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, 2014 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 */
27
28 #include <sys/zfs_context.h>
29 #include <sys/dsl_userhold.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_synctask.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/dsl_pool.h>
34 #include <sys/dsl_dir.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dsl_scan.h>
37 #include <sys/dmu_objset.h>
38 #include <sys/zap.h>
39 #include <sys/zfeature.h>
40 #include <sys/zfs_ioctl.h>
41 #include <sys/dsl_deleg.h>
42 #include <sys/dmu_impl.h>
43
44 typedef struct dmu_snapshots_destroy_arg {
45 nvlist_t *dsda_snaps;
46 nvlist_t *dsda_successful_snaps;
47 boolean_t dsda_defer;
48 nvlist_t *dsda_errlist;
49 } dmu_snapshots_destroy_arg_t;
50
51 int
dsl_destroy_snapshot_check_impl(dsl_dataset_t * ds,boolean_t defer)52 dsl_destroy_snapshot_check_impl(dsl_dataset_t *ds, boolean_t defer)
53 {
54 if (!dsl_dataset_is_snapshot(ds))
55 return (SET_ERROR(EINVAL));
56
57 if (dsl_dataset_long_held(ds))
58 return (SET_ERROR(EBUSY));
59
60 /*
61 * Only allow deferred destroy on pools that support it.
62 * NOTE: deferred destroy is only supported on snapshots.
63 */
64 if (defer) {
65 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
66 SPA_VERSION_USERREFS)
67 return (SET_ERROR(ENOTSUP));
68 return (0);
69 }
70
71 /*
72 * If this snapshot has an elevated user reference count,
73 * we can't destroy it yet.
74 */
75 if (ds->ds_userrefs > 0)
76 return (SET_ERROR(EBUSY));
77
78 /*
79 * Can't delete a branch point.
80 */
81 if (dsl_dataset_phys(ds)->ds_num_children > 1)
82 return (SET_ERROR(EEXIST));
83
84 return (0);
85 }
86
87 static int
dsl_destroy_snapshot_check(void * arg,dmu_tx_t * tx)88 dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx)
89 {
90 dmu_snapshots_destroy_arg_t *dsda = arg;
91 dsl_pool_t *dp = dmu_tx_pool(tx);
92 nvpair_t *pair;
93 int error = 0;
94
95 if (!dmu_tx_is_syncing(tx))
96 return (0);
97
98 for (pair = nvlist_next_nvpair(dsda->dsda_snaps, NULL);
99 pair != NULL; pair = nvlist_next_nvpair(dsda->dsda_snaps, pair)) {
100 dsl_dataset_t *ds;
101
102 error = dsl_dataset_hold(dp, nvpair_name(pair),
103 FTAG, &ds);
104
105 /*
106 * If the snapshot does not exist, silently ignore it
107 * (it's "already destroyed").
108 */
109 if (error == ENOENT)
110 continue;
111
112 if (error == 0) {
113 error = dsl_destroy_snapshot_check_impl(ds,
114 dsda->dsda_defer);
115 dsl_dataset_rele(ds, FTAG);
116 }
117
118 if (error == 0) {
119 fnvlist_add_boolean(dsda->dsda_successful_snaps,
120 nvpair_name(pair));
121 } else {
122 fnvlist_add_int32(dsda->dsda_errlist,
123 nvpair_name(pair), error);
124 }
125 }
126
127 pair = nvlist_next_nvpair(dsda->dsda_errlist, NULL);
128 if (pair != NULL)
129 return (fnvpair_value_int32(pair));
130
131 return (0);
132 }
133
134 struct process_old_arg {
135 dsl_dataset_t *ds;
136 dsl_dataset_t *ds_prev;
137 boolean_t after_branch_point;
138 zio_t *pio;
139 uint64_t used, comp, uncomp;
140 };
141
142 static int
process_old_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)143 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
144 {
145 struct process_old_arg *poa = arg;
146 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
147
148 ASSERT(!BP_IS_HOLE(bp));
149
150 if (bp->blk_birth <= dsl_dataset_phys(poa->ds)->ds_prev_snap_txg) {
151 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
152 if (poa->ds_prev && !poa->after_branch_point &&
153 bp->blk_birth >
154 dsl_dataset_phys(poa->ds_prev)->ds_prev_snap_txg) {
155 dsl_dataset_phys(poa->ds_prev)->ds_unique_bytes +=
156 bp_get_dsize_sync(dp->dp_spa, bp);
157 }
158 } else {
159 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
