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, 2017 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/dsl_destroy.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/dsl_pool.h>
36 #include <sys/dsl_dir.h>
37 #include <sys/dmu_traverse.h>
38 #include <sys/dsl_scan.h>
39 #include <sys/dmu_objset.h>
41 #include <sys/zfeature.h>
42 #include <sys/zfs_ioctl.h>
43 #include <sys/dsl_deleg.h>
44 #include <sys/dmu_impl.h>
49 dsl_destroy_snapshot_check_impl(dsl_dataset_t *ds, boolean_t defer)
51 if (!ds->ds_is_snapshot)
52 return (SET_ERROR(EINVAL));
54 if (dsl_dataset_long_held(ds))
55 return (SET_ERROR(EBUSY));
58 * Only allow deferred destroy on pools that support it.
59 * NOTE: deferred destroy is only supported on snapshots.
62 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
64 return (SET_ERROR(ENOTSUP));
69 * If this snapshot has an elevated user reference count,
70 * we can't destroy it yet.
72 if (ds->ds_userrefs > 0)
73 return (SET_ERROR(EBUSY));
76 * Can't delete a branch point.
78 if (dsl_dataset_phys(ds)->ds_num_children > 1)
79 return (SET_ERROR(EEXIST));
85 dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx)
87 dsl_destroy_snapshot_arg_t *ddsa = arg;
88 const char *dsname = ddsa->ddsa_name;
89 boolean_t defer = ddsa->ddsa_defer;
91 dsl_pool_t *dp = dmu_tx_pool(tx);
95 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
98 * If the snapshot does not exist, silently ignore it, and
99 * dsl_destroy_snapshot_sync() will be a no-op
100 * (it's "already destroyed").
106 error = dsl_destroy_snapshot_check_impl(ds, defer);
107 dsl_dataset_rele(ds, FTAG);
113 struct process_old_arg {
115 dsl_dataset_t *ds_prev;
116 boolean_t after_branch_point;
118 uint64_t used, comp, uncomp;
122 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
124 struct process_old_arg *poa = arg;
125 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
127 ASSERT(!BP_IS_HOLE(bp));
129 if (bp->blk_birth <= dsl_dataset_phys(poa->ds)->ds_prev_snap_txg) {
130 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
131 if (poa->ds_prev && !poa->after_branch_point &&
133 dsl_dataset_phys(poa->ds_prev)->ds_prev_snap_txg) {
134 dsl_dataset_phys(poa->ds_prev)->ds_unique_bytes +=
135 bp_get_dsize_sync(dp->dp_spa, bp);
138 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
139 poa->comp += BP_GET_PSIZE(bp);
140 poa->uncomp += BP_GET_UCSIZE(bp);
141 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
147 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
148 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
150 struct process_old_arg poa = { 0 };
151 dsl_pool_t *dp = ds->ds_dir->dd_pool;
152 objset_t *mos = dp->dp_meta_objset;
153 uint64_t deadlist_obj;
155 ASSERT(ds->ds_deadlist.dl_oldfmt);
156 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
159 poa.ds_prev = ds_prev;
160 poa.after_branch_point = after_branch_point;
161 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
162 VERIFY0(bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
163 process_old_cb, &poa, tx));
164 VERIFY0(zio_wait(poa.pio));
165 ASSERT3U(poa.used, ==, dsl_dataset_phys(ds)->ds_unique_bytes);
167 /* change snapused */
168 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
169 -poa.used, -poa.comp, -poa.uncomp, tx);
171 /* swap next's deadlist to our deadlist */
172 dsl_deadlist_close(&ds->ds_deadlist);
173 dsl_deadlist_close(&ds_next->ds_deadlist);
174 deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
175 dsl_dataset_phys(ds)->ds_deadlist_obj =
176 dsl_dataset_phys(ds_next)->ds_deadlist_obj;
177 dsl_dataset_phys(ds_next)->ds_deadlist_obj = deadlist_obj;
178 dsl_deadlist_open(&ds->ds_deadlist, mos,
179 dsl_dataset_phys(ds)->ds_deadlist_obj);
180 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
181 dsl_dataset_phys(ds_next)->ds_deadlist_obj);
185 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
