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, 2015 by Delphix. All rights reserved.
24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
25 * Copyright (c) 2013 by Joyent, Inc. All rights reserved.
26 */
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 (!ds->ds_is_snapshot)
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
266 dsl_scan_ds_destroyed(ds, tx);
267
268 obj = ds->ds_object;
269
270 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
271 if (ds->ds_feature_inuse[f]) {
272 dsl_dataset_deactivate_feature(obj, f, tx);
273 ds->ds_feature_inuse[f] = B_FALSE;
274 }
275 }
276 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
277 ASSERT3P(ds->ds_prev, ==, NULL);
278 VERIFY0(dsl_dataset_hold_obj(dp,
279 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &ds_prev));
280 after_branch_point =
281 (dsl_dataset_phys(ds_prev)->ds_next_snap_obj != obj);
282
283 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
284 if (after_branch_point &&
285 dsl_dataset_phys(ds_prev)->ds_next_clones_obj != 0) {
286 dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
287 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
288 VERIFY0(zap_add_int(mos,
289 dsl_dataset_phys(ds_prev)->
290 ds_next_clones_obj,
291 dsl_dataset_phys(ds)->ds_next_snap_obj,
292 tx));
293 }
294 }
295 if (!after_branch_point) {
296 dsl_dataset_phys(ds_prev)->ds_next_snap_obj =
297 dsl_dataset_phys(ds)->ds_next_snap_obj;
298 }
299 }
300
301 dsl_dataset_t *ds_next;
302 uint64_t old_unique;
303 uint64_t used = 0, comp = 0, uncomp = 0;
304
305 VERIFY0(dsl_dataset_hold_obj(dp,
306 dsl_dataset_phys(ds)->ds_next_snap_obj, FTAG, &ds_next));
307 ASSERT3U(dsl_dataset_phys(ds_next)->ds_prev_snap_obj, ==, obj);
308
309 old_unique = dsl_dataset_phys(ds_next)->ds_unique_bytes;
310
311 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
312 dsl_dataset_phys(ds_next)->ds_prev_snap_obj =
313 dsl_dataset_phys(ds)->ds_prev_snap_obj;
314 dsl_dataset_phys(ds_next)->ds_prev_snap_txg =
315 dsl_dataset_phys(ds)->ds_prev_snap_txg;
316 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
317 ds_prev ? dsl_dataset_phys(ds_prev)->ds_creation_txg : 0);
318
319 if (ds_next->ds_deadlist.dl_oldfmt) {
320 process_old_deadlist(ds, ds_prev, ds_next,
321 after_branch_point, tx);
322 } else {
323 /* Adjust prev's unique space. */
324 if (ds_prev && !after_branch_point) {
325 dsl_deadlist_space_range(&ds_next->ds_deadlist,
326 dsl_dataset_phys(ds_prev)->ds_prev_snap_txg,
327 dsl_dataset_phys(ds)->ds_prev_snap_txg,
328 &used, &comp, &uncomp);
329 dsl_dataset_phys(ds_prev)->ds_unique_bytes += used;
330 }
331
332 /* Adjust snapused. */
333 dsl_deadlist_space_range(&ds_next->ds_deadlist,
334 dsl_dataset_phys(ds)->ds_prev_snap_txg, UINT64_MAX,
335 &used, &comp, &uncomp);
336 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
337 -used, -comp, -uncomp, tx);
338
339 /* Move blocks to be freed to pool's free list. */
340 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
341 &dp->dp_free_bpobj, dsl_dataset_phys(ds)->ds_prev_snap_txg,
342 tx);
343 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
344 DD_USED_HEAD, used, comp, uncomp, tx);
345
346 /* Merge our deadlist into next's and free it. */
347 dsl_deadlist_merge(&ds_next->ds_deadlist,
348 dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
349 }
350 dsl_deadlist_close(&ds->ds_deadlist);
351 dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
352 dmu_buf_will_dirty(ds->ds_dbuf, tx);
353 dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
354
355 /* Collapse range in clone heads */
356 dsl_dataset_remove_clones_key(ds,
357 dsl_dataset_phys(ds)->ds_creation_txg, tx);
358
359 if (ds_next->ds_is_snapshot) {
360 dsl_dataset_t *ds_nextnext;
361
362 /*
363 * Update next's unique to include blocks which
364 * were previously shared by only this snapshot
365 * and it. Those blocks will be born after the
366 * prev snap and before this snap, and will have
367 * died after the next snap and before the one
368 * after that (ie. be on the snap after next's
369 * deadlist).
