds
->ds_longholds
);
569 kmem_free(ds
, sizeof (dsl_dataset_t
));
571 dmu_buf_rele(dbuf
, tag
);
577 unique_insert(dsl_dataset_phys(ds
)->ds_fsid_guid
);
578 if (ds
->ds_fsid_guid
!=
579 dsl_dataset_phys(ds
)->ds_fsid_guid
) {
580 zfs_dbgmsg("ds_fsid_guid changed from "
581 "%llx to %llx for pool %s dataset id %llu",
583 dsl_dataset_phys(ds
)->ds_fsid_guid
,
584 (long long)ds
->ds_fsid_guid
,
585 spa_name(dp
->dp_spa
),
590 ASSERT3P(ds
->ds_dbuf
, ==, dbuf
);
591 ASSERT3P(dsl_dataset_phys(ds
), ==, dbuf
->db_data
);
592 ASSERT(dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0 ||
593 spa_version(dp
->dp_spa
) < SPA_VERSION_ORIGIN
||
594 dp
->dp_origin_snap
== NULL
|| ds
== dp
->dp_origin_snap
);
600 dsl_dataset_hold(dsl_pool_t
*dp
, const char *name
,
601 void *tag
, dsl_dataset_t
**dsp
)
604 const char *snapname
;
609 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, &snapname
);
613 ASSERT(dsl_pool_config_held(dp
));
614 obj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
616 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
618 err
= SET_ERROR(ENOENT
);
620 /* we may be looking for a snapshot */
621 if (err
== 0 && snapname
!= NULL
) {
622 dsl_dataset_t
*snap_ds
;
624 if (*snapname
++ != '@') {
625 dsl_dataset_rele(ds
, tag
);
626 dsl_dir_rele(dd
, FTAG
);
627 return (SET_ERROR(ENOENT
));
630 dprintf("looking for snapshot '%s'\n", snapname
);
631 err
= dsl_dataset_snap_lookup(ds
, snapname
, &obj
);
633 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &snap_ds
);
634 dsl_dataset_rele(ds
, tag
);
637 mutex_enter(&snap_ds
->ds_lock
);
638 if (snap_ds
->ds_snapname
[0] == 0)
639 (void) strlcpy(snap_ds
->ds_snapname
, snapname
,
640 sizeof (snap_ds
->ds_snapname
));
641 mutex_exit(&snap_ds
->ds_lock
);
647 dsl_dir_rele(dd
, FTAG
);
652 dsl_dataset_own_obj(dsl_pool_t
*dp
, uint64_t dsobj
,
653 void *tag
, dsl_dataset_t
**dsp
)
655 int err
= dsl_dataset_hold_obj(dp
, dsobj
, tag
, dsp
);
658 if (!dsl_dataset_tryown(*dsp
, tag
)) {
659 dsl_dataset_rele(*dsp
, tag
);
661 return (SET_ERROR(EBUSY
));
667 dsl_dataset_own(dsl_pool_t
*dp
, const char *name
,
668 void *tag
, dsl_dataset_t
**dsp
)
670 int err
= dsl_dataset_hold(dp
, name
, tag
, dsp
);
673 if (!dsl_dataset_tryown(*dsp
, tag
)) {
674 dsl_dataset_rele(*dsp
, tag
);
675 return (SET_ERROR(EBUSY
));
681 * See the comment above dsl_pool_hold() for details. In summary, a long
682 * hold is used to prevent destruction of a dataset while the pool hold
683 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
685 * The dataset and pool must be held when this function is called. After it
686 * is called, the pool hold may be released while the dataset is still held
690 dsl_dataset_long_hold(dsl_dataset_t
*ds
, void *tag
)
692 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
693 (void) refcount_add(&ds
->ds_longholds
, tag
);
697 dsl_dataset_long_rele(dsl_dataset_t
*ds
, void *tag
)
699 (void) refcount_remove(&ds
->ds_longholds
, tag
);
702 /* Return B_TRUE if there are any long holds on this dataset. */
704 dsl_dataset_long_held(dsl_dataset_t
*ds
)
706 return (!refcount_is_zero(&ds
->ds_longholds
));
710 dsl_dataset_name(dsl_dataset_t
*ds
, char *name
)
713 (void) strcpy(name
, "mos");
715 dsl_dir_name(ds
->ds_dir
, name
);
716 VERIFY0(dsl_dataset_get_snapname(ds
));
717 if (ds
->ds_snapname
[0]) {
718 VERIFY3U(strlcat(name
, "@", ZFS_MAX_DATASET_NAME_LEN
),
719 <, ZFS_MAX_DATASET_NAME_LEN
);
721 * We use a "recursive" mutex so that we
722 * can call dprintf_ds() with ds_lock held.
724 if (!MUTEX_HELD(&ds
->ds_lock
)) {
725 mutex_enter(&ds
->ds_lock
);
726 VERIFY3U(strlcat(name
, ds
->ds_snapname
,
727 ZFS_MAX_DATASET_NAME_LEN
), <,
728 ZFS_MAX_DATASET_NAME_LEN
);
729 mutex_exit(&ds
->ds_lock
);
731 VERIFY3U(strlcat(name
, ds
->ds_snapname
,
732 ZFS_MAX_DATASET_NAME_LEN
), <,
733 ZFS_MAX_DATASET_NAME_LEN
);
740 dsl_dataset_namelen(dsl_dataset_t
*ds
)
742 VERIFY0(dsl_dataset_get_snapname(ds
));
743 mutex_enter(&ds
->ds_lock
);
744 int len
= dsl_dir_namelen(ds
->ds_dir
) + 1 + strlen(ds
->ds_snapname
);
745 mutex_exit(&ds
->ds_lock
);
750 dsl_dataset_rele(dsl_dataset_t
*ds
, void *tag
)
752 dmu_buf_rele(ds
->ds_dbuf
, tag
);
756 dsl_dataset_disown(dsl_dataset_t
*ds
, void *tag
)
758 ASSERT3P(ds
->ds_owner
, ==, tag
);
759 ASSERT(ds
->ds_dbuf
!= NULL
);
761 mutex_enter(&ds
->ds_lock
);
763 mutex_exit(&ds
->ds_lock
);
764 dsl_dataset_long_rele(ds
, tag
);
765 dsl_dataset_rele(ds
, tag
);
769 dsl_dataset_tryown(dsl_dataset_t
*ds
, void *tag
)
771 boolean_t gotit
= FALSE
;
773 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
774 mutex_enter(&ds
->ds_lock
);
775 if (ds
->ds_owner
== NULL
&& !DS_IS_INCONSISTENT(ds
)) {
777 dsl_dataset_long_hold(ds
, tag
);
780 mutex_exit(&ds
->ds_lock
);
785 dsl_dataset_has_owner(dsl_dataset_t
*ds
)
788 mutex_enter(&ds
->ds_lock
);
789 rv
= (ds
->ds_owner
!= NULL
);
790 mutex_exit(&ds
->ds_lock
);
795 dsl_dataset_activate_feature(uint64_t dsobj
, spa_feature_t f
, dmu_tx_t
*tx
)
797 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
798 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
801 VERIFY(spa_feature_table
[f
].fi_flags
& ZFEATURE_FLAG_PER_DATASET
);
803 spa_feature_incr(spa
, f
, tx
);
804 dmu_object_zapify(mos
, dsobj
, DMU_OT_DSL_DATASET
, tx
);
806 VERIFY0(zap_add(mos
, dsobj
, spa_feature_table
[f
].fi_guid
,
807 sizeof (zero
), 1, &zero
, tx
));
811 dsl_dataset_deactivate_feature(uint64_t dsobj
, spa_feature_t f
, dmu_tx_t
*tx
)
813 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
814 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
816 VERIFY(spa_feature_table
[f
].fi_flags
& ZFEATURE_FLAG_PER_DATASET
);
818 VERIFY0(zap_remove(mos
, dsobj
, spa_feature_table
[f
].fi_guid
, tx
));
819 spa_feature_decr(spa
, f
, tx
);
823 dsl_dataset_create_sync_dd(dsl_dir_t
*dd
, dsl_dataset_t
*origin
,
824 uint64_t flags
, dmu_tx_t
*tx
)
826 dsl_pool_t
*dp
= dd
->dd_pool
;
828 dsl_dataset_phys_t
*dsphys
;
830 objset_t
*mos
= dp
->dp_meta_objset
;
833 origin
= dp
->dp_origin_snap
;
835 ASSERT(origin
== NULL
|| origin
->ds_dir
->dd_pool
== dp
);
836 ASSERT(origin
== NULL
|| dsl_dataset_phys(origin
)->ds_num_children
> 0);
837 ASSERT(dmu_tx_is_syncing(tx
));
838 ASSERT(dsl_dir_phys(dd
)->dd_head_dataset_obj
== 0);
840 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
841 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
842 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
843 dmu_buf_will_dirty(dbuf
, tx
);
844 dsphys
= dbuf
->db_data
;
845 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
846 dsphys
->ds_dir_obj
= dd
->dd_object
;
847 dsphys
->ds_flags
= flags
;
848 dsphys
->ds_fsid_guid
= unique_create();
849 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
850 sizeof (dsphys
->ds_guid
));
851 dsphys
->ds_snapnames_zapobj
=
852 zap_create_norm(mos
, U8_TEXTPREP_TOUPPER
, DMU_OT_DSL_DS_SNAP_MAP
,
854 dsphys
->ds_creation_time
= gethrestime_sec();
855 dsphys
->ds_creation_txg
= tx
->tx_txg
== TXG_INITIAL
? 1 : tx
->tx_txg
;
857 if (origin
== NULL
) {
858 dsphys
->ds_deadlist_obj
= dsl_deadlist_alloc(mos
, tx
);
860 dsl_dataset_t
*ohds
; /* head of the origin snapshot */
862 dsphys
->ds_prev_snap_obj
= origin
->ds_object
;
863 dsphys
->ds_prev_snap_txg
=
864 dsl_dataset_phys(origin
)->ds_creation_txg
;
865 dsphys
->ds_referenced_bytes
=
866 dsl_dataset_phys(origin
)->ds_referenced_bytes
;
867 dsphys
->ds_compressed_bytes
=
868 dsl_dataset_phys(origin
)->ds_compressed_bytes
;
869 dsphys
->ds_uncompressed_bytes
=
870 dsl_dataset_phys(origin
)->ds_uncompressed_bytes
;
871 rrw_enter(&origin
->ds_bp_rwlock
, RW_READER
, FTAG
);
872 dsphys
->ds_bp
= dsl_dataset_phys(origin
)->ds_bp
;
873 rrw_exit(&origin
->ds_bp_rwlock
, FTAG
);
876 * Inherit flags that describe the dataset's contents
877 * (INCONSISTENT) or properties (Case Insensitive).
879 dsphys
->ds_flags
|= dsl_dataset_phys(origin
)->ds_flags
&
880 (DS_FLAG_INCONSISTENT
| DS_FLAG_CI_DATASET
);
882 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++) {
883 if (origin
->ds_feature_inuse
[f
])
884 dsl_dataset_activate_feature(dsobj
, f
, tx
);
887 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
888 dsl_dataset_phys(origin
)->ds_num_children
++;
890 VERIFY0(dsl_dataset_hold_obj(dp
,
891 dsl_dir_phys(origin
->ds_dir
)->dd_head_dataset_obj
,
893 dsphys
->ds_deadlist_obj
= dsl_deadlist_clone(&ohds
->ds_deadlist
,
894 dsphys
->ds_prev_snap_txg
, dsphys
->ds_prev_snap_obj
, tx
);
895 dsl_dataset_rele(ohds
, FTAG
);
897 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_NEXT_CLONES
) {
898 if (dsl_dataset_phys(origin
)->ds_next_clones_obj
== 0) {
899 dsl_dataset_phys(origin
)->ds_next_clones_obj
=
901 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
903 VERIFY0(zap_add_int(mos
,
904 dsl_dataset_phys(origin
)->ds_next_clones_obj
,
908 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
909 dsl_dir_phys(dd
)->dd_origin_obj
= origin
->ds_object
;
910 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
911 if (dsl_dir_phys(origin
->ds_dir
)->dd_clones
== 0) {
912 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
913 dsl_dir_phys(origin
->ds_dir
)->dd_clones
=
915 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
917 VERIFY0(zap_add_int(mos
,
918 dsl_dir_phys(origin
->ds_dir
)->dd_clones
,
923 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
924 dsphys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
926 dmu_buf_rele(dbuf
, FTAG
);
928 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
929 dsl_dir_phys(dd
)->dd_head_dataset_obj
= dsobj
;
935 dsl_dataset_zero_zil(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
939 VERIFY0(dmu_objset_from_ds(ds
, &os
));
940 if (bcmp(&os
->os_zil_header
, &zero_zil
, sizeof (zero_zil
)) != 0) {
941 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
944 bzero(&os
->os_zil_header
, sizeof (os
->os_zil_header
));
946 zio
= zio_root(dp
->dp_spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
947 dsl_dataset_sync(ds
, zio
, tx
);
948 VERIFY0(zio_wait(zio
));
950 /* dsl_dataset_sync_done will drop this reference. */
951 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
952 dsl_dataset_sync_done(ds
, tx
);
957 dsl_dataset_create_sync(dsl_dir_t
*pdd
, const char *lastname
,
958 dsl_dataset_t
*origin
, uint64_t flags
, cred_t
*cr
, dmu_tx_t
*tx
)
960 dsl_pool_t
*dp
= pdd
->dd_pool
;
961 uint64_t dsobj
, ddobj
;
964 ASSERT(dmu_tx_is_syncing(tx
));
965 ASSERT(lastname
[0] != '@');
967 ddobj
= dsl_dir_create_sync(dp
, pdd
, lastname
, tx
);
968 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, lastname
, FTAG
, &dd
));
970 dsobj
= dsl_dataset_create_sync_dd(dd
, origin
,
971 flags
& ~DS_CREATE_FLAG_NODIRTY
, tx
);
973 dsl_deleg_set_create_perms(dd
, tx
, cr
);
976 * Since we're creating a new node we know it's a leaf, so we can
977 * initialize the counts if the limit feature is active.
979 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
981 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
983 dsl_dir_zapify(dd
, tx
);
984 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
985 sizeof (cnt
), 1, &cnt
, tx
));
986 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
987 sizeof (cnt
), 1, &cnt
, tx
));
990 dsl_dir_rele(dd
, FTAG
);
993 * If we are creating a clone, make sure we zero out any stale
994 * data from the origin snapshots zil header.
