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, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
27 #include <sys/zfs_context.h>
28 #include <sys/dsl_userhold.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/dsl_pool.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dmu_traverse.h>
35 #include <sys/dsl_scan.h>
36 #include <sys/dmu_objset.h>
38 #include <sys/zfeature.h>
39 #include <sys/zfs_ioctl.h>
40 #include <sys/dsl_deleg.h>
41 #include <sys/dmu_impl.h>
43 typedef struct dmu_snapshots_destroy_arg
{
45 nvlist_t
*dsda_successful_snaps
;
47 nvlist_t
*dsda_errlist
;
48 } dmu_snapshots_destroy_arg_t
;
51 dsl_destroy_snapshot_check_impl(dsl_dataset_t
*ds
, boolean_t defer
)
53 if (!dsl_dataset_is_snapshot(ds
))
54 return (SET_ERROR(EINVAL
));
56 if (dsl_dataset_long_held(ds
))
57 return (SET_ERROR(EBUSY
));
60 * Only allow deferred destroy on pools that support it.
61 * NOTE: deferred destroy is only supported on snapshots.
64 if (spa_version(ds
->ds_dir
->dd_pool
->dp_spa
) <
66 return (SET_ERROR(ENOTSUP
));
71 * If this snapshot has an elevated user reference count,
72 * we can't destroy it yet.
74 if (ds
->ds_userrefs
> 0)
75 return (SET_ERROR(EBUSY
));
78 * Can't delete a branch point.
80 if (ds
->ds_phys
->ds_num_children
> 1)
81 return (SET_ERROR(EEXIST
));
87 dsl_destroy_snapshot_check(void *arg
, dmu_tx_t
*tx
)
89 dmu_snapshots_destroy_arg_t
*dsda
= arg
;
90 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
94 if (!dmu_tx_is_syncing(tx
))
97 for (pair
= nvlist_next_nvpair(dsda
->dsda_snaps
, NULL
);
98 pair
!= NULL
; pair
= nvlist_next_nvpair(dsda
->dsda_snaps
, pair
)) {
101 error
= dsl_dataset_hold(dp
, nvpair_name(pair
),
105 * If the snapshot does not exist, silently ignore it
106 * (it's "already destroyed").
112 error
= dsl_destroy_snapshot_check_impl(ds
,
114 dsl_dataset_rele(ds
, FTAG
);
118 fnvlist_add_boolean(dsda
->dsda_successful_snaps
,
121 fnvlist_add_int32(dsda
->dsda_errlist
,
122 nvpair_name(pair
), error
);
126 pair
= nvlist_next_nvpair(dsda
->dsda_errlist
, NULL
);
128 return (fnvpair_value_int32(pair
));
133 struct process_old_arg
{
135 dsl_dataset_t
*ds_prev
;
136 boolean_t after_branch_point
;
138 uint64_t used
, comp
, uncomp
;
142 process_old_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
144 struct process_old_arg
*poa
= arg
;
145 dsl_pool_t
*dp
= poa
->ds
->ds_dir
->dd_pool
;
147 ASSERT(!BP_IS_HOLE(bp
));
149 if (bp
->blk_birth
<= poa
->ds
->ds_phys
->ds_prev_snap_txg
) {
150 dsl_deadlist_insert(&poa
->ds
->ds_deadlist
, bp
, tx
);
151 if (poa
->ds_prev
&& !poa
->after_branch_point
&&
153 poa
->ds_prev
->ds_phys
->ds_prev_snap_txg
) {
154 poa
->ds_prev
->ds_phys
->ds_unique_bytes
+=
155 bp_get_dsize_sync(dp
->dp_spa
, bp
);
158 poa
->used
+= bp_get_dsize_sync(dp
->dp_spa
, bp
);
159 poa
->comp
+= BP_GET_PSIZE(bp
);
160 poa
->uncomp
+= BP_GET_UCSIZE(bp
);
161 dsl_free_sync(poa
->pio
, dp
, tx
->tx_txg
, bp
);
167 process_old_deadlist(dsl_dataset_t
*ds
, dsl_dataset_t
*ds_prev
,
168 dsl_dataset_t
*ds_next
, boolean_t after_branch_point
, dmu_tx_t
*tx
)
170 struct process_old_arg poa
= { 0 };
171 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
172 objset_t
*mos
= dp
->dp_meta_objset
;
173 uint64_t deadlist_obj
