1 /*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2018 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) 2014 Integros [integros.com]
27 */
28
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>
40 #include <sys/zap.h>
41 #include <sys/zfeature.h>
42 #include <sys/zfs_ioctl.h>
43 #include <sys/dsl_deleg.h>
44 #include <sys/dmu_impl.h>
45 #include <sys/zcp.h>
46
47 int
dsl_destroy_snapshot_check_impl(dsl_dataset_t * ds,boolean_t defer)48 dsl_destroy_snapshot_check_impl(dsl_dataset_t *ds, boolean_t defer)
49 {
50 if (!ds->ds_is_snapshot)
51 return (SET_ERROR(EINVAL));
52
53 if (dsl_dataset_long_held(ds))
54 return (SET_ERROR(EBUSY));
55
56 /*
57 * Only allow deferred destroy on pools that support it.
58 * NOTE: deferred destroy is only supported on snapshots.
59 */
60 if (defer) {
61 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
62 SPA_VERSION_USERREFS)
63 return (SET_ERROR(ENOTSUP));
64 return (0);
65 }
66
67 /*
68 * If this snapshot has an elevated user reference count,
69 * we can't destroy it yet.
70 */
71 if (ds->ds_userrefs > 0)
72 return (SET_ERROR(EBUSY));
73
74 /*
75 * Can't delete a branch point.
76 */
77 if (dsl_dataset_phys(ds)->ds_num_children > 1)
78 return (SET_ERROR(EEXIST));
79
80 return (0);
81 }
82
83 int
dsl_destroy_snapshot_check(void * arg,dmu_tx_t * tx)84 dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx)
85 {
86 dsl_destroy_snapshot_arg_t *ddsa = arg;
87 const char *dsname = ddsa->ddsa_name;
88 boolean_t defer = ddsa->ddsa_defer;
89
90 dsl_pool_t *dp = dmu_tx_pool(tx);
91 int error = 0;
92 dsl_dataset_t *ds;
93
94 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
95
96 /*
97 * If the snapshot does not exist, silently ignore it, and
98 * dsl_destroy_snapshot_sync() will be a no-op
99 * (it's "already destroyed").
100 */
101 if (error == ENOENT)
102 return (0);
103
104 if (error == 0) {
105 error = dsl_destroy_snapshot_check_impl(ds, defer);
106 dsl_dataset_rele(ds, FTAG);
107 }
108
109 return (error);
110 }
111
112 struct process_old_arg {
113 dsl_dataset_t *ds;
114 dsl_dataset_t *ds_prev;
115 boolean_t after_branch_point;
116 zio_t *pio;
117 uint64_t used, comp, uncomp;
118 };
119
120 static int
process_old_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)121 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
122 {
123 struct process_old_arg *poa = arg;
124 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
125
126 ASSERT(!BP_IS_HOLE(bp));
127
128 if (bp->blk_birth <= dsl_dataset_phys(poa->ds)->ds_prev_snap_txg) {
129 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
130 if (poa->ds_prev && !poa->after_branch_point &&
131 bp->blk_birth >
132 dsl_dataset_phys(poa->ds_prev)->ds_prev_snap_txg) {
133 dsl_dataset_phys(poa->ds_prev)->ds_unique_bytes +=
134 bp_get_dsize_sync(dp->dp_spa, bp);
135 }
136 } else {
137 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
138 poa->comp += BP_GET_PSIZE(bp);
139 poa->uncomp += BP_GET_UCSIZE(bp);
140 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
141 }
142 return (0);
143 }
144
145 static void
process_old_deadlist(dsl_dataset_t * ds,dsl_dataset_t * ds_prev,dsl_dataset_t * ds_next,boolean_t after_branch_point,dmu_tx_t * tx)146 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
147 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
148 {
149 struct process_old_arg poa = { 0 };
150 dsl_pool_t *dp = ds->ds_dir->dd_pool;
151 objset_t *mos = dp->dp_meta_objset;
152 uint64_t deadlist_obj;
153
154 ASSERT(ds->ds_deadlist.dl_oldfmt);
155 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
156
157 poa.ds = ds;
158 poa.ds_prev = ds_prev;
159 poa.after_branch_point = after_branch_point;
160 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
161 VERIFY0(bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
162 process_old_cb, &poa, tx));
163 VERIFY0(zio_wait(poa.pio));
164 ASSERT3U(poa.used, ==, dsl_dataset_phys(ds)->ds_unique_bytes);
165
166 /* change snapused */
167 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
168 -poa.used, -poa.comp, -poa.uncomp, tx);
169
170 /* swap next's deadlist to our deadlist */
171 dsl_deadlist_close(&ds->ds_deadlist);
172 dsl_deadlist_close(&ds_next->ds_deadlist);
173 deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
174 dsl_dataset_phys(ds)->ds_deadlist_obj =
175 dsl_dataset_phys(ds_next)->ds_deadlist_obj;
176 dsl_dataset_phys(ds_next)->ds_deadlist_obj = deadlist_obj;
177 dsl_deadlist_open(&ds->ds_deadlist, mos,
178 dsl_dataset_phys(ds)->ds_deadlist_obj);
179 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
180 dsl_dataset_phys(ds_next)->ds_deadlist_obj);
181 }
182
183 static void
dsl_dataset_remove_clones_key(dsl_dataset_t * ds,uint64_t mintxg,dmu_tx_t * tx)184 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
185 {
186 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
187 zap_cursor_t zc;
188 zap_attribute_t za;
189
190 /*
191 * If it is the old version, dd_clones doesn't exist so we can't
192 * find the clones, but dsl_deadlist_remove_key() is a no-op so it
193 * doesn't matter.
