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) 2012, 2015 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
25 * Copyright (c) 2013 by Joyent, Inc. All rights reserved.
26 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
29 #include <sys/zfs_context.h>
30 #include <sys/dsl_userhold.h>
31 #include <sys/dsl_dataset.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/dsl_pool.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dsl_scan.h>
38 #include <sys/dmu_objset.h>
40 #include <sys/zfeature.h>
41 #include <sys/zfs_ioctl.h>
42 #include <sys/dsl_deleg.h>
43 #include <sys/dmu_impl.h>
46 typedef struct dmu_snapshots_destroy_arg
{
48 nvlist_t
*dsda_successful_snaps
;
50 nvlist_t
*dsda_errlist
;
51 } dmu_snapshots_destroy_arg_t
;
54 dsl_destroy_snapshot_check_impl(dsl_dataset_t
*ds
, boolean_t defer
)
56 if (!ds
->ds_is_snapshot
)
57 return (SET_ERROR(EINVAL
));
59 if (dsl_dataset_long_held(ds
))
60 return (SET_ERROR(EBUSY
));
63 * Only allow deferred destroy on pools that support it.
64 * NOTE: deferred destroy is only supported on snapshots.
67 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) <
69 return (SET_ERROR(ENOTSUP
));
74 * If this snapshot has an elevated user reference count,
75 * we can't destroy it yet.
77 if (ds
->ds_userrefs
> 0)
78 return (SET_ERROR(EBUSY
));
81 * Can't delete a branch point.
83 if (dsl_dataset_phys(ds
)->ds_num_children
> 1)
84 return (SET_ERROR(EEXIST
));
90 dsl_destroy_snapshot_check(void *arg
, dmu_tx_t
*tx
)
92 dmu_snapshots_destroy_arg_t
*dsda
= arg
;
93 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
97 if (!dmu_tx_is_syncing(tx
))
100 for (pair
= nvlist_next_nvpair(dsda
->dsda_snaps
, NULL
);
101 pair
!= NULL
; pair
= nvlist_next_nvpair(dsda
->dsda_snaps
, pair
)) {
104 error
= dsl_dataset_hold(dp
, nvpair_name(pair
),
108 * If the snapshot does not exist, silently ignore it
109 * (it's "already destroyed").
115 error
= dsl_destroy_snapshot_check_impl(ds
,
117 dsl_dataset_rele(ds
, FTAG
);
121 fnvlist_add_boolean(dsda
->dsda_successful_snaps
,
124 fnvlist_add_int32(dsda
->dsda_errlist
,
125 nvpair_name(pair
), error
);
129 pair
= nvlist_next_nvpair(dsda
->dsda_errlist
, NULL
);
131 return (fnvpair_value_int32(pair
));
136 struct process_old_arg
{
138 dsl_dataset_t
*ds_prev
;
139 boolean_t after_branch_point
;
141 uint64_t used
, comp
, uncomp
;
145 process_old_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
147 struct process_old_arg
*poa
= arg
;
148 dsl_pool_t
*dp
= poa
->ds
->ds_dir
->dd_pool
;
150 ASSERT(!BP_IS_HOLE(bp
));
152 if (bp
->blk_birth
<= dsl_dataset_phys(poa
->ds
)->ds_prev_snap_txg
) {
153 dsl_deadlist_insert(&poa
->ds
->ds_deadlist
, bp
, tx
);
154 if (poa
->ds_prev
&& !poa
->after_branch_point
&&
156 dsl_dataset_phys(poa
->ds_prev
)->ds_prev_snap_txg
) {
157 dsl_dataset_phys(poa
->ds_prev
)->ds_unique_bytes
+=
158 bp_get_dsize_sync(dp
->dp_spa
, bp
);
161 poa
->used
+= bp_get_dsize_sync(dp
->dp_spa
, bp
);
162 poa
->comp
+= BP_GET_PSIZE(bp
);
163 poa
->uncomp
+= BP_GET_UCSIZE(bp
);
164 dsl_free_sync(poa
->pio
, dp
, tx
->tx_txg
, bp
);
170 process_old_deadlist(dsl_dataset_t
*ds
, dsl_dataset_t
*ds_prev
,
171 dsl_dataset_t
*ds_next
, boolean_t after_branch_point
, dmu_tx_t
*tx
)
173 struct process_old_arg poa
= { 0 };
174 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
175 objset_t
*mos
= dp
->dp_meta_objset
;
176 uint64_t deadlist_obj
;
178 ASSERT(ds
->ds_deadlist
.dl_oldfmt