;
175 ASSERT(ds
->ds_deadlist
.dl_oldfmt
);
176 ASSERT(ds_next
->ds_deadlist
.dl_oldfmt
);
179 poa
.ds_prev
= ds_prev
;
180 poa
.after_branch_point
= after_branch_point
;
181 poa
.pio
= zio_root(dp
->dp_spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
182 VERIFY0(bpobj_iterate(&ds_next
->ds_deadlist
.dl_bpobj
,
183 process_old_cb
, &poa
, tx
));
184 VERIFY0(zio_wait(poa
.pio
));
185 ASSERT3U(poa
.used
, ==, ds
->ds_phys
->ds_unique_bytes
);
187 /* change snapused */
188 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_SNAP
,
189 -poa
.used
, -poa
.comp
, -poa
.uncomp
, tx
);
191 /* swap next's deadlist to our deadlist */
192 dsl_deadlist_close(&ds
->ds_deadlist
);
193 dsl_deadlist_close(&ds_next
->ds_deadlist
);
194 deadlist_obj
= ds
->ds_phys
->ds_deadlist_obj
;
195 ds
->ds_phys
->ds_deadlist_obj
= ds_next
->ds_phys
->ds_deadlist_obj
;
196 ds_next
->ds_phys
->ds_deadlist_obj
= deadlist_obj
;
197 dsl_deadlist_open(&ds
->ds_deadlist
, mos
, ds
->ds_phys
->ds_deadlist_obj
);
198 dsl_deadlist_open(&ds_next
->ds_deadlist
, mos
,
199 ds_next
->ds_phys
->ds_deadlist_obj
);
203 dsl_dataset_remove_clones_key(dsl_dataset_t
*ds
, uint64_t mintxg
, dmu_tx_t
*tx
)
205 objset_t
*mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
210 * If it is the old version, dd_clones doesn't exist so we can't
211 * find the clones, but dsl_deadlist_remove_key() is a no-op so it
214 if (ds
->ds_dir
->dd_phys
->dd_clones
== 0)
217 zc
= kmem_alloc(sizeof (zap_cursor_t
), KM_SLEEP
);
218 za
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
220 for (zap_cursor_init(zc
, mos
, ds
->ds_dir
->dd_phys
->dd_clones
);
221 zap_cursor_retrieve(zc
, za
) == 0;
222 zap_cursor_advance(zc
)) {
223 dsl_dataset_t
*clone
;
225 VERIFY0(dsl_dataset_hold_obj(ds
->ds_dir
->dd_pool
,
226 za
->za_first_integer
, FTAG
, &clone
));
227 if (clone
->ds_dir
->dd_origin_txg
> mintxg
) {
228 dsl_deadlist_remove_key(&clone
->ds_deadlist
,
230 dsl_dataset_remove_clones_key(clone
, mintxg
, tx
);
232 dsl_dataset_rele(clone
, FTAG
);
236 kmem_free(za
, sizeof (zap_attribute_t
));
237 kmem_free(zc
, sizeof (zap_cursor_t
));
241 dsl_destroy_snapshot_sync_impl(dsl_dataset_t
*ds
, boolean_t defer
, dmu_tx_t
*tx
)
246 int after_branch_point
= FALSE
;
247 dsl_pool_t
*dp
= ds
->ds_dir
->dd_pool
;
248 objset_t
*mos
= dp
->dp_meta_objset
;
249 dsl_dataset_t
*ds_prev
= NULL
;
250 uint64_t obj
, old_unique
, used
= 0, comp
= 0, uncomp
= 0;
251 dsl_dataset_t
*ds_next
, *ds_head
, *hds
;
254 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
255 ASSERT3U(ds
->ds_phys
->ds_bp
.blk_birth
, <=, tx
->tx_txg
);
256 ASSERT(refcount_is_zero(&ds
->ds_longholds
));
259 (ds
->ds_userrefs
> 0 || ds
->ds_phys
->ds_num_children
> 1)) {
260 ASSERT(spa_version(dp
->dp_spa
) >= SPA_VERSION_USERREFS
);
261 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
262 ds
->ds_phys
->ds_flags
|= DS_FLAG_DEFER_DESTROY
;
263 spa_history_log_internal_ds(ds
, "defer_destroy", tx
, "");
267 ASSERT3U(ds
->ds_phys
->ds_num_children
, <=, 1);
269 /* We need to log before removing it from the namespace. */
270 spa_history_log_internal_ds(ds
, "destroy", tx
, "");
272 dsl_scan_ds_destroyed(ds
, tx
);
276 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0) {
