4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
25 * Copyright (c) 2014 RackTop Systems.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
30 #include <sys/dmu_objset.h>
31 #include <sys/dsl_dataset.h>
32 #include <sys/dsl_dir.h>
33 #include <sys/dsl_prop.h>
34 #include <sys/dsl_synctask.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dmu_impl.h>
37 #include <sys/dmu_tx.h>
41 #include <sys/zfeature.h>
42 #include <sys/unique.h>
43 #include <sys/zfs_context.h>
44 #include <sys/zfs_ioctl.h>
46 #include <sys/zfs_znode.h>
47 #include <sys/zfs_onexit.h>
49 #include <sys/dsl_scan.h>
50 #include <sys/dsl_deadlist.h>
51 #include <sys/dsl_destroy.h>
52 #include <sys/dsl_userhold.h>
53 #include <sys/dsl_bookmark.h>
56 * The SPA supports block sizes up to 16MB. However, very large blocks
57 * can have an impact on i/o latency (e.g. tying up a spinning disk for
58 * ~300ms), and also potentially on the memory allocator. Therefore,
59 * we do not allow the recordsize to be set larger than zfs_max_recordsize
60 * (default 1MB). Larger blocks can be created by changing this tunable,
61 * and pools with larger blocks can always be imported and used, regardless
64 int zfs_max_recordsize
= 1 * 1024 * 1024;
66 #define SWITCH64(x, y) \
68 uint64_t __tmp = (x); \
73 #define DS_REF_MAX (1ULL << 62)
75 extern inline dsl_dataset_phys_t
*dsl_dataset_phys(dsl_dataset_t
*ds
);
78 * Figure out how much of this delta should be propogated to the dsl_dir
79 * layer. If there's a refreservation, that space has already been
80 * partially accounted for in our ancestors.
83 parent_delta(dsl_dataset_t
*ds
, int64_t delta
)
85 dsl_dataset_phys_t
*ds_phys
;
86 uint64_t old_bytes
, new_bytes
;
88 if (ds
->ds_reserved
== 0)
91 ds_phys
= dsl_dataset_phys(ds
);
92 old_bytes
= MAX(ds_phys
->ds_unique_bytes
, ds
->ds_reserved
);
93 new_bytes
= MAX(ds_phys
->ds_unique_bytes
+ delta
, ds
->ds_reserved
);
95 ASSERT3U(ABS((int64_t)(new_bytes
- old_bytes
)), <=, ABS(delta
));
96 return (new_bytes
- old_bytes
);
100 dsl_dataset_block_born(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
102 int used
, compressed
, uncompressed
;
105 used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
106 compressed
= BP_GET_PSIZE(bp
);
107 uncompressed
= BP_GET_UCSIZE(bp
);
109 dprintf_bp(bp
, "ds=%p", ds
);
111 ASSERT(dmu_tx_is_syncing(tx
));
112 /* It could have been compressed away to nothing */
115 ASSERT(BP_GET_TYPE(bp
) != DMU_OT_NONE
);
116 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp
)));
118 dsl_pool_mos_diduse_space(tx
->tx_pool
,
119 used
, compressed
, uncompressed
);
123 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
124 mutex_enter(&ds
->ds_lock
);
125 delta
= parent_delta(ds
, used
);
126 dsl_dataset_phys(ds
)->ds_referenced_bytes
+= used
;
127 dsl_dataset_phys(ds
)->ds_compressed_bytes
+= compressed
;
128 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
+= uncompressed
;
129 dsl_dataset_phys(ds
)->ds_unique_bytes
+= used
;
130 if (BP_GET_LSIZE(bp
) > SPA_OLD_MAXBLOCKSIZE
) {
131 ds
->ds_feature_activation_needed
[SPA_FEATURE_LARGE_BLOCKS
] =
134 mutex_exit(&ds
->ds_lock
);
135 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
, delta
,
136 compressed
, uncompressed
, tx
);
137 dsl_dir_transfer_space(ds
->ds_dir
, used
- delta
,
138 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
142 dsl_dataset_block_kill(dsl_dataset_t
*ds
, const blkptr_t
*bp
, dmu_tx_t
*tx
,
145 int used
= bp_get_dsize_sync(tx
->tx_pool
->dp_spa
, bp
);
146 int compressed
= BP_GET_PSIZE(bp
);
147 int uncompressed
= BP_GET_UCSIZE(bp
);
152 ASSERT(dmu_tx_is_syncing(tx
));
153 ASSERT(bp
->blk_birth
<= tx
->tx_txg
);
156 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
157 dsl_pool_mos_diduse_space(tx
->tx_pool
,
158 -used
, -compressed
, -uncompressed
);
161 ASSERT3P(tx
->tx_pool
, ==, ds
->ds_dir
->dd_pool
);
163 ASSERT(!ds
->ds_is_snapshot
);
164 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
166 if (bp
->blk_birth
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
169 dprintf_bp(bp
, "freeing ds=%llu", ds
->ds_object
);
170 dsl_free(tx
->tx_pool
, tx
->tx_txg
, bp
);
172 mutex_enter(&ds
->ds_lock
);
173 ASSERT(dsl_dataset_phys(ds
)->ds_unique_bytes
>= used
||
174 !DS_UNIQUE_IS_ACCURATE(ds
));
175 delta
= parent_delta(ds
, -used
);
176 dsl_dataset_phys(ds
)->ds_unique_bytes
-= used
;
177 mutex_exit(&ds
->ds_lock
);
178 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
179 delta
, -compressed
, -uncompressed
, tx
);
180 dsl_dir_transfer_space(ds
->ds_dir
, -used
- delta
,
181 DD_USED_REFRSRV
, DD_USED_HEAD
, tx
);
183 dprintf_bp(bp
, "putting on dead list: %s", "");
186 * We are here as part of zio's write done callback,
187 * which means we're a zio interrupt thread. We can't
188 * call dsl_deadlist_insert() now because it may block
189 * waiting for I/O. Instead, put bp on the deferred
190 * queue and let dsl_pool_sync() finish the job.
192 bplist_append(&ds
->ds_pending_deadlist
, bp
);
194 dsl_deadlist_insert(&ds
->ds_deadlist
, bp
, tx
);
196 ASSERT3U(ds
->ds_prev
->ds_object
, ==,
197 dsl_dataset_phys(ds
)->ds_prev_snap_obj
);
198 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 0);
199 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
200 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
201 ds
->ds_object
&& bp
->blk_birth
>
202 dsl_dataset_phys(ds
->ds_prev
)->ds_prev_snap_txg
) {
203 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
204 mutex_enter(&ds
->ds_prev
->ds_lock
);
205 dsl_dataset_phys(ds
->ds_prev
)->ds_unique_bytes
+= used
;
206 mutex_exit(&ds
->ds_prev
->ds_lock
);
208 if (bp
->blk_birth
> ds
->ds_dir
->dd_origin_txg
) {
209 dsl_dir_transfer_space(ds
->ds_dir
, used
,
210 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
213 mutex_enter(&ds
->ds_lock
);
214 ASSERT3U(dsl_dataset_phys(ds
)->ds_referenced_bytes
, >=, used
);
215 dsl_dataset_phys(ds
)->ds_referenced_bytes
-= used
;
216 ASSERT3U(dsl_dataset_phys(ds
)->ds_compressed_bytes
, >=, compressed
);
217 dsl_dataset_phys(ds
)->ds_compressed_bytes
-= compressed
;
218 ASSERT3U(dsl_dataset_phys(ds
)->ds_uncompressed_bytes
, >=, uncompressed
);
219 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
-= uncompressed
;
220 mutex_exit(&ds
->ds_lock
);
226 dsl_dataset_prev_snap_txg(dsl_dataset_t
*ds
)
228 uint64_t trysnap
= 0;
233 * The snapshot creation could fail, but that would cause an
234 * incorrect FALSE return, which would only result in an
235 * overestimation of the amount of space that an operation would
236 * consume, which is OK.
238 * There's also a small window where we could miss a pending
239 * snapshot, because we could set the sync task in the quiescing
240 * phase. So this should only be used as a guess.
242 if (ds
->ds_trysnap_txg
>
243 spa_last_synced_txg(ds
->ds_dir
->dd_pool
->dp_spa
))
244 trysnap
= ds
->ds_trysnap_txg
;
245 return (MAX(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, trysnap
));
249 dsl_dataset_block_freeable(dsl_dataset_t
*ds
, const blkptr_t
*bp
,
252 if (blk_birth
<= dsl_dataset_prev_snap_txg(ds
) ||
253 (bp
!= NULL
&& BP_IS_HOLE(bp
)))
256 ddt_prefetch(dsl_dataset_get_spa(ds
), bp
);
262 dsl_dataset_evict(void *dbu
)
264 dsl_dataset_t
*ds
= dbu
;
266 ASSERT(ds
->ds_owner
== NULL
);
270 unique_remove(ds
->ds_fsid_guid
);
272 if (ds
->ds_objset
!= NULL
)
273 dmu_objset_evict(ds
->ds_objset
);
276 dsl_dataset_rele(ds
->ds_prev
, ds
);
280 bplist_destroy(&ds
->ds_pending_deadlist
);
281 if (ds
->ds_deadlist
.dl_os
!= NULL
)
282 dsl_deadlist_close(&ds
->ds_deadlist
);
284 dsl_dir_async_rele(ds
->ds_dir
, ds
);
286 ASSERT(!list_link_active(&ds
->ds_synced_link
));
288 list_destroy(&ds
->ds_prop_cbs
);
289 mutex_destroy(&ds
->ds_lock
);
290 mutex_destroy(&ds
->ds_opening_lock
);
291 mutex_destroy(&ds
->ds_sendstream_lock
);
292 refcount_destroy(&ds
->ds_longholds
);
294 kmem_free(ds
, sizeof (dsl_dataset_t
));
298 dsl_dataset_get_snapname(dsl_dataset_t
*ds
)
300 dsl_dataset_phys_t
*headphys
;
303 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
304 objset_t
*mos
= dp
->dp_meta_objset
;
306 if (ds
->ds_snapname
[0])
308 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0)
311 err
= dmu_bonus_hold(mos
, dsl_dir_phys(ds
->ds_dir
)->dd_head_dataset_obj
,
315 headphys
= headdbuf
->db_data
;
316 err
= zap_value_search(dp
->dp_meta_objset
,
317 headphys
->ds_snapnames_zapobj
, ds
->ds_object
, 0, ds
->ds_snapname
);
318 if (err
!= 0 && zfs_recover
== B_TRUE
) {
320 (void) snprintf(ds
->ds_snapname
, sizeof (ds
->ds_snapname
),
321 "SNAPOBJ=%llu-ERR=%d",
322 (unsigned long long)ds
->ds_object
, err
);
324 dmu_buf_rele(headdbuf
, FTAG
);
329 dsl_dataset_snap_lookup(dsl_dataset_t
*ds
, const char *name
, uint64_t *value
)
331 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
332 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
336 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
341 err
= zap_lookup_norm(mos
, snapobj
, name
, 8, 1,
342 value
, mt
, NULL
, 0, NULL
);
343 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
344 err
= zap_lookup(mos
, snapobj
, name
, 8, 1, value
);
349 dsl_dataset_snap_remove(dsl_dataset_t
*ds
, const char *name
, dmu_tx_t
*tx
,
352 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
353 uint64_t snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
357 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
359 if (dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_CI_DATASET
)
364 err
= zap_remove_norm(mos
, snapobj
, name
, mt
, tx
);
365 if (err
== ENOTSUP
&& mt
== MT_FIRST
)
366 err
= zap_remove(mos
, snapobj
, name
, tx
);
368 if (err
== 0 && adj_cnt
)
369 dsl_fs_ss_count_adjust(ds
->ds_dir
, -1,
370 DD_FIELD_SNAPSHOT_COUNT
, tx
);
376 dsl_dataset_try_add_ref(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *tag
)
378 dmu_buf_t
*dbuf
= ds
->ds_dbuf
;
379 boolean_t result
= B_FALSE
;
381 if (dbuf
!= NULL
&& dmu_buf_try_add_ref(dbuf
, dp
->dp_meta_objset
,
382 ds
->ds_object
, DMU_BONUS_BLKID
, tag
)) {
384 if (ds
== dmu_buf_get_user(dbuf
))
387 dmu_buf_rele(dbuf
, tag
);
394 dsl_dataset_hold_obj(dsl_pool_t
*dp
, uint64_t dsobj
, void *tag
,
397 objset_t
*mos
= dp
->dp_meta_objset
;
401 dmu_object_info_t doi
;
403 ASSERT(dsl_pool_config_held(dp
));
405 err
= dmu_bonus_hold(mos
, dsobj
, tag
, &dbuf
);
409 /* Make sure dsobj has the correct object type. */
410 dmu_object_info_from_db(dbuf
, &doi
);
411 if (doi
.doi_bonus_type
!= DMU_OT_DSL_DATASET
) {
412 dmu_buf_rele(dbuf
, tag
);
413 return (SET_ERROR(EINVAL
));
416 ds
= dmu_buf_get_user(dbuf
);
418 dsl_dataset_t
*winner
= NULL
;
420 ds
= kmem_zalloc(sizeof (dsl_dataset_t
), KM_SLEEP
);
422 ds
->ds_object
= dsobj
;
423 ds
->ds_is_snapshot
= dsl_dataset_phys(ds
)->ds_num_children
!= 0;
424 list_link_init(&ds
->ds_synced_link
);
426 mutex_init(&ds
->ds_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
427 mutex_init(&ds
->ds_opening_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
428 mutex_init(&ds
->ds_sendstream_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
429 refcount_create(&ds
->ds_longholds
);
431 bplist_create(&ds
->ds_pending_deadlist
);
432 dsl_deadlist_open(&ds
->ds_deadlist
,
433 mos
, dsl_dataset_phys(ds
)->ds_deadlist_obj
);
435 list_create(&ds
->ds_sendstreams
, sizeof (dmu_sendarg_t
),
436 offsetof(dmu_sendarg_t
, dsa_link
));
438 list_create(&ds
->ds_prop_cbs
, sizeof (dsl_prop_cb_record_t
),
439 offsetof(dsl_prop_cb_record_t
, cbr_ds_node
));
441 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