194 */
195 if (dsl_dir_phys(ds->ds_dir)->dd_clones == 0)
196 return;
197
198 for (zap_cursor_init(&zc, mos, dsl_dir_phys(ds->ds_dir)->dd_clones);
199 zap_cursor_retrieve(&zc, &za) == 0;
200 zap_cursor_advance(&zc)) {
201 dsl_dataset_t *clone;
202
203 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
204 za.za_first_integer, FTAG, &clone));
205 if (clone->ds_dir->dd_origin_txg > mintxg) {
206 dsl_deadlist_remove_key(&clone->ds_deadlist,
207 mintxg, tx);
208 if (dsl_dataset_remap_deadlist_exists(clone)) {
209 dsl_deadlist_remove_key(
210 &clone->ds_remap_deadlist, mintxg, tx);
211 }
212 dsl_dataset_remove_clones_key(clone, mintxg, tx);
213 }
214 dsl_dataset_rele(clone, FTAG);
215 }
216 zap_cursor_fini(&zc);
217 }
218
219 static void
dsl_destroy_snapshot_handle_remaps(dsl_dataset_t * ds,dsl_dataset_t * ds_next,dmu_tx_t * tx)220 dsl_destroy_snapshot_handle_remaps(dsl_dataset_t *ds, dsl_dataset_t *ds_next,
221 dmu_tx_t *tx)
222 {
223 dsl_pool_t *dp = ds->ds_dir->dd_pool;
224
225 /* Move blocks to be obsoleted to pool's obsolete list. */
226 if (dsl_dataset_remap_deadlist_exists(ds_next)) {
227 if (!bpobj_is_open(&dp->dp_obsolete_bpobj))
228 dsl_pool_create_obsolete_bpobj(dp, tx);
229
230 dsl_deadlist_move_bpobj(&ds_next->ds_remap_deadlist,
231 &dp->dp_obsolete_bpobj,
232 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
233 }
234
235 /* Merge our deadlist into next's and free it. */
236 if (dsl_dataset_remap_deadlist_exists(ds)) {
237 uint64_t remap_deadlist_object =
238 dsl_dataset_get_remap_deadlist_object(ds);
239 ASSERT(remap_deadlist_object != 0);
240
241 mutex_enter(&ds_next->ds_remap_deadlist_lock);
242 if (!dsl_dataset_remap_deadlist_exists(ds_next))
243 dsl_dataset_create_remap_deadlist(ds_next, tx);
244 mutex_exit(&ds_next->ds_remap_deadlist_lock);
245
246 dsl_deadlist_merge(&ds_next->ds_remap_deadlist,
247 remap_deadlist_object, tx);
248 dsl_dataset_destroy_remap_deadlist(ds, tx);
249 }
250 }
251
252 void
dsl_destroy_snapshot_sync_impl(dsl_dataset_t * ds,boolean_t defer,dmu_tx_t * tx)253 dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
254 {
255 int err;
256 int after_branch_point = FALSE;
257 dsl_pool_t *dp = ds->ds_dir->dd_pool;
258 objset_t *mos = dp->dp_meta_objset;
259 dsl_dataset_t *ds_prev = NULL;
260 uint64_t obj;
261
262 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
263 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
264 ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
265 rrw_exit(&ds->ds_bp_rwlock, FTAG);
266 ASSERT(zfs_refcount_is_zero(&ds->ds_longholds));
267
268 if (defer &&
269 (ds->ds_userrefs > 0 ||
270 dsl_dataset_phys(ds)->ds_num_children > 1)) {
271 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
272 dmu_buf_will_dirty(ds->ds_dbuf, tx);
273 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_DEFER_DESTROY;
274 spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
275 return;
276 }
277
278 ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
279
280 /* We need to log before removing it from the namespace. */
281 spa_history_log_internal_ds(ds, "destroy", tx, "");
282
283 dsl_scan_ds_destroyed(ds, tx);
284
285 obj = ds->ds_object;
286
287 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
288 if (dsl_dataset_feature_is_active(ds, f))
289 dsl_dataset_deactivate_feature(ds, f, tx);
290 }
291 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
292 ASSERT3P(ds->ds_prev, ==, NULL);
293 VERIFY0(dsl_dataset_hold_obj(dp,
294 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &ds_prev));
295 after_branch_point =
296 (dsl_dataset_phys(ds_prev)->ds_next_snap_obj != obj);
297
298 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
299 if (after_branch_point &&
300 dsl_dataset_phys(ds_prev)->ds_next_clones_obj != 0) {
301 dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
302 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
303 VERIFY0(zap_add_int(mos,
304 dsl_dataset_phys(ds_prev)->
305 ds_next_clones_obj,
306 dsl_dataset_phys(ds)->ds_next_snap_obj,
307 tx));
308 }
309 }
310 if (!after_branch_point) {
311 dsl_dataset_phys(ds_prev)->ds_next_snap_obj =