277 ASSERT3P(ds
->ds_prev
, ==, NULL
);
278 VERIFY0(dsl_dataset_hold_obj(dp
,
279 ds
->ds_phys
->ds_prev_snap_obj
, FTAG
, &ds_prev
));
281 (ds_prev
->ds_phys
->ds_next_snap_obj
!= obj
);
283 dmu_buf_will_dirty(ds_prev
->ds_dbuf
, tx
);
284 if (after_branch_point
&&
285 ds_prev
->ds_phys
->ds_next_clones_obj
!= 0) {
286 dsl_dataset_remove_from_next_clones(ds_prev
, obj
, tx
);
287 if (ds
->ds_phys
->ds_next_snap_obj
!= 0) {
288 VERIFY0(zap_add_int(mos
,
289 ds_prev
->ds_phys
->ds_next_clones_obj
,
290 ds
->ds_phys
->ds_next_snap_obj
, tx
));
293 if (!after_branch_point
) {
294 ds_prev
->ds_phys
->ds_next_snap_obj
=
295 ds
->ds_phys
->ds_next_snap_obj
;
299 VERIFY0(dsl_dataset_hold_obj(dp
,
300 ds
->ds_phys
->ds_next_snap_obj
, FTAG
, &ds_next
));
301 ASSERT3U(ds_next
->ds_phys
->ds_prev_snap_obj
, ==, obj
);
303 old_unique
= ds_next
->ds_phys
->ds_unique_bytes
;
305 dmu_buf_will_dirty(ds_next
->ds_dbuf
, tx
);
306 ds_next
->ds_phys
->ds_prev_snap_obj
=
307 ds
->ds_phys
->ds_prev_snap_obj
;
308 ds_next
->ds_phys
->ds_prev_snap_txg
=
309 ds
->ds_phys
->ds_prev_snap_txg
;
310 ASSERT3U(ds
->ds_phys
->ds_prev_snap_txg
, ==,
311 ds_prev
? ds_prev
->ds_phys
->ds_creation_txg
: 0);
313 if (ds_next
->ds_deadlist
.dl_oldfmt
) {
314 process_old_deadlist(ds
, ds_prev
, ds_next
,
315 after_branch_point
, tx
);
317 /* Adjust prev's unique space. */
318 if (ds_prev
&& !after_branch_point
) {
319 dsl_deadlist_space_range(&ds_next
->ds_deadlist
,
320 ds_prev
->ds_phys
->ds_prev_snap_txg
,
321 ds
->ds_phys
->ds_prev_snap_txg
,
322 &used
, &comp
, &uncomp
);
323 ds_prev
->ds_phys
->ds_unique_bytes
+= used
;
326 /* Adjust snapused. */
327 dsl_deadlist_space_range(&ds_next
->ds_deadlist
,
328 ds
->ds_phys
->ds_prev_snap_txg
, UINT64_MAX
,
329 &used
, &comp
, &uncomp
);
330 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_SNAP
,
331 -used
, -comp
, -uncomp
, tx
);
333 /* Move blocks to be freed to pool's free list. */
334 dsl_deadlist_move_bpobj(&ds_next
->ds_deadlist
,
335 &dp
->dp_free_bpobj
, ds
->ds_phys
->ds_prev_snap_txg
,
337 dsl_dir_diduse_space(tx
->tx_pool
->dp_free_dir
,
338 DD_USED_HEAD
, used
, comp
, uncomp
, tx
);
340 /* Merge our deadlist into next's and free it. */
341 dsl_deadlist_merge(&ds_next
->ds_deadlist
,
342 ds
->ds_phys
->ds_deadlist_obj
, tx
);
344 dsl_deadlist_close(&ds
->ds_deadlist
);
345 dsl_deadlist_free(mos
, ds
->ds_phys
->ds_deadlist_obj
, tx
);
346 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
347 ds
->ds_phys
->ds_deadlist_obj
= 0;
349 /* Collapse range in clone heads */
350 dsl_dataset_remove_clones_key(ds
,
351 ds
->ds_phys
->ds_creation_txg
, tx
);
353 if (dsl_dataset_is_snapshot(ds_next
)) {
354 dsl_dataset_t
*ds_nextnext
;
357 * Update next's unique to include blocks which
358 * were previously shared by only this snapshot
359 * and it. Those blocks will be born after the
360 * prev snap and before this snap, and will have
361 * died after the next snap and before the one
362 * after that (ie. be on the snap after next's
365 VERIFY0(dsl_dataset_hold_obj(dp
,
366 ds_next
->ds_phys
->ds_next_snap_obj
, FTAG
, &ds_nextnext
));
367 dsl_deadlist_space_range(&ds_nextnext
->ds_deadlist
,
368 ds
->ds_phys
->ds_prev_snap_txg