444 for (f
= 0; f
< SPA_FEATURES
; f
++) {
445 if (!(spa_feature_table
[f
].fi_flags
&
446 ZFEATURE_FLAG_PER_DATASET
))
448 err
= zap_contains(mos
, dsobj
,
449 spa_feature_table
[f
].fi_guid
);
451 ds
->ds_feature_inuse
[f
] = B_TRUE
;
453 ASSERT3U(err
, ==, ENOENT
);
459 err
= dsl_dir_hold_obj(dp
,
460 dsl_dataset_phys(ds
)->ds_dir_obj
, NULL
, ds
, &ds
->ds_dir
);
462 mutex_destroy(&ds
->ds_lock
);
463 mutex_destroy(&ds
->ds_opening_lock
);
464 mutex_destroy(&ds
->ds_sendstream_lock
);
465 refcount_destroy(&ds
->ds_longholds
);
466 bplist_destroy(&ds
->ds_pending_deadlist
);
467 dsl_deadlist_close(&ds
->ds_deadlist
);
468 kmem_free(ds
, sizeof (dsl_dataset_t
));
469 dmu_buf_rele(dbuf
, tag
);
473 if (!ds
->ds_is_snapshot
) {
474 ds
->ds_snapname
[0] = '\0';
475 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
476 err
= dsl_dataset_hold_obj(dp
,
477 dsl_dataset_phys(ds
)->ds_prev_snap_obj
,
480 if (doi
.doi_type
== DMU_OTN_ZAP_METADATA
) {
481 int zaperr
= zap_lookup(mos
, ds
->ds_object
,
482 DS_FIELD_BOOKMARK_NAMES
,
483 sizeof (ds
->ds_bookmarks
), 1,
485 if (zaperr
!= ENOENT
)
489 if (zfs_flags
& ZFS_DEBUG_SNAPNAMES
)
490 err
= dsl_dataset_get_snapname(ds
);
492 dsl_dataset_phys(ds
)->ds_userrefs_obj
!= 0) {
494 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
495 dsl_dataset_phys(ds
)->ds_userrefs_obj
,
500 if (err
== 0 && !ds
->ds_is_snapshot
) {
501 err
= dsl_prop_get_int_ds(ds
,
502 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
505 err
= dsl_prop_get_int_ds(ds
,
506 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
510 ds
->ds_reserved
= ds
->ds_quota
= 0;
513 dmu_buf_init_user(&ds
->ds_dbu
, dsl_dataset_evict
, &ds
->ds_dbuf
);
515 winner
= dmu_buf_set_user_ie(dbuf
, &ds
->ds_dbu
);
517 if (err
!= 0 || winner
!= NULL
) {
518 bplist_destroy(&ds
->ds_pending_deadlist
);
519 dsl_deadlist_close(&ds
->ds_deadlist
);
521 dsl_dataset_rele(ds
->ds_prev
, ds
);
522 dsl_dir_rele(ds
->ds_dir
, ds
);
523 mutex_destroy(&ds
->ds_lock
);
524 mutex_destroy(&ds
->ds_opening_lock
);
525 mutex_destroy(&ds
->ds_sendstream_lock
);
526 refcount_destroy(&ds
->ds_longholds
);
527 kmem_free(ds
, sizeof (dsl_dataset_t
));
529 dmu_buf_rele(dbuf
, tag
);
535 unique_insert(dsl_dataset_phys(ds
)->ds_fsid_guid
);
538 ASSERT3P(ds
->ds_dbuf
, ==, dbuf
);
539 ASSERT3P(dsl_dataset_phys(ds
), ==, dbuf
->db_data
);
540 ASSERT(dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0 ||
541 spa_version(dp
->dp_spa
) < SPA_VERSION_ORIGIN
||
542 dp
->dp_origin_snap
== NULL
|| ds
== dp
->dp_origin_snap
);
548 dsl_dataset_hold(dsl_pool_t
*dp
, const char *name
,
549 void *tag
, dsl_dataset_t
**dsp
)
552 const char *snapname
;
557 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, &snapname
);
561 ASSERT(dsl_pool_config_held(dp
));
562 obj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
564 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
566 err
= SET_ERROR(ENOENT
);
568 /* we may be looking for a snapshot */
569 if (err
== 0 && snapname
!= NULL
) {
570 dsl_dataset_t
*snap_ds
;
572 if (*snapname
++ != '@') {
573 dsl_dataset_rele(ds
, tag
);
574 dsl_dir_rele(dd
, FTAG
);
575 return (SET_ERROR(ENOENT
));
578 dprintf("looking for snapshot '%s'\n", snapname
);
579 err
= dsl_dataset_snap_lookup(ds
, snapname
, &obj
);
581 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &snap_ds
);
582 dsl_dataset_rele(ds
, tag
);
585 mutex_enter(&snap_ds
->ds_lock
);
586 if (snap_ds
->ds_snapname
[0] == 0)
587 (void) strlcpy(snap_ds
->ds_snapname
, snapname
,
588 sizeof (snap_ds
->ds_snapname
));
589 mutex_exit(&snap_ds
->ds_lock
);
595 dsl_dir_rele(dd
, FTAG
);
600 dsl_dataset_own_obj(dsl_pool_t
*dp
, uint64_t dsobj
,
601 void *tag
, dsl_dataset_t
**dsp
)
603 int err
= dsl_dataset_hold_obj(dp
, dsobj
, tag
, dsp
);
606 if (!dsl_dataset_tryown(*dsp
, tag
)) {
607 dsl_dataset_rele(*dsp
, tag
);
609 return (SET_ERROR(EBUSY
));
615 dsl_dataset_own(dsl_pool_t
*dp
, const char *name
,
616 void *tag
, dsl_dataset_t
**dsp
)
618 int err
= dsl_dataset_hold(dp
, name
, tag
, dsp
);
621 if (!dsl_dataset_tryown(*dsp
, tag
)) {
622 dsl_dataset_rele(*dsp
, tag
);
623 return (SET_ERROR(EBUSY
));
629 * See the comment above dsl_pool_hold() for details. In summary, a long
630 * hold is used to prevent destruction of a dataset while the pool hold
631 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
633 * The dataset and pool must be held when this function is called. After it
634 * is called, the pool hold may be released while the dataset is still held
638 dsl_dataset_long_hold(dsl_dataset_t
*ds
, void *tag
)
640 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
641 (void) refcount_add(&ds
->ds_longholds
, tag
);
645 dsl_dataset_long_rele(dsl_dataset_t
*ds
, void *tag
)
647 (void) refcount_remove(&ds
->ds_longholds
, tag
);
650 /* Return B_TRUE if there are any long holds on this dataset. */
652 dsl_dataset_long_held(dsl_dataset_t
*ds
)
654 return (!refcount_is_zero(&ds
->ds_longholds
));
658 dsl_dataset_name(dsl_dataset_t
*ds
, char *name
)
661 (void) strcpy(name
, "mos");
663 dsl_dir_name(ds
->ds_dir
, name
);
664 VERIFY0(dsl_dataset_get_snapname(ds
));
665 if (ds
->ds_snapname
[0]) {
666 (void) strcat(name
, "@");
668 * We use a "recursive" mutex so that we
669 * can call dprintf_ds() with ds_lock held.
671 if (!MUTEX_HELD(&ds
->ds_lock
)) {
672 mutex_enter(&ds
->ds_lock
);
673 (void) strcat(name
, ds
->ds_snapname
);
674 mutex_exit(&ds
->ds_lock
);
676 (void) strcat(name
, ds
->ds_snapname
);
683 dsl_dataset_rele(dsl_dataset_t
*ds
, void *tag
)
685 dmu_buf_rele(ds
->ds_dbuf
, tag
);
689 dsl_dataset_disown(dsl_dataset_t
*ds
, void *tag
)
691 ASSERT3P(ds
->ds_owner
, ==, tag
);
692 ASSERT(ds
->ds_dbuf
!= NULL
);
694 mutex_enter(&ds
->ds_lock
);
696 mutex_exit(&ds
->ds_lock
);
697 dsl_dataset_long_rele(ds
, tag
);
698 dsl_dataset_rele(ds
, tag
);
702 dsl_dataset_tryown(dsl_dataset_t
*ds
, void *tag
)
704 boolean_t gotit
= FALSE
;
706 mutex_enter(&ds
->ds_lock
);
707 if (ds
->ds_owner
== NULL
&& !DS_IS_INCONSISTENT(ds
)) {
709 dsl_dataset_long_hold(ds
, tag
);
712 mutex_exit(&ds
->ds_lock
);
717 dsl_dataset_activate_feature(uint64_t dsobj
, spa_feature_t f
, dmu_tx_t
*tx
)
719 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
720 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
723 VERIFY(spa_feature_table
[f
].fi_flags
& ZFEATURE_FLAG_PER_DATASET
);
725 spa_feature_incr(spa
, f
, tx
);
726 dmu_object_zapify(mos
, dsobj
, DMU_OT_DSL_DATASET
, tx
);
728 VERIFY0(zap_add(mos
, dsobj
, spa_feature_table
[f
].fi_guid
,
729 sizeof (zero
), 1, &zero
, tx
));
733 dsl_dataset_deactivate_feature(uint64_t dsobj
, spa_feature_t f
, dmu_tx_t
*tx
)
735 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
736 objset_t
*mos
= dmu_tx_pool(tx
)->dp_meta_objset
;
738 VERIFY(spa_feature_table
[f
].fi_flags
& ZFEATURE_FLAG_PER_DATASET
);
740 VERIFY0(zap_remove(mos
, dsobj
, spa_feature_table
[f
].fi_guid
, tx
));
741 spa_feature_decr(spa
, f
, tx
);
745 dsl_dataset_create_sync_dd(dsl_dir_t
*dd
, dsl_dataset_t
*origin
,
746 uint64_t flags
, dmu_tx_t
*tx
)
748 dsl_pool_t
*dp
= dd
->dd_pool
;
750 dsl_dataset_phys_t
*dsphys
;
752 objset_t
*mos
= dp
->dp_meta_objset
;
755 origin
= dp
->dp_origin_snap
;
757 ASSERT(origin
== NULL
|| origin
->ds_dir
->dd_pool
== dp
);
758 ASSERT(origin
== NULL
|| dsl_dataset_phys(origin
)->ds_num_children
> 0);
759 ASSERT(dmu_tx_is_syncing(tx
));
760 ASSERT(dsl_dir_phys(dd
)->dd_head_dataset_obj
== 0);
762 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
763 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
764 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
765 dmu_buf_will_dirty(dbuf
, tx
);
766 dsphys
= dbuf
->db_data
;
767 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
768 dsphys
->ds_dir_obj
= dd
->dd_object
;
769 dsphys
->ds_flags
= flags
;
770 dsphys
->ds_fsid_guid
= unique_create();
771 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
772 sizeof (dsphys
->ds_guid
));
773 dsphys
->ds_snapnames_zapobj
=
774 zap_create_norm(mos
, U8_TEXTPREP_TOUPPER
, DMU_OT_DSL_DS_SNAP_MAP
,
776 dsphys
->ds_creation_time
= gethrestime_sec();
777 dsphys
->ds_creation_txg
= tx
->tx_txg
== TXG_INITIAL
? 1 : tx
->tx_txg
;
779 if (origin
== NULL
) {
780 dsphys
->ds_deadlist_obj
= dsl_deadlist_alloc(mos
, tx
);
783 dsl_dataset_t
*ohds
; /* head of the origin snapshot */
785 dsphys
->ds_prev_snap_obj
= origin
->ds_object
;
786 dsphys
->ds_prev_snap_txg
=
787 dsl_dataset_phys(origin
)->ds_creation_txg
;
788 dsphys
->ds_referenced_bytes
=
789 dsl_dataset_phys(origin
)->ds_referenced_bytes
;
790 dsphys
->ds_compressed_bytes
=
791 dsl_dataset_phys(origin
)->ds_compressed_bytes
;
792 dsphys
->ds_uncompressed_bytes
=
793 dsl_dataset_phys(origin
)->ds_uncompressed_bytes
;
794 dsphys
->ds_bp
= dsl_dataset_phys(origin
)->ds_bp
;
797 * Inherit flags that describe the dataset's contents
798 * (INCONSISTENT) or properties (Case Insensitive).
800 dsphys
->ds_flags
|= dsl_dataset_phys(origin
)->ds_flags
&
801 (DS_FLAG_INCONSISTENT
| DS_FLAG_CI_DATASET
);
803 for (f
= 0; f
< SPA_FEATURES
; f
++) {
804 if (origin
->ds_feature_inuse
[f
])
805 dsl_dataset_activate_feature(dsobj
, f
, tx
);
808 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
809 dsl_dataset_phys(origin
)->ds_num_children
++;
811 VERIFY0(dsl_dataset_hold_obj(dp
,
812 dsl_dir_phys(origin
->ds_dir
)->dd_head_dataset_obj
,
814 dsphys
->ds_deadlist_obj
= dsl_deadlist_clone(&ohds
->ds_deadlist
,
815 dsphys
->ds_prev_snap_txg
, dsphys
->ds_prev_snap_obj
, tx
);
816 dsl_dataset_rele(ohds
, FTAG
);
818 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_NEXT_CLONES
) {
819 if (dsl_dataset_phys(origin
)->ds_next_clones_obj
== 0) {
820 dsl_dataset_phys(origin
)->ds_next_clones_obj
=
822 DMU_OT_NEXT_CLONES
, DMU_OT_NONE
, 0, tx
);
824 VERIFY0(zap_add_int(mos
,
825 dsl_dataset_phys(origin
)->ds_next_clones_obj
,
829 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
830 dsl_dir_phys(dd
)->dd_origin_obj
= origin
->ds_object
;
831 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
832 if (dsl_dir_phys(origin
->ds_dir
)->dd_clones
== 0) {
833 dmu_buf_will_dirty(origin
->ds_dir
->dd_dbuf
, tx
);
834 dsl_dir_phys(origin
->ds_dir
)->dd_clones
=
836 DMU_OT_DSL_CLONES
, DMU_OT_NONE
, 0, tx
);
838 VERIFY0(zap_add_int(mos
,
839 dsl_dir_phys(origin
->ds_dir
)->dd_clones
,
844 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
845 dsphys
->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
847 dmu_buf_rele(dbuf
, FTAG
);
849 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
850 dsl_dir_phys(dd
)->dd_head_dataset_obj
= dsobj
;
856 dsl_dataset_zero_zil(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
860 VERIFY0(dmu_objset_from_ds(ds
, &os
));
861 bzero(&os
->os_zil_header
, sizeof (os
->os_zil_header
));
862 dsl_dataset_dirty(ds
, tx
);
866 dsl_dataset_create_sync(dsl_dir_t
*pdd
, const char *lastname
,
867 dsl_dataset_t
*origin
, uint64_t flags
, cred_t
*cr
, dmu_tx_t
*tx
)
869 dsl_pool_t
*dp
= pdd
->dd_pool
;
870 uint64_t dsobj
, ddobj
;
873 ASSERT(dmu_tx_is_syncing(tx
));
874 ASSERT(lastname
[0] != '@');
876 ddobj
= dsl_dir_create_sync(dp
, pdd
, lastname
, tx
);
877 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, lastname
, FTAG
, &dd
));
879 dsobj
= dsl_dataset_create_sync_dd(dd
, origin
,
880 flags
& ~DS_CREATE_FLAG_NODIRTY
, tx
);
882 dsl_deleg_set_create_perms(dd
, tx
, cr
);
885 * Since we're creating a new node we know it's a leaf, so we can
886 * initialize the counts if the limit feature is active.