);
179 ASSERT(ds_next
->ds_deadlist
.dl_oldfmt
);
182 poa
.ds_prev
= ds_prev
;
183 poa
.after_branch_point
= after_branch_point
;
184 poa
.pio
= zio_root(dp
->dp_spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
185 VERIFY0(bpobj_iterate(&ds_next
->ds_deadlist
.dl_bpobj
,
186 process_old_cb
, &poa
, tx
));
187 VERIFY0(zio_wait(poa
.pio
));
188 ASSERT3U(poa
.used
, ==, dsl_dataset_phys(ds
)->ds_unique_bytes
);
190 /* change snapused */
191 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_SNAP
,
192 -poa
.used
, -poa
.comp
, -poa
.uncomp
, tx
);
194 /* swap next's deadlist to our deadlist */
195 dsl_deadlist_close(&ds
->ds_deadlist
);
196 dsl_deadlist_close(&ds_next
->ds_deadlist
);
197 deadlist_obj
= dsl_dataset_phys(ds
)->ds_deadlist_obj
;
198 dsl_dataset_phys(ds
)->ds_deadlist_obj
=
199 dsl_dataset_phys(ds_next
)->ds_deadlist_obj
;
200 dsl_dataset_phys(ds_next
)->ds_deadlist_obj
= deadlist_obj
;
201 dsl_deadlist_open(&ds
->ds_deadlist
, mos
,
202 dsl_dataset_phys(ds
)->ds_deadlist_obj
);
203 dsl_deadlist_open(&ds_next
->ds_deadlist
, mos
,
204 dsl_dataset_phys(ds_next
)->ds_deadlist_obj
);
208 dsl_dataset_remove_clones_key(dsl_dataset_t
*ds
, uint64_t mintxg
, dmu_tx_t
*tx
)
210 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
215 * If it is the old version, dd_clones doesn't exist so we can't
216 * find the clones, but dsl_deadlist_remove_key() is a no-op so it
219 if (dsl_dir_phys(ds
->ds_dir
)->dd_clones
== 0)
222 zc
= kmem_alloc(sizeof (zap_cursor_t
), KM_SLEEP
);
223 za
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
225 for (zap_cursor_init(zc
, mos
, dsl_dir_phys(ds
->ds_dir
)->dd_clones
);
226 zap_cursor_retrieve(zc
, za
) == 0;
227 zap_cursor_advance(zc
)) {
228 dsl_dataset_t
*clone
;
230 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
231 za
->za_first_integer
, FTAG
, &clone
));
232 if (clone
->ds_dir
->dd_origin_txg
> mintxg
) {
233 dsl_deadlist_remove_key(&clone
->ds_deadlist
,
235 dsl_dataset_remove_clones_key(clone
, mintxg
, tx
);
237 dsl_dataset_rele(clone
, FTAG
);
241 kmem_free(za
, sizeof (zap_attribute_t
));
242 kmem_free(zc
, sizeof (zap_cursor_t
));
246 dsl_destroy_snapshot_sync_impl(dsl_dataset_t
*ds
, boolean_t defer
, dmu_tx_t
*tx
)
249 int after_branch_point
= FALSE
;
250 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
251 objset_t
*mos
= dp
->dp_meta_objset
;
252 dsl_dataset_t
*ds_prev
= NULL
;
253 uint64_t obj
, old_unique
, used
= 0, comp
= 0, uncomp
= 0;
254 dsl_dataset_t
*ds_next
, *ds_head
, *hds
;
257 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
258 ASSERT3U(dsl_dataset_phys(ds
)->ds_bp
.blk_birth
, <=, tx
->tx_txg
);
259 ASSERT(refcount_is_zero(&ds
->ds_longholds
));
262 (ds
->ds_userrefs
> 0 ||
263 dsl_dataset_phys(ds
)->ds_num_children
> 1)) {
264 ASSERT(spa_version(dp
->dp_spa
) >= SPA_VERSION_USERREFS
);
265 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
266 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_DEFER_DESTROY
;
267 spa_history_log_internal_ds(ds
, "defer_destroy", tx
, "");
271 ASSERT3U(dsl_dataset_phys(ds
)->ds_num_children
, <=, 1);
273 /* We need to log before removing it from the namespace. */
274 spa_history_log_internal_ds(ds
, "destroy", tx
, "");
276 dsl_scan_ds_destroyed(ds
, tx
);
280 for (f
= 0; f
< SPA_FEATURES
; f
++) {
281 if (ds
->ds_feature_inuse
[f
]) {
282 dsl_dataset_deactivate_feature(obj
, f
, tx
);
283 ds
->ds_feature_inuse
[f
] = B_FALSE
;