187 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
192 * If it is the old version, dd_clones doesn't exist so we can't
193 * find the clones, but dsl_deadlist_remove_key() is a no-op so it
196 if (dsl_dir_phys(ds->ds_dir)->dd_clones == 0)
199 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
200 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
202 for (zap_cursor_init(zc, mos, dsl_dir_phys(ds->ds_dir)->dd_clones);
203 zap_cursor_retrieve(zc, za) == 0;
204 zap_cursor_advance(zc)) {
205 dsl_dataset_t *clone;
207 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
208 za->za_first_integer, FTAG, &clone));
209 if (clone->ds_dir->dd_origin_txg > mintxg) {
210 dsl_deadlist_remove_key(&clone->ds_deadlist,
212 if (dsl_dataset_remap_deadlist_exists(clone)) {
213 dsl_deadlist_remove_key(
214 &clone->ds_remap_deadlist, mintxg, tx);
216 dsl_dataset_remove_clones_key(clone, mintxg, tx);
218 dsl_dataset_rele(clone, FTAG);
222 kmem_free(za, sizeof (zap_attribute_t));
223 kmem_free(zc, sizeof (zap_cursor_t));
227 dsl_destroy_snapshot_handle_remaps(dsl_dataset_t *ds, dsl_dataset_t *ds_next,
230 dsl_pool_t *dp = ds->ds_dir->dd_pool;
232 /* Move blocks to be obsoleted to pool's obsolete list. */
233 if (dsl_dataset_remap_deadlist_exists(ds_next)) {
234 if (!bpobj_is_open(&dp->dp_obsolete_bpobj))
235 dsl_pool_create_obsolete_bpobj(dp, tx);
237 dsl_deadlist_move_bpobj(&ds_next->ds_remap_deadlist,
238 &dp->dp_obsolete_bpobj,
239 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
242 /* Merge our deadlist into next's and free it. */
243 if (dsl_dataset_remap_deadlist_exists(ds)) {
244 uint64_t remap_deadlist_object =
245 dsl_dataset_get_remap_deadlist_object(ds);
246 ASSERT(remap_deadlist_object != 0);
248 mutex_enter(&ds_next->ds_remap_deadlist_lock);
249 if (!dsl_dataset_remap_deadlist_exists(ds_next))
250 dsl_dataset_create_remap_deadlist(ds_next, tx);
251 mutex_exit(&ds_next->ds_remap_deadlist_lock);
253 dsl_deadlist_merge(&ds_next->ds_remap_deadlist,
254 remap_deadlist_object, tx);
255 dsl_dataset_destroy_remap_deadlist(ds, tx);
260 dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
262 int after_branch_point = FALSE;
263 dsl_pool_t *dp = ds->ds_dir->dd_pool;
264 objset_t *mos = dp->dp_meta_objset;
265 dsl_dataset_t *ds_prev = NULL;
268 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
269 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
270 ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
271 rrw_exit(&ds->ds_bp_rwlock, FTAG);
272 ASSERT(refcount_is_zero(&ds->ds_longholds));
275 (ds->ds_userrefs > 0 ||
276 dsl_dataset_phys(ds)->ds_num_children > 1)) {
277 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
278 dmu_buf_will_dirty(ds->ds_dbuf, tx);
279 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_DEFER_DESTROY;
280 spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
284 ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
286 /* We need to log before removing it from the namespace. */
287 spa_history_log_internal_ds(ds, "destroy", tx, "");
289 dsl_scan_ds_destroyed(ds, tx);
293 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
294 if (ds->ds_feature_inuse[f]) {
295 dsl_dataset_deactivate_feature(obj, f, tx);
296 ds->ds_feature_inuse[f] = B_FALSE;
299 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
300 ASSERT3P(ds->ds_prev, ==, NULL);
301 VERIFY0(dsl_dataset_hold_obj(dp,
302 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &ds_prev));
304 (dsl_dataset_phys(ds_prev)->ds_next_snap_obj != obj);
306 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
307 if (after_branch_point &&
308 dsl_dataset_phys(ds_prev)->ds_next_clones_obj != 0) {
309 dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
310 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
311 VERIFY0(zap_add_int(mos,
312 dsl_dataset_phys(ds_prev)->
314 dsl_dataset_phys(ds)->ds_next_snap_obj,
318 if (!after_branch_point) {