888 if (spa_feature_is_active(dp
->dp_spa
, SPA_FEATURE_FS_SS_LIMIT
)) {
890 objset_t
*os
= dd
->dd_pool
->dp_meta_objset
;
892 dsl_dir_zapify(dd
, tx
);
893 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_FILESYSTEM_COUNT
,
894 sizeof (cnt
), 1, &cnt
, tx
));
895 VERIFY0(zap_add(os
, dd
->dd_object
, DD_FIELD_SNAPSHOT_COUNT
,
896 sizeof (cnt
), 1, &cnt
, tx
));
899 dsl_dir_rele(dd
, FTAG
);
902 * If we are creating a clone, make sure we zero out any stale
903 * data from the origin snapshots zil header.
905 if (origin
!= NULL
&& !(flags
& DS_CREATE_FLAG_NODIRTY
)) {
908 VERIFY0(dsl_dataset_hold_obj(dp
, dsobj
, FTAG
, &ds
));
909 dsl_dataset_zero_zil(ds
, tx
);
910 dsl_dataset_rele(ds
, FTAG
);
917 * The unique space in the head dataset can be calculated by subtracting
918 * the space used in the most recent snapshot, that is still being used
919 * in this file system, from the space currently in use. To figure out
920 * the space in the most recent snapshot still in use, we need to take
921 * the total space used in the snapshot and subtract out the space that
922 * has been freed up since the snapshot was taken.
925 dsl_dataset_recalc_head_uniq(dsl_dataset_t
*ds
)
928 uint64_t dlused
, dlcomp
, dluncomp
;
930 ASSERT(!ds
->ds_is_snapshot
);
932 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0)
933 mrs_used
= dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
;
937 dsl_deadlist_space(&ds
->ds_deadlist
, &dlused
, &dlcomp
, &dluncomp
);
939 ASSERT3U(dlused
, <=, mrs_used
);
940 dsl_dataset_phys(ds
)->ds_unique_bytes
=
941 dsl_dataset_phys(ds
)->ds_referenced_bytes
- (mrs_used
- dlused
);
943 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) >=
944 SPA_VERSION_UNIQUE_ACCURATE
)
945 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
949 dsl_dataset_remove_from_next_clones(dsl_dataset_t
*ds
, uint64_t obj
,
952 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
954 ASSERTV(uint64_t count
);
956 ASSERT(dsl_dataset_phys(ds
)->ds_num_children
>= 2);
957 err
= zap_remove_int(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
960 * The err should not be ENOENT, but a bug in a previous version
961 * of the code could cause upgrade_clones_cb() to not set
962 * ds_next_snap_obj when it should, leading to a missing entry.
963 * If we knew that the pool was created after
964 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
965 * ENOENT. However, at least we can check that we don't have
966 * too many entries in the next_clones_obj even after failing to
971 ASSERT0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
973 ASSERT3U(count
, <=, dsl_dataset_phys(ds
)->ds_num_children
- 2);
978 dsl_dataset_get_blkptr(dsl_dataset_t
*ds
)
980 return (&dsl_dataset_phys(ds
)->ds_bp
);
984 dsl_dataset_set_blkptr(dsl_dataset_t
*ds
, blkptr_t
*bp
, dmu_tx_t
*tx
)
986 ASSERT(dmu_tx_is_syncing(tx
));
987 /* If it's the meta-objset, set dp_meta_rootbp */
989 tx
->tx_pool
->dp_meta_rootbp
= *bp
;
991 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
992 dsl_dataset_phys(ds
)->ds_bp
= *bp
;
997 dsl_dataset_get_spa(dsl_dataset_t
*ds
)
999 return (ds
->ds_dir
->dd_pool
->dp_spa
);
1003 dsl_dataset_dirty(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1007 if (ds
== NULL
) /* this is the meta-objset */
1010 ASSERT(ds
->ds_objset
!= NULL
);
1012 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
!= 0)
1013 panic("dirtying snapshot!");
1015 dp
= ds
->ds_dir
->dd_pool
;
1017 if (txg_list_add(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
1018 /* up the hold count until we can be written out */
1019 dmu_buf_add_ref(ds
->ds_dbuf
, ds
);
1024 dsl_dataset_is_dirty(dsl_dataset_t
*ds
)
1028 for (t
= 0; t
< TXG_SIZE
; t
++) {
1029 if (txg_list_member(&ds
->ds_dir
->dd_pool
->dp_dirty_datasets
,
1037 dsl_dataset_snapshot_reserve_space(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
1041 if (!dmu_tx_is_syncing(tx
))
1045 * If there's an fs-only reservation, any blocks that might become
1046 * owned by the snapshot dataset must be accommodated by space
1047 * outside of the reservation.
1049 ASSERT(ds
->ds_reserved
== 0 || DS_UNIQUE_IS_ACCURATE(ds
));
1050 asize
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
, ds
->ds_reserved
);
1051 if (asize
> dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
))
1052 return (SET_ERROR(ENOSPC
));
1055 * Propagate any reserved space for this snapshot to other
1056 * snapshot checks in this sync group.
1059 dsl_dir_willuse_space(ds
->ds_dir
, asize
, tx
);
1064 typedef struct dsl_dataset_snapshot_arg
{
1065 nvlist_t
*ddsa_snaps
;
1066 nvlist_t
*ddsa_props
;
1067 nvlist_t
*ddsa_errors
;
1069 } dsl_dataset_snapshot_arg_t
;
1072 dsl_dataset_snapshot_check_impl(dsl_dataset_t
*ds
, const char *snapname
,
1073 dmu_tx_t
*tx
, boolean_t recv
, uint64_t cnt
, cred_t
*cr
)
1078 ds
->ds_trysnap_txg
= tx
->tx_txg
;
1080 if (!dmu_tx_is_syncing(tx
))
1084 * We don't allow multiple snapshots of the same txg. If there
1085 * is already one, try again.
1087 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
>= tx
->tx_txg
)
1088 return (SET_ERROR(EAGAIN
));
1091 * Check for conflicting snapshot name.
1093 error
= dsl_dataset_snap_lookup(ds
, snapname
, &value
);
1095 return (SET_ERROR(EEXIST
));
1096 if (error
!= ENOENT
)
1100 * We don't allow taking snapshots of inconsistent datasets, such as
1101 * those into which we are currently receiving. However, if we are
1102 * creating this snapshot as part of a receive, this check will be
1103 * executed atomically with respect to the completion of the receive
1104 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1105 * case we ignore this, knowing it will be fixed up for us shortly in
1106 * dmu_recv_end_sync().
1108 if (!recv
&& DS_IS_INCONSISTENT(ds
))
1109 return (SET_ERROR(EBUSY
));
1112 * Skip the check for temporary snapshots or if we have already checked
1113 * the counts in dsl_dataset_snapshot_check. This means we really only
1114 * check the count here when we're receiving a stream.
1116 if (cnt
!= 0 && cr
!= NULL
) {
1117 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1118 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
, cr
);
1123 error
= dsl_dataset_snapshot_reserve_space(ds
, tx
);
1131 dsl_dataset_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1133 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1134 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1139 * Pre-compute how many total new snapshots will be created for each
1140 * level in the tree and below. This is needed for validating the
1141 * snapshot limit when either taking a recursive snapshot or when
1142 * taking multiple snapshots.
1144 * The problem is that the counts are not actually adjusted when
1145 * we are checking, only when we finally sync. For a single snapshot,
1146 * this is easy, the count will increase by 1 at each node up the tree,
1147 * but its more complicated for the recursive/multiple snapshot case.
1149 * The dsl_fs_ss_limit_check function does recursively check the count
1150 * at each level up the tree but since it is validating each snapshot
1151 * independently we need to be sure that we are validating the complete
1152 * count for the entire set of snapshots. We do this by rolling up the
1153 * counts for each component of the name into an nvlist and then
1154 * checking each of those cases with the aggregated count.
1156 * This approach properly handles not only the recursive snapshot
1157 * case (where we get all of those on the ddsa_snaps list) but also
1158 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1159 * validate the limit on 'a' using a count of 2).
1161 * We validate the snapshot names in the third loop and only report
1164 if (dmu_tx_is_syncing(tx
)) {
1166 nvlist_t
*cnt_track
= NULL
;
1167 cnt_track
= fnvlist_alloc();
1169 nm
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1171 /* Rollup aggregated counts into the cnt_track list */
1172 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1174 pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1178 (void) strlcpy(nm
, nvpair_name(pair
), MAXPATHLEN
);
1179 pdelim
= strchr(nm
, '@');
1185 if (nvlist_lookup_uint64(cnt_track
, nm
,
1187 /* update existing entry */
1188 fnvlist_add_uint64(cnt_track
, nm
,
1192 fnvlist_add_uint64(cnt_track
, nm
, 1);
1195 pdelim
= strrchr(nm
, '/');
1198 } while (pdelim
!= NULL
);
1201 kmem_free(nm
, MAXPATHLEN
);
1203 /* Check aggregated counts at each level */
1204 for (pair
= nvlist_next_nvpair(cnt_track
, NULL
);
1205 pair
!= NULL
; pair
= nvlist_next_nvpair(cnt_track
, pair
)) {
1211 name
= nvpair_name(pair
);
1212 cnt
= fnvpair_value_uint64(pair
);
1215 error
= dsl_dataset_hold(dp
, name
, FTAG
, &ds
);
1217 error
= dsl_fs_ss_limit_check(ds
->ds_dir
, cnt
,
1218 ZFS_PROP_SNAPSHOT_LIMIT
, NULL
,
1220 dsl_dataset_rele(ds
, FTAG
);
1224 if (ddsa
->ddsa_errors
!= NULL
)
1225 fnvlist_add_int32(ddsa
->ddsa_errors
,
1228 /* only report one error for this check */
1232 nvlist_free(cnt_track
);
1235 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1236 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1240 char dsname
[MAXNAMELEN
];
1242 name
= nvpair_name(pair
);
1243 if (strlen(name
) >= MAXNAMELEN
)
1244 error
= SET_ERROR(ENAMETOOLONG
);
1246 atp
= strchr(name
, '@');
1248 error
= SET_ERROR(EINVAL
);
1250 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1253 error
= dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
);
1255 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1256 error
= dsl_dataset_snapshot_check_impl(ds
,
1257 atp
+ 1, tx
, B_FALSE
, 0, NULL
);
1258 dsl_dataset_rele(ds
, FTAG
);
1262 if (ddsa
->ddsa_errors
!= NULL
) {
1263 fnvlist_add_int32(ddsa
->ddsa_errors
,
1274 dsl_dataset_snapshot_sync_impl(dsl_dataset_t
*ds
, const char *snapname
,
1277 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1279 dsl_dataset_phys_t
*dsphys
;
1280 uint64_t dsobj
, crtxg
;
1281 objset_t
*mos
= dp
->dp_meta_objset
;
1283 ASSERTV(static zil_header_t zero_zil
);
1284 ASSERTV(objset_t
*os
);
1286 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
1289 * If we are on an old pool, the zil must not be active, in which
1290 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1292 ASSERT(spa_version(dmu_tx_pool(tx
)->dp_spa
) >= SPA_VERSION_FAST_SNAP
||
1293 dmu_objset_from_ds(ds
, &os
) != 0 ||
1294 bcmp(&os
->os_phys
->os_zil_header
, &zero_zil
,
1295 sizeof (zero_zil
)) == 0);
1297 dsl_fs_ss_count_adjust(ds
->ds_dir
, 1, DD_FIELD_SNAPSHOT_COUNT
, tx
);
1300 * The origin's ds_creation_txg has to be < TXG_INITIAL
1302 if (strcmp(snapname
, ORIGIN_DIR_NAME
) == 0)
1307 dsobj
= dmu_object_alloc(mos
, DMU_OT_DSL_DATASET
, 0,
1308 DMU_OT_DSL_DATASET
, sizeof (dsl_dataset_phys_t
), tx
);
1309 VERIFY0(dmu_bonus_hold(mos
, dsobj
, FTAG
, &dbuf
));
1310 dmu_buf_will_dirty(dbuf
, tx
);
1311 dsphys
= dbuf
->db_data
;
1312 bzero(dsphys
, sizeof (dsl_dataset_phys_t
));
1313 dsphys
->ds_dir_obj
= ds
->ds_dir
->dd_object
;
1314 dsphys
->ds_fsid_guid
= unique_create();
1315 (void) random_get_pseudo_bytes((void*)&dsphys
->ds_guid
,
1316 sizeof (dsphys
->ds_guid
));
1317 dsphys
->ds_prev_snap_obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
1318 dsphys
->ds_prev_snap_txg
= dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
1319 dsphys
->ds_next_snap_obj
= ds
->ds_object
;
1320 dsphys
->ds_num_children
= 1;
1321 dsphys
->ds_creation_time
= gethrestime_sec();
1322 dsphys
->ds_creation_txg
= crtxg
;
1323 dsphys
->ds_deadlist_obj
= dsl_dataset_phys(ds
)->ds_deadlist_obj
;
1324 dsphys
->ds_referenced_bytes
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1325 dsphys
->ds_compressed_bytes
= dsl_dataset_phys(ds
)->ds_compressed_bytes
;
1326 dsphys
->ds_uncompressed_bytes
=
1327 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
;
1328 dsphys
->ds_flags
= dsl_dataset_phys(ds
)->ds_flags
;
1329 dsphys
->ds_bp
= dsl_dataset_phys(ds
)->ds_bp
;
1330 dmu_buf_rele(dbuf
, FTAG
);
1332 for (f
= 0; f
< SPA_FEATURES
; f
++) {
1333 if (ds
->ds_feature_inuse
[f
])
1334 dsl_dataset_activate_feature(dsobj
, f
, tx
);
1337 ASSERT3U(ds
->ds_prev
!= 0, ==,
1338 dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0);
1340 uint64_t next_clones_obj
=
1341 dsl_dataset_phys(ds
->ds_prev
)->ds_next_clones_obj
;
1342 ASSERT(dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1344 dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
> 1);
1345 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
==
1347 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
1348 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, ==,
1349 dsl_dataset_phys(ds
->ds_prev
)->ds_creation_txg
);
1350 dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
= dsobj
;
1351 } else if (next_clones_obj
!= 0) {
1352 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
1353 dsphys
->ds_next_snap_obj
, tx
);
1354 VERIFY0(zap_add_int(mos
,
1355 next_clones_obj
, dsobj
, tx
));
1360 * If we have a reference-reservation on this dataset, we will
1361 * need to increase the amount of refreservation being charged
1362 * since our unique space is going to zero.