996 if (origin
!= NULL
&& !(flags
& DS_CREATE_FLAG_NODIRTY
)) {
999 VERIFY0(dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
1000 dsl_dataset_zero_zil(ds
, tx
);
1001 dsl_dataset_rele(ds
, FTAG
);
1008 * The unique space in the head dataset can be calculated by subtracting
1009 * the space used in the most recent snapshot, that is still being used
1010 * in this file system, from the space currently in use. To figure out
1011 * the space in the most recent snapshot still in use, we need to take
1012 * the total space used in the snapshot and subtract out the space that
1013 * has been freed up since the snapshot was taken.
1016 dsl_dataset_recalc_head_uniq(dsl_dataset_t
*ds
)
1019 uint64_t dlused
, dlcomp
, dluncomp
;
1021 ASSERT(!ds
->ds_is_snapshot
);
1023 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0)
1024 mrs_used
= dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
;
1028 dsl_deadlist_space(&ds
->ds_deadlist
, &dlused
, &dlcomp
, &dluncomp
);
1030 ASSERT3U(dlused
, <=, mrs_used
);
1031 dsl_dataset_phys(ds
)->ds_unique_bytes
=
1032 dsl_dataset_phys(ds
)->ds_referenced_bytes
- (mrs_used
- dlused
);
1034 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) >=
1035 SPA_VERSION_UNIQUE_ACCURATE
)
1036 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1040 dsl_dataset_remove_from_next_clones(dsl_dataset_t
*ds
, uint64_t obj
,
1043 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1047 ASSERT(dsl_dataset_phys(ds
)->ds_num_children
>= 2);
1048 err
= zap_remove_int(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1051 * The err should not be ENOENT, but a bug in a previous version
1052 * of the code could cause upgrade_clones_cb() to not set
1053 * ds_next_snap_obj when it should, leading to a missing entry.
1054 * If we knew that the pool was created after
1055 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1056 * ENOENT. However, at least we can check that we don't have
1057 * too many entries in the next_clones_obj even after failing to
1062 ASSERT0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1064 ASSERT3U(count
, <=, dsl_dataset_phys(ds
)->ds_num_children
- 2);
1069 dsl_dataset_get_blkptr(dsl_dataset_t
*ds
)
1071 return (&dsl_dataset_phys(ds
)->ds_bp
);
1075 dsl_dataset_get_spa(dsl_dataset_t
*ds
)
1077 return (ds
->ds_dir
->dd_pool
->dp_spa
);
1081 dsl_dataset_dirty(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1085 if (ds
== NULL
) /* this is the meta-objset */
1088 ASSERT(ds
->ds_objset
!= NULL
);
1090 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
!= 0)
1091 panic("dirtying snapshot!");
1093 /* Must not dirty a dataset in the same txg where it got snapshotted. */
1094 ASSERT3U(tx
->tx_txg
, >, dsl_dataset_phys(ds
)->ds_prev_snap_txg
);
1096 dp
= ds
->ds_dir
->dd_pool
;
1097 if (txg_list_add(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
1098 /* up the hold count until we can be written out */
1099 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
1104 dsl_dataset_is_dirty(dsl_dataset_t
*ds
)
1106 for (int t
= 0; t
< TXG_SIZE
; t
++) {
1107 if (txg_list_member(&ds
->ds_dir
->dd_pool
->dp_dirty_datasets
,
1115 dsl_dataset_snapshot_reserve_space(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1119 if (!dmu_tx_is_syncing(tx
))
1123 * If there's an fs-only reservation, any blocks that might become
1124 * owned by the snapshot dataset must be accommodated by space
1125 * outside of the reservation.
1127 ASSERT(ds
->ds_reserved
== 0 || DS_UNIQUE_IS_ACCURATE(ds
));
1128 asize
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
, ds
->ds_reserved
);
1129 if (asize
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
1130 return (SET_ERROR(ENOSPC
));
1133 * Propagate any reserved space for this snapshot to other
1134 * snapshot checks in this sync group.
1137 dsl_dir_willuse_space(ds
->ds_dir
, asize
, tx
);
1143 dsl_dataset_snapshot_check_impl(dsl_dataset_t
*ds
, const char *snapname
,
1144 dmu_tx_t
*tx
, boolean_t recv
, uint64_t cnt
, cred_t
*cr
)
1149 ds
->ds_trysnap_txg
= tx
->tx_txg
;
1151 if (!dmu_tx_is_syncing(tx
))
1155 * We don't allow multiple snapshots of the same txg. If there
1156 * is already one, try again.
1158 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
>= tx
->tx_txg
)
1159 return (SET_ERROR(EAGAIN
));
1162 * Check for conflicting snapshot name.
1164 error
= dsl_dataset_snap_lookup(ds
, snapname
, &value
);
1166 return (SET_ERROR(EEXIST
));
1167 if (error
!= ENOENT
)
1171 * We don't allow taking snapshots of inconsistent datasets, such as
1172 * those into which we are currently receiving. However, if we are
1173 * creating this snapshot as part of a receive, this check will be
1174 * executed atomically with respect to the completion of the receive
1175 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1176 * case we ignore this, knowing it will be fixed up for us shortly in
1177 * dmu_recv_end_sync().
1179 if (!recv
&& DS_IS_INCONSISTENT(ds
))
1180 return (SET_ERROR(EBUSY
));
1183 * Skip the check for temporary snapshots or if we have already checked
1184 * the counts in dsl_dataset_snapshot_check. This means we really only
1185 * check the count here when we're receiving a stream.
1187 if (cnt
!= 0 && cr
!= NULL
) {
1188 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1189 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, cr
);
1194 error
= dsl_dataset_snapshot_reserve_space(ds
, tx
);
1202 dsl_dataset_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1204 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1205 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1210 * Pre-compute how many total new snapshots will be created for each
1211 * level in the tree and below. This is needed for validating the
1212 * snapshot limit when either taking a recursive snapshot or when
1213 * taking multiple snapshots.
1215 * The problem is that the counts are not actually adjusted when
1216 * we are checking, only when we finally sync. For a single snapshot,
1217 * this is easy, the count will increase by 1 at each node up the tree,
1218 * but its more complicated for the recursive/multiple snapshot case.
1220 * The dsl_fs_ss_limit_check function does recursively check the count
1221 * at each level up the tree but since it is validating each snapshot
1222 * independently we need to be sure that we are validating the complete
1223 * count for the entire set of snapshots. We do this by rolling up the
1224 * counts for each component of the name into an nvlist and then
1225 * checking each of those cases with the aggregated count.
1227 * This approach properly handles not only the recursive snapshot
1228 * case (where we get all of those on the ddsa_snaps list) but also
1229 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1230 * validate the limit on 'a' using a count of 2).
1232 * We validate the snapshot names in the third loop and only report
1235 if (dmu_tx_is_syncing(tx
)) {
1236 nvlist_t
*cnt_track
= NULL
;
1237 cnt_track
= fnvlist_alloc();
1239 /* Rollup aggregated counts into the cnt_track list */
1240 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1242 pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1245 char nm
[MAXPATHLEN
];
1247 (void) strlcpy(nm
, nvpair_name(pair
), sizeof (nm
));
1248 pdelim
= strchr(nm
, '@');
1254 if (nvlist_lookup_uint64(cnt_track
, nm
,
1256 /* update existing entry */
1257 fnvlist_add_uint64(cnt_track
, nm
,
1261 fnvlist_add_uint64(cnt_track
, nm
, 1);
1264 pdelim
= strrchr(nm
, '/');
1267 } while (pdelim
!= NULL
);
1270 /* Check aggregated counts at each level */
1271 for (pair
= nvlist_next_nvpair(cnt_track
, NULL
);
1272 pair
!= NULL
; pair
= nvlist_next_nvpair(cnt_track
, pair
)) {
1278 name
= nvpair_name(pair
);
1279 cnt
= fnvpair_value_uint64(pair
);
1282 error
= dsl_dataset_hold(dp
, name
, FTAG
, &ds
);
1284 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1285 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
,
1287 dsl_dataset_rele(ds
, FTAG
);
1291 if (ddsa
->ddsa_errors
!= NULL
)
1292 fnvlist_add_int32(ddsa
->ddsa_errors
,
1295 /* only report one error for this check */
1299 nvlist_free(cnt_track
);
1302 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1303 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1307 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
1309 name
= nvpair_name(pair
);
1310 if (strlen(name
) >= ZFS_MAX_DATASET_NAME_LEN
)
1311 error
= SET_ERROR(ENAMETOOLONG
);
1313 atp
= strchr(name
, '@');
1315 error
= SET_ERROR(EINVAL
);
1317 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1320 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
1322 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1323 error
= dsl_dataset_snapshot_check_impl(ds
,
1324 atp
+ 1, tx
, B_FALSE
, 0, NULL
);
1325 dsl_dataset_rele(ds
, FTAG
);
1329 if (ddsa
->ddsa_errors
!= NULL
) {
1330 fnvlist_add_int32(ddsa
->ddsa_errors
,
1341 dsl_dataset_snapshot_sync_impl(dsl_dataset_t
*ds
, const char *snapname
,
1344 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1346 dsl_dataset_phys_t
*dsphys
;
1347 uint64_t dsobj
, crtxg
;
1348 objset_t
*mos
= dp
->dp_meta_objset
;
1351 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
1354 * If we are on an old pool, the zil must not be active, in which
1355 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1357 ASSERT(spa_version(dmu_tx_pool(tx
)->dp_spa
) >= SPA_VERSION_FAST_SNAP
||
1358 dmu_objset_from_ds(ds
, &os
) != 0 ||
1359 bcmp(&os
->os_phys
->os_zil_header
, &zero_zil
,
1360 sizeof (zero_zil
)) == 0);
1362 /* Should not snapshot a dirty dataset. */
1363 ASSERT(!txg_list_member(&ds
->ds_dir
->dd_pool
->dp_dirty_datasets
,
1366 dsl_fs_ss_count_adjust(ds
->ds_dir
, 1, DD_FIELD_SNAPSHOT_COUNT
, tx
);
1369 * The origin's ds_creation_txg has to be < TXG_INITIAL
1371 if (strcmp(snapname
, ORIGIN_DIR_NAME
) == 0)
1376 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
1377 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
1378 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
1379 dmu_buf_will_dirty(dbuf
, tx
);
1380 dsphys
= dbuf
->db_data
;
1381 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
1382 dsphys
->ds_dir_obj
= ds
->ds_dir
->dd_object
;
1383 dsphys
->ds_fsid_guid
= unique_create();
1384 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
1385 sizeof (dsphys
->ds_guid
));
1386 dsphys
->ds_prev_snap_obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1387 dsphys
->ds_prev_snap_txg
= dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
1388 dsphys
->ds_next_snap_obj
= ds
->ds_object
;
1389 dsphys
->ds_num_children
= 1;
1390 dsphys
->ds_creation_time
= gethrestime_sec();
1391 dsphys
->ds_creation_txg
= crtxg
;
1392 dsphys
->ds_deadlist_obj
= dsl_dataset_phys(ds
)->ds_deadlist_obj
;
1393 dsphys
->ds_referenced_bytes
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1394 dsphys
->ds_compressed_bytes
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1395 dsphys
->ds_uncompressed_bytes
=
1396 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1397 dsphys
->ds_flags
= dsl_dataset_phys(ds
)->ds_flags
;
1398 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1399 dsphys
->ds_bp
= dsl_dataset_phys(ds
)->ds_bp
;
1400 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
1401 dmu_buf_rele(dbuf
, FTAG
);
1403 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++) {
1404 if (ds
->ds_feature_inuse
[f
])
1405 dsl_dataset_activate_feature(dsobj
, f
, tx
);
1408 ASSERT3U(ds
->ds_prev
!= 0, ==,
1409 dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0);
1411 uint64_t next_clones_obj
=
1412 dsl_dataset_phys(ds
->ds_prev
)->ds_next_clones_obj
;
1413 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1415 dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 1);
1416 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1418 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1419 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, ==,
1420 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_txg
);
1421 dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
= dsobj
;
1422 } else if (next_clones_obj
!= 0) {
1423 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
1424 dsphys
->ds_next_snap_obj
, tx
);
1425 VERIFY0(zap_add_int(mos
,
1426 next_clones_obj
, dsobj
, tx
));
1431 * If we have a reference-reservation on this dataset, we will
1432 * need to increase the amount of refreservation being charged
1433 * since our unique space is going to zero.
1435 if (ds
->ds_reserved
) {
1437 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
1438 delta
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
,
1440 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
,
1444 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1445 dsl_dataset_phys(ds
)->ds_deadlist_obj
=
1446 dsl_deadlist_clone(&ds
->ds_deadlist
, UINT64_MAX
,
1447 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, tx
);
1448 dsl_deadlist_close(&ds
->ds_deadlist
);
1449 dsl_deadlist_open(&ds
->ds_deadlist
, mos
,
1450 dsl_dataset_phys(ds
)->ds_deadlist_obj
);
1451 dsl_deadlist_add_key(&ds
->ds_deadlist
,
1452 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
);
1454 if (dsl_dataset_remap_deadlist_exists(ds
)) {
1455 uint64_t remap_deadlist_obj
=
1456 dsl_dataset_get_remap_deadlist_object(ds
);
1458 * Move the remap_deadlist to the snapshot. The head
1459 * will create a new remap deadlist on demand, from
1460 * dsl_dataset_block_remapped().
1462 dsl_dataset_unset_remap_deadlist_object(ds
, tx
);
1463 dsl_deadlist_close(&ds
->ds_remap_deadlist
);
1465 dmu_object_zapify(mos
, dsobj
, DMU_OT_DSL_DATASET
, tx
);
1466 VERIFY0(zap_add(mos
, dsobj
, DS_FIELD_REMAP_DEADLIST
,
1467 sizeof (remap_deadlist_obj
), 1, &remap_deadlist_obj
, tx
));
1470 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, <, tx
->tx_txg
);
1471 dsl_dataset_phys(ds
)->ds_prev_snap_obj
= dsobj
;
1472 dsl_dataset_phys(ds
)->ds_prev_snap_txg
= crtxg
;
1473 dsl_dataset_phys(ds
)->ds_unique_bytes
= 0;
1475 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
1476 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1478 VERIFY0(zap_add(mos
, dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1479 snapname
, 8, 1, &dsobj
, tx
));
1482 dsl_dataset_rele(ds
->ds_prev
, ds
);
1483 VERIFY0(dsl_dataset_hold_obj(dp
,
1484 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, ds
, &ds
->ds_prev
));
1486 dsl_scan_ds_snapshotted(ds
, tx
);
1488 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
1490 spa_history_log_internal_ds(ds
->ds_prev
, "snapshot", tx
, "");
1494 dsl_dataset_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1496 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1497 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1500 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1501 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1504 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
1506 name
= nvpair_name(pair
);
1507 atp
= strchr(name
, '@');
1508 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1509 VERIFY0(dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
));
1511 dsl_dataset_snapshot_sync_impl(ds
, atp
+ 1, tx
);
1512 if (ddsa
->ddsa_props
!= NULL
) {
1513 dsl_props_set_sync_impl(ds
->ds_prev
,
1514 ZPROP_SRC_LOCAL
, ddsa
->ddsa_props
, tx
);
1516 dsl_dataset_rele(ds
, FTAG
);
1521 * The snapshots must all be in the same pool.