286 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
287 ASSERT3P(ds
->ds_prev
, ==, NULL
);
288 VERIFY0(dsl_dataset_hold_obj(dp
,
289 dsl_dataset_phys(ds
)->ds_prev_snap_obj
, FTAG
, &ds_prev
));
291 (dsl_dataset_phys(ds_prev
)->ds_next_snap_obj
!= obj
);
293 dmu_buf_will_dirty(ds_prev
->ds_dbuf
, tx
);
294 if (after_branch_point
&&
295 dsl_dataset_phys(ds_prev
)->ds_next_clones_obj
!= 0) {
296 dsl_dataset_remove_from_next_clones(ds_prev
, obj
, tx
);
297 if (dsl_dataset_phys(ds
)->ds_next_snap_obj
!= 0) {
298 VERIFY0(zap_add_int(mos
,
299 dsl_dataset_phys(ds_prev
)->
301 dsl_dataset_phys(ds
)->ds_next_snap_obj
,
305 if (!after_branch_point
) {
306 dsl_dataset_phys(ds_prev
)->ds_next_snap_obj
=
307 dsl_dataset_phys(ds
)->ds_next_snap_obj
;
311 VERIFY0(dsl_dataset_hold_obj(dp
,
312 dsl_dataset_phys(ds
)->ds_next_snap_obj
, FTAG
, &ds_next
));
313 ASSERT3U(dsl_dataset_phys(ds_next
)->ds_prev_snap_obj
, ==, obj
);
315 old_unique
= dsl_dataset_phys(ds_next
)->ds_unique_bytes
;
317 dmu_buf_will_dirty(ds_next
->ds_dbuf
, tx
);
318 dsl_dataset_phys(ds_next
)->ds_prev_snap_obj
=
319 dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
320 dsl_dataset_phys(ds_next
)->ds_prev_snap_txg
=
321 dsl_dataset_phys(ds
)->ds_prev_snap_txg
;
322 ASSERT3U(dsl_dataset_phys(ds
)->ds_prev_snap_txg
, ==,
323 ds_prev
? dsl_dataset_phys(ds_prev
)->ds_creation_txg
: 0);
325 if (ds_next
->ds_deadlist
.dl_oldfmt
) {
326 process_old_deadlist(ds
, ds_prev
, ds_next
,
327 after_branch_point
, tx
);
329 /* Adjust prev's unique space. */
330 if (ds_prev
&& !after_branch_point
) {
331 dsl_deadlist_space_range(&ds_next
->ds_deadlist
,
332 dsl_dataset_phys(ds_prev
)->ds_prev_snap_txg
,
333 dsl_dataset_phys(ds
)->ds_prev_snap_txg
,
334 &used
, &comp
, &uncomp
);
335 dsl_dataset_phys(ds_prev
)->ds_unique_bytes
+= used
;
338 /* Adjust snapused. */
339 dsl_deadlist_space_range(&ds_next
->ds_deadlist
,
340 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, UINT64_MAX
,
341 &used
, &comp
, &uncomp
);
342 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_SNAP
,
343 -used
, -comp
, -uncomp
, tx
);
345 /* Move blocks to be freed to pool's free list. */
346 dsl_deadlist_move_bpobj(&ds_next
->ds_deadlist
,
347 &dp
->dp_free_bpobj
, dsl_dataset_phys(ds
)->ds_prev_snap_txg
,
349 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
,
350 DD_USED_HEAD
, used
, comp
, uncomp
, tx
);
352 /* Merge our deadlist into next's and free it. */
353 dsl_deadlist_merge(&ds_next
->ds_deadlist
,
354 dsl_dataset_phys(ds
)->ds_deadlist_obj
, tx
);
356 dsl_deadlist_close(&ds
->ds_deadlist
);
357 dsl_deadlist_free(mos
, dsl_dataset_phys(ds
)->ds_deadlist_obj
, tx
);
358 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
359 dsl_dataset_phys(ds
)->ds_deadlist_obj
= 0;
361 /* Collapse range in clone heads */
362 dsl_dataset_remove_clones_key(ds
,
363 dsl_dataset_phys(ds
)->ds_creation_txg
, tx
);
365 if (ds_next
->ds_is_snapshot
) {
366 dsl_dataset_t
*ds_nextnext
;
369 * Update next's unique to include blocks which
370 * were previously shared by only this snapshot
371 * and it. Those blocks will be born after the
372 * prev snap and before this snap, and will have
373 * died after the next snap and before the one
374 * after that (ie. be on the snap after next's
377 VERIFY0(dsl_dataset_hold_obj(dp
,
378 dsl_dataset_phys(ds_next
)->ds_next_snap_obj
,
379 FTAG
, &ds_nextnext
));
380 dsl_deadlist_space_range(&ds_nextnext