160 poa->comp += BP_GET_PSIZE(bp);
161 poa->uncomp += BP_GET_UCSIZE(bp);
162 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
163 }
164 return (0);
165 }
166
167 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)168 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
169 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
170 {
171 struct process_old_arg poa = { 0 };
172 dsl_pool_t *dp = ds->ds_dir->dd_pool;
173 objset_t *mos = dp->dp_meta_objset;
174 uint64_t deadlist_obj;
175
176 ASSERT(ds->ds_deadlist.dl_oldfmt);
177 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
178
179 poa.ds = ds;
180 poa.ds_prev = ds_prev;
181 poa.after_branch_point = after_branch_point;
182 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
183 VERIFY0(bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
184 process_old_cb, &poa, tx));
185 VERIFY0(zio_wait(poa.pio));
186 ASSERT3U(poa.used, ==, dsl_dataset_phys(ds)->ds_unique_bytes);
187
188 /* change snapused */
189 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
190 -poa.used, -poa.comp, -poa.uncomp, tx);
191
192 /* swap next's deadlist to our deadlist */
193 dsl_deadlist_close(&ds->ds_deadlist);
194 dsl_deadlist_close(&ds_next->ds_deadlist);
195 deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
196 dsl_dataset_phys(ds)->ds_deadlist_obj =
197 dsl_dataset_phys(ds_next)->ds_deadlist_obj;
198 dsl_dataset_phys(ds_next)->ds_deadlist_obj = deadlist_obj;
199 dsl_deadlist_open(&ds->ds_deadlist, mos,
200 dsl_dataset_phys(ds)->ds_deadlist_obj);
201 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
202 dsl_dataset_phys(ds_next)->ds_deadlist_obj);
203 }
204
205 static void
dsl_dataset_remove_clones_key(dsl_dataset_t * ds,uint64_t mintxg,dmu_tx_t * tx)206 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
207 {
208 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
209 zap_cursor_t zc;
210 zap_attribute_t za;
211
212 /*
213 * If it is the old version, dd_clones doesn't exist so we can't
214 * find the clones, but dsl_deadlist_remove_key() is a no-op so it
215 * doesn't matter.
216 */
217 if (dsl_dir_phys(ds->ds_dir)->dd_clones == 0)
218 return;
219
220 for (zap_cursor_init(&zc, mos, dsl_dir_phys(ds->ds_dir)->dd_clones);
221 zap_cursor_retrieve(&zc, &za) == 0;
222 zap_cursor_advance(&zc)) {
223 dsl_dataset_t *clone;
224
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,
229 mintxg, tx);
230 dsl_dataset_remove_clones_key(clone, mintxg, tx);
231 }
232 dsl_dataset_rele(clone, FTAG);
233 }
234 zap_cursor_fini(&zc);
235 }
236
237 void
dsl_destroy_snapshot_sync_impl(dsl_dataset_t * ds,boolean_t defer,dmu_tx_t * tx)238 dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
239 {
240 int err;
241 int after_branch_point = FALSE;
242 dsl_pool_t *dp = ds->ds_dir->dd_pool;
243 objset_t *mos = dp->dp_meta_objset;
244 dsl_dataset_t *ds_prev = NULL;
245 uint64_t obj;
246
247 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
248 ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
249 ASSERT(refcount_is_zero(&ds->ds_longholds));
250
251 if (defer &&
252 (ds->ds_userrefs > 0 ||
253 dsl_dataset_phys(ds)->ds_num_children > 1)) {
254 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
255 dmu_buf_will_dirty(ds->ds_dbuf, tx);
256 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_DEFER_DESTROY;
257 spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
258 return;
259 }
260
261 ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
262
263 /* We need to log before removing it from the namespace. */
264 spa_history_log_internal_ds(ds, "destroy", tx, "");
265 dsl_event_notify(ds, ESC_ZFS_DATASET_DELETE);
266
267 dsl_scan_ds_destroyed(ds, tx);