1522 * All-or-nothing: if there are any failures, nothing will be modified.
1525 dsl_dataset_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
*errors
)
1527 dsl_dataset_snapshot_arg_t ddsa
;
1529 boolean_t needsuspend
;
1533 nvlist_t
*suspended
= NULL
;
1535 pair
= nvlist_next_nvpair(snaps
, NULL
);
1538 firstname
= nvpair_name(pair
);
1540 error
= spa_open(firstname
, &spa
, FTAG
);
1543 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1544 spa_close(spa
, FTAG
);
1547 suspended
= fnvlist_alloc();
1548 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1549 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1550 char fsname
[ZFS_MAX_DATASET_NAME_LEN
];
1551 char *snapname
= nvpair_name(pair
);
1555 atp
= strchr(snapname
, '@');
1557 error
= SET_ERROR(EINVAL
);
1560 (void) strlcpy(fsname
, snapname
, atp
- snapname
+ 1);
1562 error
= zil_suspend(fsname
, &cookie
);
1565 fnvlist_add_uint64(suspended
, fsname
,
1570 ddsa
.ddsa_snaps
= snaps
;
1571 ddsa
.ddsa_props
= props
;
1572 ddsa
.ddsa_errors
= errors
;
1573 ddsa
.ddsa_cr
= CRED();
1576 error
= dsl_sync_task(firstname
, dsl_dataset_snapshot_check
,
1577 dsl_dataset_snapshot_sync
, &ddsa
,
1578 fnvlist_num_pairs(snaps
) * 3, ZFS_SPACE_CHECK_NORMAL
);
1581 if (suspended
!= NULL
) {
1582 for (pair
= nvlist_next_nvpair(suspended
, NULL
); pair
!= NULL
;
1583 pair
= nvlist_next_nvpair(suspended
, pair
)) {
1584 zil_resume((void *)(uintptr_t)
1585 fnvpair_value_uint64(pair
));
1587 fnvlist_free(suspended
);
1593 typedef struct dsl_dataset_snapshot_tmp_arg
{
1594 const char *ddsta_fsname
;
1595 const char *ddsta_snapname
;
1596 minor_t ddsta_cleanup_minor
;
1597 const char *ddsta_htag
;
1598 } dsl_dataset_snapshot_tmp_arg_t
;
1601 dsl_dataset_snapshot_tmp_check(void *arg
, dmu_tx_t
*tx
)
1603 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1604 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1608 error
= dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
);
1612 /* NULL cred means no limit check for tmp snapshot */
1613 error
= dsl_dataset_snapshot_check_impl(ds
, ddsta
->ddsta_snapname
,
1614 tx
, B_FALSE
, 0, NULL
);
1616 dsl_dataset_rele(ds
, FTAG
);
1620 if (spa_version(dp
->dp_spa
) < SPA_VERSION_USERREFS
) {
1621 dsl_dataset_rele(ds
, FTAG
);
1622 return (SET_ERROR(ENOTSUP
));
1624 error
= dsl_dataset_user_hold_check_one(NULL
, ddsta
->ddsta_htag
,
1627 dsl_dataset_rele(ds
, FTAG
);
1631 dsl_dataset_rele(ds
, FTAG
);
1636 dsl_dataset_snapshot_tmp_sync(void *arg
, dmu_tx_t
*tx
)
1638 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1639 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1642 VERIFY0(dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
));
1644 dsl_dataset_snapshot_sync_impl(ds
, ddsta
->ddsta_snapname
, tx
);
1645 dsl_dataset_user_hold_sync_one(ds
->ds_prev
, ddsta
->ddsta_htag
,
1646 ddsta
->ddsta_cleanup_minor
, gethrestime_sec(), tx
);
1647 dsl_destroy_snapshot_sync_impl(ds
->ds_prev
, B_TRUE
, tx
);
1649 dsl_dataset_rele(ds
, FTAG
);
1653 dsl_dataset_snapshot_tmp(const char *fsname
, const char *snapname
,
1654 minor_t cleanup_minor
, const char *htag
)
1656 dsl_dataset_snapshot_tmp_arg_t ddsta
;
1659 boolean_t needsuspend
;
1662 ddsta
.ddsta_fsname
= fsname
;
1663 ddsta
.ddsta_snapname
= snapname
;
1664 ddsta
.ddsta_cleanup_minor
= cleanup_minor
;
1665 ddsta
.ddsta_htag
= htag
;
1667 error
= spa_open(fsname
, &spa
, FTAG
);
1670 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1671 spa_close(spa
, FTAG
);
1674 error
= zil_suspend(fsname
, &cookie
);
1679 error
= dsl_sync_task(fsname
, dsl_dataset_snapshot_tmp_check
,
1680 dsl_dataset_snapshot_tmp_sync
, &ddsta
, 3, ZFS_SPACE_CHECK_RESERVED
);
1688 dsl_dataset_sync(dsl_dataset_t
*ds
, zio_t
*zio
, dmu_tx_t
*tx
)
1690 ASSERT(dmu_tx_is_syncing(tx
));
1691 ASSERT(ds
->ds_objset
!= NULL
);
1692 ASSERT(dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0);
1695 * in case we had to change ds_fsid_guid when we opened it,
1698 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1699 dsl_dataset_phys(ds
)->ds_fsid_guid
= ds
->ds_fsid_guid
;
1701 if (ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
] != 0) {
1702 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1703 ds
->ds_object
, DS_FIELD_RESUME_OBJECT
, 8, 1,
1704 &ds
->ds_resume_object
[tx
->tx_txg
& TXG_MASK
], tx
));
1705 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1706 ds
->ds_object
, DS_FIELD_RESUME_OFFSET
, 8, 1,
1707 &ds
->ds_resume_offset
[tx
->tx_txg
& TXG_MASK
], tx
));
1708 VERIFY0(zap_update(tx
->tx_pool
->dp_meta_objset
,
1709 ds
->ds_object
, DS_FIELD_RESUME_BYTES
, 8, 1,
1710 &ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
], tx
));
1711 ds
->ds_resume_object
[tx
->tx_txg
& TXG_MASK
] = 0;
1712 ds
->ds_resume_offset
[tx
->tx_txg
& TXG_MASK
] = 0;
1713 ds
->ds_resume_bytes
[tx
->tx_txg
& TXG_MASK
] = 0;
1716 dmu_objset_sync(ds
->ds_objset
, zio
, tx
);
1718 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++) {
1719 if (ds
->ds_feature_activation_needed
[f
]) {
1720 if (ds
->ds_feature_inuse
[f
])
1722 dsl_dataset_activate_feature(ds
->ds_object
, f
, tx
);
1723 ds
->ds_feature_inuse
[f
] = B_TRUE
;
1729 deadlist_enqueue_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
1731 dsl_deadlist_t
*dl
= arg
;
1732 dsl_deadlist_insert(dl
, bp
, tx
);
1737 dsl_dataset_sync_done(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1739 objset_t
*os
= ds
->ds_objset
;
1741 bplist_iterate(&ds
->ds_pending_deadlist
,
1742 deadlist_enqueue_cb
, &ds
->ds_deadlist
, tx
);
1744 if (os
->os_synced_dnodes
!= NULL
) {
1745 multilist_destroy(os
->os_synced_dnodes
);
1746 os
->os_synced_dnodes
= NULL
;
1749 ASSERT(!dmu_objset_is_dirty(os
, dmu_tx_get_txg(tx
)));
1751 dmu_buf_rele(ds
->ds_dbuf
, ds
);
1755 get_clones_stat_impl(dsl_dataset_t
*ds
, nvlist_t
*val
)
1758 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1762 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1765 * There may be missing entries in ds_next_clones_obj
1766 * due to a bug in a previous version of the code.
1767 * Only trust it if it has the right number of entries.
1769 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
!= 0) {
1770 VERIFY0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1773 if (count
!= dsl_dataset_phys(ds
)->ds_num_children
- 1) {
1776 for (zap_cursor_init(&zc
, mos
,
1777 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
1778 zap_cursor_retrieve(&zc
, &za
) == 0;
1779 zap_cursor_advance(&zc
)) {
1780 dsl_dataset_t
*clone
;
1781 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
1782 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
1783 za
.za_first_integer
, FTAG
, &clone
));
1784 dsl_dir_name(clone
->ds_dir
, buf
);
1785 fnvlist_add_boolean(val
, buf
);
1786 dsl_dataset_rele(clone
, FTAG
);
1788 zap_cursor_fini(&zc
);
1793 get_clones_stat(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1795 nvlist_t
*propval
= fnvlist_alloc();
1799 * We use nvlist_alloc() instead of fnvlist_alloc() because the
1800 * latter would allocate the list with NV_UNIQUE_NAME flag.
1801 * As a result, every time a clone name is appended to the list
1802 * it would be (linearly) searched for for a duplicate name.
1803 * We already know that all clone names must be unique and we
1804 * want avoid the quadratic complexity of double-checking that
1805 * because we can have a large number of clones.
1807 VERIFY0(nvlist_alloc(&val
, 0, KM_SLEEP
));
1809 if (get_clones_stat_impl(ds
, val
) == 0) {
1810 fnvlist_add_nvlist(propval
, ZPROP_VALUE
, val
);
1811 fnvlist_add_nvlist(nv
, zfs_prop_to_name(ZFS_PROP_CLONES
),
1816 nvlist_free(propval
);
1820 * Returns a string that represents the receive resume stats token. It should
1821 * be freed with strfree().
1824 get_receive_resume_stats_impl(dsl_dataset_t
*ds
)
1826 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1828 if (dsl_dataset_has_resume_receive_state(ds
)) {
1831 uint8_t *compressed
;
1833 nvlist_t
*token_nv
= fnvlist_alloc();
1834 size_t packed_size
, compressed_size
;
1836 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1837 DS_FIELD_RESUME_FROMGUID
, sizeof (val
), 1, &val
) == 0) {
1838 fnvlist_add_uint64(token_nv
, "fromguid", val
);
1840 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1841 DS_FIELD_RESUME_OBJECT
, sizeof (val
), 1, &val
) == 0) {
1842 fnvlist_add_uint64(token_nv
, "object", val
);
1844 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1845 DS_FIELD_RESUME_OFFSET
, sizeof (val
), 1, &val
) == 0) {
1846 fnvlist_add_uint64(token_nv
, "offset", val
);
1848 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1849 DS_FIELD_RESUME_BYTES
, sizeof (val
), 1, &val
) == 0) {
1850 fnvlist_add_uint64(token_nv
, "bytes", val
);
1852 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1853 DS_FIELD_RESUME_TOGUID
, sizeof (val
), 1, &val
) == 0) {
1854 fnvlist_add_uint64(token_nv
, "toguid", val
);
1857 if (zap_lookup(dp
->dp_meta_objset
, ds
->ds_object
,
1858 DS_FIELD_RESUME_TONAME
, 1, sizeof (buf
), buf
) == 0) {
1859 fnvlist_add_string(token_nv
, "toname", buf
);
1861 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1862 DS_FIELD_RESUME_LARGEBLOCK
) == 0) {
1863 fnvlist_add_boolean(token_nv
, "largeblockok");
1865 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1866 DS_FIELD_RESUME_EMBEDOK
) == 0) {
1867 fnvlist_add_boolean(token_nv
, "embedok");
1869 if (zap_contains(dp
->dp_meta_objset
, ds
->ds_object
,
1870 DS_FIELD_RESUME_COMPRESSOK
) == 0) {
1871 fnvlist_add_boolean(token_nv
, "compressok");
1873 packed
= fnvlist_pack(token_nv
, &packed_size
);
1874 fnvlist_free(token_nv
);
1875 compressed
= kmem_alloc(packed_size
, KM_SLEEP
);
1877 compressed_size
= gzip_compress(packed
, compressed
,
1878 packed_size
, packed_size
, 6);
1881 fletcher_4_native(compressed
, compressed_size
, NULL
, &cksum
);
1883 str
= kmem_alloc(compressed_size
* 2 + 1, KM_SLEEP
);
1884 for (int i
= 0; i
< compressed_size
; i
++) {
1885 (void) sprintf(str
+ i
* 2, "%02x", compressed
[i
]);
1887 str
[compressed_size
* 2] = '\0';
1888 char *propval
= kmem_asprintf("%u-%llx-%llx-%s",
1889 ZFS_SEND_RESUME_TOKEN_VERSION
,
1890 (longlong_t
)cksum
.zc_word
[0],
1891 (longlong_t
)packed_size
, str
);
1892 kmem_free(packed
, packed_size
);
1893 kmem_free(str
, compressed_size
* 2 + 1);
1894 kmem_free(compressed
, packed_size
);
1897 return (strdup(""));
1901 * Returns a string that represents the receive resume stats token of the
1902 * dataset's child. It should be freed with strfree().