319 dsl_dataset_phys(ds_prev)->ds_next_snap_obj =
320 dsl_dataset_phys(ds)->ds_next_snap_obj;
324 dsl_dataset_t *ds_next;
326 uint64_t used = 0, comp = 0, uncomp = 0;
328 VERIFY0(dsl_dataset_hold_obj(dp,
329 dsl_dataset_phys(ds)->ds_next_snap_obj, FTAG, &ds_next));
330 ASSERT3U(dsl_dataset_phys(ds_next)->ds_prev_snap_obj, ==, obj);
332 old_unique = dsl_dataset_phys(ds_next)->ds_unique_bytes;
334 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
335 dsl_dataset_phys(ds_next)->ds_prev_snap_obj =
336 dsl_dataset_phys(ds)->ds_prev_snap_obj;
337 dsl_dataset_phys(ds_next)->ds_prev_snap_txg =
338 dsl_dataset_phys(ds)->ds_prev_snap_txg;
339 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
340 ds_prev ? dsl_dataset_phys(ds_prev)->ds_creation_txg : 0);
342 if (ds_next->ds_deadlist.dl_oldfmt) {
343 process_old_deadlist(ds, ds_prev, ds_next,
344 after_branch_point, tx);
346 /* Adjust prev's unique space. */
347 if (ds_prev && !after_branch_point) {
348 dsl_deadlist_space_range(&ds_next->ds_deadlist,
349 dsl_dataset_phys(ds_prev)->ds_prev_snap_txg,
350 dsl_dataset_phys(ds)->ds_prev_snap_txg,
351 &used, &comp, &uncomp);
352 dsl_dataset_phys(ds_prev)->ds_unique_bytes += used;
355 /* Adjust snapused. */
356 dsl_deadlist_space_range(&ds_next->ds_deadlist,
357 dsl_dataset_phys(ds)->ds_prev_snap_txg, UINT64_MAX,
358 &used, &comp, &uncomp);
359 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
360 -used, -comp, -uncomp, tx);
362 /* Move blocks to be freed to pool's free list. */
363 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
364 &dp->dp_free_bpobj, dsl_dataset_phys(ds)->ds_prev_snap_txg,
366 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
367 DD_USED_HEAD, used, comp, uncomp, tx);
369 /* Merge our deadlist into next's and free it. */
370 dsl_deadlist_merge(&ds_next->ds_deadlist,
371 dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
374 dsl_deadlist_close(&ds->ds_deadlist);
375 dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
376 dmu_buf_will_dirty(ds->ds_dbuf, tx);
377 dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
379 dsl_destroy_snapshot_handle_remaps(ds, ds_next, tx);
381 /* Collapse range in clone heads */
382 dsl_dataset_remove_clones_key(ds,
383 dsl_dataset_phys(ds)->ds_creation_txg, tx);
385 if (ds_next->ds_is_snapshot) {
386 dsl_dataset_t *ds_nextnext;
389 * Update next's unique to include blocks which
390 * were previously shared by only this snapshot
391 * and it. Those blocks will be born after the
392 * prev snap and before this snap, and will have
393 * died after the next snap and before the one
394 * after that (ie. be on the snap after next's
397 VERIFY0(dsl_dataset_hold_obj(dp,
398 dsl_dataset_phys(ds_next)->ds_next_snap_obj,
399 FTAG, &ds_nextnext));
400 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
401 dsl_dataset_phys(ds)->ds_prev_snap_txg,
402 dsl_dataset_phys(ds)->ds_creation_txg,
403 &used, &comp, &uncomp);
404 dsl_dataset_phys(ds_next)->ds_unique_bytes += used;
405 dsl_dataset_rele(ds_nextnext, FTAG);
406 ASSERT3P(ds_next->ds_prev, ==, NULL);
408 /* Collapse range in this head. */
410 VERIFY0(dsl_dataset_hold_obj(dp,
411 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &hds));
412 dsl_deadlist_remove_key(&hds->ds_deadlist,
413 dsl_dataset_phys(ds)->ds_creation_txg, tx);
414 if (dsl_dataset_remap_deadlist_exists(hds)) {
415 dsl_deadlist_remove_key(&hds->ds_remap_deadlist,
416 dsl_dataset_phys(ds)->ds_creation_txg, tx);
418 dsl_dataset_rele(hds, FTAG);
421 ASSERT3P(ds_next->ds_prev, ==, ds);
422 dsl_dataset_rele(ds_next->ds_prev, ds_next);
423 ds_next->ds_prev = NULL;
425 VERIFY0(dsl_dataset_hold_obj(dp,
426 dsl_dataset_phys(ds)->ds_prev_snap_obj,
427 ds_next, &ds_next->ds_prev));
430 dsl_dataset_recalc_head_uniq(ds_next);
433 * Reduce the amount of our unconsumed refreservation
434 * being charged to our parent by the amount of
435 * new unique data we have gained.