268
269 obj = ds->ds_object;
270
271 if (ds->ds_large_blocks) {
272 ASSERT0(zap_contains(mos, obj, DS_FIELD_LARGE_BLOCKS));
273 spa_feature_decr(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS, tx);
274 }
275 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
276 ASSERT3P(ds->ds_prev, ==, NULL);
277 VERIFY0(dsl_dataset_hold_obj(dp,
278 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &ds_prev));
279 after_branch_point =
280 (dsl_dataset_phys(ds_prev)->ds_next_snap_obj != obj);
281
282 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
283 if (after_branch_point &&
284 dsl_dataset_phys(ds_prev)->ds_next_clones_obj != 0) {
285 dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
286 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
287 VERIFY0(zap_add_int(mos,
288 dsl_dataset_phys(ds_prev)->
289 ds_next_clones_obj,
290 dsl_dataset_phys(ds)->ds_next_snap_obj,
291 tx));
292 }
293 }
294 if (!after_branch_point) {
295 dsl_dataset_phys(ds_prev)->ds_next_snap_obj =
296 dsl_dataset_phys(ds)->ds_next_snap_obj;
297 }
298 }
299
300 dsl_dataset_t *ds_next;
301 uint64_t old_unique;
302 uint64_t used = 0, comp = 0, uncomp = 0;
303
304 VERIFY0(dsl_dataset_hold_obj(dp,
305 dsl_dataset_phys(ds)->ds_next_snap_obj, FTAG, &ds_next));
306 ASSERT3U(dsl_dataset_phys(ds_next)->ds_prev_snap_obj, ==, obj);
307
308 old_unique = dsl_dataset_phys(ds_next)->ds_unique_bytes;
309
310 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
311 dsl_dataset_phys(ds_next)->ds_prev_snap_obj =
312 dsl_dataset_phys(ds)->ds_prev_snap_obj;
313 dsl_dataset_phys(ds_next)->ds_prev_snap_txg =
314 dsl_dataset_phys(ds)->ds_prev_snap_txg;
315 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
316 ds_prev ? dsl_dataset_phys(ds_prev)->ds_creation_txg : 0);
317
318 if (ds_next->ds_deadlist.dl_oldfmt) {
319 process_old_deadlist(ds, ds_prev, ds_next,
320 after_branch_point, tx);
321 } else {
322 /* Adjust prev's unique space. */
323 if (ds_prev && !after_branch_point) {
324 dsl_deadlist_space_range(&ds_next->ds_deadlist,
325 dsl_dataset_phys(ds_prev)->ds_prev_snap_txg,
326 dsl_dataset_phys(ds)->ds_prev_snap_txg,
327 &used, &comp, &uncomp);
328 dsl_dataset_phys(ds_prev)->ds_unique_bytes += used;
329 }
330
331 /* Adjust snapused. */
332 dsl_deadlist_space_range(&ds_next->ds_deadlist,
333 dsl_dataset_phys(ds)->ds_prev_snap_txg, UINT64_MAX,
334 &used, &comp, &uncomp);
335 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
336 -used, -comp, -uncomp, tx);
337
338 /* Move blocks to be freed to pool's free list. */
339 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
340 &dp->dp_free_bpobj, dsl_dataset_phys(ds)->ds_prev_snap_txg,
341 tx);
342 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
343 DD_USED_HEAD, used, comp, uncomp, tx);
344
345 /* Merge our deadlist into next's and free it. */
346 dsl_deadlist_merge(&ds_next->ds_deadlist,
347 dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
348 }
349 dsl_deadlist_close(&ds->ds_deadlist);
350 dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
351 dmu_buf_will_dirty(ds->ds_dbuf, tx);
352 dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
353
354 /* Collapse range in clone heads */
355 dsl_dataset_remove_clones_key(ds,
356 dsl_dataset_phys(ds)->ds_creation_txg, tx);
357
358 if (dsl_dataset_is_snapshot(ds_next)) {
359 dsl_dataset_t *ds_nextnext;
360
361 /*
362 * Update next's unique to include blocks which
363 * were previously shared by only this snapshot
364 * and it. Those blocks will be born after the
365 * prev snap and before this snap, and will have
366 * died after the next snap and before the one
367 * after that (ie. be on the snap after next's
368 * deadlist).