->ds_deadlist
,
381 dsl_dataset_phys(ds
)->ds_prev_snap_txg
,
382 dsl_dataset_phys(ds
)->ds_creation_txg
,
383 &used
, &comp
, &uncomp
);
384 dsl_dataset_phys(ds_next
)->ds_unique_bytes
+= used
;
385 dsl_dataset_rele(ds_nextnext
, FTAG
);
386 ASSERT3P(ds_next
->ds_prev
, ==, NULL
);
388 /* Collapse range in this head. */
389 VERIFY0(dsl_dataset_hold_obj(dp
,
390 dsl_dir_phys(ds
->ds_dir
)->dd_head_dataset_obj
, FTAG
, &hds
));
391 dsl_deadlist_remove_key(&hds
->ds_deadlist
,
392 dsl_dataset_phys(ds
)->ds_creation_txg
, tx
);
393 dsl_dataset_rele(hds
, FTAG
);
396 ASSERT3P(ds_next
->ds_prev
, ==, ds
);
397 dsl_dataset_rele(ds_next
->ds_prev
, ds_next
);
398 ds_next
->ds_prev
= NULL
;
400 VERIFY0(dsl_dataset_hold_obj(dp
,
401 dsl_dataset_phys(ds
)->ds_prev_snap_obj
,
402 ds_next
, &ds_next
->ds_prev
));
405 dsl_dataset_recalc_head_uniq(ds_next
);
408 * Reduce the amount of our unconsumed refreservation
409 * being charged to our parent by the amount of
410 * new unique data we have gained.
412 if (old_unique
< ds_next
->ds_reserved
) {
414 uint64_t new_unique
=
415 dsl_dataset_phys(ds_next
)->ds_unique_bytes
;
417 ASSERT(old_unique
<= new_unique
);
418 mrsdelta
= MIN(new_unique
- old_unique
,
419 ds_next
->ds_reserved
- old_unique
);
420 dsl_dir_diduse_space(ds
->ds_dir
,
421 DD_USED_REFRSRV
, -mrsdelta
, 0, 0, tx
);
424 dsl_dataset_rele(ds_next
, FTAG
);
427 * This must be done after the dsl_traverse(), because it will
428 * re-open the objset.
431 dmu_objset_evict(ds
->ds_objset
);
432 ds
->ds_objset
= NULL
;
435 /* remove from snapshot namespace */
436 ASSERT(dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0);
437 VERIFY0(dsl_dataset_hold_obj(dp
,
438 dsl_dir_phys(ds
->ds_dir
)->dd_head_dataset_obj
, FTAG
, &ds_head
));
439 VERIFY0(dsl_dataset_get_snapname(ds
));
445 err
= dsl_dataset_snap_lookup(ds_head
,
446 ds
->ds_snapname
, &val
);
448 ASSERT3U(val
, ==, obj
);
451 VERIFY0(dsl_dataset_snap_remove(ds_head
, ds
->ds_snapname
, tx
, B_TRUE
));
452 dsl_dataset_rele(ds_head
, FTAG
);
455 dsl_dataset_rele(ds_prev
, FTAG
);
457 spa_prop_clear_bootfs(dp
->dp_spa
, ds
->ds_object
, tx
);
459 if (dsl_dataset_phys(ds
)->ds_next_clones_obj
!= 0) {
460 ASSERTV(uint64_t count
);
461 ASSERT0(zap_count(mos
,
462 dsl_dataset_phys(ds
)->ds_next_clones_obj
, &count
) &&
464 VERIFY0(dmu_object_free(mos
,
465 dsl_dataset_phys(ds
)->ds_next_clones_obj
, tx
));
467 if (dsl_dataset_phys(ds
)->ds_props_obj
!= 0)
468 VERIFY0(zap_destroy(mos
, dsl_dataset_phys(ds
)->ds_props_obj
,
470 if (dsl_dataset_phys(ds
)->ds_userrefs_obj
!= 0)
471 VERIFY0(zap_destroy(mos
, dsl_dataset_phys(ds
)->ds_userrefs_obj
,
473 dsl_dir_rele(ds
->ds_dir
, ds
);
475 dmu_object_free_zapified(mos
, obj
, tx
);
479 dsl_destroy_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
481 dmu_snapshots_destroy_arg_t
*dsda
= arg
;
482 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
485 for (pair
= nvlist_next_nvpair(dsda
->dsda_successful_snaps
, NULL
);
487 pair
= nvlist_next_nvpair(dsda
->dsda_successful_snaps
, pair
)) {
490 VERIFY0(dsl_dataset_hold(dp
, nvpair_name(pair
), FTAG
, &ds
));
492 dsl_destroy_snapshot_sync_impl(ds
, dsda
->dsda_defer
, tx
);
493 zvol_remove_minors(dp
->dp_spa
, nvpair_name(pair
), B_TRUE
);
494 dsl_dataset_rele(ds
, FTAG
);
499 * The semantics of this function are described in the comment above
500 * lzc_destroy_snaps(). To summarize:
502 * The snapshots must all be in the same pool.