1905 get_child_receive_stats(dsl_dataset_t
*ds
)
1907 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
1908 dsl_dataset_t
*recv_ds
;
1909 dsl_dataset_name(ds
, recvname
);
1910 if (strlcat(recvname
, "/", sizeof (recvname
)) <
1911 sizeof (recvname
) &&
1912 strlcat(recvname
, recv_clone_name
, sizeof (recvname
)) <
1913 sizeof (recvname
) &&
1914 dsl_dataset_hold(ds
->ds_dir
->dd_pool
, recvname
, FTAG
,
1916 char *propval
= get_receive_resume_stats_impl(recv_ds
);
1917 dsl_dataset_rele(recv_ds
, FTAG
);
1920 return (strdup(""));
1924 get_receive_resume_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1926 char *propval
= get_receive_resume_stats_impl(ds
);
1927 if (strcmp(propval
, "") != 0) {
1928 dsl_prop_nvlist_add_string(nv
,
1929 ZFS_PROP_RECEIVE_RESUME_TOKEN
, propval
);
1931 char *childval
= get_child_receive_stats(ds
);
1932 if (strcmp(childval
, "") != 0) {
1933 dsl_prop_nvlist_add_string(nv
,
1934 ZFS_PROP_RECEIVE_RESUME_TOKEN
, childval
);
1942 dsl_get_refratio(dsl_dataset_t
*ds
)
1944 uint64_t ratio
= dsl_dataset_phys(ds
)->ds_compressed_bytes
== 0 ? 100 :
1945 (dsl_dataset_phys(ds
)->ds_uncompressed_bytes
* 100 /
1946 dsl_dataset_phys(ds
)->ds_compressed_bytes
);
1951 dsl_get_logicalreferenced(dsl_dataset_t
*ds
)
1953 return (dsl_dataset_phys(ds
)->ds_uncompressed_bytes
);
1957 dsl_get_compressratio(dsl_dataset_t
*ds
)
1959 if (ds
->ds_is_snapshot
) {
1960 return (dsl_get_refratio(ds
));
1962 dsl_dir_t
*dd
= ds
->ds_dir
;
1963 mutex_enter(&dd
->dd_lock
);
1964 uint64_t val
= dsl_dir_get_compressratio(dd
);
1965 mutex_exit(&dd
->dd_lock
);
1971 dsl_get_used(dsl_dataset_t
*ds
)
1973 if (ds
->ds_is_snapshot
) {
1974 return (dsl_dataset_phys(ds
)->ds_unique_bytes
);
1976 dsl_dir_t
*dd
= ds
->ds_dir
;
1977 mutex_enter(&dd
->dd_lock
);
1978 uint64_t val
= dsl_dir_get_used(dd
);
1979 mutex_exit(&dd
->dd_lock
);
1985 dsl_get_creation(dsl_dataset_t
*ds
)
1987 return (dsl_dataset_phys(ds
)->ds_creation_time
);
1991 dsl_get_creationtxg(dsl_dataset_t
*ds
)
1993 return (dsl_dataset_phys(ds
)->ds_creation_txg
);
1997 dsl_get_refquota(dsl_dataset_t
*ds
)
1999 return (ds
->ds_quota
);
2003 dsl_get_refreservation(dsl_dataset_t
*ds
)
2005 return (ds
->ds_reserved
);
2009 dsl_get_guid(dsl_dataset_t
*ds
)
2011 return (dsl_dataset_phys(ds
)->ds_guid
);
2015 dsl_get_unique(dsl_dataset_t
*ds
)
2017 return (dsl_dataset_phys(ds
)->ds_unique_bytes
);
2021 dsl_get_objsetid(dsl_dataset_t
*ds
)
2023 return (ds
->ds_object
);
2027 dsl_get_userrefs(dsl_dataset_t
*ds
)
2029 return (ds
->ds_userrefs
);
2033 dsl_get_defer_destroy(dsl_dataset_t
*ds
)
2035 return (DS_IS_DEFER_DESTROY(ds
) ? 1 : 0);
2039 dsl_get_referenced(dsl_dataset_t
*ds
)
2041 return (dsl_dataset_phys(ds
)->ds_referenced_bytes
);
2045 dsl_get_numclones(dsl_dataset_t
*ds
)
2047 ASSERT(ds
->ds_is_snapshot
);
2048 return (dsl_dataset_phys(ds
)->ds_num_children
- 1);
2052 dsl_get_inconsistent(dsl_dataset_t
*ds
)
2054 return ((dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_INCONSISTENT
) ?
2059 dsl_get_available(dsl_dataset_t
*ds
)
2061 uint64_t refdbytes
= dsl_get_referenced(ds
);
2062 uint64_t availbytes
= dsl_dir_space_available(ds
->ds_dir
,
2064 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
) {
2066 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
;
2068 if (ds
->ds_quota
!= 0) {
2070 * Adjust available bytes according to refquota
2072 if (refdbytes
< ds
->ds_quota
) {
2073 availbytes
= MIN(availbytes
,
2074 ds
->ds_quota
- refdbytes
);
2079 return (availbytes
);
2083 dsl_get_written(dsl_dataset_t
*ds
, uint64_t *written
)
2085 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2086 dsl_dataset_t
*prev
;
2087 int err
= dsl_dataset_hold_obj(dp
,
2088 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
2090 uint64_t comp
, uncomp
;
2091 err
= dsl_dataset_space_written(prev
, ds
, written
,
2093 dsl_dataset_rele(prev
, FTAG
);
2099 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN.
2102 dsl_get_prev_snap(dsl_dataset_t
*ds
, char *snap
)
2104 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2105 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
!= dp
->dp_origin_snap
) {
2106 dsl_dataset_name(ds
->ds_prev
, snap
);
2114 * Returns the mountpoint property and source for the given dataset in the value
2115 * and source buffers. The value buffer must be at least as large as MAXPATHLEN
2116 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN.
2117 * Returns 0 on success and an error on failure.
2120 dsl_get_mountpoint(dsl_dataset_t
*ds
, const char *dsname
, char *value
,
2124 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2126 /* Retrieve the mountpoint value stored in the zap opbject */
2127 error
= dsl_prop_get_ds(ds
, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT
), 1,
2128 ZAP_MAXVALUELEN
, value
, source
);
2133 /* Process the dsname and source to find the full mountpoint string */
2134 if (value
[0] == '/') {
2135 char *buf
= kmem_alloc(ZAP_MAXVALUELEN
, KM_SLEEP
);
2137 const char *relpath
;
2140 * If we inherit the mountpoint, even from a dataset
2141 * with a received value, the source will be the path of
2142 * the dataset we inherit from. If source is
2143 * ZPROP_SOURCE_VAL_RECVD, the received value is not
2146 if (strcmp(source
, ZPROP_SOURCE_VAL_RECVD
) == 0) {
2149 ASSERT0(strncmp(dsname
, source
, strlen(source
)));
2150 relpath
= dsname
+ strlen(source
);
2151 if (relpath
[0] == '/')
2155 spa_altroot(dp
->dp_spa
, root
, ZAP_MAXVALUELEN
);
2158 * Special case an alternate root of '/'. This will
2159 * avoid having multiple leading slashes in the
2162 if (strcmp(root
, "/") == 0)
2166 * If the mountpoint is '/' then skip over this
2167 * if we are obtaining either an alternate root or
2168 * an inherited mountpoint.
2171 if (value
[1] == '\0' && (root
[0] != '\0' ||
2172 relpath
[0] != '\0'))
2175 if (relpath
[0] == '\0') {
2176 (void) snprintf(value
, ZAP_MAXVALUELEN
, "%s%s",
2179 (void) snprintf(value
, ZAP_MAXVALUELEN
, "%s%s%s%s",
2180 root
, mnt
, relpath
[0] == '@' ? "" : "/",
2183 kmem_free(buf
, ZAP_MAXVALUELEN
);
2185 /* 'legacy' or 'none' */
2186 (void) snprintf(value
, ZAP_MAXVALUELEN
, "%s", value
);
2192 dsl_dataset_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
2194 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2196 ASSERT(dsl_pool_config_held(dp
));
2198 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRATIO
,
2199 dsl_get_refratio(ds
));
2200 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALREFERENCED
,
2201 dsl_get_logicalreferenced(ds
));
2202 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
,
2203 dsl_get_compressratio(ds
));
2204 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
2207 if (ds
->ds_is_snapshot
) {
2208 get_clones_stat(ds
, nv
);
2210 char buf
[ZFS_MAX_DATASET_NAME_LEN
];
2211 if (dsl_get_prev_snap(ds
, buf
) == 0)
2212 dsl_prop_nvlist_add_string(nv
, ZFS_PROP_PREV_SNAP
,
2214 dsl_dir_stats(ds
->ds_dir
, nv
);
2217 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_AVAILABLE
,
2218 dsl_get_available(ds
));
2219 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFERENCED
,
2220 dsl_get_referenced(ds
));
2221 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATION
,
2222 dsl_get_creation(ds
));
2223 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATETXG
,
2224 dsl_get_creationtxg(ds
));
2225 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFQUOTA
,
2226 dsl_get_refquota(ds
));
2227 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRESERVATION
,
2228 dsl_get_refreservation(ds
));
2229 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_GUID
,
2231 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_UNIQUE
,
2232 dsl_get_unique(ds
));
2233 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_OBJSETID
,
2234 dsl_get_objsetid(ds
));
2235 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERREFS
,
2236 dsl_get_userrefs(ds
));
2237 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_DEFER_DESTROY
,
2238 dsl_get_defer_destroy(ds
));
2240 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
2242 if (dsl_get_written(ds
, &written
) == 0) {
2243 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_WRITTEN
,
2248 if (!dsl_dataset_is_snapshot(ds
)) {
2250 * A failed "newfs" (e.g. full) resumable receive leaves
2251 * the stats set on this dataset. Check here for the prop.
2253 get_receive_resume_stats(ds
, nv
);
2256 * A failed incremental resumable receive leaves the
2257 * stats set on our child named "%recv". Check the child
2260 /* 6 extra bytes for /%recv */
2261 char recvname
[ZFS_MAX_DATASET_NAME_LEN
+ 6];
2262 dsl_dataset_t
*recv_ds
;
2263 dsl_dataset_name(ds
, recvname
);
2264 if (strlcat(recvname
, "/", sizeof (recvname
)) <
2265 sizeof (recvname
) &&
2266 strlcat(recvname
, recv_clone_name
, sizeof (recvname
)) <
2267 sizeof (recvname
) &&
2268 dsl_dataset_hold(dp
, recvname
, FTAG
, &recv_ds
) == 0) {
2269 get_receive_resume_stats(recv_ds
, nv
);
2270 dsl_dataset_rele(recv_ds
, FTAG
);
2276 dsl_dataset_fast_stat(dsl_dataset_t
*ds
, dmu_objset_stats_t
*stat
)
2278 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2279 ASSERT(dsl_pool_config_held(dp
));
2281 stat
->dds_creation_txg
= dsl_get_creationtxg(ds
);
2282 stat
->dds_inconsistent
= dsl_get_inconsistent(ds
);
2283 stat
->dds_guid
= dsl_get_guid(ds
);
2284 stat
->dds_origin
[0] = '\0';
2285 if (ds
->ds_is_snapshot
) {
2286 stat
->dds_is_snapshot
= B_TRUE
;
2287 stat
->dds_num_clones
= dsl_get_numclones(ds
);
2289 stat
->dds_is_snapshot
= B_FALSE
;
2290 stat
->dds_num_clones
= 0;
2292 if (dsl_dir_is_clone(ds
->ds_dir
)) {
2293 dsl_dir_get_origin(ds
->ds_dir
, stat
->dds_origin
);
2299 dsl_dataset_fsid_guid(dsl_dataset_t
*ds
)
2301 return (ds
->ds_fsid_guid
);
2305 dsl_dataset_space(dsl_dataset_t
*ds
,
2306 uint64_t *refdbytesp
, uint64_t *availbytesp
,
2307 uint64_t *usedobjsp
, uint64_t *availobjsp
)
2309 *refdbytesp
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
2310 *availbytesp
= dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
);
2311 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
)
2313 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
;
2314 if (ds
->ds_quota
!= 0) {
2316 * Adjust available bytes according to refquota
2318 if (*refdbytesp
< ds
->ds_quota
)
2319 *availbytesp
= MIN(*availbytesp
,
2320 ds
->ds_quota
- *refdbytesp
);
2324 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
2325 *usedobjsp
= BP_GET_FILL(&dsl_dataset_phys(ds
)->ds_bp
);
2326 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2327 *availobjsp
= DN_MAX_OBJECT
- *usedobjsp
;
2331 dsl_dataset_modified_since_snap(dsl_dataset_t
*ds
, dsl_dataset_t
*snap
)
2333 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
2336 ASSERT(dsl_pool_config_held(dp
));
2339 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
2340 birth
= dsl_dataset_get_blkptr(ds
)->blk_birth
;
2341 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
2342 if (birth
> dsl_dataset_phys(snap
)->ds_creation_txg
) {
2343 objset_t
*os
, *os_snap
;
2345 * It may be that only the ZIL differs, because it was
2346 * reset in the head. Don't count that as being
2349 if (dmu_objset_from_ds(ds
, &os
) != 0)
2351 if (dmu_objset_from_ds(snap
, &os_snap
) != 0)
2353 return (bcmp(&os
->os_phys
->os_meta_dnode
,
2354 &os_snap
->os_phys
->os_meta_dnode
,
2355 sizeof (os
->os_phys
->os_meta_dnode
)) != 0);
2360 typedef struct dsl_dataset_rename_snapshot_arg
{
2361 const char *ddrsa_fsname
;
2362 const char *ddrsa_oldsnapname
;
2363 const char *ddrsa_newsnapname
;
2364 boolean_t ddrsa_recursive
;
2366 } dsl_dataset_rename_snapshot_arg_t
;
2370 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t
*dp
,
2371 dsl_dataset_t
*hds
, void *arg
)
2373 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2377 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
2379 /* ignore nonexistent snapshots */
2380 return (error
== ENOENT
? 0 : error
);
2383 /* new name should not exist */
2384 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_newsnapname
, &val
);
2386 error
= SET_ERROR(EEXIST
);
2387 else if (error
== ENOENT
)
2390 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2391 if (dsl_dir_namelen(hds
->ds_dir
) + 1 +
2392 strlen(ddrsa
->ddrsa_newsnapname
) >= ZFS_MAX_DATASET_NAME_LEN
)
2393 error
= SET_ERROR(ENAMETOOLONG
);
2399 dsl_dataset_rename_snapshot_check(void *arg
, dmu_tx_t
*tx
)
2401 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2402 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2406 error
= dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
);
2410 if (ddrsa
->ddrsa_recursive
) {
2411 error
= dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
2412 dsl_dataset_rename_snapshot_check_impl
, ddrsa
,
2415 error
= dsl_dataset_rename_snapshot_check_impl(dp
, hds
, ddrsa
);
2417 dsl_dataset_rele(hds
, FTAG
);
2422 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t
*dp
,
2423 dsl_dataset_t
*hds
, void *arg
)
2425 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2428 dmu_tx_t
*tx
= ddrsa
->ddrsa_tx
;
2431 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
2432 ASSERT(error
== 0 || error
== ENOENT
);
2433 if (error
== ENOENT
) {
2434 /* ignore nonexistent snapshots */
2438 VERIFY0(dsl_dataset_hold_obj(dp
, val
, FTAG
, &ds
));
2440 /* log before we change the name */
2441 spa_history_log_internal_ds(ds
, "rename", tx
,
2442 "-> @%s", ddrsa
->ddrsa_newsnapname
);
2444 VERIFY0(dsl_dataset_snap_remove(hds
, ddrsa
->ddrsa_oldsnapname
, tx
,
2446 mutex_enter(&ds
->ds_lock
);
2447 (void) strcpy(ds
->ds_snapname
, ddrsa
->ddrsa_newsnapname
);
2448 mutex_exit(&ds
->ds_lock
);
2449 VERIFY0(zap_add(dp
->dp_meta_objset
,
2450 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
,
2451 ds
->ds_snapname
, 8, 1, &ds
->ds_object
, tx
));
2453 dsl_dataset_rele(ds
, FTAG
);
2458 dsl_dataset_rename_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
2460 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
2461 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2464 VERIFY0(dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
));
2465 ddrsa
->ddrsa_tx
= tx
;
2466 if (ddrsa
->ddrsa_recursive
) {
2467 VERIFY0(dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
2468 dsl_dataset_rename_snapshot_sync_impl
, ddrsa
,
2471 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp
, hds
, ddrsa
));
2473 dsl_dataset_rele(hds
, FTAG
);
2477 dsl_dataset_rename_snapshot(const char *fsname
,
2478 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
)
2480 dsl_dataset_rename_snapshot_arg_t ddrsa
;
2482 ddrsa
.ddrsa_fsname
= fsname
;
2483 ddrsa
.ddrsa_oldsnapname
= oldsnapname
;
2484 ddrsa
.ddrsa_newsnapname
= newsnapname
;
2485 ddrsa
.ddrsa_recursive
= recursive
;
2487 return (dsl_sync_task(fsname
, dsl_dataset_rename_snapshot_check
,
2488 dsl_dataset_rename_snapshot_sync
, &ddrsa
,
2489 1, ZFS_SPACE_CHECK_RESERVED
));
2493 * If we're doing an ownership handoff, we need to make sure that there is
2494 * only one long hold on the dataset. We're not allowed to change anything here
2495 * so we don't permanently release the long hold or regular hold here. We want
2496 * to do this only when syncing to avoid the dataset unexpectedly going away
2497 * when we release the long hold.