369 */
370 VERIFY0(dsl_dataset_hold_obj(dp,
371 dsl_dataset_phys(ds_next)->ds_next_snap_obj,
372 FTAG, &ds_nextnext));
373 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
374 dsl_dataset_phys(ds)->ds_prev_snap_txg,
375 dsl_dataset_phys(ds)->ds_creation_txg,
376 &used, &comp, &uncomp);
377 dsl_dataset_phys(ds_next)->ds_unique_bytes += used;
378 dsl_dataset_rele(ds_nextnext, FTAG);
379 ASSERT3P(ds_next->ds_prev, ==, NULL);
380
381 /* Collapse range in this head. */
382 dsl_dataset_t *hds;
383 VERIFY0(dsl_dataset_hold_obj(dp,
384 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &hds));
385 dsl_deadlist_remove_key(&hds->ds_deadlist,
386 dsl_dataset_phys(ds)->ds_creation_txg, tx);
387 dsl_dataset_rele(hds, FTAG);
388
389 } else {
390 ASSERT3P(ds_next->ds_prev, ==, ds);
391 dsl_dataset_rele(ds_next->ds_prev, ds_next);
392 ds_next->ds_prev = NULL;
393 if (ds_prev) {
394 VERIFY0(dsl_dataset_hold_obj(dp,
395 dsl_dataset_phys(ds)->ds_prev_snap_obj,
396 ds_next, &ds_next->ds_prev));
397 }
398
399 dsl_dataset_recalc_head_uniq(ds_next);
400
401 /*
402 * Reduce the amount of our unconsumed refreservation
403 * being charged to our parent by the amount of
404 * new unique data we have gained.
405 */
406 if (old_unique < ds_next->ds_reserved) {
407 int64_t mrsdelta;
408 uint64_t new_unique =
409 dsl_dataset_phys(ds_next)->ds_unique_bytes;
410
411 ASSERT(old_unique <= new_unique);
412 mrsdelta = MIN(new_unique - old_unique,
413 ds_next->ds_reserved - old_unique);
414 dsl_dir_diduse_space(ds->ds_dir,
415 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
416 }
417 }
418 dsl_dataset_rele(ds_next, FTAG);
419
420 /*
421 * This must be done after the dsl_traverse(), because it will
422 * re-open the objset.
423 */
424 if (ds->ds_objset) {
425 dmu_objset_evict(ds->ds_objset);
426 ds->ds_objset = NULL;
427 }
428
429 /* remove from snapshot namespace */
430 dsl_dataset_t *ds_head;
431 ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0);
432 VERIFY0(dsl_dataset_hold_obj(dp,
433 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &ds_head));
434 VERIFY0(dsl_dataset_get_snapname(ds));
435 #ifdef ZFS_DEBUG
436 {
437 uint64_t val;
438
439 err = dsl_dataset_snap_lookup(ds_head,
440 ds->ds_snapname, &val);
441 ASSERT0(err);
442 ASSERT3U(val, ==, obj);
443 }
444 #endif
445 VERIFY0(dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx, B_TRUE));
446 dsl_dataset_rele(ds_head, FTAG);
447
448 if (ds_prev != NULL)
449 dsl_dataset_rele(ds_prev, FTAG);
450
451 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
452
453 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
454 uint64_t count;
455 ASSERT0(zap_count(mos,
456 dsl_dataset_phys(ds)->ds_next_clones_obj, &count) &&
457 count == 0);
458 VERIFY0(dmu_object_free(mos,
459 dsl_dataset_phys(ds)->ds_next_clones_obj, tx));
460 }
461 if (dsl_dataset_phys(ds)->ds_props_obj != 0)
462 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_props_obj,
463 tx));
464 if (dsl_dataset_phys(ds)->ds_userrefs_obj != 0)
465 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_userrefs_obj,
466 tx));
467 dsl_dir_rele(ds->ds_dir, ds);
468 ds->ds_dir = NULL;
469 dmu_object_free_zapified(mos, obj, tx);
470 }
471
472 static void
dsl_destroy_snapshot_sync(void * arg,dmu_tx_t * tx)473 dsl_destroy_snapshot_sync(void *arg, dmu_tx_t *tx)
474 {
475 dmu_snapshots_destroy_arg_t *dsda = arg;
476 dsl_pool_t *dp = dmu_tx_pool(tx);
477 nvpair_t *pair;
478
479 for (pair = nvlist_next_nvpair(dsda->dsda_successful_snaps, NULL);
480 pair != NULL;
481 pair = nvlist_next_nvpair(dsda->dsda_successful_snaps, pair)) {
482 dsl_dataset_t *ds;
483
484 VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
485
486 dsl_destroy_snapshot_sync_impl(ds, dsda->dsda_defer, tx);
487 dsl_dataset_rele(ds, FTAG);
488 }
489 }
490
491 /*
492 * The semantics of this function are described in the comment above
493 * lzc_destroy_snaps(). To summarize:
494 *
495 * The snapshots must all be in the same pool.