437 if (old_unique < ds_next->ds_reserved) {
439 uint64_t new_unique =
440 dsl_dataset_phys(ds_next)->ds_unique_bytes;
442 ASSERT(old_unique <= new_unique);
443 mrsdelta = MIN(new_unique - old_unique,
444 ds_next->ds_reserved - old_unique);
445 dsl_dir_diduse_space(ds->ds_dir,
446 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
449 dsl_dataset_rele(ds_next, FTAG);
452 * This must be done after the dsl_traverse(), because it will
453 * re-open the objset.
456 dmu_objset_evict(ds->ds_objset);
457 ds->ds_objset = NULL;
460 /* remove from snapshot namespace */
461 dsl_dataset_t *ds_head;
462 ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0);
463 VERIFY0(dsl_dataset_hold_obj(dp,
464 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &ds_head));
465 VERIFY0(dsl_dataset_get_snapname(ds));
471 err = dsl_dataset_snap_lookup(ds_head,
472 ds->ds_snapname, &val);
474 ASSERT3U(val, ==, obj);
477 VERIFY0(dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx, B_TRUE));
478 dsl_dataset_rele(ds_head, FTAG);
481 dsl_dataset_rele(ds_prev, FTAG);
483 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
485 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
486 ASSERTV(uint64_t count);
487 ASSERT0(zap_count(mos,
488 dsl_dataset_phys(ds)->ds_next_clones_obj, &count) &&
490 VERIFY0(dmu_object_free(mos,
491 dsl_dataset_phys(ds)->ds_next_clones_obj, tx));
493 if (dsl_dataset_phys(ds)->ds_props_obj != 0)
494 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_props_obj,
496 if (dsl_dataset_phys(ds)->ds_userrefs_obj != 0)
497 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_userrefs_obj,
499 dsl_dir_rele(ds->ds_dir, ds);
501 dmu_object_free_zapified(mos, obj, tx);
505 dsl_destroy_snapshot_sync(void *arg, dmu_tx_t *tx)
507 dsl_destroy_snapshot_arg_t *ddsa = arg;
508 const char *dsname = ddsa->ddsa_name;
509 boolean_t defer = ddsa->ddsa_defer;
511 dsl_pool_t *dp = dmu_tx_pool(tx);
514 int error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
518 dsl_destroy_snapshot_sync_impl(ds, defer, tx);
519 zvol_remove_minors(dp->dp_spa, dsname, B_TRUE);
520 dsl_dataset_rele(ds, FTAG);
524 * The semantics of this function are described in the comment above
525 * lzc_destroy_snaps(). To summarize:
527 * The snapshots must all be in the same pool.
529 * Snapshots that don't exist will be silently ignored (considered to be
530 * "already deleted").
532 * On success, all snaps will be destroyed and this will return 0.
533 * On failure, no snaps will be destroyed, the errlist will be filled in,
534 * and this will return an errno.
537 dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t defer,
540 if (nvlist_next_nvpair(snaps, NULL) == NULL)
544 * lzc_destroy_snaps() is documented to take an nvlist whose
545 * values "don't matter". We need to convert that nvlist to
546 * one that we know can be converted to LUA. We also don't
547 * care about any duplicate entries because the nvlist will
548 * be converted to a LUA table which should take care of this.