,
369 ds
->ds_phys
->ds_creation_txg
,
370 &used
, &comp
, &uncomp
);
371 ds_next
->ds_phys
->ds_unique_bytes
+= used
;
372 dsl_dataset_rele(ds_nextnext
, FTAG
);
373 ASSERT3P(ds_next
->ds_prev
, ==, NULL
);
375 /* Collapse range in this head. */
376 VERIFY0(dsl_dataset_hold_obj(dp
,
377 ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
, FTAG
, &hds
));
378 dsl_deadlist_remove_key(&hds
->ds_deadlist
,
379 ds
->ds_phys
->ds_creation_txg
, tx
);
380 dsl_dataset_rele(hds
, FTAG
);
383 ASSERT3P(ds_next
->ds_prev
, ==, ds
);
384 dsl_dataset_rele(ds_next
->ds_prev
, ds_next
);
385 ds_next
->ds_prev
= NULL
;
387 VERIFY0(dsl_dataset_hold_obj(dp
,
388 ds
->ds_phys
->ds_prev_snap_obj
,
389 ds_next
, &ds_next
->ds_prev
));
392 dsl_dataset_recalc_head_uniq(ds_next
);
395 * Reduce the amount of our unconsumed refreservation
396 * being charged to our parent by the amount of
397 * new unique data we have gained.
399 if (old_unique
< ds_next
->ds_reserved
) {
401 uint64_t new_unique
=
402 ds_next
->ds_phys
->ds_unique_bytes
;
404 ASSERT(old_unique
<= new_unique
);
405 mrsdelta
= MIN(new_unique
- old_unique
,
406 ds_next
->ds_reserved
- old_unique
);
407 dsl_dir_diduse_space(ds
->ds_dir
,
408 DD_USED_REFRSRV
, -mrsdelta
, 0, 0, tx
);
411 dsl_dataset_rele(ds_next
, FTAG
);
414 * This must be done after the dsl_traverse(), because it will
415 * re-open the objset.
418 dmu_objset_evict(ds
->ds_objset
);
419 ds
->ds_objset
= NULL
;
422 /* remove from snapshot namespace */
423 ASSERT(ds
->ds_phys
->ds_snapnames_zapobj
== 0);
424 VERIFY0(dsl_dataset_hold_obj(dp
,
425 ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
, FTAG
, &ds_head
));
426 VERIFY0(dsl_dataset_get_snapname(ds
));
431 err
= dsl_dataset_snap_lookup(ds_head
,
432 ds
->ds_snapname
, &val
);
434 ASSERT3U(val
, ==, obj
);
437 VERIFY0(dsl_dataset_snap_remove(ds_head
, ds
->ds_snapname
, tx
));
438 dsl_dataset_rele(ds_head
, FTAG
);
441 dsl_dataset_rele(ds_prev
, FTAG
);
443 spa_prop_clear_bootfs(dp
->dp_spa
, ds
->ds_object
, tx
);
445 if (ds
->ds_phys
->ds_next_clones_obj
!= 0) {
446 ASSERTV(uint64_t count
);
447 ASSERT0(zap_count(mos
,
448 ds
->ds_phys
->ds_next_clones_obj
, &count
) && count
== 0);
449 VERIFY0(dmu_object_free(mos
,
450 ds
->ds_phys
->ds_next_clones_obj
, tx
));
452 if (ds
->ds_phys
->ds_props_obj
!= 0)
453 VERIFY0(zap_destroy(mos
, ds
->ds_phys
->ds_props_obj
, tx
));
454 if (ds
->ds_phys
->ds_userrefs_obj
!= 0)
455 VERIFY0(zap_destroy(mos
, ds
->ds_phys
->ds_userrefs_obj
, tx
));
456 dsl_dir_rele(ds
->ds_dir
, ds
);
458 dmu_object_free_zapified(mos
, obj
, tx
);
462 dsl_destroy_snapshot_sync(void *arg
, dmu_tx_t
*tx
)
464 dmu_snapshots_destroy_arg_t
*dsda
= arg
;
465 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
468 for (pair
= nvlist_next_nvpair(dsda
->dsda_successful_snaps
, NULL
);
470 pair
= nvlist_next_nvpair(dsda
->dsda_successful_snaps
, pair
)) {
473 VERIFY0(dsl_dataset_hold(dp
, nvpair_name(pair
), FTAG
, &ds
));
475 dsl_destroy_snapshot_sync_impl(ds
, dsda
->dsda_defer
, tx
);
476 dsl_dataset_rele(ds
, FTAG
);
481 * The semantics of this function are described in the comment above
482 * lzc_destroy_snaps(). To summarize:
484 * The snapshots must all be in the same pool.