496 *
497 * Snapshots that don't exist will be silently ignored (considered to be
498 * "already deleted").
499 *
500 * On success, all snaps will be destroyed and this will return 0.
501 * On failure, no snaps will be destroyed, the errlist will be filled in,
502 * and this will return an errno.
503 */
504 int
dsl_destroy_snapshots_nvl(nvlist_t * snaps,boolean_t defer,nvlist_t * errlist)505 dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t defer,
506 nvlist_t *errlist)
507 {
508 dmu_snapshots_destroy_arg_t dsda;
509 int error;
510 nvpair_t *pair;
511
512 pair = nvlist_next_nvpair(snaps, NULL);
513 if (pair == NULL)
514 return (0);
515
516 dsda.dsda_snaps = snaps;
517 dsda.dsda_successful_snaps = fnvlist_alloc();
518 dsda.dsda_defer = defer;
519 dsda.dsda_errlist = errlist;
520
521 error = dsl_sync_task(nvpair_name(pair),
522 dsl_destroy_snapshot_check, dsl_destroy_snapshot_sync,
523 &dsda, 0, ZFS_SPACE_CHECK_NONE);
524 fnvlist_free(dsda.dsda_successful_snaps);
525
526 return (error);
527 }
528
529 int
dsl_destroy_snapshot(const char * name,boolean_t defer)530 dsl_destroy_snapshot(const char *name, boolean_t defer)
531 {
532 int error;
533 nvlist_t *nvl = fnvlist_alloc();
534 nvlist_t *errlist = fnvlist_alloc();
535
536 fnvlist_add_boolean(nvl, name);
537 error = dsl_destroy_snapshots_nvl(nvl, defer, errlist);
538 fnvlist_free(errlist);
539 fnvlist_free(nvl);
540 return (error);
541 }
542
543 struct killarg {
544 dsl_dataset_t *ds;
545 dmu_tx_t *tx;
546 };
547
548 /* ARGSUSED */
549 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)550 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
551 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
552 {
553 struct killarg *ka = arg;
554 dmu_tx_t *tx = ka->tx;
555
556 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
557 return (0);
558
559 if (zb->zb_level == ZB_ZIL_LEVEL) {
560 ASSERT(zilog != NULL);
561 /*
562 * It's a block in the intent log. It has no
563 * accounting, so just free it.
564 */
565 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
566 } else {
567 ASSERT(zilog == NULL);
568 ASSERT3U(bp->blk_birth, >,
569 dsl_dataset_phys(ka->ds)->ds_prev_snap_txg);
570 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
571 }
572
573 return (0);
574 }
575
576 static void
old_synchronous_dataset_destroy(dsl_dataset_t * ds,dmu_tx_t * tx)577 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
578 {
579 struct killarg ka;
580
581 /*
582 * Free everything that we point to (that's born after
583 * the previous snapshot, if we are a clone)
584 *
585 * NB: this should be very quick, because we already
586 * freed all the objects in open context.
587 */
588 ka.ds = ds;
589 ka.tx = tx;
590 VERIFY0(traverse_dataset(ds,
591 dsl_dataset_phys(ds)->ds_prev_snap_txg, TRAVERSE_POST,
592 kill_blkptr, &ka));
593 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
594 dsl_dataset_phys(ds)->ds_unique_bytes == 0);
595 }
596
597 typedef struct dsl_destroy_head_arg {
598 const char *ddha_name;
599 } dsl_destroy_head_arg_t;
600
601 int
dsl_destroy_head_check_impl(dsl_dataset_t * ds,int expected_holds)602 dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
603 {
604 int error;
605 uint64_t count;
606 objset_t *mos;
607
608 ASSERT(!dsl_dataset_is_snapshot(ds));
609 if (dsl_dataset_is_snapshot(ds))
610 return (SET_ERROR(EINVAL));
611
612 if (refcount_count(&ds->ds_longholds) != expected_holds)
613 return (SET_ERROR(EBUSY));
614
615 mos = ds->ds_dir->dd_pool->dp_meta_objset;
616
617 /*
618 * Can't delete a head dataset if there are snapshots of it.
619 * (Except if the only snapshots are from the branch we cloned
620 * from.)
621 */
622 if (ds->ds_prev != NULL &&
623 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object)
624 return (SET_ERROR(EBUSY));
625
626 /*
627 * Can't delete if there are children of this fs.