312 dsl_dataset_phys(ds)->ds_next_snap_obj;
313 }
314 }
315
316 dsl_dataset_t *ds_next;
317 uint64_t old_unique;
318 uint64_t used = 0, comp = 0, uncomp = 0;
319
320 VERIFY0(dsl_dataset_hold_obj(dp,
321 dsl_dataset_phys(ds)->ds_next_snap_obj, FTAG, &ds_next));
322 ASSERT3U(dsl_dataset_phys(ds_next)->ds_prev_snap_obj, ==, obj);
323
324 old_unique = dsl_dataset_phys(ds_next)->ds_unique_bytes;
325
326 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
327 dsl_dataset_phys(ds_next)->ds_prev_snap_obj =
328 dsl_dataset_phys(ds)->ds_prev_snap_obj;
329 dsl_dataset_phys(ds_next)->ds_prev_snap_txg =
330 dsl_dataset_phys(ds)->ds_prev_snap_txg;
331 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
332 ds_prev ? dsl_dataset_phys(ds_prev)->ds_creation_txg : 0);
333
334 if (ds_next->ds_deadlist.dl_oldfmt) {
335 process_old_deadlist(ds, ds_prev, ds_next,
336 after_branch_point, tx);
337 } else {
338 /* Adjust prev's unique space. */
339 if (ds_prev && !after_branch_point) {
340 dsl_deadlist_space_range(&ds_next->ds_deadlist,
341 dsl_dataset_phys(ds_prev)->ds_prev_snap_txg,
342 dsl_dataset_phys(ds)->ds_prev_snap_txg,
343 &used, &comp, &uncomp);
344 dsl_dataset_phys(ds_prev)->ds_unique_bytes += used;
345 }
346
347 /* Adjust snapused. */
348 dsl_deadlist_space_range(&ds_next->ds_deadlist,
349 dsl_dataset_phys(ds)->ds_prev_snap_txg, UINT64_MAX,
350 &used, &comp, &uncomp);
351 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
352 -used, -comp, -uncomp, tx);
353
354 /* Move blocks to be freed to pool's free list. */
355 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
356 &dp->dp_free_bpobj, dsl_dataset_phys(ds)->ds_prev_snap_txg,
357 tx);
358 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
359 DD_USED_HEAD, used, comp, uncomp, tx);
360
361 /* Merge our deadlist into next's and free it. */
362 dsl_deadlist_merge(&ds_next->ds_deadlist,
363 dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
364 }
365
366 dsl_deadlist_close(&ds->ds_deadlist);
367 dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
368 dmu_buf_will_dirty(ds->ds_dbuf, tx);
369 dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
370
371 dsl_destroy_snapshot_handle_remaps(ds, ds_next, tx);
372
373 /* Collapse range in clone heads */
374 dsl_dataset_remove_clones_key(ds,
375 dsl_dataset_phys(ds)->ds_creation_txg, tx);
376
377 if (ds_next->ds_is_snapshot) {
378 dsl_dataset_t *ds_nextnext;
379
380 /*
381 * Update next's unique to include blocks which
382 * were previously shared by only this snapshot
383 * and it. Those blocks will be born after the
384 * prev snap and before this snap, and will have
385 * died after the next snap and before the one
386 * after that (ie. be on the snap after next's
387 * deadlist).
388 */
389 VERIFY0(dsl_dataset_hold_obj(dp,
390 dsl_dataset_phys(ds_next)->ds_next_snap_obj,
391 FTAG, &ds_nextnext));
392 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
393 dsl_dataset_phys(ds)->ds_prev_snap_txg,
394 dsl_dataset_phys(ds)->ds_creation_txg,
395 &used, &comp, &uncomp);
396 dsl_dataset_phys(ds_next)->ds_unique_bytes += used;
397 dsl_dataset_rele(ds_nextnext, FTAG);
398 ASSERT3P(ds_next->ds_prev, ==, NULL);
399
400 /* Collapse range in this head. */
401 dsl_dataset_t *hds;
402 VERIFY0(dsl_dataset_hold_obj(dp,
403 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &hds));
404 dsl_deadlist_remove_key(&hds->ds_deadlist,
405 dsl_dataset_phys(ds)->ds_creation_txg, tx);
406 if (dsl_dataset_remap_deadlist_exists(hds)) {
407 dsl_deadlist_remove_key(&hds->ds_remap_deadlist,
408 dsl_dataset_phys(ds)->ds_creation_txg, tx);
409 }
410 dsl_dataset_rele(hds, FTAG);
411
412 } else {
413 ASSERT3P(ds_next->ds_prev, ==, ds);
414 dsl_dataset_rele(ds_next->ds_prev, ds_next);
415 ds_next->ds_prev = NULL;
416 if (ds_prev) {
417 VERIFY0(dsl_dataset_hold_obj(dp,
418 dsl_dataset_phys(ds)->ds_prev_snap_obj,
419 ds_next, &ds_next->ds_prev));
420 }
421
422 dsl_dataset_recalc_head_uniq(ds_next);
423
424 /*
425 * Reduce the amount of our unconsumed refreservation
426 * being charged to our parent by the amount of
427 * new unique data we have gained.