486 * Snapshots that don't exist will be silently ignored (considered to be
487 * "already deleted").
489 * On success, all snaps will be destroyed and this will return 0.
490 * On failure, no snaps will be destroyed, the errlist will be filled in,
491 * and this will return an errno.
494 dsl_destroy_snapshots_nvl(nvlist_t
*snaps
, boolean_t defer
,
497 dmu_snapshots_destroy_arg_t dsda
;
501 pair
= nvlist_next_nvpair(snaps
, NULL
);
505 dsda
.dsda_snaps
= snaps
;
506 VERIFY0(nvlist_alloc(&dsda
.dsda_successful_snaps
,
507 NV_UNIQUE_NAME
, KM_SLEEP
));
508 dsda
.dsda_defer
= defer
;
509 dsda
.dsda_errlist
= errlist
;
511 error
= dsl_sync_task(nvpair_name(pair
),
512 dsl_destroy_snapshot_check
, dsl_destroy_snapshot_sync
,
514 fnvlist_free(dsda
.dsda_successful_snaps
);
520 dsl_destroy_snapshot(const char *name
, boolean_t defer
)
523 nvlist_t
*nvl
= fnvlist_alloc();
524 nvlist_t
*errlist
= fnvlist_alloc();
526 fnvlist_add_boolean(nvl
, name
);
527 error
= dsl_destroy_snapshots_nvl(nvl
, defer
, errlist
);
528 fnvlist_free(errlist
);
540 kill_blkptr(spa_t
*spa
, zilog_t
*zilog
, const blkptr_t
*bp
,
541 const zbookmark_phys_t
*zb
, const dnode_phys_t
*dnp
, void *arg
)
543 struct killarg
*ka
= arg
;
544 dmu_tx_t
*tx
= ka
->tx
;
546 if (BP_IS_HOLE(bp
) || BP_IS_EMBEDDED(bp
))
549 if (zb
->zb_level
== ZB_ZIL_LEVEL
) {
550 ASSERT(zilog
!= NULL
);
552 * It's a block in the intent log. It has no
553 * accounting, so just free it.
555 dsl_free(ka
->tx
->tx_pool
, ka
->tx
->tx_txg
, bp
);
557 ASSERT(zilog
== NULL
);
558 ASSERT3U(bp
->blk_birth
, >, ka
->ds
->ds_phys
->ds_prev_snap_txg
);
559 (void) dsl_dataset_block_kill(ka
->ds
, bp
, tx
, B_FALSE
);
566 old_synchronous_dataset_destroy(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
571 * Free everything that we point to (that's born after
572 * the previous snapshot, if we are a clone)
574 * NB: this should be very quick, because we already
575 * freed all the objects in open context.
579 VERIFY0(traverse_dataset(ds
,
580 ds
->ds_phys
->ds_prev_snap_txg
, TRAVERSE_POST
,
582 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds
) || ds
->ds_phys
->ds_unique_bytes
== 0);
585 typedef struct dsl_destroy_head_arg
{
586 const char *ddha_name
;
587 } dsl_destroy_head_arg_t
;
590 dsl_destroy_head_check_impl(dsl_dataset_t
*ds
, int expected_holds
)
596 ASSERT(!dsl_dataset_is_snapshot(ds
));
597 if (dsl_dataset_is_snapshot(ds
))
598 return (SET_ERROR(EINVAL
));
600 if (refcount_count(&ds
->ds_longholds
) != expected_holds
)
601 return (SET_ERROR(EBUSY
));
603 mos
= ds
->ds_dir
->dd_pool
->dp_meta_objset
;
606 * Can't delete a head dataset if there are snapshots of it.
607 * (Except if the only snapshots are from the branch we cloned
610 if (ds
->ds_prev
!= NULL
&&
611 ds
->ds_prev
->ds_phys
->ds_next_snap_obj
== ds
->ds_object
)
612 return (SET_ERROR(EBUSY
));
615 * Can't delete if there are children of this fs.