2500 dsl_dataset_handoff_check(dsl_dataset_t
*ds
, void *owner
, dmu_tx_t
*tx
)
2504 if (!dmu_tx_is_syncing(tx
))
2507 if (owner
!= NULL
) {
2508 VERIFY3P(ds
->ds_owner
, ==, owner
);
2509 dsl_dataset_long_rele(ds
, owner
);
2512 held
= dsl_dataset_long_held(ds
);
2515 dsl_dataset_long_hold(ds
, owner
);
2518 return (SET_ERROR(EBUSY
));
2524 dsl_dataset_rollback_check(void *arg
, dmu_tx_t
*tx
)
2526 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2527 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2529 int64_t unused_refres_delta
;
2532 error
= dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
);
2536 /* must not be a snapshot */
2537 if (ds
->ds_is_snapshot
) {
2538 dsl_dataset_rele(ds
, FTAG
);
2539 return (SET_ERROR(EINVAL
));
2542 /* must have a most recent snapshot */
2543 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
< TXG_INITIAL
) {
2544 dsl_dataset_rele(ds
, FTAG
);
2545 return (SET_ERROR(ESRCH
));
2549 * No rollback to a snapshot created in the current txg, because
2550 * the rollback may dirty the dataset and create blocks that are
2551 * not reachable from the rootbp while having a birth txg that
2552 * falls into the snapshot's range.
2554 if (dmu_tx_is_syncing(tx
) &&
2555 dsl_dataset_phys(ds
)->ds_prev_snap_txg
>= tx
->tx_txg
) {
2556 dsl_dataset_rele(ds
, FTAG
);
2557 return (SET_ERROR(EAGAIN
));
2561 * If the expected target snapshot is specified, then check that
2562 * the latest snapshot is it.
2564 if (ddra
->ddra_tosnap
!= NULL
) {
2565 dsl_dataset_t
*snapds
;
2567 /* Check if the target snapshot exists at all. */
2568 error
= dsl_dataset_hold(dp
, ddra
->ddra_tosnap
, FTAG
, &snapds
);
2571 * ESRCH is used to signal that the target snapshot does
2572 * not exist, while ENOENT is used to report that
2573 * the rolled back dataset does not exist.
2574 * ESRCH is also used to cover other cases where the
2575 * target snapshot is not related to the dataset being
2576 * rolled back such as being in a different pool.
2578 if (error
== ENOENT
|| error
== EXDEV
)
2579 error
= SET_ERROR(ESRCH
);
2580 dsl_dataset_rele(ds
, FTAG
);
2583 ASSERT(snapds
->ds_is_snapshot
);
2585 /* Check if the snapshot is the latest snapshot indeed. */
2586 if (snapds
!= ds
->ds_prev
) {
2588 * Distinguish between the case where the only problem
2589 * is intervening snapshots (EEXIST) vs the snapshot
2590 * not being a valid target for rollback (ESRCH).
2592 if (snapds
->ds_dir
== ds
->ds_dir
||
2593 (dsl_dir_is_clone(ds
->ds_dir
) &&
2594 dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
==
2595 snapds
->ds_object
)) {
2596 error
= SET_ERROR(EEXIST
);
2598 error
= SET_ERROR(ESRCH
);
2600 dsl_dataset_rele(snapds
, FTAG
);
2601 dsl_dataset_rele(ds
, FTAG
);
2604 dsl_dataset_rele(snapds
, FTAG
);
2607 /* must not have any bookmarks after the most recent snapshot */
2608 nvlist_t
*proprequest
= fnvlist_alloc();
2609 fnvlist_add_boolean(proprequest
, zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2610 nvlist_t
*bookmarks
= fnvlist_alloc();
2611 error
= dsl_get_bookmarks_impl(ds
, proprequest
, bookmarks
);
2612 fnvlist_free(proprequest
);
2614 dsl_dataset_rele(ds
, FTAG
);
2617 for (nvpair_t
*pair
= nvlist_next_nvpair(bookmarks
, NULL
);
2618 pair
!= NULL
; pair
= nvlist_next_nvpair(bookmarks
, pair
)) {
2620 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair
),
2621 zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2622 uint64_t createtxg
= fnvlist_lookup_uint64(valuenv
, "value");
2623 if (createtxg
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
2624 fnvlist_free(bookmarks
);
2625 dsl_dataset_rele(ds
, FTAG
);
2626 return (SET_ERROR(EEXIST
));
2629 fnvlist_free(bookmarks
);
2631 error
= dsl_dataset_handoff_check(ds
, ddra
->ddra_owner
, tx
);
2633 dsl_dataset_rele(ds
, FTAG
);
2638 * Check if the snap we are rolling back to uses more than
2641 if (ds
->ds_quota
!= 0 &&
2642 dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
> ds
->ds_quota
) {
2643 dsl_dataset_rele(ds
, FTAG
);
2644 return (SET_ERROR(EDQUOT
));
2648 * When we do the clone swap, we will temporarily use more space
2649 * due to the refreservation (the head will no longer have any
2650 * unique space, so the entire amount of the refreservation will need
2651 * to be free). We will immediately destroy the clone, freeing
2652 * this space, but the freeing happens over many txg's.
2654 unused_refres_delta
= (int64_t)MIN(ds
->ds_reserved
,
2655 dsl_dataset_phys(ds
)->ds_unique_bytes
);
2657 if (unused_refres_delta
> 0 &&
2658 unused_refres_delta
>
2659 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
)) {
2660 dsl_dataset_rele(ds
, FTAG
);
2661 return (SET_ERROR(ENOSPC
));
2664 dsl_dataset_rele(ds
, FTAG
);
2669 dsl_dataset_rollback_sync(void *arg
, dmu_tx_t
*tx
)
2671 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2672 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2673 dsl_dataset_t
*ds
, *clone
;
2675 char namebuf
[ZFS_MAX_DATASET_NAME_LEN
];
2677 VERIFY0(dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
));
2679 dsl_dataset_name(ds
->ds_prev
, namebuf
);
2680 fnvlist_add_string(ddra
->ddra_result
, "target", namebuf
);
2682 cloneobj
= dsl_dataset_create_sync(ds
->ds_dir
, "%rollback",
2683 ds
->ds_prev
, DS_CREATE_FLAG_NODIRTY
, kcred
, tx
);
2685 VERIFY0(dsl_dataset_hold_obj(dp
, cloneobj
, FTAG
, &clone
));
2687 dsl_dataset_clone_swap_sync_impl(clone
, ds
, tx
);
2688 dsl_dataset_zero_zil(ds
, tx
);
2690 dsl_destroy_head_sync_impl(clone
, tx
);
2692 dsl_dataset_rele(clone
, FTAG
);
2693 dsl_dataset_rele(ds
, FTAG
);
2697 * Rolls back the given filesystem or volume to the most recent snapshot.
2698 * The name of the most recent snapshot will be returned under key "target"
2699 * in the result nvlist.
2702 * - The existing dataset MUST be owned by the specified owner at entry
2703 * - Upon return, dataset will still be held by the same owner, whether we
2706 * This mode is required any time the existing filesystem is mounted. See
2707 * notes above zfs_suspend_fs() for further details.
2710 dsl_dataset_rollback(const char *fsname
, const char *tosnap
, void *owner
,
2713 dsl_dataset_rollback_arg_t ddra
;
2715 ddra
.ddra_fsname
= fsname
;
2716 ddra
.ddra_tosnap
= tosnap
;
2717 ddra
.ddra_owner
= owner
;
2718 ddra
.ddra_result
= result
;
2720 return (dsl_sync_task(fsname
, dsl_dataset_rollback_check
,
2721 dsl_dataset_rollback_sync
, &ddra
,
2722 1, ZFS_SPACE_CHECK_RESERVED
));
2725 struct promotenode
{
2730 static int snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
);
2731 static int promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
,
2733 static void promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
);
2736 dsl_dataset_promote_check(void *arg
, dmu_tx_t
*tx
)
2738 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2739 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2741 struct promotenode
*snap
;
2742 dsl_dataset_t
*origin_ds
;
2746 size_t max_snap_len
;
2747 boolean_t conflicting_snaps
;
2749 err
= promote_hold(ddpa
, dp
, FTAG
);
2753 hds
= ddpa
->ddpa_clone
;
2754 snap
= list_head(&ddpa
->shared_snaps
);
2755 origin_ds
= snap
->ds
;
2756 max_snap_len
= MAXNAMELEN
- strlen(ddpa
->ddpa_clonename
) - 1;
2758 snap
= list_head(&ddpa
->origin_snaps
);
2760 if (dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
) {
2761 promote_rele(ddpa
, FTAG
);
2762 return (SET_ERROR(EXDEV
));
2766 * Compute and check the amount of space to transfer. Since this is
2767 * so expensive, don't do the preliminary check.
2769 if (!dmu_tx_is_syncing(tx
)) {
2770 promote_rele(ddpa
, FTAG
);
2774 /* compute origin's new unique space */
2775 snap
= list_tail(&ddpa
->clone_snaps
);
2776 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2777 origin_ds
->ds_object
);
2778 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2779 dsl_dataset_phys(origin_ds
)->ds_prev_snap_txg
, UINT64_MAX
,
2780 &ddpa
->unique
, &unused
, &unused
);
2783 * Walk the snapshots that we are moving
2785 * Compute space to transfer. Consider the incremental changes
2786 * to used by each snapshot:
2787 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2788 * So each snapshot gave birth to:
2789 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2790 * So a sequence would look like:
2791 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2792 * Which simplifies to:
2793 * uN + kN + kN-1 + ... + k1 + k0
2794 * Note however, if we stop before we reach the ORIGIN we get:
2795 * uN + kN + kN-1 + ... + kM - uM-1
2797 conflicting_snaps
= B_FALSE
;
2799 ddpa
->used
= dsl_dataset_phys(origin_ds
)->ds_referenced_bytes
;
2800 ddpa
->comp
= dsl_dataset_phys(origin_ds
)->ds_compressed_bytes
;
2801 ddpa
->uncomp
= dsl_dataset_phys(origin_ds
)->ds_uncompressed_bytes
;
2802 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2803 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2804 uint64_t val
, dlused
, dlcomp
, dluncomp
;
2805 dsl_dataset_t
*ds
= snap
->ds
;
2810 * If there are long holds, we won't be able to evict
2813 if (dsl_dataset_long_held(ds
)) {
2814 err
= SET_ERROR(EBUSY
);
2818 /* Check that the snapshot name does not conflict */
2819 VERIFY0(dsl_dataset_get_snapname(ds
));
2820 if (strlen(ds
->ds_snapname
) >= max_snap_len
) {
2821 err
= SET_ERROR(ENAMETOOLONG
);
2824 err
= dsl_dataset_snap_lookup(hds
, ds
->ds_snapname
, &val
);
2826 fnvlist_add_boolean(ddpa
->err_ds
,
2827 snap
->ds
->ds_snapname
);
2828 conflicting_snaps
= B_TRUE
;
2829 } else if (err
!= ENOENT
) {
2833 /* The very first snapshot does not have a deadlist */
2834 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
== 0)
2837 dsl_deadlist_space(&ds
->ds_deadlist
,
2838 &dlused
, &dlcomp
, &dluncomp
);
2839 ddpa
->used
+= dlused
;
2840 ddpa
->comp
+= dlcomp
;
2841 ddpa
->uncomp
+= dluncomp
;
2845 * In order to return the full list of conflicting snapshots, we check
2846 * whether there was a conflict after traversing all of them.
2848 if (conflicting_snaps
) {
2849 err
= SET_ERROR(EEXIST
);
2854 * If we are a clone of a clone then we never reached ORIGIN,
2855 * so we need to subtract out the clone origin's used space.