428 */
429 if (old_unique < ds_next->ds_reserved) {
430 int64_t mrsdelta;
431 uint64_t new_unique =
432 dsl_dataset_phys(ds_next)->ds_unique_bytes;
433
434 ASSERT(old_unique <= new_unique);
435 mrsdelta = MIN(new_unique - old_unique,
436 ds_next->ds_reserved - old_unique);
437 dsl_dir_diduse_space(ds->ds_dir,
438 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
439 }
440 }
441 dsl_dataset_rele(ds_next, FTAG);
442
443 /*
444 * This must be done after the dsl_traverse(), because it will
445 * re-open the objset.
446 */
447 if (ds->ds_objset) {
448 dmu_objset_evict(ds->ds_objset);
449 ds->ds_objset = NULL;
450 }
451
452 /* remove from snapshot namespace */
453 dsl_dataset_t *ds_head;
454 ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0);
455 VERIFY0(dsl_dataset_hold_obj(dp,
456 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &ds_head));
457 VERIFY0(dsl_dataset_get_snapname(ds));
458 #ifdef ZFS_DEBUG
459 {
460 uint64_t val;
461
462 err = dsl_dataset_snap_lookup(ds_head,
463 ds->ds_snapname, &val);
464 ASSERT0(err);
465 ASSERT3U(val, ==, obj);
466 }
467 #endif
468 VERIFY0(dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx, B_TRUE));
469 dsl_dataset_rele(ds_head, FTAG);
470
471 if (ds_prev != NULL)
472 dsl_dataset_rele(ds_prev, FTAG);
473
474 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
475
476 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
477 uint64_t count;
478 ASSERT0(zap_count(mos,
479 dsl_dataset_phys(ds)->ds_next_clones_obj, &count) &&
480 count == 0);
481 VERIFY0(dmu_object_free(mos,
482 dsl_dataset_phys(ds)->ds_next_clones_obj, tx));
483 }
484 if (dsl_dataset_phys(ds)->ds_props_obj != 0)
485 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_props_obj,
486 tx));
487 if (dsl_dataset_phys(ds)->ds_userrefs_obj != 0)
488 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_userrefs_obj,
489 tx));
490 dsl_dir_rele(ds->ds_dir, ds);
491 ds->ds_dir = NULL;
492 dmu_object_free_zapified(mos, obj, tx);
493 }
494
495 void
dsl_destroy_snapshot_sync(void * arg,dmu_tx_t * tx)496 dsl_destroy_snapshot_sync(void *arg, dmu_tx_t *tx)
497 {
498 dsl_destroy_snapshot_arg_t *ddsa = arg;
499 const char *dsname = ddsa->ddsa_name;
500 boolean_t defer = ddsa->ddsa_defer;
501
502 dsl_pool_t *dp = dmu_tx_pool(tx);
503 dsl_dataset_t *ds;
504
505 int error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
506 if (error == ENOENT)
507 return;
508 ASSERT0(error);
509 dsl_destroy_snapshot_sync_impl(ds, defer, tx);
510 dsl_dataset_rele(ds, FTAG);
511 }
512
513 /*
514 * The semantics of this function are described in the comment above
515 * lzc_destroy_snaps(). To summarize:
516 *
517 * The snapshots must all be in the same pool.
518 *
519 * Snapshots that don't exist will be silently ignored (considered to be
520 * "already deleted").
521 *
522 * On success, all snaps will be destroyed and this will return 0.
523 * On failure, no snaps will be destroyed, the errlist will be filled in,
524 * and this will return an errno.
525 */
526 int
dsl_destroy_snapshots_nvl(nvlist_t * snaps,boolean_t defer,nvlist_t * errlist)527 dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t defer,
528 nvlist_t *errlist)
529 {
530 if (nvlist_next_nvpair(snaps, NULL) == NULL)
531 return (0);
532
533 /*
534 * lzc_destroy_snaps() is documented to take an nvlist whose
535 * values "don't matter". We need to convert that nvlist to
536 * one that we know can be converted to LUA. We also don't
537 * care about any duplicate entries because the nvlist will
538 * be converted to a LUA table which should take care of this.