617 error
= zap_count(mos
,
618 ds
->ds_dir
->dd_phys
->dd_child_dir_zapobj
, &count
);
622 return (SET_ERROR(EEXIST
));
624 if (dsl_dir_is_clone(ds
->ds_dir
) && DS_IS_DEFER_DESTROY(ds
->ds_prev
) &&
625 ds
->ds_prev
->ds_phys
->ds_num_children
== 2 &&
626 ds
->ds_prev
->ds_userrefs
== 0) {
627 /* We need to remove the origin snapshot as well. */
628 if (!refcount_is_zero(&ds
->ds_prev
->ds_longholds
))
629 return (SET_ERROR(EBUSY
));
635 dsl_destroy_head_check(void *arg
, dmu_tx_t
*tx
)
637 dsl_destroy_head_arg_t
*ddha
= arg
;
638 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
642 error
= dsl_dataset_hold(dp
, ddha
->ddha_name
, FTAG
, &ds
);
646 error
= dsl_destroy_head_check_impl(ds
, 0);
647 dsl_dataset_rele(ds
, FTAG
);
652 dsl_dir_destroy_sync(uint64_t ddobj
, dmu_tx_t
*tx
)
655 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
656 objset_t
*mos
= dp
->dp_meta_objset
;
659 ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx
)->dp_config_rwlock
));
661 VERIFY0(dsl_dir_hold_obj(dp
, ddobj
, NULL
, FTAG
, &dd
));
663 ASSERT0(dd
->dd_phys
->dd_head_dataset_obj
);
666 * Remove our reservation. The impl() routine avoids setting the
667 * actual property, which would require the (already destroyed) ds.
669 dsl_dir_set_reservation_sync_impl(dd
, 0, tx
);
671 ASSERT0(dd
->dd_phys
->dd_used_bytes
);
672 ASSERT0(dd
->dd_phys
->dd_reserved
);
673 for (t
= 0; t
< DD_USED_NUM
; t
++)
674 ASSERT0(dd
->dd_phys
->dd_used_breakdown
[t
]);
676 VERIFY0(zap_destroy(mos
, dd
->dd_phys
->dd_child_dir_zapobj
, tx
));
677 VERIFY0(zap_destroy(mos
, dd
->dd_phys
->dd_props_zapobj
, tx
));
678 VERIFY0(dsl_deleg_destroy(mos
, dd
->dd_phys
->dd_deleg_zapobj
, tx
));
679 VERIFY0(zap_remove(mos
,
680 dd
->dd_parent
->dd_phys
->dd_child_dir_zapobj
, dd
->dd_myname
, tx
));
682 dsl_dir_rele(dd
, FTAG
);
683 dmu_object_free_zapified(mos
, ddobj
, tx
);
687 dsl_destroy_head_sync_impl(dsl_dataset_t
*ds
, dmu_tx_t
*tx
)
689 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
690 objset_t
*mos
= dp
->dp_meta_objset
;
691 uint64_t obj
, ddobj
, prevobj
= 0;
695 ASSERT3U(ds
->ds_phys
->ds_num_children
, <=, 1);
696 ASSERT(ds
->ds_prev
== NULL
||
697 ds
->ds_prev
->ds_phys
->ds_next_snap_obj
!= ds
->ds_object
);
698 ASSERT3U(ds
->ds_phys
->ds_bp
.blk_birth
, <=, tx
->tx_txg
);
699 ASSERT(RRW_WRITE_HELD(&dp
->dp_config_rwlock
));
701 /* We need to log before removing it from the namespace. */
702 spa_history_log_internal_ds(ds
, "destroy", tx
, "");
704 rmorigin
= (dsl_dir_is_clone(ds
->ds_dir
) &&
705 DS_IS_DEFER_DESTROY(ds
->ds_prev
) &&
706 ds
->ds_prev
->ds_phys
->ds_num_children
== 2 &&
707 ds
->ds_prev
->ds_userrefs
== 0);
709 /* Remove our reservation. */
710 if (ds
->ds_reserved
!= 0) {
711 dsl_dataset_set_refreservation_sync_impl(ds
,
712 (ZPROP_SRC_NONE
| ZPROP_SRC_LOCAL
| ZPROP_SRC_RECEIVED
),
714 ASSERT0(ds
->ds_reserved
);
717 dsl_scan_ds_destroyed(ds
, tx
);
721 if (ds
->ds_phys
->ds_prev_snap_obj
!= 0) {
722 /* This is a clone */
723 ASSERT(ds
->ds_prev
!= NULL
);
724 ASSERT3U(ds
->ds_prev
->ds_phys
->ds_next_snap_obj
, !=, obj
);
725 ASSERT0(ds
->ds_phys
->ds_next_snap_obj
);
727 dmu_buf_will_dirty(ds
->ds_prev
->ds_dbuf
, tx
);
728 if (ds
->ds_prev
->ds_phys
->ds_next_clones_obj
!= 0) {
729 dsl_dataset_remove_from_next_clones(ds
->ds_prev
,
733 ASSERT3U(ds
->ds_prev
->ds_phys
->ds_num_children
, >, 1);
734 ds
->ds_prev
->ds_phys
->ds_num_children
--;
738 * Destroy the deadlist. Unless it's a clone, the
739 * deadlist should be empty. (If it's a clone, it's
740 * safe to ignore the deadlist contents.)