1364 if (ds
->ds_reserved
) {
1366 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
1367 delta
= MIN(dsl_dataset_phys(ds
)->ds_unique_bytes
,
1369 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
,
1373 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1374 dsl_dataset_phys(ds
)->ds_deadlist_obj
=
1375 dsl_deadlist_clone(&ds
->ds_deadlist
, UINT64_MAX
,
1376 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, tx
);
1377 dsl_deadlist_close(&ds
->ds_deadlist
);
1378 dsl_deadlist_open(&ds
->ds_deadlist
, mos
,
1379 dsl_dataset_phys(ds
)->ds_deadlist_obj
);
1380 dsl_deadlist_add_key(&ds
->ds_deadlist
,
1381 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
);
1383 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, <, tx
->tx_txg
);
1384 dsl_dataset_phys(ds
)->ds_prev_snap_obj
= dsobj
;
1385 dsl_dataset_phys(ds
)->ds_prev_snap_txg
= crtxg
;
1386 dsl_dataset_phys(ds
)->ds_unique_bytes
= 0;
1387 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_UNIQUE_ACCURATE
)
1388 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_UNIQUE_ACCURATE
;
1390 VERIFY0(zap_add(mos
, dsl_dataset_phys(ds
)->ds_snapnames_zapobj
,
1391 snapname
, 8, 1, &dsobj
, tx
));
1394 dsl_dataset_rele(ds
->ds_prev
, ds
);
1395 VERIFY0(dsl_dataset_hold_obj(dp
,
1396 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, ds
, &ds
->ds_prev
));
1398 dsl_scan_ds_snapshotted(ds
, tx
);
1400 dsl_dir_snap_cmtime_update(ds
->ds_dir
);
1402 spa_history_log_internal_ds(ds
->ds_prev
, "snapshot", tx
, "");
1406 dsl_dataset_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1408 dsl_dataset_snapshot_arg_t
*ddsa
= arg
;
1409 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1412 for (pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, NULL
);
1413 pair
!= NULL
; pair
= nvlist_next_nvpair(ddsa
->ddsa_snaps
, pair
)) {
1416 char dsname
[MAXNAMELEN
];
1418 name
= nvpair_name(pair
);
1419 atp
= strchr(name
, '@');
1420 (void) strlcpy(dsname
, name
, atp
- name
+ 1);
1421 VERIFY0(dsl_dataset_hold(dp
, dsname
, FTAG
, &ds
));
1423 dsl_dataset_snapshot_sync_impl(ds
, atp
+ 1, tx
);
1424 if (ddsa
->ddsa_props
!= NULL
) {
1425 dsl_props_set_sync_impl(ds
->ds_prev
,
1426 ZPROP_SRC_LOCAL
, ddsa
->ddsa_props
, tx
);
1428 zvol_create_minors(dp
->dp_spa
, nvpair_name(pair
), B_TRUE
);
1429 dsl_dataset_rele(ds
, FTAG
);
1434 * The snapshots must all be in the same pool.
1435 * All-or-nothing: if there are any failures, nothing will be modified.
1438 dsl_dataset_snapshot(nvlist_t
*snaps
, nvlist_t
*props
, nvlist_t
*errors
)
1440 dsl_dataset_snapshot_arg_t ddsa
;
1442 boolean_t needsuspend
;
1446 nvlist_t
*suspended
= NULL
;
1448 pair
= nvlist_next_nvpair(snaps
, NULL
);
1451 firstname
= nvpair_name(pair
);
1453 error
= spa_open(firstname
, &spa
, FTAG
);
1456 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1457 spa_close(spa
, FTAG
);
1460 suspended
= fnvlist_alloc();
1461 for (pair
= nvlist_next_nvpair(snaps
, NULL
); pair
!= NULL
;
1462 pair
= nvlist_next_nvpair(snaps
, pair
)) {
1463 char fsname
[MAXNAMELEN
];
1464 char *snapname
= nvpair_name(pair
);
1468 atp
= strchr(snapname
, '@');
1470 error
= SET_ERROR(EINVAL
);
1473 (void) strlcpy(fsname
, snapname
, atp
- snapname
+ 1);
1475 error
= zil_suspend(fsname
, &cookie
);
1478 fnvlist_add_uint64(suspended
, fsname
,
1483 ddsa
.ddsa_snaps
= snaps
;
1484 ddsa
.ddsa_props
= props
;
1485 ddsa
.ddsa_errors
= errors
;
1486 ddsa
.ddsa_cr
= CRED();
1489 error
= dsl_sync_task(firstname
, dsl_dataset_snapshot_check
,
1490 dsl_dataset_snapshot_sync
, &ddsa
,
1491 fnvlist_num_pairs(snaps
) * 3, ZFS_SPACE_CHECK_NORMAL
);
1494 if (suspended
!= NULL
) {
1495 for (pair
= nvlist_next_nvpair(suspended
, NULL
); pair
!= NULL
;
1496 pair
= nvlist_next_nvpair(suspended
, pair
)) {
1497 zil_resume((void *)(uintptr_t)
1498 fnvpair_value_uint64(pair
));
1500 fnvlist_free(suspended
);
1506 typedef struct dsl_dataset_snapshot_tmp_arg
{
1507 const char *ddsta_fsname
;
1508 const char *ddsta_snapname
;
1509 minor_t ddsta_cleanup_minor
;
1510 const char *ddsta_htag
;
1511 } dsl_dataset_snapshot_tmp_arg_t
;
1514 dsl_dataset_snapshot_tmp_check(void *arg
, dmu_tx_t
*tx
)
1516 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1517 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1521 error
= dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
);
1525 /* NULL cred means no limit check for tmp snapshot */
1526 error
= dsl_dataset_snapshot_check_impl(ds
, ddsta
->ddsta_snapname
,
1527 tx
, B_FALSE
, 0, NULL
);
1529 dsl_dataset_rele(ds
, FTAG
);
1533 if (spa_version(dp
->dp_spa
) < SPA_VERSION_USERREFS
) {
1534 dsl_dataset_rele(ds
, FTAG
);
1535 return (SET_ERROR(ENOTSUP
));
1537 error
= dsl_dataset_user_hold_check_one(NULL
, ddsta
->ddsta_htag
,
1540 dsl_dataset_rele(ds
, FTAG
);
1544 dsl_dataset_rele(ds
, FTAG
);
1549 dsl_dataset_snapshot_tmp_sync(void *arg
, dmu_tx_t
*tx
)
1551 dsl_dataset_snapshot_tmp_arg_t
*ddsta
= arg
;
1552 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1555 VERIFY0(dsl_dataset_hold(dp
, ddsta
->ddsta_fsname
, FTAG
, &ds
));
1557 dsl_dataset_snapshot_sync_impl(ds
, ddsta
->ddsta_snapname
, tx
);
1558 dsl_dataset_user_hold_sync_one(ds
->ds_prev
, ddsta
->ddsta_htag
,
1559 ddsta
->ddsta_cleanup_minor
, gethrestime_sec(), tx
);
1560 dsl_destroy_snapshot_sync_impl(ds
->ds_prev
, B_TRUE
, tx
);
1562 dsl_dataset_rele(ds
, FTAG
);
1566 dsl_dataset_snapshot_tmp(const char *fsname
, const char *snapname
,
1567 minor_t cleanup_minor
, const char *htag
)
1569 dsl_dataset_snapshot_tmp_arg_t ddsta
;
1572 boolean_t needsuspend
;
1575 ddsta
.ddsta_fsname
= fsname
;
1576 ddsta
.ddsta_snapname
= snapname
;
1577 ddsta
.ddsta_cleanup_minor
= cleanup_minor
;
1578 ddsta
.ddsta_htag
= htag
;
1580 error
= spa_open(fsname
, &spa
, FTAG
);
1583 needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
1584 spa_close(spa
, FTAG
);
1587 error
= zil_suspend(fsname
, &cookie
);
1592 error
= dsl_sync_task(fsname
, dsl_dataset_snapshot_tmp_check
,
1593 dsl_dataset_snapshot_tmp_sync
, &ddsta
, 3, ZFS_SPACE_CHECK_RESERVED
);
1602 dsl_dataset_sync(dsl_dataset_t
*ds
, zio_t
*zio
, dmu_tx_t
*tx
)
1606 ASSERT(dmu_tx_is_syncing(tx
));
1607 ASSERT(ds
->ds_objset
!= NULL
);
1608 ASSERT(dsl_dataset_phys(ds
)->ds_next_snap_obj
== 0);
1611 * in case we had to change ds_fsid_guid when we opened it,
1614 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
1615 dsl_dataset_phys(ds
)->ds_fsid_guid
= ds
->ds_fsid_guid
;
1617 dmu_objset_sync(ds
->ds_objset
, zio
, tx
);
1619 for (f
= 0; f
< SPA_FEATURES
; f
++) {
1620 if (ds
->ds_feature_activation_needed
[f
]) {
1621 if (ds
->ds_feature_inuse
[f
])
1623 dsl_dataset_activate_feature(ds
->ds_object
, f
, tx
);
1624 ds
->ds_feature_inuse
[f
] = B_TRUE
;
1630 get_clones_stat(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1633 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
1636 nvlist_t
*propval
= fnvlist_alloc();
1637 nvlist_t
*val
= fnvlist_alloc();
1639 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1642 * There may be missing entries in ds_next_clones_obj
1643 * due to a bug in a previous version of the code.
1644 * Only trust it if it has the right number of entries.
1646 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
!= 0) {
1647 VERIFY0(zap_count(mos
, dsl_dataset_phys(ds
)->ds_next_clones_obj
,
1650 if (count
!= dsl_dataset_phys(ds
)->ds_num_children
- 1)
1652 for (zap_cursor_init(&zc
, mos
,
1653 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
1654 zap_cursor_retrieve(&zc
, &za
) == 0;
1655 zap_cursor_advance(&zc
)) {
1656 dsl_dataset_t
*clone
;
1657 char buf
[ZFS_MAXNAMELEN
];
1658 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
1659 za
.za_first_integer
, FTAG
, &clone
));
1660 dsl_dir_name(clone
->ds_dir
, buf
);
1661 fnvlist_add_boolean(val
, buf
);
1662 dsl_dataset_rele(clone
, FTAG
);
1664 zap_cursor_fini(&zc
);
1665 fnvlist_add_nvlist(propval
, ZPROP_VALUE
, val
);
1666 fnvlist_add_nvlist(nv
, zfs_prop_to_name(ZFS_PROP_CLONES
), propval
);
1669 nvlist_free(propval
);
1673 dsl_dataset_stats(dsl_dataset_t
*ds
, nvlist_t
*nv
)
1675 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1676 uint64_t refd
, avail
, uobjs
, aobjs
, ratio
;
1678 ASSERT(dsl_pool_config_held(dp
));
1680 ratio
= dsl_dataset_phys(ds
)->ds_compressed_bytes
== 0 ? 100 :
1681 (dsl_dataset_phys(ds
)->ds_uncompressed_bytes
* 100 /
1682 dsl_dataset_phys(ds
)->ds_compressed_bytes
);
1684 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRATIO
, ratio
);
1685 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_LOGICALREFERENCED
,
1686 dsl_dataset_phys(ds
)->ds_uncompressed_bytes
);
1688 if (ds
->ds_is_snapshot
) {
1689 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_COMPRESSRATIO
, ratio
);
1690 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USED
,
1691 dsl_dataset_phys(ds
)->ds_unique_bytes
);
1692 get_clones_stat(ds
, nv
);
1694 if (ds
->ds_prev
!= NULL
&& ds
->ds_prev
!= dp
->dp_origin_snap
) {
1695 char buf
[MAXNAMELEN
];
1696 dsl_dataset_name(ds
->ds_prev
, buf
);
1697 dsl_prop_nvlist_add_string(nv
, ZFS_PROP_PREV_SNAP
, buf
);
1700 dsl_dir_stats(ds
->ds_dir
, nv
);
1703 dsl_dataset_space(ds
, &refd
, &avail
, &uobjs
, &aobjs
);
1704 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_AVAILABLE
, avail
);
1705 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFERENCED
, refd
);
1707 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATION
,
1708 dsl_dataset_phys(ds
)->ds_creation_time
);
1709 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_CREATETXG
,
1710 dsl_dataset_phys(ds
)->ds_creation_txg
);
1711 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFQUOTA
,
1713 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_REFRESERVATION
,
1715 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_GUID
,
1716 dsl_dataset_phys(ds
)->ds_guid
);
1717 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_UNIQUE
,
1718 dsl_dataset_phys(ds
)->ds_unique_bytes
);
1719 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_OBJSETID
,
1721 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERREFS
,
1723 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_DEFER_DESTROY
,
1724 DS_IS_DEFER_DESTROY(ds
) ? 1 : 0);
1726 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
1727 uint64_t written
, comp
, uncomp
;
1728 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1729 dsl_dataset_t
*prev
;
1732 err
= dsl_dataset_hold_obj(dp
,
1733 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &prev
);
1735 err
= dsl_dataset_space_written(prev
, ds
, &written
,
1737 dsl_dataset_rele(prev
, FTAG
);
1739 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_WRITTEN
,
1748 dsl_dataset_fast_stat(dsl_dataset_t
*ds
, dmu_objset_stats_t
*stat
)
1750 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
1751 ASSERT(dsl_pool_config_held(dp
));
1753 stat
->dds_creation_txg
= dsl_dataset_phys(ds
)->ds_creation_txg
;
1754 stat
->dds_inconsistent
=
1755 dsl_dataset_phys(ds
)->ds_flags
& DS_FLAG_INCONSISTENT
;
1756 stat
->dds_guid
= dsl_dataset_phys(ds
)->ds_guid
;
1757 stat
->dds_origin
[0] = '\0';
1758 if (ds
->ds_is_snapshot
) {
1759 stat
->dds_is_snapshot
= B_TRUE
;
1760 stat
->dds_num_clones
=
1761 dsl_dataset_phys(ds
)->ds_num_children
- 1;
1763 stat
->dds_is_snapshot
= B_FALSE
;
1764 stat
->dds_num_clones
= 0;
1766 if (dsl_dir_is_clone(ds
->ds_dir
)) {
1769 VERIFY0(dsl_dataset_hold_obj(dp
,
1770 dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
,
1772 dsl_dataset_name(ods
, stat
->dds_origin
);
1773 dsl_dataset_rele(ods
, FTAG
);
1779 dsl_dataset_fsid_guid(dsl_dataset_t
*ds
)
1781 return (ds
->ds_fsid_guid
);
1785 dsl_dataset_space(dsl_dataset_t
*ds
,
1786 uint64_t *refdbytesp
, uint64_t *availbytesp
,
1787 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1789 *refdbytesp
= dsl_dataset_phys(ds
)->ds_referenced_bytes
;
1790 *availbytesp
= dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
);
1791 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
)
1793 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
;
1794 if (ds
->ds_quota
!= 0) {
1796 * Adjust available bytes according to refquota
1798 if (*refdbytesp
< ds
->ds_quota
)
1799 *availbytesp
= MIN(*availbytesp
,
1800 ds
->ds_quota
- *refdbytesp
);
1804 *usedobjsp
= BP_GET_FILL(&dsl_dataset_phys(ds
)->ds_bp
);
1805 *availobjsp
= DN_MAX_OBJECT
- *usedobjsp
;
1809 dsl_dataset_modified_since_snap(dsl_dataset_t
*ds
, dsl_dataset_t
*snap
)
1811 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
1814 if (dsl_dataset_phys(ds
)->ds_bp
.blk_birth
>
1815 dsl_dataset_phys(snap
)->ds_creation_txg
) {
1816 objset_t
*os
, *os_snap
;