2857 if (ddpa
->origin_origin
) {
2859 dsl_dataset_phys(ddpa
->origin_origin
)->ds_referenced_bytes
;
2861 dsl_dataset_phys(ddpa
->origin_origin
)->ds_compressed_bytes
;
2863 dsl_dataset_phys(ddpa
->origin_origin
)->
2864 ds_uncompressed_bytes
;
2867 /* Check that there is enough space and limit headroom here */
2868 err
= dsl_dir_transfer_possible(origin_ds
->ds_dir
, hds
->ds_dir
,
2869 0, ss_mv_cnt
, ddpa
->used
, ddpa
->cr
);
2874 * Compute the amounts of space that will be used by snapshots
2875 * after the promotion (for both origin and clone). For each,
2876 * it is the amount of space that will be on all of their
2877 * deadlists (that was not born before their new origin).
2879 if (dsl_dir_phys(hds
->ds_dir
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2883 * Note, typically this will not be a clone of a clone,
2884 * so dd_origin_txg will be < TXG_INITIAL, so
2885 * these snaplist_space() -> dsl_deadlist_space_range()
2886 * calls will be fast because they do not have to
2887 * iterate over all bps.
2889 snap
= list_head(&ddpa
->origin_snaps
);
2890 err
= snaplist_space(&ddpa
->shared_snaps
,
2891 snap
->ds
->ds_dir
->dd_origin_txg
, &ddpa
->cloneusedsnap
);
2895 err
= snaplist_space(&ddpa
->clone_snaps
,
2896 snap
->ds
->ds_dir
->dd_origin_txg
, &space
);
2899 ddpa
->cloneusedsnap
+= space
;
2901 if (dsl_dir_phys(origin_ds
->ds_dir
)->dd_flags
&
2902 DD_FLAG_USED_BREAKDOWN
) {
2903 err
= snaplist_space(&ddpa
->origin_snaps
,
2904 dsl_dataset_phys(origin_ds
)->ds_creation_txg
,
2905 &ddpa
->originusedsnap
);
2911 promote_rele(ddpa
, FTAG
);
2916 dsl_dataset_promote_sync(void *arg
, dmu_tx_t
*tx
)
2918 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2919 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2921 struct promotenode
*snap
;
2922 dsl_dataset_t
*origin_ds
;
2923 dsl_dataset_t
*origin_head
;
2925 dsl_dir_t
*odd
= NULL
;
2926 uint64_t oldnext_obj
;
2929 VERIFY0(promote_hold(ddpa
, dp
, FTAG
));
2930 hds
= ddpa
->ddpa_clone
;
2932 ASSERT0(dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
);
2934 snap
= list_head(&ddpa
->shared_snaps
);
2935 origin_ds
= snap
->ds
;
2938 snap
= list_head(&ddpa
->origin_snaps
);
2939 origin_head
= snap
->ds
;
2942 * We need to explicitly open odd, since origin_ds's dd will be
2945 VERIFY0(dsl_dir_hold_obj(dp
, origin_ds
->ds_dir
->dd_object
,
2948 /* change origin's next snap */
2949 dmu_buf_will_dirty(origin_ds
->ds_dbuf
, tx
);
2950 oldnext_obj
= dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
;
2951 snap
= list_tail(&ddpa
->clone_snaps
);
2952 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2953 origin_ds
->ds_object
);
2954 dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
= snap
->ds
->ds_object
;
2956 /* change the origin's next clone */
2957 if (dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
) {
2958 dsl_dataset_remove_from_next_clones(origin_ds
,
2959 snap
->ds
->ds_object
, tx
);
2960 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2961 dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
,
2966 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
2967 ASSERT3U(dsl_dir_phys(dd
)->dd_origin_obj
, ==, origin_ds
->ds_object
);
2968 dsl_dir_phys(dd
)->dd_origin_obj
= dsl_dir_phys(odd
)->dd_origin_obj
;
2969 dd
->dd_origin_txg
= origin_head
->ds_dir
->dd_origin_txg
;
2970 dmu_buf_will_dirty(odd
->dd_dbuf
, tx
);
2971 dsl_dir_phys(odd
)->dd_origin_obj
= origin_ds
->ds_object
;
2972 origin_head
->ds_dir
->dd_origin_txg
=
2973 dsl_dataset_phys(origin_ds
)->ds_creation_txg
;
2975 /* change dd_clone entries */
2976 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2977 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2978 dsl_dir_phys(odd
)->dd_clones
, hds
->ds_object
, tx
));
2979 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2980 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2981 hds
->ds_object
, tx
));
2983 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2984 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2985 origin_head
->ds_object
, tx
));
2986 if (dsl_dir_phys(dd
)->dd_clones
== 0) {
2987 dsl_dir_phys(dd
)->dd_clones
=
2988 zap_create(dp
->dp_meta_objset
, DMU_OT_DSL_CLONES
,
2989 DMU_OT_NONE
, 0, tx
);
2991 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2992 dsl_dir_phys(dd
)->dd_clones
, origin_head
->ds_object
, tx
));
2995 /* move snapshots to this dir */
2996 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2997 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2998 dsl_dataset_t
*ds
= snap
->ds
;
3001 * Property callbacks are registered to a particular
3002 * dsl_dir. Since ours is changing, evict the objset
3003 * so that they will be unregistered from the old dsl_dir.
3005 if (ds
->ds_objset
) {
3006 dmu_objset_evict(ds
->ds_objset
);
3007 ds
->ds_objset
= NULL
;
3010 /* move snap name entry */
3011 VERIFY0(dsl_dataset_get_snapname(ds
));
3012 VERIFY0(dsl_dataset_snap_remove(origin_head
,
3013 ds
->ds_snapname
, tx
, B_TRUE
));
3014 VERIFY0(zap_add(dp
->dp_meta_objset
,
3015 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
, ds
->ds_snapname
,
3016 8, 1, &ds
->ds_object
, tx
));
3017 dsl_fs_ss_count_adjust(hds
->ds_dir
, 1,
3018 DD_FIELD_SNAPSHOT_COUNT
, tx
);
3020 /* change containing dsl_dir */
3021 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3022 ASSERT3U(dsl_dataset_phys(ds
)->ds_dir_obj
, ==, odd
->dd_object
);
3023 dsl_dataset_phys(ds
)->ds_dir_obj
= dd
->dd_object
;
3024 ASSERT3P(ds
->ds_dir
, ==, odd
);
3025 dsl_dir_rele(ds
->ds_dir
, ds
);
3026 VERIFY0(dsl_dir_hold_obj(dp
, dd
->dd_object
,
3027 NULL
, ds
, &ds
->ds_dir
));
3029 /* move any clone references */
3030 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
&&
3031 spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
3035 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
3036 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
3037 zap_cursor_retrieve(&zc
, &za
) == 0;
3038 zap_cursor_advance(&zc
)) {
3039 dsl_dataset_t
*cnds
;
3042 if (za
.za_first_integer
== oldnext_obj
) {
3044 * We've already moved the
3045 * origin's reference.
3050 VERIFY0(dsl_dataset_hold_obj(dp
,
3051 za
.za_first_integer
, FTAG
, &cnds
));
3052 o
= dsl_dir_phys(cnds
->ds_dir
)->
3053 dd_head_dataset_obj
;
3055 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
3056 dsl_dir_phys(odd
)->dd_clones
, o
, tx
));
3057 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
3058 dsl_dir_phys(dd
)->dd_clones
, o
, tx
));
3059 dsl_dataset_rele(cnds
, FTAG
);
3061 zap_cursor_fini(&zc
);
3064 ASSERT(!dsl_prop_hascb(ds
));
3068 * Change space accounting.
3069 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
3070 * both be valid, or both be 0 (resulting in delta == 0). This
3071 * is true for each of {clone,origin} independently.
3074 delta
= ddpa
->cloneusedsnap
-
3075 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_SNAP
];
3076 ASSERT3S(delta
, >=, 0);
3077 ASSERT3U(ddpa
->used
, >=, delta
);
3078 dsl_dir_diduse_space(dd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
3079 dsl_dir_diduse_space(dd
, DD_USED_HEAD
,
3080 ddpa
->used
- delta
, ddpa
->comp
, ddpa
->uncomp
, tx
);
3082 delta
= ddpa
->originusedsnap
-
3083 dsl_dir_phys(odd
)->dd_used_breakdown
[DD_USED_SNAP
];
3084 ASSERT3S(delta
, <=, 0);
3085 ASSERT3U(ddpa
->used
, >=, -delta
);
3086 dsl_dir_diduse_space(odd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
3087 dsl_dir_diduse_space(odd
, DD_USED_HEAD
,
3088 -ddpa
->used
- delta
, -ddpa
->comp
, -ddpa
->uncomp
, tx
);
3090 dsl_dataset_phys(origin_ds
)->ds_unique_bytes
= ddpa
->unique
;
3092 /* log history record */
3093 spa_history_log_internal_ds(hds
, "promote", tx
, "");
3095 dsl_dir_rele(odd
, FTAG
);
3096 promote_rele(ddpa
, FTAG
);
3100 * Make a list of dsl_dataset_t's for the snapshots between first_obj
3101 * (exclusive) and last_obj (inclusive). The list will be in reverse
3102 * order (last_obj will be the list_head()). If first_obj == 0, do all
3103 * snapshots back to this dataset's origin.
3106 snaplist_make(dsl_pool_t
*dp
,
3107 uint64_t first_obj
, uint64_t last_obj
, list_t
*l
, void *tag
)
3109 uint64_t obj
= last_obj
;
3111 list_create(l
, sizeof (struct promotenode
),
3112 offsetof(struct promotenode
, link
));
3114 while (obj
!= first_obj
) {
3116 struct promotenode
*snap
;
3119 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
3120 ASSERT(err
!= ENOENT
);
3125 first_obj
= dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
;
3127 snap
= kmem_alloc(sizeof (*snap
), KM_SLEEP
);
3129 list_insert_tail(l
, snap
);
3130 obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
3137 snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
)
3139 struct promotenode
*snap
;
3142 for (snap
= list_head(l
); snap
; snap
= list_next(l
, snap
)) {
3143 uint64_t used
, comp
, uncomp
;
3144 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
3145 mintxg
, UINT64_MAX
, &used
, &comp
, &uncomp
);
3152 snaplist_destroy(list_t
*l
, void *tag
)
3154 struct promotenode
*snap
;
3156 if (l
== NULL
|| !list_link_active(&l
->list_head
))
3159 while ((snap
= list_tail(l
)) != NULL
) {
3160 list_remove(l
, snap
);
3161 dsl_dataset_rele(snap
->ds
, tag
);
3162 kmem_free(snap
, sizeof (*snap
));
3168 promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
, void *tag
)
3172 struct promotenode
*snap
;
3174 error
= dsl_dataset_hold(dp
, ddpa
->ddpa_clonename
, tag
,
3178 dd
= ddpa
->ddpa_clone
->ds_dir
;
3180 if (ddpa
->ddpa_clone
->ds_is_snapshot
||
3181 !dsl_dir_is_clone(dd
)) {
3182 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
3183 return (SET_ERROR(EINVAL
));
3186 error
= snaplist_make(dp
, 0, dsl_dir_phys(dd
)->dd_origin_obj
,
3187 &ddpa
->shared_snaps
, tag
);
3191 error
= snaplist_make(dp
, 0, ddpa
->ddpa_clone
->ds_object
,
3192 &ddpa
->clone_snaps
, tag
);
3196 snap
= list_head(&ddpa
->shared_snaps
);
3197 ASSERT3U(snap
->ds
->ds_object
, ==, dsl_dir_phys(dd
)->dd_origin_obj
);
3198 error
= snaplist_make(dp
, dsl_dir_phys(dd
)->dd_origin_obj
,
3199 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_head_dataset_obj
,
3200 &ddpa
->origin_snaps
, tag
);
3204 if (dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
!= 0) {
3205 error
= dsl_dataset_hold_obj(dp
,
3206 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
,
3207 tag
, &ddpa
->origin_origin
);
3213 promote_rele(ddpa
, tag
);
3218 promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
)
3220 snaplist_destroy(&ddpa
->shared_snaps
, tag
);
3221 snaplist_destroy(&ddpa
->clone_snaps
, tag
);
3222 snaplist_destroy(&ddpa
->origin_snaps
, tag
);
3223 if (ddpa
->origin_origin
!= NULL
)
3224 dsl_dataset_rele(ddpa
->origin_origin
, tag
);
3225 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
3231 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
3232 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
3235 dsl_dataset_promote(const char *name
, char *conflsnap
)
3237 dsl_dataset_promote_arg_t ddpa
= { 0 };
3240 nvpair_t
*snap_pair
;
3244 * We will modify space proportional to the number of
3245 * snapshots. Compute numsnaps.
3247 error
= dmu_objset_hold(name
, FTAG
, &os
);
3250 error
= zap_count(dmu_objset_pool(os
)->dp_meta_objset
,
3251 dsl_dataset_phys(dmu_objset_ds(os
))->ds_snapnames_zapobj
,
3253 dmu_objset_rele(os
, FTAG
);
3257 ddpa
.ddpa_clonename
= name
;
3258 ddpa
.err_ds
= fnvlist_alloc();
3261 error
= dsl_sync_task(name
, dsl_dataset_promote_check
,
3262 dsl_dataset_promote_sync
, &ddpa
,
3263 2 + numsnaps
, ZFS_SPACE_CHECK_RESERVED
);
3266 * Return the first conflicting snapshot found.
3268 snap_pair
= nvlist_next_nvpair(ddpa
.err_ds
, NULL
);
3269 if (snap_pair
!= NULL
&& conflsnap
!= NULL
)
3270 (void) strcpy(conflsnap
, nvpair_name(snap_pair
));
3272 fnvlist_free(ddpa
.err_ds
);
3277 dsl_dataset_clone_swap_check_impl(dsl_dataset_t
*clone
,
3278 dsl_dataset_t
*origin_head
, boolean_t force
, void *owner
, dmu_tx_t
*tx
)
3281 * "slack" factor for received datasets with refquota set on them.
3282 * See the bottom of this function for details on its use.