742 dsl_deadlist_close(&ds
->ds_deadlist
);
743 dsl_deadlist_free(mos
, ds
->ds_phys
->ds_deadlist_obj
, tx
);
744 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
745 ds
->ds_phys
->ds_deadlist_obj
= 0;
747 VERIFY0(dmu_objset_from_ds(ds
, &os
));
749 if (!spa_feature_is_enabled(dp
->dp_spa
, SPA_FEATURE_ASYNC_DESTROY
)) {
750 old_synchronous_dataset_destroy(ds
, tx
);
753 * Move the bptree into the pool's list of trees to
754 * clean up and update space accounting information.
756 uint64_t used
, comp
, uncomp
;
758 zil_destroy_sync(dmu_objset_zil(os
), tx
);
760 if (!spa_feature_is_active(dp
->dp_spa
,
761 SPA_FEATURE_ASYNC_DESTROY
)) {
762 dsl_scan_t
*scn
= dp
->dp_scan
;
763 spa_feature_incr(dp
->dp_spa
, SPA_FEATURE_ASYNC_DESTROY
,
765 dp
->dp_bptree_obj
= bptree_alloc(mos
, tx
);
767 DMU_POOL_DIRECTORY_OBJECT
,
768 DMU_POOL_BPTREE_OBJ
, sizeof (uint64_t), 1,
769 &dp
->dp_bptree_obj
, tx
));
770 ASSERT(!scn
->scn_async_destroying
);
771 scn
->scn_async_destroying
= B_TRUE
;
774 used
= ds
->ds_dir
->dd_phys
->dd_used_bytes
;
775 comp
= ds
->ds_dir
->dd_phys
->dd_compressed_bytes
;
776 uncomp
= ds
->ds_dir
->dd_phys
->dd_uncompressed_bytes
;
778 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds
) ||
779 ds
->ds_phys
->ds_unique_bytes
== used
);
781 bptree_add(mos
, dp
->dp_bptree_obj
,
782 &ds
->ds_phys
->ds_bp
, ds
->ds_phys
->ds_prev_snap_txg
,
783 used
, comp
, uncomp
, tx
);
784 dsl_dir_diduse_space(ds
->ds_dir
, DD_USED_HEAD
,
785 -used
, -comp
, -uncomp
, tx
);
786 dsl_dir_diduse_space(dp
->dp_free_dir
, DD_USED_HEAD
,
787 used
, comp
, uncomp
, tx
);
790 if (ds
->ds_prev
!= NULL
) {
791 if (spa_version(dp
->dp_spa
) >= SPA_VERSION_DIR_CLONES
) {
792 VERIFY0(zap_remove_int(mos
,
793 ds
->ds_prev
->ds_dir
->dd_phys
->dd_clones
,
796 prevobj
= ds
->ds_prev
->ds_object
;
797 dsl_dataset_rele(ds
->ds_prev
, ds
);
802 * This must be done after the dsl_traverse(), because it will
803 * re-open the objset.