1818 * It may be that only the ZIL differs, because it was
1819 * reset in the head. Don't count that as being
1822 if (dmu_objset_from_ds(ds
, &os
) != 0)
1824 if (dmu_objset_from_ds(snap
, &os_snap
) != 0)
1826 return (bcmp(&os
->os_phys
->os_meta_dnode
,
1827 &os_snap
->os_phys
->os_meta_dnode
,
1828 sizeof (os
->os_phys
->os_meta_dnode
)) != 0);
1833 typedef struct dsl_dataset_rename_snapshot_arg
{
1834 const char *ddrsa_fsname
;
1835 const char *ddrsa_oldsnapname
;
1836 const char *ddrsa_newsnapname
;
1837 boolean_t ddrsa_recursive
;
1839 } dsl_dataset_rename_snapshot_arg_t
;
1843 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t
*dp
,
1844 dsl_dataset_t
*hds
, void *arg
)
1846 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1850 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
1852 /* ignore nonexistent snapshots */
1853 return (error
== ENOENT
? 0 : error
);
1856 /* new name should not exist */
1857 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_newsnapname
, &val
);
1859 error
= SET_ERROR(EEXIST
);
1860 else if (error
== ENOENT
)
1863 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1864 if (dsl_dir_namelen(hds
->ds_dir
) + 1 +
1865 strlen(ddrsa
->ddrsa_newsnapname
) >= MAXNAMELEN
)
1866 error
= SET_ERROR(ENAMETOOLONG
);
1872 dsl_dataset_rename_snapshot_check(void *arg
, dmu_tx_t
*tx
)
1874 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1875 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1879 error
= dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
);
1883 if (ddrsa
->ddrsa_recursive
) {
1884 error
= dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
1885 dsl_dataset_rename_snapshot_check_impl
, ddrsa
,
1888 error
= dsl_dataset_rename_snapshot_check_impl(dp
, hds
, ddrsa
);
1890 dsl_dataset_rele(hds
, FTAG
);
1895 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t
*dp
,
1896 dsl_dataset_t
*hds
, void *arg
)
1898 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1901 dmu_tx_t
*tx
= ddrsa
->ddrsa_tx
;
1904 error
= dsl_dataset_snap_lookup(hds
, ddrsa
->ddrsa_oldsnapname
, &val
);
1905 ASSERT(error
== 0 || error
== ENOENT
);
1906 if (error
== ENOENT
) {
1907 /* ignore nonexistent snapshots */
1911 VERIFY0(dsl_dataset_hold_obj(dp
, val
, FTAG
, &ds
));
1913 /* log before we change the name */
1914 spa_history_log_internal_ds(ds
, "rename", tx
,
1915 "-> @%s", ddrsa
->ddrsa_newsnapname
);
1917 VERIFY0(dsl_dataset_snap_remove(hds
, ddrsa
->ddrsa_oldsnapname
, tx
,
1919 mutex_enter(&ds
->ds_lock
);
1920 (void) strcpy(ds
->ds_snapname
, ddrsa
->ddrsa_newsnapname
);
1921 mutex_exit(&ds
->ds_lock
);
1922 VERIFY0(zap_add(dp
->dp_meta_objset
,
1923 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
,
1924 ds
->ds_snapname
, 8, 1, &ds
->ds_object
, tx
));
1925 zvol_rename_minors(dp
->dp_spa
, ddrsa
->ddrsa_oldsnapname
,
1926 ddrsa
->ddrsa_newsnapname
, B_TRUE
);
1928 dsl_dataset_rele(ds
, FTAG
);
1933 dsl_dataset_rename_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
1935 dsl_dataset_rename_snapshot_arg_t
*ddrsa
= arg
;
1936 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1939 VERIFY0(dsl_dataset_hold(dp
, ddrsa
->ddrsa_fsname
, FTAG
, &hds
));
1940 ddrsa
->ddrsa_tx
= tx
;
1941 if (ddrsa
->ddrsa_recursive
) {
1942 VERIFY0(dmu_objset_find_dp(dp
, hds
->ds_dir
->dd_object
,
1943 dsl_dataset_rename_snapshot_sync_impl
, ddrsa
,
1946 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp
, hds
, ddrsa
));
1948 dsl_dataset_rele(hds
, FTAG
);
1952 dsl_dataset_rename_snapshot(const char *fsname
,
1953 const char *oldsnapname
, const char *newsnapname
, boolean_t recursive
)
1955 dsl_dataset_rename_snapshot_arg_t ddrsa
;
1957 ddrsa
.ddrsa_fsname
= fsname
;
1958 ddrsa
.ddrsa_oldsnapname
= oldsnapname
;
1959 ddrsa
.ddrsa_newsnapname
= newsnapname
;
1960 ddrsa
.ddrsa_recursive
= recursive
;
1962 return (dsl_sync_task(fsname
, dsl_dataset_rename_snapshot_check
,
1963 dsl_dataset_rename_snapshot_sync
, &ddrsa
,
1964 1, ZFS_SPACE_CHECK_RESERVED
));
1968 * If we're doing an ownership handoff, we need to make sure that there is
1969 * only one long hold on the dataset. We're not allowed to change anything here
1970 * so we don't permanently release the long hold or regular hold here. We want
1971 * to do this only when syncing to avoid the dataset unexpectedly going away
1972 * when we release the long hold.
1975 dsl_dataset_handoff_check(dsl_dataset_t
*ds
, void *owner
, dmu_tx_t
*tx
)
1979 if (!dmu_tx_is_syncing(tx
))
1982 if (owner
!= NULL
) {
1983 VERIFY3P(ds
->ds_owner
, ==, owner
);
1984 dsl_dataset_long_rele(ds
, owner
);
1987 held
= dsl_dataset_long_held(ds
);
1990 dsl_dataset_long_hold(ds
, owner
);
1993 return (SET_ERROR(EBUSY
));
1998 typedef struct dsl_dataset_rollback_arg
{
1999 const char *ddra_fsname
;
2001 nvlist_t
*ddra_result
;
2002 } dsl_dataset_rollback_arg_t
;
2005 dsl_dataset_rollback_check(void *arg
, dmu_tx_t
*tx
)
2007 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2008 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2010 int64_t unused_refres_delta
;
2013 nvlist_t
*proprequest
, *bookmarks
;
2015 error
= dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
);
2019 /* must not be a snapshot */
2020 if (ds
->ds_is_snapshot
) {
2021 dsl_dataset_rele(ds
, FTAG
);
2022 return (SET_ERROR(EINVAL
));
2025 /* must have a most recent snapshot */
2026 if (dsl_dataset_phys(ds
)->ds_prev_snap_txg
< TXG_INITIAL
) {
2027 dsl_dataset_rele(ds
, FTAG
);
2028 return (SET_ERROR(EINVAL
));
2031 /* must not have any bookmarks after the most recent snapshot */
2032 proprequest
= fnvlist_alloc();
2033 fnvlist_add_boolean(proprequest
, zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2034 bookmarks
= fnvlist_alloc();
2035 error
= dsl_get_bookmarks_impl(ds
, proprequest
, bookmarks
);
2036 fnvlist_free(proprequest
);
2039 for (pair
= nvlist_next_nvpair(bookmarks
, NULL
);
2040 pair
!= NULL
; pair
= nvlist_next_nvpair(bookmarks
, pair
)) {
2042 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair
),
2043 zfs_prop_to_name(ZFS_PROP_CREATETXG
));
2044 uint64_t createtxg
= fnvlist_lookup_uint64(valuenv
, "value");
2045 if (createtxg
> dsl_dataset_phys(ds
)->ds_prev_snap_txg
) {
2046 fnvlist_free(bookmarks
);
2047 dsl_dataset_rele(ds
, FTAG
);
2048 return (SET_ERROR(EEXIST
));
2051 fnvlist_free(bookmarks
);
2053 error
= dsl_dataset_handoff_check(ds
, ddra
->ddra_owner
, tx
);
2055 dsl_dataset_rele(ds
, FTAG
);
2060 * Check if the snap we are rolling back to uses more than
2063 if (ds
->ds_quota
!= 0 &&
2064 dsl_dataset_phys(ds
->ds_prev
)->ds_referenced_bytes
> ds
->ds_quota
) {
2065 dsl_dataset_rele(ds
, FTAG
);
2066 return (SET_ERROR(EDQUOT
));
2070 * When we do the clone swap, we will temporarily use more space
2071 * due to the refreservation (the head will no longer have any
2072 * unique space, so the entire amount of the refreservation will need
2073 * to be free). We will immediately destroy the clone, freeing
2074 * this space, but the freeing happens over many txg's.
2076 unused_refres_delta
= (int64_t)MIN(ds
->ds_reserved
,
2077 dsl_dataset_phys(ds
)->ds_unique_bytes
);
2079 if (unused_refres_delta
> 0 &&
2080 unused_refres_delta
>
2081 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, TRUE
)) {
2082 dsl_dataset_rele(ds
, FTAG
);
2083 return (SET_ERROR(ENOSPC
));
2086 dsl_dataset_rele(ds
, FTAG
);
2091 dsl_dataset_rollback_sync(void *arg
, dmu_tx_t
*tx
)
2093 dsl_dataset_rollback_arg_t
*ddra
= arg
;
2094 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2095 dsl_dataset_t
*ds
, *clone
;
2097 char namebuf
[ZFS_MAXNAMELEN
];
2099 VERIFY0(dsl_dataset_hold(dp
, ddra
->ddra_fsname
, FTAG
, &ds
));
2101 dsl_dataset_name(ds
->ds_prev
, namebuf
);
2102 fnvlist_add_string(ddra
->ddra_result
, "target", namebuf
);
2104 cloneobj
= dsl_dataset_create_sync(ds
->ds_dir
, "%rollback",
2105 ds
->ds_prev
, DS_CREATE_FLAG_NODIRTY
, kcred
, tx
);
2107 VERIFY0(dsl_dataset_hold_obj(dp
, cloneobj
, FTAG
, &clone
));
2109 dsl_dataset_clone_swap_sync_impl(clone
, ds
, tx
);
2110 dsl_dataset_zero_zil(ds
, tx
);
2112 dsl_destroy_head_sync_impl(clone
, tx
);
2114 dsl_dataset_rele(clone
, FTAG
);
2115 dsl_dataset_rele(ds
, FTAG
);
2119 * Rolls back the given filesystem or volume to the most recent snapshot.
2120 * The name of the most recent snapshot will be returned under key "target"
2121 * in the result nvlist.
2124 * - The existing dataset MUST be owned by the specified owner at entry
2125 * - Upon return, dataset will still be held by the same owner, whether we
2128 * This mode is required any time the existing filesystem is mounted. See
2129 * notes above zfs_suspend_fs() for further details.
2132 dsl_dataset_rollback(const char *fsname
, void *owner
, nvlist_t
*result
)
2134 dsl_dataset_rollback_arg_t ddra
;
2136 ddra
.ddra_fsname
= fsname
;
2137 ddra
.ddra_owner
= owner
;
2138 ddra
.ddra_result
= result
;
2140 return (dsl_sync_task(fsname
, dsl_dataset_rollback_check
,
2141 dsl_dataset_rollback_sync
, &ddra
,
2142 1, ZFS_SPACE_CHECK_RESERVED
));
2145 struct promotenode
{
2150 typedef struct dsl_dataset_promote_arg
{
2151 const char *ddpa_clonename
;
2152 dsl_dataset_t
*ddpa_clone
;
2153 list_t shared_snaps
, origin_snaps
, clone_snaps
;
2154 dsl_dataset_t
*origin_origin
; /* origin of the origin */
2155 uint64_t used
, comp
, uncomp
, unique
, cloneusedsnap
, originusedsnap
;
2158 } dsl_dataset_promote_arg_t
;
2160 static int snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
);
2161 static int promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
,
2163 static void promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
);
2166 dsl_dataset_promote_check(void *arg
, dmu_tx_t
*tx
)
2168 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2169 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2171 struct promotenode
*snap
;
2172 dsl_dataset_t
*origin_ds
;
2176 size_t max_snap_len
;
2178 err
= promote_hold(ddpa
, dp
, FTAG
);
2182 hds
= ddpa
->ddpa_clone
;
2183 max_snap_len
= MAXNAMELEN
- strlen(ddpa
->ddpa_clonename
) - 1;
2185 if (dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
) {
2186 promote_rele(ddpa
, FTAG
);
2187 return (SET_ERROR(EXDEV
));
2191 * Compute and check the amount of space to transfer. Since this is
2192 * so expensive, don't do the preliminary check.
2194 if (!dmu_tx_is_syncing(tx
)) {
2195 promote_rele(ddpa
, FTAG
);
2199 snap
= list_head(&ddpa
->shared_snaps
);
2200 origin_ds
= snap
->ds
;
2202 /* compute origin's new unique space */
2203 snap
= list_tail(&ddpa
->clone_snaps
);
2204 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2205 origin_ds
->ds_object
);
2206 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2207 dsl_dataset_phys(origin_ds
)->ds_prev_snap_txg
, UINT64_MAX
,
2208 &ddpa
->unique
, &unused
, &unused
);
2211 * Walk the snapshots that we are moving
2213 * Compute space to transfer. Consider the incremental changes
2214 * to used by each snapshot:
2215 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2216 * So each snapshot gave birth to:
2217 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2218 * So a sequence would look like:
2219 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2220 * Which simplifies to:
2221 * uN + kN + kN-1 + ... + k1 + k0
2222 * Note however, if we stop before we reach the ORIGIN we get:
2223 * uN + kN + kN-1 + ... + kM - uM-1
2226 ddpa
->used
= dsl_dataset_phys(origin_ds
)->ds_referenced_bytes
;
2227 ddpa
->comp
= dsl_dataset_phys(origin_ds
)->ds_compressed_bytes
;
2228 ddpa
->uncomp
= dsl_dataset_phys(origin_ds
)->ds_uncompressed_bytes
;
2229 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2230 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2231 uint64_t val
, dlused
, dlcomp
, dluncomp
;
2232 dsl_dataset_t
*ds
= snap
->ds
;
2237 * If there are long holds, we won't be able to evict
2240 if (dsl_dataset_long_held(ds
)) {
2241 err
= SET_ERROR(EBUSY
);
2245 /* Check that the snapshot name does not conflict */
2246 VERIFY0(dsl_dataset_get_snapname(ds
));
2247 if (strlen(ds
->ds_snapname
) >= max_snap_len
) {
2248 err
= SET_ERROR(ENAMETOOLONG
);
2251 err
= dsl_dataset_snap_lookup(hds
, ds
->ds_snapname
, &val
);
2253 (void) strcpy(ddpa
->err_ds
, snap
->ds
->ds_snapname
);
2254 err
= SET_ERROR(EEXIST
);
2260 /* The very first snapshot does not have a deadlist */
2261 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
== 0)
2264 dsl_deadlist_space(&ds
->ds_deadlist
,
2265 &dlused
, &dlcomp
, &dluncomp
);
2266 ddpa
->used
+= dlused
;
2267 ddpa
->comp
+= dlcomp
;
2268 ddpa
->uncomp
+= dluncomp
;
2272 * If we are a clone of a clone then we never reached ORIGIN,
2273 * so we need to subtract out the clone origin's used space.