504 * Snapshots that don't exist will be silently ignored (considered to be
505 * "already deleted").
507 * On success, all snaps will be destroyed and this will return 0.
508 * On failure, no snaps will be destroyed, the errlist will be filled in,
509 * and this will return an errno.
512 dsl_destroy_snapshots_nvl(nvlist_t
*snaps
, boolean_t defer
,
515 dmu_snapshots_destroy_arg_t dsda
;
519 pair
= nvlist_next_nvpair(snaps
, NULL
);
523 dsda
.dsda_snaps
= snaps
;
524 VERIFY0(nvlist_alloc(&dsda
.dsda_successful_snaps
,
525 NV_UNIQUE_NAME
, KM_SLEEP
));
526 dsda
.dsda_defer
= defer
;
527 dsda
.dsda_errlist
= errlist
;
529 error
= dsl_sync_task(nvpair_name(pair
),
530 dsl_destroy_snapshot_check
, dsl_destroy_snapshot_sync
,
531 &dsda
, 0, ZFS_SPACE_CHECK_NONE
);
532 fnvlist_free(dsda
.dsda_successful_snaps
);
538 dsl_destroy_snapshot(const char *name
, boolean_t defer
)
541 nvlist_t
*nvl
= fnvlist_alloc();
542 nvlist_t
*errlist
= fnvlist_alloc();
544 fnvlist_add_boolean(nvl
, name
);
545 error
= dsl_destroy_snapshots_nvl(nvl
, defer
, errlist
);
546 fnvlist_free(errlist
);
558 kill_blkptr(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
559 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
561 struct killarg
*ka
= arg
;
562 dmu_tx_t
*tx
= ka
->tx
;
564 if (bp
== NULL
|| BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
567 if (zb
->zb_level
== ZB_ZIL_LEVEL
) {
568 ASSERT(zilog
!= NULL
);
570 * It's a block in the intent log. It has no
571 * accounting, so just free it.
573 dsl_free(ka
->tx
->tx_pool
, ka
->tx
->tx_txg
, bp
);
575 ASSERT(zilog
== NULL
);
576 ASSERT3U(bp
->blk_birth
, >,
577 dsl_dataset_phys(ka
->ds
)->ds_prev_snap_txg
);
578 (void) dsl_dataset_block_kill(ka
->ds
, bp
, tx
, B_FALSE
);
585 old_synchronous_dataset_destroy(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
590 * Free everything that we point to (that's born after
591 * the previous snapshot, if we are a clone)
593 * NB: this should be very quick, because we already
594 * freed all the objects in open context.
598 VERIFY0(traverse_dataset(ds
,
599 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, TRAVERSE_POST
,
601 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds
) ||
602 dsl_dataset_phys(ds
)->ds_unique_bytes
== 0);
605 typedef struct dsl_destroy_head_arg
{
606 const char *ddha_name
;
607 } dsl_destroy_head_arg_t
;
610 dsl_destroy_head_check_impl(dsl_dataset_t
*ds
, int expected_holds
)
616 ASSERT(!ds
->ds_is_snapshot
);
617 if (ds
->ds_is_snapshot
)
618 return (SET_ERROR(EINVAL
));
620 if (refcount_count(&ds
->ds_longholds
) != expected_holds
)
621 return (SET_ERROR(EBUSY
));
623 mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
626 * Can't delete a head dataset if there are snapshots of it.
627 * (Except if the only snapshots are from the branch we cloned
630 if (ds
->ds_prev
!= NULL
&&
631 dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
== ds
->ds_object
)
632 return (SET_ERROR(EBUSY
));
635 * Can't delete if there are children of this fs.