550 nvlist_t *snaps_normalized;
551 VERIFY0(nvlist_alloc(&snaps_normalized, 0, KM_SLEEP));
552 for (nvpair_t *pair = nvlist_next_nvpair(snaps, NULL);
553 pair != NULL; pair = nvlist_next_nvpair(snaps, pair)) {
554 fnvlist_add_boolean_value(snaps_normalized,
555 nvpair_name(pair), B_TRUE);
559 VERIFY0(nvlist_alloc(&arg, 0, KM_SLEEP));
560 fnvlist_add_nvlist(arg, "snaps", snaps_normalized);
561 fnvlist_free(snaps_normalized);
562 fnvlist_add_boolean_value(arg, "defer", defer);
565 VERIFY0(nvlist_alloc(&wrapper, 0, KM_SLEEP));
566 fnvlist_add_nvlist(wrapper, ZCP_ARG_ARGLIST, arg);
569 const char *program =
571 "snaps = arg['snaps']\n"
572 "defer = arg['defer']\n"
574 "has_errors = false\n"
575 "for snap, v in pairs(snaps) do\n"
576 " errno = zfs.check.destroy{snap, defer=defer}\n"
577 " zfs.debug('snap: ' .. snap .. ' errno: ' .. errno)\n"
578 " if errno == ENOENT then\n"
579 " snaps[snap] = nil\n"
580 " elseif errno ~= 0 then\n"
581 " errors[snap] = errno\n"
582 " has_errors = true\n"
585 "if has_errors then\n"
588 "for snap, v in pairs(snaps) do\n"
589 " errno = zfs.sync.destroy{snap, defer=defer}\n"
590 " assert(errno == 0)\n"
594 nvlist_t *result = fnvlist_alloc();
595 int error = zcp_eval(nvpair_name(nvlist_next_nvpair(snaps, NULL)),
599 zfs_lua_max_memlimit,
600 nvlist_next_nvpair(wrapper, NULL), result);
602 char *errorstr = NULL;
603 (void) nvlist_lookup_string(result, ZCP_RET_ERROR, &errorstr);
604 if (errorstr != NULL) {
605 zfs_dbgmsg(errorstr);
609 fnvlist_free(wrapper);
612 * lzc_destroy_snaps() is documented to fill the errlist with
613 * int32 values, so we need to covert the int64 values that are
617 nvlist_t *errlist_raw = fnvlist_lookup_nvlist(result, ZCP_RET_RETURN);
618 for (nvpair_t *pair = nvlist_next_nvpair(errlist_raw, NULL);
619 pair != NULL; pair = nvlist_next_nvpair(errlist_raw, pair)) {
620 int32_t val = (int32_t)fnvpair_value_int64(pair);
623 fnvlist_add_int32(errlist, nvpair_name(pair), val);
625 fnvlist_free(result);
630 dsl_destroy_snapshot(const char *name, boolean_t defer)
633 nvlist_t *nvl = fnvlist_alloc();
634 nvlist_t *errlist = fnvlist_alloc();
636 fnvlist_add_boolean(nvl, name);
637 error = dsl_destroy_snapshots_nvl(nvl, defer, errlist);
638 fnvlist_free(errlist);
650 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
651 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
653 struct killarg *ka = arg;
654 dmu_tx_t *tx = ka->tx;
656 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
659 if (zb->zb_level == ZB_ZIL_LEVEL) {
660 ASSERT(zilog != NULL);
662 * It's a block in the intent log. It has no
663 * accounting, so just free it.
665 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
667 ASSERT(zilog == NULL);
668 ASSERT3U(bp->blk_birth, >,
669 dsl_dataset_phys(ka->ds)->ds_prev_snap_txg);
670 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
677 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
682 * Free everything that we point to (that's born after
683 * the previous snapshot, if we are a clone)
685 * NB: this should be very quick, because we already
686 * freed all the objects in open context.
690 VERIFY0(traverse_dataset(ds,
691 dsl_dataset_phys(ds)->ds_prev_snap_txg, TRAVERSE_POST |
692 TRAVERSE_NO_DECRYPT, kill_blkptr, &ka));
693 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
694 dsl_dataset_phys(ds)->ds_unique_bytes == 0);
698 dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
704 ASSERT(!ds->ds_is_snapshot);
705 if (ds->ds_is_snapshot)
706 return (SET_ERROR(EINVAL));
708 if (refcount_count(&ds->ds_longholds) != expected_holds)
709 return (SET_ERROR(EBUSY));
711 mos = ds->ds_dir->dd_pool->dp_meta_objset;
714 * Can't delete a head dataset if there are snapshots of it.