2275 if (ddpa
->origin_origin
) {
2277 dsl_dataset_phys(ddpa
->origin_origin
)->ds_referenced_bytes
;
2279 dsl_dataset_phys(ddpa
->origin_origin
)->ds_compressed_bytes
;
2281 dsl_dataset_phys(ddpa
->origin_origin
)->
2282 ds_uncompressed_bytes
;
2285 /* Check that there is enough space and limit headroom here */
2286 err
= dsl_dir_transfer_possible(origin_ds
->ds_dir
, hds
->ds_dir
,
2287 0, ss_mv_cnt
, ddpa
->used
, ddpa
->cr
);
2292 * Compute the amounts of space that will be used by snapshots
2293 * after the promotion (for both origin and clone). For each,
2294 * it is the amount of space that will be on all of their
2295 * deadlists (that was not born before their new origin).
2297 if (dsl_dir_phys(hds
->ds_dir
)->dd_flags
& DD_FLAG_USED_BREAKDOWN
) {
2301 * Note, typically this will not be a clone of a clone,
2302 * so dd_origin_txg will be < TXG_INITIAL, so
2303 * these snaplist_space() -> dsl_deadlist_space_range()
2304 * calls will be fast because they do not have to
2305 * iterate over all bps.
2307 snap
= list_head(&ddpa
->origin_snaps
);
2308 err
= snaplist_space(&ddpa
->shared_snaps
,
2309 snap
->ds
->ds_dir
->dd_origin_txg
, &ddpa
->cloneusedsnap
);
2313 err
= snaplist_space(&ddpa
->clone_snaps
,
2314 snap
->ds
->ds_dir
->dd_origin_txg
, &space
);
2317 ddpa
->cloneusedsnap
+= space
;
2319 if (dsl_dir_phys(origin_ds
->ds_dir
)->dd_flags
&
2320 DD_FLAG_USED_BREAKDOWN
) {
2321 err
= snaplist_space(&ddpa
->origin_snaps
,
2322 dsl_dataset_phys(origin_ds
)->ds_creation_txg
,
2323 &ddpa
->originusedsnap
);
2329 promote_rele(ddpa
, FTAG
);
2334 dsl_dataset_promote_sync(void *arg
, dmu_tx_t
*tx
)
2336 dsl_dataset_promote_arg_t
*ddpa
= arg
;
2337 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2339 struct promotenode
*snap
;
2340 dsl_dataset_t
*origin_ds
;
2341 dsl_dataset_t
*origin_head
;
2343 dsl_dir_t
*odd
= NULL
;
2344 uint64_t oldnext_obj
;
2347 VERIFY0(promote_hold(ddpa
, dp
, FTAG
));
2348 hds
= ddpa
->ddpa_clone
;
2350 ASSERT0(dsl_dataset_phys(hds
)->ds_flags
& DS_FLAG_NOPROMOTE
);
2352 snap
= list_head(&ddpa
->shared_snaps
);
2353 origin_ds
= snap
->ds
;
2356 snap
= list_head(&ddpa
->origin_snaps
);
2357 origin_head
= snap
->ds
;
2360 * We need to explicitly open odd, since origin_ds's dd will be
2363 VERIFY0(dsl_dir_hold_obj(dp
, origin_ds
->ds_dir
->dd_object
,
2366 /* change origin's next snap */
2367 dmu_buf_will_dirty(origin_ds
->ds_dbuf
, tx
);
2368 oldnext_obj
= dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
;
2369 snap
= list_tail(&ddpa
->clone_snaps
);
2370 ASSERT3U(dsl_dataset_phys(snap
->ds
)->ds_prev_snap_obj
, ==,
2371 origin_ds
->ds_object
);
2372 dsl_dataset_phys(origin_ds
)->ds_next_snap_obj
= snap
->ds
->ds_object
;
2374 /* change the origin's next clone */
2375 if (dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
) {
2376 dsl_dataset_remove_from_next_clones(origin_ds
,
2377 snap
->ds
->ds_object
, tx
);
2378 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2379 dsl_dataset_phys(origin_ds
)->ds_next_clones_obj
,
2384 dmu_buf_will_dirty(dd
->dd_dbuf
, tx
);
2385 ASSERT3U(dsl_dir_phys(dd
)->dd_origin_obj
, ==, origin_ds
->ds_object
);
2386 dsl_dir_phys(dd
)->dd_origin_obj
= dsl_dir_phys(odd
)->dd_origin_obj
;
2387 dd
->dd_origin_txg
= origin_head
->ds_dir
->dd_origin_txg
;
2388 dmu_buf_will_dirty(odd
->dd_dbuf
, tx
);
2389 dsl_dir_phys(odd
)->dd_origin_obj
= origin_ds
->ds_object
;
2390 origin_head
->ds_dir
->dd_origin_txg
=
2391 dsl_dataset_phys(origin_ds
)->ds_creation_txg
;
2393 /* change dd_clone entries */
2394 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2395 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2396 dsl_dir_phys(odd
)->dd_clones
, hds
->ds_object
, tx
));
2397 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2398 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2399 hds
->ds_object
, tx
));
2401 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2402 dsl_dir_phys(ddpa
->origin_origin
->ds_dir
)->dd_clones
,
2403 origin_head
->ds_object
, tx
));
2404 if (dsl_dir_phys(dd
)->dd_clones
== 0) {
2405 dsl_dir_phys(dd
)->dd_clones
=
2406 zap_create(dp
->dp_meta_objset
, DMU_OT_DSL_CLONES
,
2407 DMU_OT_NONE
, 0, tx
);
2409 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2410 dsl_dir_phys(dd
)->dd_clones
, origin_head
->ds_object
, tx
));
2413 /* move snapshots to this dir */
2414 for (snap
= list_head(&ddpa
->shared_snaps
); snap
;
2415 snap
= list_next(&ddpa
->shared_snaps
, snap
)) {
2416 dsl_dataset_t
*ds
= snap
->ds
;
2419 * Property callbacks are registered to a particular
2420 * dsl_dir. Since ours is changing, evict the objset
2421 * so that they will be unregistered from the old dsl_dir.
2423 if (ds
->ds_objset
) {
2424 dmu_objset_evict(ds
->ds_objset
);
2425 ds
->ds_objset
= NULL
;
2428 /* move snap name entry */
2429 VERIFY0(dsl_dataset_get_snapname(ds
));
2430 VERIFY0(dsl_dataset_snap_remove(origin_head
,
2431 ds
->ds_snapname
, tx
, B_TRUE
));
2432 VERIFY0(zap_add(dp
->dp_meta_objset
,
2433 dsl_dataset_phys(hds
)->ds_snapnames_zapobj
, ds
->ds_snapname
,
2434 8, 1, &ds
->ds_object
, tx
));
2435 dsl_fs_ss_count_adjust(hds
->ds_dir
, 1,
2436 DD_FIELD_SNAPSHOT_COUNT
, tx
);
2438 /* change containing dsl_dir */
2439 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
2440 ASSERT3U(dsl_dataset_phys(ds
)->ds_dir_obj
, ==, odd
->dd_object
);
2441 dsl_dataset_phys(ds
)->ds_dir_obj
= dd
->dd_object
;
2442 ASSERT3P(ds
->ds_dir
, ==, odd
);
2443 dsl_dir_rele(ds
->ds_dir
, ds
);
2444 VERIFY0(dsl_dir_hold_obj(dp
, dd
->dd_object
,
2445 NULL
, ds
, &ds
->ds_dir
));
2447 /* move any clone references */
2448 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
&&
2449 spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
2453 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2454 dsl_dataset_phys(ds
)->ds_next_clones_obj
);
2455 zap_cursor_retrieve(&zc
, &za
) == 0;
2456 zap_cursor_advance(&zc
)) {
2457 dsl_dataset_t
*cnds
;
2460 if (za
.za_first_integer
== oldnext_obj
) {
2462 * We've already moved the
2463 * origin's reference.
2468 VERIFY0(dsl_dataset_hold_obj(dp
,
2469 za
.za_first_integer
, FTAG
, &cnds
));
2470 o
= dsl_dir_phys(cnds
->ds_dir
)->
2471 dd_head_dataset_obj
;
2473 VERIFY0(zap_remove_int(dp
->dp_meta_objset
,
2474 dsl_dir_phys(odd
)->dd_clones
, o
, tx
));
2475 VERIFY0(zap_add_int(dp
->dp_meta_objset
,
2476 dsl_dir_phys(dd
)->dd_clones
, o
, tx
));
2477 dsl_dataset_rele(cnds
, FTAG
);
2479 zap_cursor_fini(&zc
);
2482 ASSERT(!dsl_prop_hascb(ds
));
2486 * Change space accounting.
2487 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2488 * both be valid, or both be 0 (resulting in delta == 0). This
2489 * is true for each of {clone,origin} independently.
2492 delta
= ddpa
->cloneusedsnap
-
2493 dsl_dir_phys(dd
)->dd_used_breakdown
[DD_USED_SNAP
];
2494 ASSERT3S(delta
, >=, 0);
2495 ASSERT3U(ddpa
->used
, >=, delta
);
2496 dsl_dir_diduse_space(dd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2497 dsl_dir_diduse_space(dd
, DD_USED_HEAD
,
2498 ddpa
->used
- delta
, ddpa
->comp
, ddpa
->uncomp
, tx
);
2500 delta
= ddpa
->originusedsnap
-
2501 dsl_dir_phys(odd
)->dd_used_breakdown
[DD_USED_SNAP
];
2502 ASSERT3S(delta
, <=, 0);
2503 ASSERT3U(ddpa
->used
, >=, -delta
);
2504 dsl_dir_diduse_space(odd
, DD_USED_SNAP
, delta
, 0, 0, tx
);
2505 dsl_dir_diduse_space(odd
, DD_USED_HEAD
,
2506 -ddpa
->used
- delta
, -ddpa
->comp
, -ddpa
->uncomp
, tx
);
2508 dsl_dataset_phys(origin_ds
)->ds_unique_bytes
= ddpa
->unique
;
2510 /* log history record */
2511 spa_history_log_internal_ds(hds
, "promote", tx
, "");
2513 dsl_dir_rele(odd
, FTAG
);
2514 promote_rele(ddpa
, FTAG
);
2518 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2519 * (exclusive) and last_obj (inclusive). The list will be in reverse
2520 * order (last_obj will be the list_head()). If first_obj == 0, do all
2521 * snapshots back to this dataset's origin.
2524 snaplist_make(dsl_pool_t
*dp
,
2525 uint64_t first_obj
, uint64_t last_obj
, list_t
*l
, void *tag
)
2527 uint64_t obj
= last_obj
;
2529 list_create(l
, sizeof (struct promotenode
),
2530 offsetof(struct promotenode
, link
));
2532 while (obj
!= first_obj
) {
2534 struct promotenode
*snap
;
2537 err
= dsl_dataset_hold_obj(dp
, obj
, tag
, &ds
);
2538 ASSERT(err
!= ENOENT
);
2543 first_obj
= dsl_dir_phys(ds
->ds_dir
)->dd_origin_obj
;
2545 snap
= kmem_alloc(sizeof (*snap
), KM_SLEEP
);
2547 list_insert_tail(l
, snap
);
2548 obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
2555 snaplist_space(list_t
*l
, uint64_t mintxg
, uint64_t *spacep
)
2557 struct promotenode
*snap
;
2560 for (snap
= list_head(l
); snap
; snap
= list_next(l
, snap
)) {
2561 uint64_t used
, comp
, uncomp
;
2562 dsl_deadlist_space_range(&snap
->ds
->ds_deadlist
,
2563 mintxg
, UINT64_MAX
, &used
, &comp
, &uncomp
);
2570 snaplist_destroy(list_t
*l
, void *tag
)
2572 struct promotenode
*snap
;
2574 if (l
== NULL
|| !list_link_active(&l
->list_head
))
2577 while ((snap
= list_tail(l
)) != NULL
) {
2578 list_remove(l
, snap
);
2579 dsl_dataset_rele(snap
->ds
, tag
);
2580 kmem_free(snap
, sizeof (*snap
));
2586 promote_hold(dsl_dataset_promote_arg_t
*ddpa
, dsl_pool_t
*dp
, void *tag
)
2590 struct promotenode
*snap
;
2592 error
= dsl_dataset_hold(dp
, ddpa
->ddpa_clonename
, tag
,
2596 dd
= ddpa
->ddpa_clone
->ds_dir
;
2598 if (ddpa
->ddpa_clone
->ds_is_snapshot
||
2599 !dsl_dir_is_clone(dd
)) {
2600 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2601 return (SET_ERROR(EINVAL
));
2604 error
= snaplist_make(dp
, 0, dsl_dir_phys(dd
)->dd_origin_obj
,
2605 &ddpa
->shared_snaps
, tag
);
2609 error
= snaplist_make(dp
, 0, ddpa
->ddpa_clone
->ds_object
,
2610 &ddpa
->clone_snaps
, tag
);
2614 snap
= list_head(&ddpa
->shared_snaps
);
2615 ASSERT3U(snap
->ds
->ds_object
, ==, dsl_dir_phys(dd
)->dd_origin_obj
);
2616 error
= snaplist_make(dp
, dsl_dir_phys(dd
)->dd_origin_obj
,
2617 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_head_dataset_obj
,
2618 &ddpa
->origin_snaps
, tag
);
2622 if (dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
!= 0) {
2623 error
= dsl_dataset_hold_obj(dp
,
2624 dsl_dir_phys(snap
->ds
->ds_dir
)->dd_origin_obj
,
2625 tag
, &ddpa
->origin_origin
);
2631 promote_rele(ddpa
, tag
);
2636 promote_rele(dsl_dataset_promote_arg_t
*ddpa
, void *tag
)
2638 snaplist_destroy(&ddpa
->shared_snaps
, tag
);
2639 snaplist_destroy(&ddpa
->clone_snaps
, tag
);
2640 snaplist_destroy(&ddpa
->origin_snaps
, tag
);
2641 if (ddpa
->origin_origin
!= NULL
)
2642 dsl_dataset_rele(ddpa
->origin_origin
, tag
);
2643 dsl_dataset_rele(ddpa
->ddpa_clone
, tag
);
2649 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2650 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2653 dsl_dataset_promote(const char *name
, char *conflsnap
)
2655 dsl_dataset_promote_arg_t ddpa
= { 0 };
2661 * We will modify space proportional to the number of
2662 * snapshots. Compute numsnaps.