628 */
629 error = zap_count(mos,
630 dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, &count);
631 if (error != 0)
632 return (error);
633 if (count != 0)
634 return (SET_ERROR(EEXIST));
635
636 if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev) &&
637 dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
638 ds->ds_prev->ds_userrefs == 0) {
639 /* We need to remove the origin snapshot as well. */
640 if (!refcount_is_zero(&ds->ds_prev->ds_longholds))
641 return (SET_ERROR(EBUSY));
642 }
643 return (0);
644 }
645
646 static int
dsl_destroy_head_check(void * arg,dmu_tx_t * tx)647 dsl_destroy_head_check(void *arg, dmu_tx_t *tx)
648 {
649 dsl_destroy_head_arg_t *ddha = arg;
650 dsl_pool_t *dp = dmu_tx_pool(tx);
651 dsl_dataset_t *ds;
652 int error;
653
654 error = dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds);
655 if (error != 0)
656 return (error);
657
658 error = dsl_destroy_head_check_impl(ds, 0);
659 dsl_dataset_rele(ds, FTAG);
660 return (error);
661 }
662
663 static void
dsl_dir_destroy_sync(uint64_t ddobj,dmu_tx_t * tx)664 dsl_dir_destroy_sync(uint64_t ddobj, dmu_tx_t *tx)
665 {
666 dsl_dir_t *dd;
667 dsl_pool_t *dp = dmu_tx_pool(tx);
668 objset_t *mos = dp->dp_meta_objset;
669 dd_used_t t;
670
671 ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx)->dp_config_rwlock));
672
673 VERIFY0(dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd));
674
675 ASSERT0(dsl_dir_phys(dd)->dd_head_dataset_obj);
676
677 /*
678 * Decrement the filesystem count for all parent filesystems.
679 *
680 * When we receive an incremental stream into a filesystem that already
681 * exists, a temporary clone is created. We never count this temporary
682 * clone, whose name begins with a '%'.
683 */
684 if (dd->dd_myname[0] != '%' && dd->dd_parent != NULL)
685 dsl_fs_ss_count_adjust(dd->dd_parent, -1,
686 DD_FIELD_FILESYSTEM_COUNT, tx);
687
688 /*
689 * Remove our reservation. The impl() routine avoids setting the
690 * actual property, which would require the (already destroyed) ds.
691 */
692 dsl_dir_set_reservation_sync_impl(dd, 0, tx);
693
694 ASSERT0(dsl_dir_phys(dd)->dd_used_bytes);
695 ASSERT0(dsl_dir_phys(dd)->dd_reserved);
696 for (t = 0; t < DD_USED_NUM; t++)
697 ASSERT0(dsl_dir_phys(dd)->dd_used_breakdown[t]);
698
699 VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_child_dir_zapobj, tx));
700 VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_props_zapobj, tx));
701 VERIFY0(dsl_deleg_destroy(mos, dsl_dir_phys(dd)->dd_deleg_zapobj, tx));
702 VERIFY0(zap_remove(mos,
703 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
704 dd->dd_myname, tx));
705
706 dsl_dir_rele(dd, FTAG);
707 dmu_object_free_zapified(mos, ddobj, tx);
708 }
709
710 void
dsl_destroy_head_sync_impl(dsl_dataset_t * ds,dmu_tx_t * tx)711 dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
712 {
713 dsl_pool_t *dp = dmu_tx_pool(tx);
714 objset_t *mos = dp->dp_meta_objset;
715 uint64_t obj, ddobj, prevobj = 0;
716 boolean_t rmorigin;
717
718 ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
719 ASSERT(ds->ds_prev == NULL ||
720 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj != ds->ds_object);
721 ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
722 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
723
724 /* We need to log before removing it from the namespace. */
725 spa_history_log_internal_ds(ds, "destroy", tx, "");
726 dsl_event_notify(ds, ESC_ZFS_DATASET_DELETE);
727
728 rmorigin = (dsl_dir_is_clone(ds->ds_dir) &&
729 DS_IS_DEFER_DESTROY(ds->ds_prev) &&
730 dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
731 ds->ds_prev->ds_userrefs == 0);
732
733 /* Remove our reservation. */
734 if (ds->ds_reserved != 0) {
735 dsl_dataset_set_refreservation_sync_impl(ds,
736 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
737 0, tx);
738 ASSERT0(ds->ds_reserved);
739 }
740
741 if (ds->ds_large_blocks)
742 spa_feature_decr(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS, tx);
743
744 dsl_scan_ds_destroyed(ds, tx);
745
746 obj = ds->ds_object;
747
748 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
749 /* This is a clone */
750 ASSERT(ds->ds_prev != NULL);
751 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj, !=,
752 obj);
753 ASSERT0(dsl_dataset_phys(ds)->ds_next_snap_obj);
754
755 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
756 if (dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj != 0) {
757 dsl_dataset_remove_from_next_clones(ds->ds_prev,
758 obj, tx);
759 }
760
761 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_num_children, >, 1);
762 dsl_dataset_phys(ds->ds_prev)->ds_num_children--;
763 }
764
765 /*
766 * Destroy the deadlist. Unless it's a clone, the
767 * deadlist should be empty. (If it's a clone, it's
768 * safe to ignore the deadlist contents.)