806 dmu_objset_evict(ds
->ds_objset
);
807 ds
->ds_objset
= NULL
;
810 /* Erase the link in the dir */
811 dmu_buf_will_dirty(ds
->ds_dir
->dd_dbuf
, tx
);
812 ds
->ds_dir
->dd_phys
->dd_head_dataset_obj
= 0;
813 ddobj
= ds
->ds_dir
->dd_object
;
814 ASSERT(ds
->ds_phys
->ds_snapnames_zapobj
!= 0);
815 VERIFY0(zap_destroy(mos
, ds
->ds_phys
->ds_snapnames_zapobj
, tx
));
817 if (ds
->ds_bookmarks
!= 0) {
818 VERIFY0(zap_destroy(mos
,
819 ds
->ds_bookmarks
, tx
));
820 spa_feature_decr(dp
->dp_spa
, SPA_FEATURE_BOOKMARKS
, tx
);
823 spa_prop_clear_bootfs(dp
->dp_spa
, ds
->ds_object
, tx
);
825 ASSERT0(ds
->ds_phys
->ds_next_clones_obj
);
826 ASSERT0(ds
->ds_phys
->ds_props_obj
);
827 ASSERT0(ds
->ds_phys
->ds_userrefs_obj
);
828 dsl_dir_rele(ds
->ds_dir
, ds
);
830 dmu_object_free_zapified(mos
, obj
, tx
);
832 dsl_dir_destroy_sync(ddobj
, tx
);
836 VERIFY0(dsl_dataset_hold_obj(dp
, prevobj
, FTAG
, &prev
));
837 dsl_destroy_snapshot_sync_impl(prev
, B_FALSE
, tx
);
838 dsl_dataset_rele(prev
, FTAG
);
843 dsl_destroy_head_sync(void *arg
, dmu_tx_t
*tx
)
845 dsl_destroy_head_arg_t
*ddha
= arg
;
846 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
849 VERIFY0(dsl_dataset_hold(dp
, ddha
->ddha_name
, FTAG
, &ds
));
850 dsl_destroy_head_sync_impl(ds
, tx
);
851 dsl_dataset_rele(ds
, FTAG
);
855 dsl_destroy_head_begin_sync(void *arg
, dmu_tx_t
*tx
)
857 dsl_destroy_head_arg_t
*ddha
= arg
;
858 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
861 VERIFY0(dsl_dataset_hold(dp
, ddha
->ddha_name
, FTAG
, &ds
));
863 /* Mark it as inconsistent on-disk, in case we crash */
864 dmu_buf_will_dirty(ds
->ds_dbuf
, tx
);
865 ds
->ds_phys
->ds_flags
|= DS_FLAG_INCONSISTENT
;
867 spa_history_log_internal_ds(ds
, "destroy begin", tx
, "");
868 dsl_dataset_rele(ds
, FTAG
);
872 dsl_destroy_head(const char *name
)
874 dsl_destroy_head_arg_t ddha
;
880 zfs_destroy_unmount_origin(name
);
883 error
= spa_open(name
, &spa
, FTAG
);
886 isenabled
= spa_feature_is_enabled(spa
, SPA_FEATURE_ASYNC_DESTROY
);
887 spa_close(spa
, FTAG
);
889 ddha
.ddha_name
= name
;
894 error
= dsl_sync_task(name
, dsl_destroy_head_check
,
895 dsl_destroy_head_begin_sync
, &ddha
, 0);
900 * Head deletion is processed in one txg on old pools;
901 * remove the objects from open context so that the txg sync
904 error
= dmu_objset_own(name
, DMU_OST_ANY
, B_FALSE
, FTAG
, &os
);
907 uint64_t prev_snap_txg
=
908 dmu_objset_ds(os
)->ds_phys
->ds_prev_snap_txg
;
909 for (obj
= 0; error
== 0;
910 error
= dmu_object_next(os
, &obj
, FALSE
,
912 (void) dmu_free_long_object(os
, obj
);
913 /* sync out all frees */
914 txg_wait_synced(dmu_objset_pool(os
), 0);
915 dmu_objset_disown(os
, FTAG
);
919 return (dsl_sync_task(name
, dsl_destroy_head_check
,
920 dsl_destroy_head_sync
, &ddha
, 0));
924 * Note, this function is used as the callback for dmu_objset_find(). We
925 * always return 0 so that we will continue to find and process
926 * inconsistent datasets, even if we encounter an error trying to
927 * process one of them.
931 dsl_destroy_inconsistent(const char *dsname
, void *arg
)
935 if (dmu_objset_hold(dsname
, FTAG
, &os
) == 0) {
936 boolean_t inconsistent
= DS_IS_INCONSISTENT(dmu_objset_ds(os
));
937 dmu_objset_rele(os
, FTAG
);
939 (void) dsl_destroy_head(dsname
);
945 #if defined(_KERNEL) && defined(HAVE_SPL)
946 EXPORT_SYMBOL(dsl_destroy_head
);
947 EXPORT_SYMBOL(dsl_destroy_head_sync_impl
);
948 EXPORT_SYMBOL(dsl_dataset_user_hold_check_one
);
949 EXPORT_SYMBOL(dsl_destroy_snapshot_sync_impl
);
950 EXPORT_SYMBOL(dsl_destroy_inconsistent
);
951 EXPORT_SYMBOL(dsl_dataset_user_release_tmp
);
952 EXPORT_SYMBOL(dsl_destroy_head_check_impl
);