2664 error
= dmu_objset_hold(name
, FTAG
, &os
);
2667 error
= zap_count(dmu_objset_pool(os
)->dp_meta_objset
,
2668 dsl_dataset_phys(dmu_objset_ds(os
))->ds_snapnames_zapobj
,
2670 dmu_objset_rele(os
, FTAG
);
2674 ddpa
.ddpa_clonename
= name
;
2675 ddpa
.err_ds
= conflsnap
;
2678 return (dsl_sync_task(name
, dsl_dataset_promote_check
,
2679 dsl_dataset_promote_sync
, &ddpa
,
2680 2 + numsnaps
, ZFS_SPACE_CHECK_RESERVED
));
2684 dsl_dataset_clone_swap_check_impl(dsl_dataset_t
*clone
,
2685 dsl_dataset_t
*origin_head
, boolean_t force
, void *owner
, dmu_tx_t
*tx
)
2687 int64_t unused_refres_delta
;
2689 /* they should both be heads */
2690 if (clone
->ds_is_snapshot
||
2691 origin_head
->ds_is_snapshot
)
2692 return (SET_ERROR(EINVAL
));
2694 /* if we are not forcing, the branch point should be just before them */
2695 if (!force
&& clone
->ds_prev
!= origin_head
->ds_prev
)
2696 return (SET_ERROR(EINVAL
));
2698 /* clone should be the clone (unless they are unrelated) */
2699 if (clone
->ds_prev
!= NULL
&&
2700 clone
->ds_prev
!= clone
->ds_dir
->dd_pool
->dp_origin_snap
&&
2701 origin_head
->ds_dir
!= clone
->ds_prev
->ds_dir
)
2702 return (SET_ERROR(EINVAL
));
2704 /* the clone should be a child of the origin */
2705 if (clone
->ds_dir
->dd_parent
!= origin_head
->ds_dir
)
2706 return (SET_ERROR(EINVAL
));
2708 /* origin_head shouldn't be modified unless 'force' */
2710 dsl_dataset_modified_since_snap(origin_head
, origin_head
->ds_prev
))
2711 return (SET_ERROR(ETXTBSY
));
2713 /* origin_head should have no long holds (e.g. is not mounted) */
2714 if (dsl_dataset_handoff_check(origin_head
, owner
, tx
))
2715 return (SET_ERROR(EBUSY
));
2717 /* check amount of any unconsumed refreservation */
2718 unused_refres_delta
=
2719 (int64_t)MIN(origin_head
->ds_reserved
,
2720 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
2721 (int64_t)MIN(origin_head
->ds_reserved
,
2722 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2724 if (unused_refres_delta
> 0 &&
2725 unused_refres_delta
>
2726 dsl_dir_space_available(origin_head
->ds_dir
, NULL
, 0, TRUE
))
2727 return (SET_ERROR(ENOSPC
));
2729 /* clone can't be over the head's refquota */
2730 if (origin_head
->ds_quota
!= 0 &&
2731 dsl_dataset_phys(clone
)->ds_referenced_bytes
>
2732 origin_head
->ds_quota
)
2733 return (SET_ERROR(EDQUOT
));
2739 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t
*clone
,
2740 dsl_dataset_t
*origin_head
, dmu_tx_t
*tx
)
2743 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2744 int64_t unused_refres_delta
;
2746 ASSERT(clone
->ds_reserved
== 0);
2747 ASSERT(origin_head
->ds_quota
== 0 ||
2748 dsl_dataset_phys(clone
)->ds_unique_bytes
<= origin_head
->ds_quota
);
2749 ASSERT3P(clone
->ds_prev
, ==, origin_head
->ds_prev
);
2752 * Swap per-dataset feature flags.
2754 for (f
= 0; f
< SPA_FEATURES
; f
++) {
2755 boolean_t clone_inuse
;
2756 boolean_t origin_head_inuse
;
2758 if (!(spa_feature_table
[f
].fi_flags
&
2759 ZFEATURE_FLAG_PER_DATASET
)) {
2760 ASSERT(!clone
->ds_feature_inuse
[f
]);
2761 ASSERT(!origin_head
->ds_feature_inuse
[f
]);
2765 clone_inuse
= clone
->ds_feature_inuse
[f
];
2766 origin_head_inuse
= origin_head
->ds_feature_inuse
[f
];
2769 dsl_dataset_deactivate_feature(clone
->ds_object
, f
, tx
);
2770 clone
->ds_feature_inuse
[f
] = B_FALSE
;
2772 if (origin_head_inuse
) {
2773 dsl_dataset_deactivate_feature(origin_head
->ds_object
,
2775 origin_head
->ds_feature_inuse
[f
] = B_FALSE
;
2778 dsl_dataset_activate_feature(origin_head
->ds_object
,
2780 origin_head
->ds_feature_inuse
[f
] = B_TRUE
;
2782 if (origin_head_inuse
) {
2783 dsl_dataset_activate_feature(clone
->ds_object
, f
, tx
);
2784 clone
->ds_feature_inuse
[f
] = B_TRUE
;
2788 dmu_buf_will_dirty(clone
->ds_dbuf
, tx
);
2789 dmu_buf_will_dirty(origin_head
->ds_dbuf
, tx
);
2791 if (clone
->ds_objset
!= NULL
) {
2792 dmu_objset_evict(clone
->ds_objset
);
2793 clone
->ds_objset
= NULL
;
2796 if (origin_head
->ds_objset
!= NULL
) {
2797 dmu_objset_evict(origin_head
->ds_objset
);
2798 origin_head
->ds_objset
= NULL
;
2801 unused_refres_delta
=
2802 (int64_t)MIN(origin_head
->ds_reserved
,
2803 dsl_dataset_phys(origin_head
)->ds_unique_bytes
) -
2804 (int64_t)MIN(origin_head
->ds_reserved
,
2805 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2808 * Reset origin's unique bytes, if it exists.
2810 if (clone
->ds_prev
) {
2811 dsl_dataset_t
*origin
= clone
->ds_prev
;
2812 uint64_t comp
, uncomp
;
2814 dmu_buf_will_dirty(origin
->ds_dbuf
, tx
);
2815 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2816 dsl_dataset_phys(origin
)->ds_prev_snap_txg
, UINT64_MAX
,
2817 &dsl_dataset_phys(origin
)->ds_unique_bytes
, &comp
, &uncomp
);
2823 tmp
= dsl_dataset_phys(origin_head
)->ds_bp
;
2824 dsl_dataset_phys(origin_head
)->ds_bp
=
2825 dsl_dataset_phys(clone
)->ds_bp
;
2826 dsl_dataset_phys(clone
)->ds_bp
= tmp
;
2829 /* set dd_*_bytes */
2831 int64_t dused
, dcomp
, duncomp
;
2832 uint64_t cdl_used
, cdl_comp
, cdl_uncomp
;
2833 uint64_t odl_used
, odl_comp
, odl_uncomp
;
2835 ASSERT3U(dsl_dir_phys(clone
->ds_dir
)->
2836 dd_used_breakdown
[DD_USED_SNAP
], ==, 0);
2838 dsl_deadlist_space(&clone
->ds_deadlist
,
2839 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
2840 dsl_deadlist_space(&origin_head
->ds_deadlist
,
2841 &odl_used
, &odl_comp
, &odl_uncomp
);
2843 dused
= dsl_dataset_phys(clone
)->ds_referenced_bytes
+
2845 (dsl_dataset_phys(origin_head
)->ds_referenced_bytes
+
2847 dcomp
= dsl_dataset_phys(clone
)->ds_compressed_bytes
+
2849 (dsl_dataset_phys(origin_head
)->ds_compressed_bytes
+
2851 duncomp
= dsl_dataset_phys(clone
)->ds_uncompressed_bytes
+
2853 (dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
+
2856 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_HEAD
,
2857 dused
, dcomp
, duncomp
, tx
);
2858 dsl_dir_diduse_space(clone
->ds_dir
, DD_USED_HEAD
,
2859 -dused
, -dcomp
, -duncomp
, tx
);
2862 * The difference in the space used by snapshots is the
2863 * difference in snapshot space due to the head's
2864 * deadlist (since that's the only thing that's
2865 * changing that affects the snapused).
2867 dsl_deadlist_space_range(&clone
->ds_deadlist
,
2868 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
2869 &cdl_used
, &cdl_comp
, &cdl_uncomp
);
2870 dsl_deadlist_space_range(&origin_head
->ds_deadlist
,
2871 origin_head
->ds_dir
->dd_origin_txg
, UINT64_MAX
,
2872 &odl_used
, &odl_comp
, &odl_uncomp
);
2873 dsl_dir_transfer_space(origin_head
->ds_dir
, cdl_used
- odl_used
,
2874 DD_USED_HEAD
, DD_USED_SNAP
, tx
);
2877 /* swap ds_*_bytes */
2878 SWITCH64(dsl_dataset_phys(origin_head
)->ds_referenced_bytes
,
2879 dsl_dataset_phys(clone
)->ds_referenced_bytes
);
2880 SWITCH64(dsl_dataset_phys(origin_head
)->ds_compressed_bytes
,
2881 dsl_dataset_phys(clone
)->ds_compressed_bytes
);
2882 SWITCH64(dsl_dataset_phys(origin_head
)->ds_uncompressed_bytes
,
2883 dsl_dataset_phys(clone
)->ds_uncompressed_bytes
);
2884 SWITCH64(dsl_dataset_phys(origin_head
)->ds_unique_bytes
,
2885 dsl_dataset_phys(clone
)->ds_unique_bytes
);
2887 /* apply any parent delta for change in unconsumed refreservation */
2888 dsl_dir_diduse_space(origin_head
->ds_dir
, DD_USED_REFRSRV
,
2889 unused_refres_delta
, 0, 0, tx
);
2894 dsl_deadlist_close(&clone
->ds_deadlist
);
2895 dsl_deadlist_close(&origin_head
->ds_deadlist
);
2896 SWITCH64(dsl_dataset_phys(origin_head
)->ds_deadlist_obj
,
2897 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
2898 dsl_deadlist_open(&clone
->ds_deadlist
, dp
->dp_meta_objset
,
2899 dsl_dataset_phys(clone
)->ds_deadlist_obj
);
2900 dsl_deadlist_open(&origin_head
->ds_deadlist
, dp
->dp_meta_objset
,
2901 dsl_dataset_phys(origin_head
)->ds_deadlist_obj
);
2903 dsl_scan_ds_clone_swapped(origin_head
, clone
, tx
);
2905 spa_history_log_internal_ds(clone
, "clone swap", tx
,
2906 "parent=%s", origin_head
->ds_dir
->dd_myname
);
2910 * Given a pool name and a dataset object number in that pool,
2911 * return the name of that dataset.
2914 dsl_dsobj_to_dsname(char *pname
, uint64_t obj
, char *buf
)
2920 error
= dsl_pool_hold(pname
, FTAG
, &dp
);
2924 error
= dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
);
2926 dsl_dataset_name(ds
, buf
);
2927 dsl_dataset_rele(ds
, FTAG
);
2929 dsl_pool_rele(dp
, FTAG
);
2935 dsl_dataset_check_quota(dsl_dataset_t
*ds
, boolean_t check_quota
,
2936 uint64_t asize
, uint64_t inflight
, uint64_t *used
, uint64_t *ref_rsrv
)
2940 ASSERT3S(asize
, >, 0);
2943 * *ref_rsrv is the portion of asize that will come from any
2944 * unconsumed refreservation space.
2948 mutex_enter(&ds
->ds_lock
);
2950 * Make a space adjustment for reserved bytes.
2952 if (ds
->ds_reserved
> dsl_dataset_phys(ds
)->ds_unique_bytes
) {
2954 ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
2956 (ds
->ds_reserved
- dsl_dataset_phys(ds
)->ds_unique_bytes
);
2958 asize
- MIN(asize
, parent_delta(ds
, asize
+ inflight
));
2961 if (!check_quota
|| ds
->ds_quota
== 0) {
2962 mutex_exit(&ds
->ds_lock
);
2966 * If they are requesting more space, and our current estimate
2967 * is over quota, they get to try again unless the actual
2968 * on-disk is over quota and there are no pending changes (which
2969 * may free up space for us).