637 error
= zap_count(mos
,
638 dsl_dir_phys(ds
->ds_dir
)->dd_child_dir_zapobj
, &count
);
642 return (SET_ERROR(EEXIST
));
644 if (dsl_dir_is_clone(ds
->ds_dir
) && DS_IS_DEFER_DESTROY(ds
->ds_prev
) &&
645 dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
== 2 &&
646 ds
->ds_prev
->ds_userrefs
== 0) {
647 /* We need to remove the origin snapshot as well. */
648 if (!refcount_is_zero(&ds
->ds_prev
->ds_longholds
))
649 return (SET_ERROR(EBUSY
));
655 dsl_destroy_head_check(void *arg
, dmu_tx_t
*tx
)
657 dsl_destroy_head_arg_t
*ddha
= arg
;
658 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
662 error
= dsl_dataset_hold(dp
, ddha
->ddha_name
, FTAG
, &ds
);
666 error
= dsl_destroy_head_check_impl(ds
, 0);
667 dsl_dataset_rele(ds
, FTAG
);
672 dsl_dir_destroy_sync(uint64_t ddobj
, dmu_tx_t
*tx
)
675 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
676 objset_t
*mos
= dp
->dp_meta_objset
;
679 ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx
)->dp_config_rwlock
));
681 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, NULL
, FTAG
, &dd
));
683 ASSERT0(dsl_dir_phys(dd
)->dd_head_dataset_obj
);
686 * Decrement the filesystem count for all parent filesystems.
688 * When we receive an incremental stream into a filesystem that already
689 * exists, a temporary clone is created. We never count this temporary
690 * clone, whose name begins with a '%'.
692 if (dd
->dd_myname
[0] != '%' && dd
->dd_parent
!= NULL
)
693 dsl_fs_ss_count_adjust(dd
->dd_parent
, -1,
694 DD_FIELD_FILESYSTEM_COUNT
, tx
);
697 * Remove our reservation. The impl() routine avoids setting the
698 * actual property, which would require the (already destroyed) ds.
700 dsl_dir_set_reservation_sync_impl(dd
, 0, tx
);
702 ASSERT0(dsl_dir_phys(dd
)->dd_used_bytes
);
703 ASSERT0(dsl_dir_phys(dd
)->dd_reserved
);
704 for (t
= 0; t
< DD_USED_NUM
; t
++)
705 ASSERT0(dsl_dir_phys(dd
)->dd_used_breakdown
[t
]);
707 VERIFY0(zap_destroy(mos
, dsl_dir_phys(dd
)->dd_child_dir_zapobj
, tx
));
708 VERIFY0(zap_destroy(mos
, dsl_dir_phys(dd
)->dd_props_zapobj
, tx
));
709 VERIFY0(dsl_deleg_destroy(mos
, dsl_dir_phys(dd
)->dd_deleg_zapobj
, tx
));
710 VERIFY0(zap_remove(mos
,
711 dsl_dir_phys(dd
->dd_parent
)->dd_child_dir_zapobj
,
714 dsl_dir_rele(dd
, FTAG
);
715 dmu_object_free_zapified(mos
, ddobj
, tx
);
719 dsl_destroy_head_sync_impl(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
721 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
723 objset_t
*mos
= dp
->dp_meta_objset
;
724 uint64_t obj
, ddobj
, prevobj
= 0;
728 ASSERT3U(dsl_dataset_phys(ds
)->ds_num_children
, <=, 1);
729 ASSERT(ds
->ds_prev
== NULL
||
730 dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
!= ds
->ds_object
);
731 ASSERT3U(dsl_dataset_phys(ds
)->ds_bp
.blk_birth
, <=, tx
->tx_txg
);
732 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
734 /* We need to log before removing it from the namespace. */
735 spa_history_log_internal_ds(ds
, "destroy", tx
, "");
737 rmorigin
= (dsl_dir_is_clone(ds
->ds_dir
) &&
738 DS_IS_DEFER_DESTROY(ds
->ds_prev
) &&
739 dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
== 2 &&
740 ds
->ds_prev
->ds_userrefs
== 0);
742 /* Remove our reservation. */
743 if (ds
->ds_reserved
!= 0) {
744 dsl_dataset_set_refreservation_sync_impl(ds
,
745 (ZPROP_SRC_NONE
| ZPROP_SRC_LOCAL
| ZPROP_SRC_RECEIVED
),
747 ASSERT0(ds
->ds_reserved
);
752 for (f
= 0; f
< SPA_FEATURES
; f
++) {
753 if (ds
->ds_feature_inuse
[f
]) {
754 dsl_dataset_deactivate_feature(obj
, f
, tx
);
755 ds
->ds_feature_inuse
[f
] = B_FALSE
;
759 dsl_scan_ds_destroyed(ds
, tx
);
761 if (dsl_dataset_phys(ds
)->ds_prev_snap_obj
!= 0) {
762 /* This is a clone */
763 ASSERT(ds
->ds_prev
!= NULL
);
764 ASSERT3U(dsl_dataset_phys(ds
->ds_prev
)->ds_next_snap_obj
, !=,
766 ASSERT0(dsl_dataset_phys(ds
)->ds_next_snap_obj
);
768 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
769 if (dsl_dataset_phys(ds
->ds_prev
)->ds_next_clones_obj
!= 0) {
770 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
774 ASSERT3U(dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
, >, 1);
775 dsl_dataset_phys(ds
->ds_prev
)->ds_num_children
--;
779 * Destroy the deadlist. Unless it's a clone, the
780 * deadlist should be empty. (If it's a clone, it's
781 * safe to ignore the deadlist contents.)