769 */
770 dsl_deadlist_close(&ds->ds_deadlist);
771 dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
772 dmu_buf_will_dirty(ds->ds_dbuf, tx);
773 dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
774
775 objset_t *os;
776 VERIFY0(dmu_objset_from_ds(ds, &os));
777
778 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) {
779 old_synchronous_dataset_destroy(ds, tx);
780 } else {
781 /*
782 * Move the bptree into the pool's list of trees to
783 * clean up and update space accounting information.
784 */
785 uint64_t used, comp, uncomp;
786
787 zil_destroy_sync(dmu_objset_zil(os), tx);
788
789 if (!spa_feature_is_active(dp->dp_spa,
790 SPA_FEATURE_ASYNC_DESTROY)) {
791 dsl_scan_t *scn = dp->dp_scan;
792 spa_feature_incr(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY,
793 tx);
794 dp->dp_bptree_obj = bptree_alloc(mos, tx);
795 VERIFY0(zap_add(mos,
796 DMU_POOL_DIRECTORY_OBJECT,
797 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
798 &dp->dp_bptree_obj, tx));
799 ASSERT(!scn->scn_async_destroying);
800 scn->scn_async_destroying = B_TRUE;
801 }
802
803 used = dsl_dir_phys(ds->ds_dir)->dd_used_bytes;
804 comp = dsl_dir_phys(ds->ds_dir)->dd_compressed_bytes;
805 uncomp = dsl_dir_phys(ds->ds_dir)->dd_uncompressed_bytes;
806
807 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
808 dsl_dataset_phys(ds)->ds_unique_bytes == used);
809
810 bptree_add(mos, dp->dp_bptree_obj,
811 &dsl_dataset_phys(ds)->ds_bp,
812 dsl_dataset_phys(ds)->ds_prev_snap_txg,
813 used, comp, uncomp, tx);
814 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
815 -used, -comp, -uncomp, tx);
816 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
817 used, comp, uncomp, tx);
818 }
819
820 if (ds->ds_prev != NULL) {
821 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
822 VERIFY0(zap_remove_int(mos,
823 dsl_dir_phys(ds->ds_prev->ds_dir)->dd_clones,
824 ds->ds_object, tx));
825 }
826 prevobj = ds->ds_prev->ds_object;
827 dsl_dataset_rele(ds->ds_prev, ds);
828 ds->ds_prev = NULL;
829 }
830
831 /*
832 * This must be done after the dsl_traverse(), because it will
833 * re-open the objset.