539 */
540 nvlist_t *snaps_normalized;
541 VERIFY0(nvlist_alloc(&snaps_normalized, 0, KM_SLEEP));
542 for (nvpair_t *pair = nvlist_next_nvpair(snaps, NULL);
543 pair != NULL; pair = nvlist_next_nvpair(snaps, pair)) {
544 fnvlist_add_boolean_value(snaps_normalized,
545 nvpair_name(pair), B_TRUE);
546 }
547
548 nvlist_t *arg;
549 VERIFY0(nvlist_alloc(&arg, 0, KM_SLEEP));
550 fnvlist_add_nvlist(arg, "snaps", snaps_normalized);
551 fnvlist_free(snaps_normalized);
552 fnvlist_add_boolean_value(arg, "defer", defer);
553
554 nvlist_t *wrapper;
555 VERIFY0(nvlist_alloc(&wrapper, 0, KM_SLEEP));
556 fnvlist_add_nvlist(wrapper, ZCP_ARG_ARGLIST, arg);
557 fnvlist_free(arg);
558
559 const char *program =
560 "arg = ...\n"
561 "snaps = arg['snaps']\n"
562 "defer = arg['defer']\n"
563 "errors = { }\n"
564 "has_errors = false\n"
565 "for snap, v in pairs(snaps) do\n"
566 " errno = zfs.check.destroy{snap, defer=defer}\n"
567 " zfs.debug('snap: ' .. snap .. ' errno: ' .. errno)\n"
568 " if errno == ENOENT then\n"
569 " snaps[snap] = nil\n"
570 " elseif errno ~= 0 then\n"
571 " errors[snap] = errno\n"
572 " has_errors = true\n"
573 " end\n"
574 "end\n"
575 "if has_errors then\n"
576 " return errors\n"
577 "end\n"
578 "for snap, v in pairs(snaps) do\n"
579 " errno = zfs.sync.destroy{snap, defer=defer}\n"
580 " assert(errno == 0)\n"
581 "end\n"
582 "return { }\n";
583
584 nvlist_t *result = fnvlist_alloc();
585 int error = zcp_eval(nvpair_name(nvlist_next_nvpair(snaps, NULL)),
586 program,
587 B_TRUE,
588 0,
589 zfs_lua_max_memlimit,
590 nvlist_next_nvpair(wrapper, NULL), result);
591 if (error != 0) {
592 char *errorstr = NULL;
593 (void) nvlist_lookup_string(result, ZCP_RET_ERROR, &errorstr);
594 if (errorstr != NULL) {
595 zfs_dbgmsg(errorstr);
596 }
597 return (error);
598 }
599 fnvlist_free(wrapper);
600
601 /*
602 * lzc_destroy_snaps() is documented to fill the errlist with
603 * int32 values, so we need to covert the int64 values that are
604 * returned from LUA.
605 */
606 int rv = 0;
607 nvlist_t *errlist_raw = fnvlist_lookup_nvlist(result, ZCP_RET_RETURN);
608 for (nvpair_t *pair = nvlist_next_nvpair(errlist_raw, NULL);
609 pair != NULL; pair = nvlist_next_nvpair(errlist_raw, pair)) {
610 int32_t val = (int32_t)fnvpair_value_int64(pair);
611 if (rv == 0)
612 rv = val;
613 fnvlist_add_int32(errlist, nvpair_name(pair), val);
614 }
615 fnvlist_free(result);
616 return (rv);
617 }
618
619 int
dsl_destroy_snapshot(const char * name,boolean_t defer)620 dsl_destroy_snapshot(const char *name, boolean_t defer)
621 {
622 int error;
623 nvlist_t *nvl = fnvlist_alloc();
624 nvlist_t *errlist = fnvlist_alloc();
625
626 fnvlist_add_boolean(nvl, name);
627 error = dsl_destroy_snapshots_nvl(nvl, defer, errlist);
628 fnvlist_free(errlist);
629 fnvlist_free(nvl);
630 return (error);
631 }
632
633 struct killarg {
634 dsl_dataset_t *ds;
635 dmu_tx_t *tx;
636 };
637
638 /* ARGSUSED */
639 static int
kill_blkptr(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)640 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
641 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
642 {
643 struct killarg *ka = arg;
644 dmu_tx_t *tx = ka->tx;
645
646 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
647 return (0);
648
649 if (zb->zb_level == ZB_ZIL_LEVEL) {
650 ASSERT(zilog != NULL);
651 /*
652 * It's a block in the intent log. It has no
653 * accounting, so just free it.
654 */
655 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
656 } else {
657 ASSERT(zilog == NULL);
658 ASSERT3U(bp->blk_birth, >,
659 dsl_dataset_phys(ka->ds)->ds_prev_snap_txg);
660 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
661 }
662
663 return (0);
664 }
665
666 static void
old_synchronous_dataset_destroy(dsl_dataset_t * ds,dmu_tx_t * tx)667 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
668 {
669 struct killarg ka;
670
671 /*
672 * Free everything that we point to (that's born after
673 * the previous snapshot, if we are a clone)
674 *
675 * NB: this should be very quick, because we already
676 * freed all the objects in open context.