370 */
371 VERIFY0(dsl_dataset_hold_obj(dp,
372 dsl_dataset_phys(ds_next)->ds_next_snap_obj,
373 FTAG, &ds_nextnext));
374 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
375 dsl_dataset_phys(ds)->ds_prev_snap_txg,
376 dsl_dataset_phys(ds)->ds_creation_txg,
377 &used, &comp, &uncomp);
378 dsl_dataset_phys(ds_next)->ds_unique_bytes += used;
379 dsl_dataset_rele(ds_nextnext, FTAG);
380 ASSERT3P(ds_next->ds_prev, ==, NULL);
381
382 /* Collapse range in this head. */
383 dsl_dataset_t *hds;
384 VERIFY0(dsl_dataset_hold_obj(dp,
385 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &hds));
386 dsl_deadlist_remove_key(&hds->ds_deadlist,
387 dsl_dataset_phys(ds)->ds_creation_txg, tx);
388 dsl_dataset_rele(hds, FTAG);
389
390 } else {
391 ASSERT3P(ds_next->ds_prev, ==, ds);
392 dsl_dataset_rele(ds_next->ds_prev, ds_next);
393 ds_next->ds_prev = NULL;
394 if (ds_prev) {
395 VERIFY0(dsl_dataset_hold_obj(dp,
396 dsl_dataset_phys(ds)->ds_prev_snap_obj,
397 ds_next, &ds_next->ds_prev));
398 }
399
400 dsl_dataset_recalc_head_uniq(ds_next);
401
402 /*
403 * Reduce the amount of our unconsumed refreservation
404 * being charged to our parent by the amount of
405 * new unique data we have gained.
406 */
407 if (old_unique < ds_next->ds_reserved) {
408 int64_t mrsdelta;
409 uint64_t new_unique =
410 dsl_dataset_phys(ds_next)->ds_unique_bytes;
411
412 ASSERT(old_unique <= new_unique);
413 mrsdelta = MIN(new_unique - old_unique,
414 ds_next->ds_reserved - old_unique);
415 dsl_dir_diduse_space(ds->ds_dir,
416 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
417 }
418 }
419 dsl_dataset_rele(ds_next, FTAG);
420
421 /*
422 * This must be done after the dsl_traverse(), because it will
423 * re-open the objset.