834 */
835 if (ds->ds_objset) {
836 dmu_objset_evict(ds->ds_objset);
837 ds->ds_objset = NULL;
838 }
839
840 /* Erase the link in the dir */
841 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
842 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj = 0;
843 ddobj = ds->ds_dir->dd_object;
844 ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0);
845 VERIFY0(zap_destroy(mos,
846 dsl_dataset_phys(ds)->ds_snapnames_zapobj, tx));
847
848 if (ds->ds_bookmarks != 0) {
849 VERIFY0(zap_destroy(mos, ds->ds_bookmarks, tx));
850 spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
851 }
852
853 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
854
855 ASSERT0(dsl_dataset_phys(ds)->ds_next_clones_obj);
856 ASSERT0(dsl_dataset_phys(ds)->ds_props_obj);
857 ASSERT0(dsl_dataset_phys(ds)->ds_userrefs_obj);
858 dsl_dir_rele(ds->ds_dir, ds);
859 ds->ds_dir = NULL;
860 dmu_object_free_zapified(mos, obj, tx);
861
862 dsl_dir_destroy_sync(ddobj, tx);
863
864 if (rmorigin) {
865 dsl_dataset_t *prev;
866 VERIFY0(dsl_dataset_hold_obj(dp, prevobj, FTAG, &prev));
867 dsl_destroy_snapshot_sync_impl(prev, B_FALSE, tx);
868 dsl_dataset_rele(prev, FTAG);
869 }
870 }
871
872 static void
dsl_destroy_head_sync(void * arg,dmu_tx_t * tx)873 dsl_destroy_head_sync(void *arg, dmu_tx_t *tx)
874 {
875 dsl_destroy_head_arg_t *ddha = arg;
876 dsl_pool_t *dp = dmu_tx_pool(tx);
877 dsl_dataset_t *ds;
878
879 VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
880 dsl_destroy_head_sync_impl(ds, tx);
881 dsl_dataset_rele(ds, FTAG);
882 }
883
884 static void
dsl_destroy_head_begin_sync(void * arg,dmu_tx_t * tx)885 dsl_destroy_head_begin_sync(void *arg, dmu_tx_t *tx)
886 {
887 dsl_destroy_head_arg_t *ddha = arg;
888 dsl_pool_t *dp = dmu_tx_pool(tx);
889 dsl_dataset_t *ds;
890
891 VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
892
893 /* Mark it as inconsistent on-disk, in case we crash */
894 dmu_buf_will_dirty(ds->ds_dbuf, tx);
895 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT;
896
897 spa_history_log_internal_ds(ds, "destroy begin", tx, "");
898 dsl_dataset_rele(ds, FTAG);
899 }
900
901 int
dsl_destroy_head(const char * name)902 dsl_destroy_head(const char *name)
903 {
904 dsl_destroy_head_arg_t ddha;
905 int error;
906 spa_t *spa;
907 boolean_t isenabled;
908
909 #ifdef _KERNEL
910 zfs_destroy_unmount_origin(name);
911 #endif
912
913 error = spa_open(name, &spa, FTAG);
914 if (error != 0)
915 return (error);
916 isenabled = spa_feature_is_enabled(spa, SPA_FEATURE_ASYNC_DESTROY);
917 spa_close(spa, FTAG);
918
919 ddha.ddha_name = name;
920
921 if (!isenabled) {
922 objset_t *os;
923
924 error = dsl_sync_task(name, dsl_destroy_head_check,
925 dsl_destroy_head_begin_sync, &ddha,
926 0, ZFS_SPACE_CHECK_NONE);
927 if (error != 0)
928 return (error);
929
930 /*
931 * Head deletion is processed in one txg on old pools;
932 * remove the objects from open context so that the txg sync
933 * is not too long.
934 */
935 error = dmu_objset_own(name, DMU_OST_ANY, B_FALSE, FTAG, &os);
936 if (error == 0) {
937 uint64_t prev_snap_txg =
938 dsl_dataset_phys(dmu_objset_ds(os))->
939 ds_prev_snap_txg;
940 for (uint64_t obj = 0; error == 0;
941 error = dmu_object_next(os, &obj, FALSE,
942 prev_snap_txg))
943 (void) dmu_free_long_object(os, obj);
944 /* sync out all frees */
945 txg_wait_synced(dmu_objset_pool(os), 0);
946 dmu_objset_disown(os, FTAG);
947 }
948 }
949
950 return (dsl_sync_task(name, dsl_destroy_head_check,
951 dsl_destroy_head_sync, &ddha, 0, ZFS_SPACE_CHECK_NONE));
952 }
953
954 /*
955 * Note, this function is used as the callback for dmu_objset_find(). We
956 * always return 0 so that we will continue to find and process
957 * inconsistent datasets, even if we encounter an error trying to
958 * process one of them.
959 */
960 /* ARGSUSED */
961 int
dsl_destroy_inconsistent(const char * dsname,void * arg)962 dsl_destroy_inconsistent(const char *dsname, void *arg)
963 {
964 objset_t *os;
965
966 if (dmu_objset_hold(dsname, FTAG, &os) == 0) {
967 boolean_t inconsistent = DS_IS_INCONSISTENT(dmu_objset_ds(os));
968 dmu_objset_rele(os, FTAG);
969 if (inconsistent)
970 (void) dsl_destroy_head(dsname);
971 }
972 return (0);
973 }
974