677 */
678 ka.ds = ds;
679 ka.tx = tx;
680 VERIFY0(traverse_dataset(ds,
681 dsl_dataset_phys(ds)->ds_prev_snap_txg, TRAVERSE_POST |
682 TRAVERSE_NO_DECRYPT, kill_blkptr, &ka));
683 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
684 dsl_dataset_phys(ds)->ds_unique_bytes == 0);
685 }
686
687 int
dsl_destroy_head_check_impl(dsl_dataset_t * ds,int expected_holds)688 dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
689 {
690 int error;
691 uint64_t count;
692 objset_t *mos;
693
694 ASSERT(!ds->ds_is_snapshot);
695 if (ds->ds_is_snapshot)
696 return (SET_ERROR(EINVAL));
697
698 if (zfs_refcount_count(&ds->ds_longholds) != expected_holds)
699 return (SET_ERROR(EBUSY));
700
701 mos = ds->ds_dir->dd_pool->dp_meta_objset;
702
703 /*
704 * Can't delete a head dataset if there are snapshots of it.
705 * (Except if the only snapshots are from the branch we cloned
706 * from.)
707 */
708 if (ds->ds_prev != NULL &&
709 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object)
710 return (SET_ERROR(EBUSY));
711
712 /*
713 * Can't delete if there are children of this fs.
714 */
715 error = zap_count(mos,
716 dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, &count);
717 if (error != 0)
718 return (error);
719 if (count != 0)
720 return (SET_ERROR(EEXIST));
721
722 if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev) &&
723 dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
724 ds->ds_prev->ds_userrefs == 0) {
725 /* We need to remove the origin snapshot as well. */
726 if (!zfs_refcount_is_zero(&ds->ds_prev->ds_longholds))
727 return (SET_ERROR(EBUSY));
728 }
729 return (0);
730 }
731
732 int
dsl_destroy_head_check(void * arg,dmu_tx_t * tx)733 dsl_destroy_head_check(void *arg, dmu_tx_t *tx)
734 {
735 dsl_destroy_head_arg_t *ddha = arg;
736 dsl_pool_t *dp = dmu_tx_pool(tx);
737 dsl_dataset_t *ds;
738 int error;
739
740 error = dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds);
741 if (error != 0)
742 return (error);
743
744 error = dsl_destroy_head_check_impl(ds, 0);
745 dsl_dataset_rele(ds, FTAG);
746 return (error);
747 }
748
749 static void
dsl_dir_destroy_sync(uint64_t ddobj,dmu_tx_t * tx)750 dsl_dir_destroy_sync(uint64_t ddobj, dmu_tx_t *tx)
751 {
752 dsl_dir_t *dd;
753 dsl_pool_t *dp = dmu_tx_pool(tx);
754 objset_t *mos = dp->dp_meta_objset;
755 dd_used_t t;
756
757 ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx)->dp_config_rwlock));
758
759 VERIFY0(dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd));
760
761 ASSERT0(dsl_dir_phys(dd)->dd_head_dataset_obj);
762
763 /*
764 * Decrement the filesystem count for all parent filesystems.
765 *
766 * When we receive an incremental stream into a filesystem that already
767 * exists, a temporary clone is created. We never count this temporary
768 * clone, whose name begins with a '%'.
769 */
770 if (dd->dd_myname[0] != '%' && dd->dd_parent != NULL)
771 dsl_fs_ss_count_adjust(dd->dd_parent, -1,
772 DD_FIELD_FILESYSTEM_COUNT, tx);
773
774 /*
775 * Remove our reservation. The impl() routine avoids setting the
776 * actual property, which would require the (already destroyed) ds.
777 */
778 dsl_dir_set_reservation_sync_impl(dd, 0, tx);
779
780 ASSERT0(dsl_dir_phys(dd)->dd_used_bytes);
781 ASSERT0(dsl_dir_phys(dd)->dd_reserved);
782 for (t = 0; t < DD_USED_NUM; t++)
783 ASSERT0(dsl_dir_phys(dd)->dd_used_breakdown[t]);
784
785 if (dd->dd_crypto_obj != 0) {
786 dsl_crypto_key_destroy_sync(dd->dd_crypto_obj, tx);
787 (void) spa_keystore_unload_wkey_impl(dp->dp_spa, dd->dd_object);
788 }
789
790 VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_child_dir_zapobj, tx));
791 VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_props_zapobj, tx));
792 if (dsl_dir_phys(dd)->dd_clones != 0)
793 VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_clones, tx));
794 VERIFY0(dsl_deleg_destroy(mos, dsl_dir_phys(dd)->dd_deleg_zapobj, tx));
795 VERIFY0(zap_remove(mos,
796 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
797 dd->dd_myname, tx));
798
799 dsl_dir_rele(dd, FTAG);
800 dmu_object_free_zapified(mos, ddobj, tx);
801 }
802
803 void
dsl_destroy_head_sync_impl(dsl_dataset_t * ds,dmu_tx_t * tx)804 dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
805 {
806 dsl_pool_t *dp = dmu_tx_pool(tx);
807 objset_t *mos = dp->dp_meta_objset;
808 uint64_t obj, ddobj, prevobj = 0;
809 boolean_t rmorigin;
810
811 ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
812 ASSERT(ds->ds_prev == NULL ||
813 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj != ds->ds_object);
814 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
815 ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
816 rrw_exit(&ds->ds_bp_rwlock, FTAG);
817 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
818
819 /* We need to log before removing it from the namespace. */
820 spa_history_log_internal_ds(ds, "destroy", tx, "");
821
822 rmorigin = (dsl_dir_is_clone(ds->ds_dir) &&
823 DS_IS_DEFER_DESTROY(ds->ds_prev) &&
824 dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
825 ds->ds_prev->ds_userrefs == 0);
826
827 /* Remove our reservation. */
828 if (ds->ds_reserved != 0) {
829 dsl_dataset_set_refreservation_sync_impl(ds,
830 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
831 0, tx);
832 ASSERT0(ds->ds_reserved);
833 }
834
835 obj = ds->ds_object;
836
837 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
838 if (dsl_dataset_feature_is_active(ds, f))
839 dsl_dataset_deactivate_feature(ds, f, tx);
840 }
841
842 dsl_scan_ds_destroyed(ds, tx);
843
844 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
845 /* This is a clone */
846 ASSERT(ds->ds_prev != NULL);
847 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj, !=,
848 obj);
849 ASSERT0(dsl_dataset_phys(ds)->ds_next_snap_obj);
850
851 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
852 if (dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj != 0) {
853 dsl_dataset_remove_from_next_clones(ds->ds_prev,
854 obj, tx);
855 }
856
857 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_num_children, >, 1);
858 dsl_dataset_phys(ds->ds_prev)->ds_num_children--;
859 }
860
861 /*
862 * Destroy the deadlist. Unless it's a clone, the
863 * deadlist should be empty since the dataset has no snapshots.