715 * (Except if the only snapshots are from the branch we cloned
718 if (ds->ds_prev != NULL &&
719 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object)
720 return (SET_ERROR(EBUSY));
723 * Can't delete if there are children of this fs.
725 error = zap_count(mos,
726 dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, &count);
730 return (SET_ERROR(EEXIST));
732 if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev) &&
733 dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
734 ds->ds_prev->ds_userrefs == 0) {
735 /* We need to remove the origin snapshot as well. */
736 if (!refcount_is_zero(&ds->ds_prev->ds_longholds))
737 return (SET_ERROR(EBUSY));
743 dsl_destroy_head_check(void *arg, dmu_tx_t *tx)
745 dsl_destroy_head_arg_t *ddha = arg;
746 dsl_pool_t *dp = dmu_tx_pool(tx);
750 error = dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds);
754 error = dsl_destroy_head_check_impl(ds, 0);
755 dsl_dataset_rele(ds, FTAG);
760 dsl_dir_destroy_sync(uint64_t ddobj, dmu_tx_t *tx)
763 dsl_pool_t *dp = dmu_tx_pool(tx);
764 objset_t *mos = dp->dp_meta_objset;
767 ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx)->dp_config_rwlock));
769 VERIFY0(dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd));
771 ASSERT0(dsl_dir_phys(dd)->dd_head_dataset_obj);
774 * Decrement the filesystem count for all parent filesystems.
776 * When we receive an incremental stream into a filesystem that already
777 * exists, a temporary clone is created. We never count this temporary
778 * clone, whose name begins with a '%'.
780 if (dd->dd_myname[0] != '%' && dd->dd_parent != NULL)
781 dsl_fs_ss_count_adjust(dd->dd_parent, -1,
782 DD_FIELD_FILESYSTEM_COUNT, tx);
785 * Remove our reservation. The impl() routine avoids setting the
786 * actual property, which would require the (already destroyed) ds.
788 dsl_dir_set_reservation_sync_impl(dd, 0, tx);
790 ASSERT0(dsl_dir_phys(dd)->dd_used_bytes);
791 ASSERT0(dsl_dir_phys(dd)->dd_reserved);
792 for (t = 0; t < DD_USED_NUM; t++)
793 ASSERT0(dsl_dir_phys(dd)->dd_used_breakdown[t]);
795 if (dd->dd_crypto_obj != 0) {
796 dsl_crypto_key_destroy_sync(dd->dd_crypto_obj, tx);
797 (void) spa_keystore_unload_wkey_impl(dp->dp_spa, dd->dd_object);
800 VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_child_dir_zapobj, tx));
801 VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_props_zapobj, tx));
802 VERIFY0(dsl_deleg_destroy(mos, dsl_dir_phys(dd)->dd_deleg_zapobj, tx));
803 VERIFY0(zap_remove(mos,
804 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
807 dsl_dir_rele(dd, FTAG);
808 dmu_object_free_zapified(mos, ddobj, tx);
812 dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
814 dsl_pool_t *dp = dmu_tx_pool(tx);
815 objset_t *mos = dp->dp_meta_objset;
816 uint64_t obj, ddobj, prevobj = 0;
819 ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
820 ASSERT(ds->ds_prev == NULL ||
821 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj != ds->ds_object);
822 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
823 ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
824 rrw_exit(&ds->ds_bp_rwlock, FTAG);
825 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
827 /* We need to log before removing it from the namespace. */
828 spa_history_log_internal_ds(ds, "destroy", tx, "");
830 rmorigin = (dsl_dir_is_clone(ds->ds_dir) &&
831 DS_IS_DEFER_DESTROY(ds->ds_prev) &&
832 dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
833 ds->ds_prev->ds_userrefs == 0);
835 /* Remove our reservation. */
836 if (ds->ds_reserved != 0) {
837 dsl_dataset_set_refreservation_sync_impl(ds,
838 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
840 ASSERT0(ds->ds_reserved);
845 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
846 if (ds->ds_feature_inuse[f]) {
847 dsl_dataset_deactivate_feature(obj, f, tx);
848 ds->ds_feature_inuse[f] = B_FALSE;
852 dsl_scan_ds_destroyed(ds, tx);
854 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
855 /* This is a clone */
856 ASSERT(ds->ds_prev != NULL);
857 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj, !=,
859 ASSERT0(dsl_dataset_phys(ds)->ds_next_snap_obj);
861 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
862 if (dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj != 0) {
863 dsl_dataset_remove_from_next_clones(ds->ds_prev,
867 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_num_children, >, 1);
868 dsl_dataset_phys(ds->ds_prev)->ds_num_children--;
872 * Destroy the deadlist. Unless it's a clone, the
873 * deadlist should be empty since the dataset has no snapshots.