3284 uint64_t refquota_slack
= DMU_MAX_ACCESS
* spa_asize_inflation
;
3285 int64_t unused_refres_delta
;
3287 /* they should both be heads */
3288 if (clone
->ds_is_snapshot
||
3289 origin_head
->ds_is_snapshot
)
3290 return (SET_ERROR(EINVAL
));
3292 /* if we are not forcing, the branch point should be just before them */
3293 if (!force
&& clone
->ds_prev
!= origin_head
->ds_prev
)
3294 return (SET_ERROR(EINVAL
));
3296 /* clone should be the clone (unless they are unrelated) */
3297 if (clone
->ds_prev
!= NULL
&&
3298 clone
->ds_prev
!= clone
->ds_dir
->dd_pool
->dp_origin_snap
&&
3299 origin_head
->ds_dir
!= clone
->ds_prev
->ds_dir
)
3300 return (SET_ERROR(EINVAL
));
3302 /* the clone should be a child of the origin */
3303 if (clone
->ds_dir
->dd_parent
!= origin_head
->ds_dir
)
3304 return (SET_ERROR(EINVAL
));
3306 /* origin_head shouldn't be modified unless 'force' */
3308 dsl_dataset_modified_since_snap(origin_head
, origin_head
->ds_prev
))
3309 return (SET_ERROR(ETXTBSY
));
3311 /* origin_head should have no long holds (e.g. is not mounted) */
3312 if (dsl_dataset_handoff_check(origin_head
, owner
, tx
))
3313 return (SET_ERROR(EBUSY
));
3315 /* check amount of any unconsumed refreservation */
3316 unused_refres_delta
=
3317 (int64_t)MIN(origin_head
->ds_reserved
,
3318 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
3319 (int64_t)MIN(origin_head
->ds_reserved
,
3320 dsl_dataset_phys(clone
)->ds_unique_bytes
);
3322 if (unused_refres_delta
> 0 &&
3323 unused_refres_delta
>
3324 dsl_dir_space_available(origin_head
->ds_dir
, NULL
, 0, TRUE
))
3325 return (SET_ERROR(ENOSPC
));
3328 * The clone can't be too much over the head's refquota.
3330 * To ensure that the entire refquota can be used, we allow one
3331 * transaction to exceed the the refquota. Therefore, this check
3332 * needs to also allow for the space referenced to be more than the
3333 * refquota. The maximum amount of space that one transaction can use
3334 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this
3335 * overage ensures that we are able to receive a filesystem that
3336 * exceeds the refquota on the source system.
3338 * So that overage is the refquota_slack we use below.
3340 if (origin_head
->ds_quota
!= 0 &&
3341 dsl_dataset_phys(clone
)->ds_referenced_bytes
>
3342 origin_head
->ds_quota
+ refquota_slack
)
3343 return (SET_ERROR(EDQUOT
));
3349 dsl_dataset_swap_remap_deadlists(dsl_dataset_t
*clone
,
3350 dsl_dataset_t
*origin
, dmu_tx_t
*tx
)
3352 uint64_t clone_remap_dl_obj
, origin_remap_dl_obj
;
3353 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3355 ASSERT(dsl_pool_sync_context(dp
));
3357 clone_remap_dl_obj
= dsl_dataset_get_remap_deadlist_object(clone
);
3358 origin_remap_dl_obj
= dsl_dataset_get_remap_deadlist_object(origin
);
3360 if (clone_remap_dl_obj
!= 0) {
3361 dsl_deadlist_close(&clone
->ds_remap_deadlist
);
3362 dsl_dataset_unset_remap_deadlist_object(clone
, tx
);
3364 if (origin_remap_dl_obj
!= 0) {
3365 dsl_deadlist_close(&origin
->ds_remap_deadlist
);
3366 dsl_dataset_unset_remap_deadlist_object(origin
, tx
);
3369 if (clone_remap_dl_obj
!= 0) {
3370 dsl_dataset_set_remap_deadlist_object(origin
,
3371 clone_remap_dl_obj
, tx
);
3372 dsl_deadlist_open(&origin
->ds_remap_deadlist
,
3373 dp
->dp_meta_objset
, clone_remap_dl_obj
);
3375 if (origin_remap_dl_obj
!= 0) {
3376 dsl_dataset_set_remap_deadlist_object(clone
,
3377 origin_remap_dl_obj
, tx
);
3378 dsl_deadlist_open(&clone
->ds_remap_deadlist
,
3379 dp
->dp_meta_objset
, origin_remap_dl_obj
);
3384 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t
*clone
,
3385 dsl_dataset_t
*origin_head
, dmu_tx_t
*tx
)
3387 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3388 int64_t unused_refres_delta
;
3390 ASSERT(clone
->ds_reserved
== 0);
3392 * NOTE: On DEBUG kernels there could be a race between this and
3393 * the check function if spa_asize_inflation is adjusted...
3395 ASSERT(origin_head
->ds_quota
== 0 ||
3396 dsl_dataset_phys(clone
)->ds_unique_bytes
<= origin_head
->ds_quota
+
3397 DMU_MAX_ACCESS
* spa_asize_inflation
);
3398 ASSERT3P(clone
->ds_prev
, ==, origin_head
->ds_prev
);
3401 * Swap per-dataset feature flags.
3403 for (spa_feature_t f
= 0; f
< SPA_FEATURES
; f
++) {
3404 if (!(spa_feature_table
[f
].fi_flags
&
3405 ZFEATURE_FLAG_PER_DATASET
)) {
3406 ASSERT(!clone
->ds_feature_inuse
[f
]);
3407 ASSERT(!origin_head
->ds_feature_inuse
[f
]);
3411 boolean_t clone_inuse
= clone
->ds_feature_inuse
[f
];
3412 boolean_t origin_head_inuse
= origin_head
->ds_feature_inuse
[f
];
3415 dsl_dataset_deactivate_feature(clone
->ds_object
, f
, tx
);
3416 clone
->ds_feature_inuse
[f
] = B_FALSE
;
3418 if (origin_head_inuse
) {
3419 dsl_dataset_deactivate_feature(origin_head
->ds_object
,
3421 origin_head
->ds_feature_inuse
[f
] = B_FALSE
;
3424 dsl_dataset_activate_feature(origin_head
->ds_object
,
3426 origin_head
->ds_feature_inuse
[f
] = B_TRUE
;
3428 if (origin_head_inuse
) {
3429 dsl_dataset_activate_feature(clone
->ds_object
, f
, tx
);
3430 clone
->ds_feature_inuse
[f
] = B_TRUE
;
3434 dmu_buf_will_dirty(clone
->ds_dbuf
, tx
);
3435 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
3437 if (clone
->ds_objset
!= NULL
) {
3438 dmu_objset_evict(clone
->ds_objset
);
3439 clone
->ds_objset
= NULL
;
3442 if (origin_head
->ds_objset
!= NULL
) {
3443 dmu_objset_evict(origin_head
->ds_objset
);
3444 origin_head
->ds_objset
= NULL
;
3447 unused_refres_delta
=
3448 (int64_t)MIN(origin_head
->ds_reserved
,
3449 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
3450 (int64_t)MIN(origin_head
->ds_reserved
,
3451 dsl_dataset_phys(clone
)->ds_unique_bytes
);
3454 * Reset origin's unique bytes, if it exists.
3456 if (clone
->ds_prev
) {
3457 dsl_dataset_t
*origin
= clone
->ds_prev
;
3458 uint64_t comp
, uncomp
;
3460 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
3461 dsl_deadlist_space_range(&clone
->ds_deadlist
,
3462 dsl_dataset_phys(origin
)->ds_prev_snap_txg
, UINT64_MAX
,
3463 &dsl_dataset_phys(origin
)->ds_unique_bytes
, &comp
, &uncomp
);
3468 rrw_enter(&clone
->ds_bp_rwlock
, RW_WRITER
, FTAG
);
3469 rrw_enter(&origin_head
->ds_bp_rwlock
, RW_WRITER
, FTAG
);
3471 tmp
= dsl_dataset_phys(origin_head
)->ds_bp
;
3472 dsl_dataset_phys(origin_head
)->ds_bp
=
3473 dsl_dataset_phys(clone
)->ds_bp
;
3474 dsl_dataset_phys(clone
)->ds_bp
= tmp
;
3475 rrw_exit(&origin_head
->ds_bp_rwlock
, FTAG
);
3476 rrw_exit(&clone
->ds_bp_rwlock
, FTAG
);
3479 /* set dd_*_bytes */
3481 int64_t dused
, dcomp
, duncomp
;
3482 uint64_t cdl_used
, cdl_comp
, cdl_uncomp
;
3483 uint64_t odl_used
, odl_comp
, odl_uncomp
;
3485 ASSERT3U(dsl_dir_phys(clone
->ds_dir
)->
3486 dd_used_breakdown
[DD_USED_SNAP
], ==, 0);
3488 dsl_deadlist_space(&clone
->ds_deadlist
,
3489 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
3490 dsl_deadlist_space(&origin_head
->ds_deadlist
,
3491 &odl_used
, &odl_comp
, &odl_uncomp
);
3493 dused
= dsl_dataset_phys(clone
)->ds_referenced_bytes
+
3495 (dsl_dataset_phys(origin_head
)->ds_referenced_bytes
+
3497 dcomp
= dsl_dataset_phys(clone
)->ds_compressed_bytes
+
3499 (dsl_dataset_phys(origin_head
)->ds_compressed_bytes
+
3501 duncomp
= dsl_dataset_phys(clone
)->ds_uncompressed_bytes
+
3503 (dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
+
3506 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_HEAD
,
3507 dused
, dcomp
, duncomp
, tx
);
3508 dsl_dir_diduse_space(clone
->ds_dir
, DD_USED_HEAD
,
3509 -dused
, -dcomp
, -duncomp
, tx
);
3512 * The difference in the space used by snapshots is the
3513 * difference in snapshot space due to the head's
3514 * deadlist (since that's the only thing that's
3515 * changing that affects the snapused).
3517 dsl_deadlist_space_range(&clone
->ds_deadlist
,
3518 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
3519 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
3520 dsl_deadlist_space_range(&origin_head
->ds_deadlist
,
3521 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
3522 &odl_used
, &odl_comp
, &odl_uncomp
);
3523 dsl_dir_transfer_space(origin_head
->ds_dir
, cdl_used
- odl_used
,
3524 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
3527 /* swap ds_*_bytes */
3528 SWITCH64(dsl_dataset_phys(origin_head
)->ds_referenced_bytes
,
3529 dsl_dataset_phys(clone
)->ds_referenced_bytes
);
3530 SWITCH64(dsl_dataset_phys(origin_head
)->ds_compressed_bytes
,
3531 dsl_dataset_phys(clone
)->ds_compressed_bytes
);
3532 SWITCH64(dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
,
3533 dsl_dataset_phys(clone
)->ds_uncompressed_bytes
);
3534 SWITCH64(dsl_dataset_phys(origin_head
)->ds_unique_bytes
,
3535 dsl_dataset_phys(clone
)->ds_unique_bytes
);
3537 /* apply any parent delta for change in unconsumed refreservation */
3538 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_REFRSRV
,
3539 unused_refres_delta
, 0, 0, tx
);
3544 dsl_deadlist_close(&clone
->ds_deadlist
);
3545 dsl_deadlist_close(&origin_head
->ds_deadlist
);
3546 SWITCH64(dsl_dataset_phys(origin_head
)->ds_deadlist_obj
,
3547 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
3548 dsl_deadlist_open(&clone
->ds_deadlist
, dp
->dp_meta_objset
,
3549 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
3550 dsl_deadlist_open(&origin_head
->ds_deadlist
, dp
->dp_meta_objset
,
3551 dsl_dataset_phys(origin_head
)->ds_deadlist_obj
);
3552 dsl_dataset_swap_remap_deadlists(clone
, origin_head
, tx
);
3554 dsl_scan_ds_clone_swapped(origin_head
, clone
, tx
);
3556 spa_history_log_internal_ds(clone
, "clone swap", tx
,
3557 "parent=%s", origin_head
->ds_dir
->dd_myname
);
3561 * Given a pool name and a dataset object number in that pool,
3562 * return the name of that dataset.
3565 dsl_dsobj_to_dsname(char *pname
, uint64_t obj
, char *buf
)
3571 error
= dsl_pool_hold(pname
, FTAG
, &dp
);
3575 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
3577 dsl_dataset_name(ds
, buf
);
3578 dsl_dataset_rele(ds
, FTAG
);
3580 dsl_pool_rele(dp
, FTAG
);
3586 dsl_dataset_check_quota(dsl_dataset_t
*ds
, boolean_t check_quota
,
3587 uint64_t asize
, uint64_t inflight
, uint64_t *used
, uint64_t *ref_rsrv
)
3591 ASSERT3S(asize
, >, 0);
3594 * *ref_rsrv is the portion of asize that will come from any
3595 * unconsumed refreservation space.
3599 mutex_enter(&ds
->ds_lock
);
3601 * Make a space adjustment for reserved bytes.
3603 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
) {
3605 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
3607 (ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
3609 asize
- MIN(asize
, parent_delta(ds
, asize
+ inflight
));
3612 if (!check_quota
|| ds
->ds_quota
== 0) {
3613 mutex_exit(&ds
->ds_lock
);
3617 * If they are requesting more space, and our current estimate
3618 * is over quota, they get to try again unless the actual
3619 * on-disk is over quota and there are no pending changes (which
3620 * may free up space for us).
3622 if (dsl_dataset_phys(ds
)->ds_referenced_bytes
+ inflight
>=
3625 dsl_dataset_phys(ds
)->ds_referenced_bytes
< ds
->ds_quota
)
3626 error
= SET_ERROR(ERESTART
);
3628 error
= SET_ERROR(EDQUOT
);
3630 mutex_exit(&ds
->ds_lock
);
3635 typedef struct dsl_dataset_set_qr_arg
{
3636 const char *ddsqra_name
;
3637 zprop_source_t ddsqra_source
;
3638 uint64_t ddsqra_value
;
3639 } dsl_dataset_set_qr_arg_t
;
3644 dsl_dataset_set_refquota_check(void *arg
, dmu_tx_t
*tx
)
3646 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3647 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3652 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFQUOTA
)
3653 return (SET_ERROR(ENOTSUP
));
3655 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3659 if (ds
->ds_is_snapshot
) {
3660 dsl_dataset_rele(ds
, FTAG
);
3661 return (SET_ERROR(EINVAL
));
3664 error
= dsl_prop_predict(ds
->ds_dir
,
3665 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
3666 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3668 dsl_dataset_rele(ds
, FTAG
);
3673 dsl_dataset_rele(ds
, FTAG
);
3677 if (newval
< dsl_dataset_phys(ds
)->ds_referenced_bytes
||
3678 newval
< ds
->ds_reserved
) {
3679 dsl_dataset_rele(ds
, FTAG
);
3680 return (SET_ERROR(ENOSPC
));
3683 dsl_dataset_rele(ds
, FTAG
);
3688 dsl_dataset_set_refquota_sync(void *arg
, dmu_tx_t
*tx
)
3690 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3691 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3695 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3697 dsl_prop_set_sync_impl(ds
,
3698 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
3699 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
3700 &ddsqra
->ddsqra_value
, tx
);
3702 VERIFY0(dsl_prop_get_int_ds(ds
,
3703 zfs_prop_to_name(ZFS_PROP_REFQUOTA
), &newval
));
3705 if (ds
->ds_quota
!= newval
) {
3706 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3707 ds
->ds_quota
= newval
;
3709 dsl_dataset_rele(ds
, FTAG
);
3713 dsl_dataset_set_refquota(const char *dsname
, zprop_source_t source
,
3716 dsl_dataset_set_qr_arg_t ddsqra
;
3718 ddsqra
.ddsqra_name
= dsname
;
3719 ddsqra
.ddsqra_source
= source
;
3720 ddsqra
.ddsqra_value
= refquota
;
3722 return (dsl_sync_task(dsname
, dsl_dataset_set_refquota_check
,
3723 dsl_dataset_set_refquota_sync
, &ddsqra
, 0,
3724 ZFS_SPACE_CHECK_EXTRA_RESERVED
));
3728 dsl_dataset_set_refreservation_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
);
3734 uint64_t newval
, unique
;
3736 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFRESERVATION
)
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_REFRESERVATION
),
3750 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3752 dsl_dataset_rele(ds
, FTAG
);
3757 * If we are doing the preliminary check in open context, the
3758 * space estimates may be inaccurate.