783 dsl_deadlist_close(&ds
->ds_deadlist
);
784 dsl_deadlist_free(mos
, dsl_dataset_phys(ds
)->ds_deadlist_obj
, tx
);
785 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
786 dsl_dataset_phys(ds
)->ds_deadlist_obj
= 0;
788 VERIFY0(dmu_objset_from_ds(ds
, &os
));
790 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_ASYNC_DESTROY
)) {
791 old_synchronous_dataset_destroy(ds
, tx
);
794 * Move the bptree into the pool's list of trees to
795 * clean up and update space accounting information.
797 uint64_t used
, comp
, uncomp
;
799 zil_destroy_sync(dmu_objset_zil(os
), tx
);
801 if (!spa_feature_is_active(dp
->dp_spa
,
802 SPA_FEATURE_ASYNC_DESTROY
)) {
803 dsl_scan_t
*scn
= dp
->dp_scan
;
804 spa_feature_incr(dp
->dp_spa
, SPA_FEATURE_ASYNC_DESTROY
,
806 dp
->dp_bptree_obj
= bptree_alloc(mos
, tx
);
808 DMU_POOL_DIRECTORY_OBJECT
,
809 DMU_POOL_BPTREE_OBJ
, sizeof (uint64_t), 1,
810 &dp
->dp_bptree_obj
, tx
));
811 ASSERT(!scn
->scn_async_destroying
);
812 scn
->scn_async_destroying
= B_TRUE
;
815 used
= dsl_dir_phys(ds
->ds_dir
)->dd_used_bytes
;
816 comp
= dsl_dir_phys(ds
->ds_dir
)->dd_compressed_bytes
;
817 uncomp
= dsl_dir_phys(ds
->ds_dir
)->dd_uncompressed_bytes
;
819 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds
) ||
820 dsl_dataset_phys(ds
)->ds_unique_bytes
== used
);
822 bptree_add(mos
, dp
->dp_bptree_obj
,
823 &dsl_dataset_phys(ds
)->ds_bp
,
824 dsl_dataset_phys(ds
)->ds_prev_snap_txg
,
825 used
, comp
, uncomp
, tx
);
826 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
827 -used
, -comp
, -uncomp
, tx
);
828 dsl_dir_diduse_space(dp
->dp_free_dir
, DD_USED_HEAD
,
829 used
, comp
, uncomp
, tx
);
832 if (ds
->ds_prev
!= NULL
) {
833 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
834 VERIFY0(zap_remove_int(mos
,
835 dsl_dir_phys(ds
->ds_prev
->ds_dir
)->dd_clones
,
838 prevobj
= ds
->ds_prev
->ds_object
;
839 dsl_dataset_rele(ds
->ds_prev
, ds
);
844 * This must be done after the dsl_traverse(), because it will
845 * re-open the objset.