864 * (If it's a clone, it's safe to ignore the deadlist contents
865 * since they are still referenced by the origin snapshot.)
866 */
867 dsl_deadlist_close(&ds->ds_deadlist);
868 dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
869 dmu_buf_will_dirty(ds->ds_dbuf, tx);
870 dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
871
872 if (dsl_dataset_remap_deadlist_exists(ds))
873 dsl_dataset_destroy_remap_deadlist(ds, tx);
874
875 objset_t *os;
876 VERIFY0(dmu_objset_from_ds(ds, &os));
877
878 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) {
879 old_synchronous_dataset_destroy(ds, tx);
880 } else {
881 /*
882 * Move the bptree into the pool's list of trees to
883 * clean up and update space accounting information.
884 */
885 uint64_t used, comp, uncomp;
886
887 zil_destroy_sync(dmu_objset_zil(os), tx);
888
889 if (!spa_feature_is_active(dp->dp_spa,
890 SPA_FEATURE_ASYNC_DESTROY)) {
891 dsl_scan_t *scn = dp->dp_scan;
892 spa_feature_incr(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY,
893 tx);
894 dp->dp_bptree_obj = bptree_alloc(mos, tx);
895 VERIFY0(zap_add(mos,
896 DMU_POOL_DIRECTORY_OBJECT,
897 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
898 &dp->dp_bptree_obj, tx));
899 ASSERT(!scn->scn_async_destroying);
900 scn->scn_async_destroying = B_TRUE;
901 }
902
903 used = dsl_dir_phys(ds->ds_dir)->dd_used_bytes;
904 comp = dsl_dir_phys(ds->ds_dir)->dd_compressed_bytes;
905 uncomp = dsl_dir_phys(ds->ds_dir)->dd_uncompressed_bytes;
906
907 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
908 dsl_dataset_phys(ds)->ds_unique_bytes == used);
909
910 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
911 bptree_add(mos, dp->dp_bptree_obj,
912 &dsl_dataset_phys(ds)->ds_bp,
913 dsl_dataset_phys(ds)->ds_prev_snap_txg,
914 used, comp, uncomp, tx);
915 rrw_exit(&ds->ds_bp_rwlock, FTAG);
916 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
917 -used, -comp, -uncomp, tx);
918 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
919 used, comp, uncomp, tx);
920 }
921
922 if (ds->ds_prev != NULL) {
923 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
924 VERIFY0(zap_remove_int(mos,
925 dsl_dir_phys(ds->ds_prev->ds_dir)->dd_clones,
926 ds->ds_object, tx));
927 }
928 prevobj = ds->ds_prev->ds_object;
929 dsl_dataset_rele(ds->ds_prev, ds);
930 ds->ds_prev = NULL;
931 }
932
933 /*
934 * This must be done after the dsl_traverse(), because it will
935 * re-open the objset.