2971 if (dsl_dataset_phys(ds
)->ds_referenced_bytes
+ inflight
>=
2974 dsl_dataset_phys(ds
)->ds_referenced_bytes
< ds
->ds_quota
)
2975 error
= SET_ERROR(ERESTART
);
2977 error
= SET_ERROR(EDQUOT
);
2979 mutex_exit(&ds
->ds_lock
);
2984 typedef struct dsl_dataset_set_qr_arg
{
2985 const char *ddsqra_name
;
2986 zprop_source_t ddsqra_source
;
2987 uint64_t ddsqra_value
;
2988 } dsl_dataset_set_qr_arg_t
;
2993 dsl_dataset_set_refquota_check(void *arg
, dmu_tx_t
*tx
)
2995 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
2996 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3001 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFQUOTA
)
3002 return (SET_ERROR(ENOTSUP
));
3004 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3008 if (ds
->ds_is_snapshot
) {
3009 dsl_dataset_rele(ds
, FTAG
);
3010 return (SET_ERROR(EINVAL
));
3013 error
= dsl_prop_predict(ds
->ds_dir
,
3014 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
3015 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3017 dsl_dataset_rele(ds
, FTAG
);
3022 dsl_dataset_rele(ds
, FTAG
);
3026 if (newval
< dsl_dataset_phys(ds
)->ds_referenced_bytes
||
3027 newval
< ds
->ds_reserved
) {
3028 dsl_dataset_rele(ds
, FTAG
);
3029 return (SET_ERROR(ENOSPC
));
3032 dsl_dataset_rele(ds
, FTAG
);
3037 dsl_dataset_set_refquota_sync(void *arg
, dmu_tx_t
*tx
)
3039 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3040 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3044 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3046 dsl_prop_set_sync_impl(ds
,
3047 zfs_prop_to_name(ZFS_PROP_REFQUOTA
),
3048 ddsqra
->ddsqra_source
, sizeof (ddsqra
->ddsqra_value
), 1,
3049 &ddsqra
->ddsqra_value
, tx
);
3051 VERIFY0(dsl_prop_get_int_ds(ds
,
3052 zfs_prop_to_name(ZFS_PROP_REFQUOTA
), &newval
));
3054 if (ds
->ds_quota
!= newval
) {
3055 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3056 ds
->ds_quota
= newval
;
3058 dsl_dataset_rele(ds
, FTAG
);
3062 dsl_dataset_set_refquota(const char *dsname
, zprop_source_t source
,
3065 dsl_dataset_set_qr_arg_t ddsqra
;
3067 ddsqra
.ddsqra_name
= dsname
;
3068 ddsqra
.ddsqra_source
= source
;
3069 ddsqra
.ddsqra_value
= refquota
;
3071 return (dsl_sync_task(dsname
, dsl_dataset_set_refquota_check
,
3072 dsl_dataset_set_refquota_sync
, &ddsqra
, 0, ZFS_SPACE_CHECK_NONE
));
3076 dsl_dataset_set_refreservation_check(void *arg
, dmu_tx_t
*tx
)
3078 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3079 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3082 uint64_t newval
, unique
;
3084 if (spa_version(dp
->dp_spa
) < SPA_VERSION_REFRESERVATION
)
3085 return (SET_ERROR(ENOTSUP
));
3087 error
= dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
);
3091 if (ds
->ds_is_snapshot
) {
3092 dsl_dataset_rele(ds
, FTAG
);
3093 return (SET_ERROR(EINVAL
));
3096 error
= dsl_prop_predict(ds
->ds_dir
,
3097 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3098 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, &newval
);
3100 dsl_dataset_rele(ds
, FTAG
);
3105 * If we are doing the preliminary check in open context, the
3106 * space estimates may be inaccurate.
3108 if (!dmu_tx_is_syncing(tx
)) {
3109 dsl_dataset_rele(ds
, FTAG
);
3113 mutex_enter(&ds
->ds_lock
);
3114 if (!DS_UNIQUE_IS_ACCURATE(ds
))
3115 dsl_dataset_recalc_head_uniq(ds
);
3116 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3117 mutex_exit(&ds
->ds_lock
);
3119 if (MAX(unique
, newval
) > MAX(unique
, ds
->ds_reserved
)) {
3120 uint64_t delta
= MAX(unique
, newval
) -
3121 MAX(unique
, ds
->ds_reserved
);
3124 dsl_dir_space_available(ds
->ds_dir
, NULL
, 0, B_TRUE
) ||
3125 (ds
->ds_quota
> 0 && newval
> ds
->ds_quota
)) {
3126 dsl_dataset_rele(ds
, FTAG
);
3127 return (SET_ERROR(ENOSPC
));
3131 dsl_dataset_rele(ds
, FTAG
);
3136 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t
*ds
,
3137 zprop_source_t source
, uint64_t value
, dmu_tx_t
*tx
)
3143 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_REFRESERVATION
),
3144 source
, sizeof (value
), 1, &value
, tx
);
3146 VERIFY0(dsl_prop_get_int_ds(ds
,
3147 zfs_prop_to_name(ZFS_PROP_REFRESERVATION
), &newval
));
3149 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
3150 mutex_enter(&ds
->ds_dir
->dd_lock
);
3151 mutex_enter(&ds
->ds_lock
);
3152 ASSERT(DS_UNIQUE_IS_ACCURATE(ds
));
3153 unique
= dsl_dataset_phys(ds
)->ds_unique_bytes
;
3154 delta
= MAX(0, (int64_t)(newval
- unique
)) -
3155 MAX(0, (int64_t)(ds
->ds_reserved
- unique
));
3156 ds
->ds_reserved
= newval
;
3157 mutex_exit(&ds
->ds_lock
);
3159 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_REFRSRV
, delta
, 0, 0, tx
);
3160 mutex_exit(&ds
->ds_dir
->dd_lock
);
3164 dsl_dataset_set_refreservation_sync(void *arg
, dmu_tx_t
*tx
)
3166 dsl_dataset_set_qr_arg_t
*ddsqra
= arg
;
3167 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
3170 VERIFY0(dsl_dataset_hold(dp
, ddsqra
->ddsqra_name
, FTAG
, &ds
));
3171 dsl_dataset_set_refreservation_sync_impl(ds
,
3172 ddsqra
->ddsqra_source
, ddsqra
->ddsqra_value
, tx
);
3173 dsl_dataset_rele(ds
, FTAG
);
3177 dsl_dataset_set_refreservation(const char *dsname
, zprop_source_t source
,
3178 uint64_t refreservation
)
3180 dsl_dataset_set_qr_arg_t ddsqra
;
3182 ddsqra
.ddsqra_name
= dsname
;
3183 ddsqra
.ddsqra_source
= source
;
3184 ddsqra
.ddsqra_value
= refreservation
;
3186 return (dsl_sync_task(dsname
, dsl_dataset_set_refreservation_check
,
3187 dsl_dataset_set_refreservation_sync
, &ddsqra
,
3188 0, ZFS_SPACE_CHECK_NONE
));
3192 * Return (in *usedp) the amount of space written in new that is not
3193 * present in oldsnap. New may be a snapshot or the head. Old must be
3194 * a snapshot before new, in new's filesystem (or its origin). If not then
3195 * fail and return EINVAL.
3197 * The written space is calculated by considering two components: First, we
3198 * ignore any freed space, and calculate the written as new's used space
3199 * minus old's used space. Next, we add in the amount of space that was freed
3200 * between the two snapshots, thus reducing new's used space relative to old's.
3201 * Specifically, this is the space that was born before old->ds_creation_txg,
3202 * and freed before new (ie. on new's deadlist or a previous deadlist).
3204 * space freed [---------------------]
3205 * snapshots ---O-------O--------O-------O------
3209 dsl_dataset_space_written(dsl_dataset_t
*oldsnap
, dsl_dataset_t
*new,
3210 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3214 dsl_pool_t
*dp
= new->ds_dir
->dd_pool
;
3216 ASSERT(dsl_pool_config_held(dp
));
3219 *usedp
+= dsl_dataset_phys(new)->ds_referenced_bytes
;
3220 *usedp
-= dsl_dataset_phys(oldsnap
)->ds_referenced_bytes
;
3223 *compp
+= dsl_dataset_phys(new)->ds_compressed_bytes
;
3224 *compp
-= dsl_dataset_phys(oldsnap
)->ds_compressed_bytes
;
3227 *uncompp
+= dsl_dataset_phys(new)->ds_uncompressed_bytes
;
3228 *uncompp
-= dsl_dataset_phys(oldsnap
)->ds_uncompressed_bytes
;
3230 snapobj
= new->ds_object
;
3231 while (snapobj
!= oldsnap
->ds_object
) {
3232 dsl_dataset_t
*snap
;
3233 uint64_t used
, comp
, uncomp
;
3235 if (snapobj
== new->ds_object
) {
3238 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &snap
);
3243 if (dsl_dataset_phys(snap
)->ds_prev_snap_txg
==
3244 dsl_dataset_phys(oldsnap
)->ds_creation_txg
) {
3246 * The blocks in the deadlist can not be born after
3247 * ds_prev_snap_txg, so get the whole deadlist space,
3248 * which is more efficient (especially for old-format
3249 * deadlists). Unfortunately the deadlist code
3250 * doesn't have enough information to make this
3251 * optimization itself.
3253 dsl_deadlist_space(&snap
->ds_deadlist
,
3254 &used
, &comp
, &uncomp
);
3256 dsl_deadlist_space_range(&snap
->ds_deadlist
,
3257 0, dsl_dataset_phys(oldsnap
)->ds_creation_txg
,
3258 &used
, &comp
, &uncomp
);
3265 * If we get to the beginning of the chain of snapshots
3266 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3267 * was not a snapshot of/before new.
3269 snapobj
= dsl_dataset_phys(snap
)->ds_prev_snap_obj
;
3271 dsl_dataset_rele(snap
, FTAG
);
3273 err
= SET_ERROR(EINVAL
);
3282 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3283 * lastsnap, and all snapshots in between are deleted.
3285 * blocks that would be freed [---------------------------]
3286 * snapshots ---O-------O--------O-------O--------O
3287 * firstsnap lastsnap
3289 * This is the set of blocks that were born after the snap before firstsnap,
3290 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3291 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3292 * We calculate this by iterating over the relevant deadlists (from the snap
3293 * after lastsnap, backward to the snap after firstsnap), summing up the
3294 * space on the deadlist that was born after the snap before firstsnap.
3297 dsl_dataset_space_wouldfree(dsl_dataset_t
*firstsnap
,
3298 dsl_dataset_t
*lastsnap
,
3299 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
3303 dsl_pool_t
*dp
= firstsnap
->ds_dir
->dd_pool
;
3305 ASSERT(firstsnap
->ds_is_snapshot
);
3306 ASSERT(lastsnap
->ds_is_snapshot
);
3309 * Check that the snapshots are in the same dsl_dir, and firstsnap
3310 * is before lastsnap.
3312 if (firstsnap
->ds_dir
!= lastsnap
->ds_dir
||
3313 dsl_dataset_phys(firstsnap
)->ds_creation_txg
>
3314 dsl_dataset_phys(lastsnap
)->ds_creation_txg
)
3315 return (SET_ERROR(EINVAL
));
3317 *usedp
= *compp
= *uncompp
= 0;
3319 snapobj
= dsl_dataset_phys(lastsnap
)->ds_next_snap_obj
;
3320 while (snapobj
!= firstsnap
->ds_object
) {
3322 uint64_t used
, comp
, uncomp
;
3324 err
= dsl_dataset_hold_obj(dp
, snapobj
, FTAG
, &ds
);
3328 dsl_deadlist_space_range(&ds
->ds_deadlist
,
3329 dsl_dataset_phys(firstsnap
)->ds_prev_snap_txg
, UINT64_MAX
,
3330 &used
, &comp
, &uncomp
);
3335 snapobj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
3336 ASSERT3U(snapobj
, !=, 0);
3337 dsl_dataset_rele(ds
, FTAG
);
3343 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3344 * For example, they could both be snapshots of the same filesystem, and
3345 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3346 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3347 * filesystem. Or 'earlier' could be the origin's origin.
3349 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3352 dsl_dataset_is_before(dsl_dataset_t
*later
, dsl_dataset_t
*earlier
,
3353 uint64_t earlier_txg
)
3355 dsl_pool_t
*dp
= later
->ds_dir
->dd_pool
;
3358 dsl_dataset_t
*origin
;
3360 ASSERT(dsl_pool_config_held(dp
));
3361 ASSERT(earlier
->ds_is_snapshot
|| earlier_txg
!= 0);
3363 if (earlier_txg
== 0)
3364 earlier_txg
= dsl_dataset_phys(earlier
)->ds_creation_txg
;
3366 if (later
->ds_is_snapshot
&&
3367 earlier_txg
>= dsl_dataset_phys(later
)->ds_creation_txg
)
3370 if (later
->ds_dir
== earlier
->ds_dir
)
3372 if (!dsl_dir_is_clone(later
->ds_dir
))
3375 if (dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
== earlier
->ds_object
)
3377 error
= dsl_dataset_hold_obj(dp
,
3378 dsl_dir_phys(later
->ds_dir
)->dd_origin_obj
, FTAG
, &origin
);
3381 ret
= dsl_dataset_is_before(origin
, earlier
, earlier_txg
);
3382 dsl_dataset_rele(origin
, FTAG
);
3387 dsl_dataset_zapify(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
3389 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
3390 dmu_object_zapify(mos
, ds
->ds_object
, DMU_OT_DSL_DATASET
, tx
);
3393 #if defined(_KERNEL) && defined(HAVE_SPL)
3395 module_param(zfs_max_recordsize
, int, 0644);
3396 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
3398 /* Limited to 1M on 32-bit platforms due to lack of virtual address space */
3399 module_param(zfs_max_recordsize
, int, 0444);
3400 MODULE_PARM_DESC(zfs_max_recordsize
, "Max allowed record size");
3403 EXPORT_SYMBOL(dsl_dataset_hold
);
3404 EXPORT_SYMBOL(dsl_dataset_hold_obj
);
3405 EXPORT_SYMBOL(dsl_dataset_own
);
3406 EXPORT_SYMBOL(dsl_dataset_own_obj
);
3407 EXPORT_SYMBOL(dsl_dataset_name
);
3408 EXPORT_SYMBOL(dsl_dataset_rele
);
3409 EXPORT_SYMBOL(dsl_dataset_disown
);
3410 EXPORT_SYMBOL(dsl_dataset_tryown
);
3411 EXPORT_SYMBOL(dsl_dataset_create_sync
);
3412 EXPORT_SYMBOL(dsl_dataset_create_sync_dd
);
3413 EXPORT_SYMBOL(dsl_dataset_snapshot_check
);
3414 EXPORT_SYMBOL(dsl_dataset_snapshot_sync
);
3415 EXPORT_SYMBOL(dsl_dataset_promote
);
3416 EXPORT_SYMBOL(dsl_dataset_user_hold
);
3417 EXPORT_SYMBOL(dsl_dataset_user_release
);
3418 EXPORT_SYMBOL(dsl_dataset_get_holds
);
3419 EXPORT_SYMBOL(dsl_dataset_get_blkptr
);
3420 EXPORT_SYMBOL(dsl_dataset_set_blkptr
);
3421 EXPORT_SYMBOL(dsl_dataset_get_spa
);
3422 EXPORT_SYMBOL(dsl_dataset_modified_since_snap
);
3423 EXPORT_SYMBOL(dsl_dataset_space_written
);
3424 EXPORT_SYMBOL(dsl_dataset_space_wouldfree
);
3425 EXPORT_SYMBOL(dsl_dataset_sync
);
3426 EXPORT_SYMBOL(dsl_dataset_block_born
);
3427 EXPORT_SYMBOL(dsl_dataset_block_kill
);
3428 EXPORT_SYMBOL(dsl_dataset_block_freeable
);
3429 EXPORT_SYMBOL(dsl_dataset_prev_snap_txg
);
3430 EXPORT_SYMBOL(dsl_dataset_dirty
);
3431 EXPORT_SYMBOL(dsl_dataset_stats
);
3432 EXPORT_SYMBOL(dsl_dataset_fast_stat
);
3433 EXPORT_SYMBOL(dsl_dataset_space
);
3434 EXPORT_SYMBOL(dsl_dataset_fsid_guid
);
3435 EXPORT_SYMBOL(dsl_dsobj_to_dsname
);
3436 EXPORT_SYMBOL(dsl_dataset_check_quota
);
3437 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl
);
3438 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl
);