848 dmu_objset_evict(ds
->ds_objset
);
849 ds
->ds_objset
= NULL
;
852 /* Erase the link in the dir */
853 dmu_buf_will_dirty(ds
->ds_dir
->dd_dbuf
, tx
);
854 dsl_dir_phys(ds
->ds_dir
)->dd_head_dataset_obj
= 0;
855 ddobj
= ds
->ds_dir
->dd_object
;
856 ASSERT(dsl_dataset_phys(ds
)->ds_snapnames_zapobj
!= 0);
857 VERIFY0(zap_destroy(mos
,
858 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, tx
));
860 if (ds
->ds_bookmarks
!= 0) {
861 VERIFY0(zap_destroy(mos
, ds
->ds_bookmarks
, tx
));
862 spa_feature_decr(dp
->dp_spa
, SPA_FEATURE_BOOKMARKS
, tx
);
865 spa_prop_clear_bootfs(dp
->dp_spa
, ds
->ds_object
, tx
);
867 ASSERT0(dsl_dataset_phys(ds
)->ds_next_clones_obj
);
868 ASSERT0(dsl_dataset_phys(ds
)->ds_props_obj
);
869 ASSERT0(dsl_dataset_phys(ds
)->ds_userrefs_obj
);
870 dsl_dir_rele(ds
->ds_dir
, ds
);
872 dmu_object_free_zapified(mos
, obj
, tx
);
874 dsl_dir_destroy_sync(ddobj
, tx
);
878 VERIFY0(dsl_dataset_hold_obj(dp
, prevobj
, FTAG
, &prev
));
879 dsl_destroy_snapshot_sync_impl(prev
, B_FALSE
, tx
);
880 dsl_dataset_rele(prev
, FTAG
);
885 dsl_destroy_head_sync(void *arg
, dmu_tx_t
*tx
)
887 dsl_destroy_head_arg_t
*ddha
= arg
;
888 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
891 VERIFY0(dsl_dataset_hold(dp
, ddha
->ddha_name
, FTAG
, &ds
));
892 dsl_destroy_head_sync_impl(ds
, tx
);
893 zvol_remove_minors(dp
->dp_spa
, ddha
->ddha_name
, B_TRUE
);
894 dsl_dataset_rele(ds
, FTAG
);
898 dsl_destroy_head_begin_sync(void *arg
, dmu_tx_t
*tx
)
900 dsl_destroy_head_arg_t
*ddha
= arg
;
901 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
904 VERIFY0(dsl_dataset_hold(dp
, ddha
->ddha_name
, FTAG
, &ds
));
906 /* Mark it as inconsistent on-disk, in case we crash */
907 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
908 dsl_dataset_phys(ds
)->ds_flags
|= DS_FLAG_INCONSISTENT
;
910 spa_history_log_internal_ds(ds
, "destroy begin", tx
, "");
911 dsl_dataset_rele(ds
, FTAG
);
915 dsl_destroy_head(const char *name
)
917 dsl_destroy_head_arg_t ddha
;
923 zfs_destroy_unmount_origin(name
);
926 error
= spa_open(name
, &spa
, FTAG
);
929 isenabled
= spa_feature_is_enabled(spa
, SPA_FEATURE_ASYNC_DESTROY
);
930 spa_close(spa
, FTAG
);
932 ddha
.ddha_name
= name
;
937 error
= dsl_sync_task(name
, dsl_destroy_head_check
,
938 dsl_destroy_head_begin_sync
, &ddha
,
939 0, ZFS_SPACE_CHECK_NONE
);
944 * Head deletion is processed in one txg on old pools;
945 * remove the objects from open context so that the txg sync
948 error
= dmu_objset_own(name
, DMU_OST_ANY
, B_FALSE
, FTAG
, &os
);
951 uint64_t prev_snap_txg
=
952 dsl_dataset_phys(dmu_objset_ds(os
))->
954 for (obj
= 0; error
== 0;
955 error
= dmu_object_next(os
, &obj
, FALSE
,
957 (void) dmu_free_long_object(os
, obj
);
958 /* sync out all frees */
959 txg_wait_synced(dmu_objset_pool(os
), 0);
960 dmu_objset_disown(os
, FTAG
);
964 return (dsl_sync_task(name
, dsl_destroy_head_check
,
965 dsl_destroy_head_sync
, &ddha
, 0, ZFS_SPACE_CHECK_NONE
));
969 * Note, this function is used as the callback for dmu_objset_find(). We
970 * always return 0 so that we will continue to find and process
971 * inconsistent datasets, even if we encounter an error trying to
972 * process one of them.
976 dsl_destroy_inconsistent(const char *dsname
, void *arg
)
980 if (dmu_objset_hold(dsname
, FTAG
, &os
) == 0) {
981 boolean_t inconsistent
= DS_IS_INCONSISTENT(dmu_objset_ds(os
));
982 dmu_objset_rele(os
, FTAG
);
984 (void) dsl_destroy_head(dsname
);
990 #if defined(_KERNEL) && defined(HAVE_SPL)
991 EXPORT_SYMBOL(dsl_destroy_head
);
992 EXPORT_SYMBOL(dsl_destroy_head_sync_impl
);
993 EXPORT_SYMBOL(dsl_dataset_user_hold_check_one
);
994 EXPORT_SYMBOL(dsl_destroy_snapshot_sync_impl
);
995 EXPORT_SYMBOL(dsl_destroy_inconsistent
);
996 EXPORT_SYMBOL(dsl_dataset_user_release_tmp
);
997 EXPORT_SYMBOL(dsl_destroy_head_check_impl
);