874 * (If it's a clone, it's safe to ignore the deadlist contents
875 * since they are still referenced by the origin snapshot.)
877 dsl_deadlist_close(&ds->ds_deadlist);
878 dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
879 dmu_buf_will_dirty(ds->ds_dbuf, tx);
880 dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
882 if (dsl_dataset_remap_deadlist_exists(ds))
883 dsl_dataset_destroy_remap_deadlist(ds, tx);
886 VERIFY0(dmu_objset_from_ds(ds, &os));
888 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) {
889 old_synchronous_dataset_destroy(ds, tx);
892 * Move the bptree into the pool's list of trees to
893 * clean up and update space accounting information.
895 uint64_t used, comp, uncomp;
897 zil_destroy_sync(dmu_objset_zil(os), tx);
899 if (!spa_feature_is_active(dp->dp_spa,
900 SPA_FEATURE_ASYNC_DESTROY)) {
901 dsl_scan_t *scn = dp->dp_scan;
902 spa_feature_incr(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY,
904 dp->dp_bptree_obj = bptree_alloc(mos, tx);
906 DMU_POOL_DIRECTORY_OBJECT,
907 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
908 &dp->dp_bptree_obj, tx));
909 ASSERT(!scn->scn_async_destroying);
910 scn->scn_async_destroying = B_TRUE;
913 used = dsl_dir_phys(ds->ds_dir)->dd_used_bytes;
914 comp = dsl_dir_phys(ds->ds_dir)->dd_compressed_bytes;
915 uncomp = dsl_dir_phys(ds->ds_dir)->dd_uncompressed_bytes;
917 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
918 dsl_dataset_phys(ds)->ds_unique_bytes == used);
920 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
921 bptree_add(mos, dp->dp_bptree_obj,
922 &dsl_dataset_phys(ds)->ds_bp,
923 dsl_dataset_phys(ds)->ds_prev_snap_txg,
924 used, comp, uncomp, tx);
925 rrw_exit(&ds->ds_bp_rwlock, FTAG);
926 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
927 -used, -comp, -uncomp, tx);
928 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
929 used, comp, uncomp, tx);
932 if (ds->ds_prev != NULL) {
933 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
934 VERIFY0(zap_remove_int(mos,
935 dsl_dir_phys(ds->ds_prev->ds_dir)->dd_clones,
938 prevobj = ds->ds_prev->ds_object;
939 dsl_dataset_rele(ds->ds_prev, ds);
944 * This must be done after the dsl_traverse(), because it will
945 * re-open the objset.