936 */
937 if (ds->ds_objset) {
938 dmu_objset_evict(ds->ds_objset);
939 ds->ds_objset = NULL;
940 }
941
942 /* Erase the link in the dir */
943 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
944 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj = 0;
945 ddobj = ds->ds_dir->dd_object;
946 ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0);
947 VERIFY0(zap_destroy(mos,
948 dsl_dataset_phys(ds)->ds_snapnames_zapobj, tx));
949
950 if (ds->ds_bookmarks != 0) {
951 VERIFY0(zap_destroy(mos, ds->ds_bookmarks, tx));
952 spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
953 }
954
955 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
956
957 ASSERT0(dsl_dataset_phys(ds)->ds_next_clones_obj);
958 ASSERT0(dsl_dataset_phys(ds)->ds_props_obj);
959 ASSERT0(dsl_dataset_phys(ds)->ds_userrefs_obj);
960 dsl_dir_rele(ds->ds_dir, ds);
961 ds->ds_dir = NULL;
962 dmu_object_free_zapified(mos, obj, tx);
963
964 dsl_dir_destroy_sync(ddobj, tx);
965
966 if (rmorigin) {
967 dsl_dataset_t *prev;
968 VERIFY0(dsl_dataset_hold_obj(dp, prevobj, FTAG, &prev));
969 dsl_destroy_snapshot_sync_impl(prev, B_FALSE, tx);
970 dsl_dataset_rele(prev, FTAG);
971 }
972 }
973
974 void
dsl_destroy_head_sync(void * arg,dmu_tx_t * tx)975 dsl_destroy_head_sync(void *arg, dmu_tx_t *tx)
976 {
977 dsl_destroy_head_arg_t *ddha = arg;
978 dsl_pool_t *dp = dmu_tx_pool(tx);
979 dsl_dataset_t *ds;
980
981 VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
982 dsl_destroy_head_sync_impl(ds, tx);
983 dsl_dataset_rele(ds, FTAG);
984 }
985
986 static void
dsl_destroy_head_begin_sync(void * arg,dmu_tx_t * tx)987 dsl_destroy_head_begin_sync(void *arg, dmu_tx_t *tx)
988 {
989 dsl_destroy_head_arg_t *ddha = arg;
990 dsl_pool_t *dp = dmu_tx_pool(tx);
991 dsl_dataset_t *ds;
992
993 VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
994
995 /* Mark it as inconsistent on-disk, in case we crash */
996 dmu_buf_will_dirty(ds->ds_dbuf, tx);
997 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT;
998
999 spa_history_log_internal_ds(ds, "destroy begin", tx, "");
1000 dsl_dataset_rele(ds, FTAG);
1001 }
1002
1003 int
dsl_destroy_head(const char * name)1004 dsl_destroy_head(const char *name)
1005 {
1006 dsl_destroy_head_arg_t ddha;
1007 int error;
1008 spa_t *spa;
1009 boolean_t isenabled;
1010
1011 #ifdef _KERNEL
1012 zfs_destroy_unmount_origin(name);
1013 #endif
1014
1015 error = spa_open(name, &spa, FTAG);
1016 if (error != 0)
1017 return (error);
1018 isenabled = spa_feature_is_enabled(spa, SPA_FEATURE_ASYNC_DESTROY);
1019 spa_close(spa, FTAG);
1020
1021 ddha.ddha_name = name;
1022
1023 if (!isenabled) {
1024 objset_t *os;
1025
1026 error = dsl_sync_task(name, dsl_destroy_head_check,
1027 dsl_destroy_head_begin_sync, &ddha,
1028 0, ZFS_SPACE_CHECK_DESTROY);
1029 if (error != 0)
1030 return (error);
1031
1032 /*
1033 * Head deletion is processed in one txg on old pools;
1034 * remove the objects from open context so that the txg sync
1035 * is not too long.
1036 */
1037 error = dmu_objset_own(name, DMU_OST_ANY, B_FALSE, B_FALSE,
1038 FTAG, &os);
1039 if (error == 0) {
1040 uint64_t prev_snap_txg =
1041 dsl_dataset_phys(dmu_objset_ds(os))->
1042 ds_prev_snap_txg;
1043 for (uint64_t obj = 0; error == 0;
1044 error = dmu_object_next(os, &obj, FALSE,
1045 prev_snap_txg))
1046 (void) dmu_free_long_object(os, obj);
1047 /* sync out all frees */
1048 txg_wait_synced(dmu_objset_pool(os), 0);
1049 dmu_objset_disown(os, B_FALSE, FTAG);
1050 }
1051 }
1052
1053 return (dsl_sync_task(name, dsl_destroy_head_check,
1054 dsl_destroy_head_sync, &ddha, 0, ZFS_SPACE_CHECK_DESTROY));
1055 }
1056
1057 /*
1058 * Note, this function is used as the callback for dmu_objset_find(). We
1059 * always return 0 so that we will continue to find and process
1060 * inconsistent datasets, even if we encounter an error trying to
1061 * process one of them.
1062 */
1063 /* ARGSUSED */
1064 int
dsl_destroy_inconsistent(const char * dsname,void * arg)1065 dsl_destroy_inconsistent(const char *dsname, void *arg)
1066 {
1067 objset_t *os;
1068
1069 if (dmu_objset_hold(dsname, FTAG, &os) == 0) {
1070 boolean_t need_destroy = DS_IS_INCONSISTENT(dmu_objset_ds(os));
1071
1072 /*
1073 * If the dataset is inconsistent because a resumable receive
1074 * has failed, then do not destroy it.
1075 */
1076 if (dsl_dataset_has_resume_receive_state(dmu_objset_ds(os)))
1077 need_destroy = B_FALSE;
1078
1079 dmu_objset_rele(os, FTAG);
1080 if (need_destroy)
1081 (void) dsl_destroy_head(dsname);
1082 }
1083 return (0);
1084 }
1085