3760 if (!dmu_tx_is_syncing(tx
)) {
3761 dsl_dataset_rele(ds
, FTAG
);
3765 mutex_enter(&ds
->ds_lock
);
3766 if (!DS_UNIQUE_IS_ACCURATE(ds
))
3767 dsl_dataset_recalc_head_uniq(ds
);
3768 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3769 mutex_exit(&ds
->ds_lock
);
3771 if (MAX(unique
, newval
) > MAX(unique
, ds
->ds_reserved
)) {
3772 uint64_t delta
= MAX(unique
, newval
) -
3773 MAX(unique
, ds
->ds_reserved
);
3776 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, B_TRUE
) ||
3777 (ds
->ds_quota
> 0 && newval
> ds
->ds_quota
)) {
3778 dsl_dataset_rele(ds
, FTAG
);
3779 return (SET_ERROR(ENOSPC
));
3783 dsl_dataset_rele(ds
, FTAG
);
3788 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t
*ds
,
3789 zprop_source_t source
, uint64_t value
, dmu_tx_t
*tx
)
3795 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3796 source
, sizeof (value
), 1, &value
, tx
);
3798 VERIFY0(dsl_prop_get_int_ds(ds
,
3799 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
), &newval
));
3801 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3802 mutex_enter(&ds
->ds_dir
->dd_lock
);
3803 mutex_enter(&ds
->ds_lock
);
3804 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
3805 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3806 delta
= MAX(0, (int64_t)(newval
- unique
)) -
3807 MAX(0, (int64_t)(ds
->ds_reserved
- unique
));
3808 ds
->ds_reserved
= newval
;
3809 mutex_exit(&ds
->ds_lock
);
3811 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
, delta
, 0, 0, tx
);
3812 mutex_exit(&ds
->ds_dir
->dd_lock
);
3816 dsl_dataset_set_refreservation_sync(void *arg
, dmu_tx_t
*tx
)
3818 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3819 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3822 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3823 dsl_dataset_set_refreservation_sync_impl(ds
,
3824 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, tx
);
3825 dsl_dataset_rele(ds
, FTAG
);
3829 dsl_dataset_set_refreservation(const char *dsname
, zprop_source_t source
,
3830 uint64_t refreservation
)
3832 dsl_dataset_set_qr_arg_t ddsqra
;
3834 ddsqra
.ddsqra_name
= dsname
;
3835 ddsqra
.ddsqra_source
= source
;
3836 ddsqra
.ddsqra_value
= refreservation
;
3838 return (dsl_sync_task(dsname
, dsl_dataset_set_refreservation_check
,
3839 dsl_dataset_set_refreservation_sync
, &ddsqra
, 0,
3840 ZFS_SPACE_CHECK_EXTRA_RESERVED
));
3844 * Return (in *usedp) the amount of space written in new that is not
3845 * present in oldsnap. New may be a snapshot or the head. Old must be
3846 * a snapshot before new, in new's filesystem (or its origin). If not then
3847 * fail and return EINVAL.
3849 * The written space is calculated by considering two components: First, we
3850 * ignore any freed space, and calculate the written as new's used space
3851 * minus old's used space. Next, we add in the amount of space that was freed
3852 * between the two snapshots, thus reducing new's used space relative to old's.
3853 * Specifically, this is the space that was born before old->ds_creation_txg,
3854 * and freed before new (ie. on new's deadlist or a previous deadlist).
3856 * space freed [---------------------]
3857 * snapshots ---O-------O--------O-------O------
3861 dsl_dataset_space_written(dsl_dataset_t
*oldsnap
, dsl_dataset_t
*new,
3862 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3866 dsl_pool_t
*dp
= new->ds_dir
->dd_pool
;
3868 ASSERT(dsl_pool_config_held(dp
));
3871 *usedp
+= dsl_dataset_phys(new)->ds_referenced_bytes
;
3872 *usedp
-= dsl_dataset_phys(oldsnap
)->ds_referenced_bytes
;
3875 *compp
+= dsl_dataset_phys(new)->ds_compressed_bytes
;
3876 *compp
-= dsl_dataset_phys(oldsnap
)->ds_compressed_bytes
;
3879 *uncompp
+= dsl_dataset_phys(new)->ds_uncompressed_bytes
;
3880 *uncompp
-= dsl_dataset_phys(oldsnap
)->ds_uncompressed_bytes
;
3882 snapobj
= new->ds_object
;
3883 while (snapobj
!= oldsnap
->ds_object
) {
3884 dsl_dataset_t
*snap
;
3885 uint64_t used
, comp
, uncomp
;
3887 if (snapobj
== new->ds_object
) {
3890 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &snap
);
3895 if (dsl_dataset_phys(snap
)->ds_prev_snap_txg
==
3896 dsl_dataset_phys(oldsnap
)->ds_creation_txg
) {
3898 * The blocks in the deadlist can not be born after
3899 * ds_prev_snap_txg, so get the whole deadlist space,
3900 * which is more efficient (especially for old-format
3901 * deadlists). Unfortunately the deadlist code
3902 * doesn't have enough information to make this
3903 * optimization itself.
3905 dsl_deadlist_space(&snap
->ds_deadlist
,
3906 &used
, &comp
, &uncomp
);
3908 dsl_deadlist_space_range(&snap
->ds_deadlist
,
3909 0, dsl_dataset_phys(oldsnap
)->ds_creation_txg
,
3910 &used
, &comp
, &uncomp
);
3917 * If we get to the beginning of the chain of snapshots
3918 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3919 * was not a snapshot of/before new.
3921 snapobj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3923 dsl_dataset_rele(snap
, FTAG
);
3925 err
= SET_ERROR(EINVAL
);
3934 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3935 * lastsnap, and all snapshots in between are deleted.
3937 * blocks that would be freed [---------------------------]
3938 * snapshots ---O-------O--------O-------O--------O
3939 * firstsnap lastsnap
3941 * This is the set of blocks that were born after the snap before firstsnap,
3942 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3943 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3944 * We calculate this by iterating over the relevant deadlists (from the snap
3945 * after lastsnap, backward to the snap after firstsnap), summing up the
3946 * space on the deadlist that was born after the snap before firstsnap.
3949 dsl_dataset_space_wouldfree(dsl_dataset_t
*firstsnap
,
3950 dsl_dataset_t
*lastsnap
,
3951 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3955 dsl_pool_t
*dp
= firstsnap
->ds_dir
->dd_pool
;
3957 ASSERT(firstsnap
->ds_is_snapshot
);
3958 ASSERT(lastsnap
->ds_is_snapshot
);
3961 * Check that the snapshots are in the same dsl_dir, and firstsnap
3962 * is before lastsnap.
3964 if (firstsnap
->ds_dir
!= lastsnap
->ds_dir
||
3965 dsl_dataset_phys(firstsnap
)->ds_creation_txg
>
3966 dsl_dataset_phys(lastsnap
)->ds_creation_txg
)
3967 return (SET_ERROR(EINVAL
));
3969 *usedp
= *compp
= *uncompp
= 0;
3971 snapobj
= dsl_dataset_phys(lastsnap
)->ds_next_snap_obj
;
3972 while (snapobj
!= firstsnap
->ds_object
) {
3974 uint64_t used
, comp
, uncomp
;
3976 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &ds
);
3980 dsl_deadlist_space_range(&ds
->ds_deadlist
,
3981 dsl_dataset_phys(firstsnap
)->ds_prev_snap_txg
, UINT64_MAX
,
3982 &used
, &comp
, &uncomp
);
3987 snapobj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
3988 ASSERT3U(snapobj
, !=, 0);
3989 dsl_dataset_rele(ds
, FTAG
);
3995 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3996 * For example, they could both be snapshots of the same filesystem, and
3997 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3998 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3999 * filesystem. Or 'earlier' could be the origin's origin.
4001 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
4004 dsl_dataset_is_before(dsl_dataset_t
*later
, dsl_dataset_t
*earlier
,
4005 uint64_t earlier_txg
)
4007 dsl_pool_t
*dp
= later
->ds_dir
->dd_pool
;
4011 ASSERT(dsl_pool_config_held(dp
));
4012 ASSERT(earlier
->ds_is_snapshot
|| earlier_txg
!= 0);
4014 if (earlier_txg
== 0)
4015 earlier_txg
= dsl_dataset_phys(earlier
)->ds_creation_txg
;
4017 if (later
->ds_is_snapshot
&&
4018 earlier_txg
>= dsl_dataset_phys(later
)->ds_creation_txg
)
4021 if (later
->ds_dir
== earlier
->ds_dir
)
4023 if (!dsl_dir_is_clone(later
->ds_dir
))
4026 if (dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
== earlier
->ds_object
)
4028 dsl_dataset_t
*origin
;
4029 error
= dsl_dataset_hold_obj(dp
,
4030 dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
, FTAG
, &origin
);
4033 ret
= dsl_dataset_is_before(origin
, earlier
, earlier_txg
);
4034 dsl_dataset_rele(origin
, FTAG
);
4039 dsl_dataset_zapify(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
4041 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
4042 dmu_object_zapify(mos
, ds
->ds_object
, DMU_OT_DSL_DATASET
, tx
);
4046 dsl_dataset_is_zapified(dsl_dataset_t
*ds
)
4048 dmu_object_info_t doi
;
4050 dmu_object_info_from_db(ds
->ds_dbuf
, &doi
);
4051 return (doi
.doi_type
== DMU_OTN_ZAP_METADATA
);
4055 dsl_dataset_has_resume_receive_state(dsl_dataset_t
*ds
)
4057 return (dsl_dataset_is_zapified(ds
) &&
4058 zap_contains(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
4059 ds
->ds_object
, DS_FIELD_RESUME_TOGUID
) == 0);
4063 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t
*ds
)
4065 uint64_t remap_deadlist_obj
;
4068 if (!dsl_dataset_is_zapified(ds
))
4071 err
= zap_lookup(ds
->ds_dir
->dd_pool
->dp_meta_objset
, ds
->ds_object
,
4072 DS_FIELD_REMAP_DEADLIST
, sizeof (remap_deadlist_obj
), 1,
4073 &remap_deadlist_obj
);
4076 VERIFY3S(err
, ==, ENOENT
);
4080 ASSERT(remap_deadlist_obj
!= 0);
4081 return (remap_deadlist_obj
);
4085 dsl_dataset_remap_deadlist_exists(dsl_dataset_t
*ds
)
4087 EQUIV(dsl_deadlist_is_open(&ds
->ds_remap_deadlist
),
4088 dsl_dataset_get_remap_deadlist_object(ds
) != 0);
4089 return (dsl_deadlist_is_open(&ds
->ds_remap_deadlist
));
4093 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t
*ds
, uint64_t obj
,
4097 dsl_dataset_zapify(ds
, tx
);
4098 VERIFY0(zap_add(ds
->ds_dir
->dd_pool
->dp_meta_objset
, ds
->ds_object
,
4099 DS_FIELD_REMAP_DEADLIST
, sizeof (obj
), 1, &obj
, tx
));
4103 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
4105 VERIFY0(zap_remove(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
4106 ds
->ds_object
, DS_FIELD_REMAP_DEADLIST
, tx
));
4110 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
4112 uint64_t remap_deadlist_object
;
4113 spa_t
*spa
= ds
->ds_dir
->dd_pool
->dp_spa
;
4115 ASSERT(dmu_tx_is_syncing(tx
));
4116 ASSERT(dsl_dataset_remap_deadlist_exists(ds
));
4118 remap_deadlist_object
= ds
->ds_remap_deadlist
.dl_object
;
4119 dsl_deadlist_close(&ds
->ds_remap_deadlist
);
4120 dsl_deadlist_free(spa_meta_objset(spa
), remap_deadlist_object
, tx
);
4121 dsl_dataset_unset_remap_deadlist_object(ds
, tx
);
4122 spa_feature_decr(spa
, SPA_FEATURE_OBSOLETE_COUNTS
, tx
);
4126 dsl_dataset_create_remap_deadlist(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
4128 uint64_t remap_deadlist_obj
;
4129 spa_t
*spa
= ds
->ds_dir
->dd_pool
->dp_spa
;
4131 ASSERT(dmu_tx_is_syncing(tx
));
4132 ASSERT(MUTEX_HELD(&ds
->ds_remap_deadlist_lock
));
4134 * Currently we only create remap deadlists when there are indirect
4135 * vdevs with referenced mappings.
4137 ASSERT(spa_feature_is_active(spa
, SPA_FEATURE_DEVICE_REMOVAL
));
4139 remap_deadlist_obj
= dsl_deadlist_clone(
4140 &ds
->ds_deadlist
, UINT64_MAX
,
4141 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, tx
);
4142 dsl_dataset_set_remap_deadlist_object(ds
,
4143 remap_deadlist_obj
, tx
);
4144 dsl_deadlist_open(&ds
->ds_remap_deadlist
, spa_meta_objset(spa
),
4145 remap_deadlist_obj
);
4146 spa_feature_incr(spa
, SPA_FEATURE_OBSOLETE_COUNTS
, tx
);