948 dmu_objset_evict(ds->ds_objset);
949 ds->ds_objset = NULL;
952 /* Erase the link in the dir */
953 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
954 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj = 0;
955 ddobj = ds->ds_dir->dd_object;
956 ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0);
957 VERIFY0(zap_destroy(mos,
958 dsl_dataset_phys(ds)->ds_snapnames_zapobj, tx));
960 if (ds->ds_bookmarks != 0) {
961 VERIFY0(zap_destroy(mos, ds->ds_bookmarks, tx));
962 spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
965 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
967 ASSERT0(dsl_dataset_phys(ds)->ds_next_clones_obj);
968 ASSERT0(dsl_dataset_phys(ds)->ds_props_obj);
969 ASSERT0(dsl_dataset_phys(ds)->ds_userrefs_obj);
970 dsl_dir_rele(ds->ds_dir, ds);
972 dmu_object_free_zapified(mos, obj, tx);
974 dsl_dir_destroy_sync(ddobj, tx);
978 VERIFY0(dsl_dataset_hold_obj(dp, prevobj, FTAG, &prev));
979 dsl_destroy_snapshot_sync_impl(prev, B_FALSE, tx);
980 dsl_dataset_rele(prev, FTAG);
985 dsl_destroy_head_sync(void *arg, dmu_tx_t *tx)
987 dsl_destroy_head_arg_t *ddha = arg;
988 dsl_pool_t *dp = dmu_tx_pool(tx);
991 VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
992 dsl_destroy_head_sync_impl(ds, tx);
993 zvol_remove_minors(dp->dp_spa, ddha->ddha_name, B_TRUE);
994 dsl_dataset_rele(ds, FTAG);
998 dsl_destroy_head_begin_sync(void *arg, dmu_tx_t *tx)
1000 dsl_destroy_head_arg_t *ddha = arg;
1001 dsl_pool_t *dp = dmu_tx_pool(tx);
1004 VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
1006 /* Mark it as inconsistent on-disk, in case we crash */
1007 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1008 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT;
1010 spa_history_log_internal_ds(ds, "destroy begin", tx, "");
1011 dsl_dataset_rele(ds, FTAG);
1015 dsl_destroy_head(const char *name)
1017 dsl_destroy_head_arg_t ddha;
1020 boolean_t isenabled;
1023 zfs_destroy_unmount_origin(name);
1026 error = spa_open(name, &spa, FTAG);
1029 isenabled = spa_feature_is_enabled(spa, SPA_FEATURE_ASYNC_DESTROY);
1030 spa_close(spa, FTAG);
1032 ddha.ddha_name = name;
1037 error = dsl_sync_task(name, dsl_destroy_head_check,
1038 dsl_destroy_head_begin_sync, &ddha,
1039 0, ZFS_SPACE_CHECK_DESTROY);
1044 * Head deletion is processed in one txg on old pools;
1045 * remove the objects from open context so that the txg sync
1048 error = dmu_objset_own(name, DMU_OST_ANY, B_FALSE, B_FALSE,
1051 uint64_t prev_snap_txg =
1052 dsl_dataset_phys(dmu_objset_ds(os))->
1054 for (uint64_t obj = 0; error == 0;
1055 error = dmu_object_next(os, &obj, FALSE,
1057 (void) dmu_free_long_object(os, obj);
1058 /* sync out all frees */
1059 txg_wait_synced(dmu_objset_pool(os), 0);
1060 dmu_objset_disown(os, B_FALSE, FTAG);
1064 return (dsl_sync_task(name, dsl_destroy_head_check,
1065 dsl_destroy_head_sync, &ddha, 0, ZFS_SPACE_CHECK_DESTROY));
1069 * Note, this function is used as the callback for dmu_objset_find(). We
1070 * always return 0 so that we will continue to find and process
1071 * inconsistent datasets, even if we encounter an error trying to
1072 * process one of them.
1076 dsl_destroy_inconsistent(const char *dsname, void *arg)
1080 if (dmu_objset_hold(dsname, FTAG, &os) == 0) {
1081 boolean_t need_destroy = DS_IS_INCONSISTENT(dmu_objset_ds(os));
1084 * If the dataset is inconsistent because a resumable receive
1085 * has failed, then do not destroy it.
1087 if (dsl_dataset_has_resume_receive_state(dmu_objset_ds(os)))
1088 need_destroy = B_FALSE;
1090 dmu_objset_rele(os, FTAG);
1092 (void) dsl_destroy_head(dsname);
1098 #if defined(_KERNEL)
1099 EXPORT_SYMBOL(dsl_destroy_head);
1100 EXPORT_SYMBOL(dsl_destroy_head_sync_impl);
1101 EXPORT_SYMBOL(dsl_dataset_user_hold_check_one);
1102 EXPORT_SYMBOL(dsl_destroy_snapshot_sync_impl);
1103 EXPORT_SYMBOL(dsl_destroy_inconsistent);
1104 EXPORT_SYMBOL(dsl_dataset_user_release_tmp);
1105 EXPORT_SYMBOL(dsl_destroy_head_check_impl);