424 */
425 if (ds->ds_objset) {
426 dmu_objset_evict(ds->ds_objset);
427 ds->ds_objset = NULL;
428 }
429
430 /* remove from snapshot namespace */
431 dsl_dataset_t *ds_head;
432 ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0);
433 VERIFY0(dsl_dataset_hold_obj(dp,
434 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &ds_head));
435 VERIFY0(dsl_dataset_get_snapname(ds));
436 #ifdef ZFS_DEBUG
437 {
438 uint64_t val;
439
440 err = dsl_dataset_snap_lookup(ds_head,
441 ds->ds_snapname, &val);
442 ASSERT0(err);
443 ASSERT3U(val, ==, obj);
444 }
445 #endif
446 VERIFY0(dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx, B_TRUE));
447 dsl_dataset_rele(ds_head, FTAG);
448
449 if (ds_prev != NULL)
450 dsl_dataset_rele(ds_prev, FTAG);
451
452 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
453
454 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
455 uint64_t count;
456 ASSERT0(zap_count(mos,
457 dsl_dataset_phys(ds)->ds_next_clones_obj, &count) &&
458 count == 0);
459 VERIFY0(dmu_object_free(mos,
460 dsl_dataset_phys(ds)->ds_next_clones_obj, tx));
461 }
462 if (dsl_dataset_phys(ds)->ds_props_obj != 0)
463 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_props_obj,
464 tx));
465 if (dsl_dataset_phys(ds)->ds_userrefs_obj != 0)
466 VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_userrefs_obj,
467 tx));
468 dsl_dir_rele(ds->ds_dir, ds);
469 ds->ds_dir = NULL;
470 dmu_object_free_zapified(mos, obj, tx);
471 }
472
473 static void
dsl_destroy_snapshot_sync(void * arg,dmu_tx_t * tx)474 dsl_destroy_snapshot_sync(void *arg, dmu_tx_t *tx)
475 {
476 dmu_snapshots_destroy_arg_t *dsda = arg;
477 dsl_pool_t *dp = dmu_tx_pool(tx);
478 nvpair_t *pair;
479
480 for (pair = nvlist_next_nvpair(dsda->dsda_successful_snaps, NULL);
481 pair != NULL;
482 pair = nvlist_next_nvpair(dsda->dsda_successful_snaps, pair)) {
483 dsl_dataset_t *ds;
484
485 VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
486
487 dsl_destroy_snapshot_sync_impl(ds, dsda->dsda_defer, tx);
488 dsl_dataset_rele(ds, FTAG);
489 }
490 }
491
492 /*
493 * The semantics of this function are described in the comment above
494 * lzc_destroy_snaps(). To summarize:
495 *
496 * The snapshots must all be in the same pool.
497 *
498 * Snapshots that don't exist will be silently ignored (considered to be
499 * "already deleted").
500 *
501 * On success, all snaps will be destroyed and this will return 0.
502 * On failure, no snaps will be destroyed, the errlist will be filled in,
503 * and this will return an errno.
504 */
505 int
dsl_destroy_snapshots_nvl(nvlist_t * snaps,boolean_t defer,nvlist_t * errlist)506 dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t defer,
507 nvlist_t *errlist)
508 {
509 dmu_snapshots_destroy_arg_t dsda;
510 int error;
511 nvpair_t *pair;
512
513 pair = nvlist_next_nvpair(snaps, NULL);
514 if (pair == NULL)
515 return (0);
516
517 dsda.dsda_snaps = snaps;
518 dsda.dsda_successful_snaps = fnvlist_alloc();
519 dsda.dsda_defer = defer;
520 dsda.dsda_errlist = errlist;
521
522 error = dsl_sync_task(nvpair_name(pair),
523 dsl_destroy_snapshot_check, dsl_destroy_snapshot_sync,
524 &dsda, 0, ZFS_SPACE_CHECK_NONE);
525 fnvlist_free(dsda.dsda_successful_snaps);
526
527 return (error);
528 }
529
530 int
dsl_destroy_snapshot(const char * name,boolean_t defer)531 dsl_destroy_snapshot(const char *name, boolean_t defer)
532 {
533 int error;
534 nvlist_t *nvl = fnvlist_alloc();
535 nvlist_t *errlist = fnvlist_alloc();
536
537 fnvlist_add_boolean(nvl, name);
538 error = dsl_destroy_snapshots_nvl(nvl, defer, errlist);
539 fnvlist_free(errlist);
540 fnvlist_free(nvl);
541 return (error);
542 }
543
544 struct killarg {
545 dsl_dataset_t *ds;
546 dmu_tx_t *tx;
547 };
548
549 /* ARGSUSED */
550 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)551 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
552 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
553 {
554 struct killarg *ka = arg;
555 dmu_tx_t *tx = ka->tx;
556
557 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
558 return (0);
559
560 if (zb->zb_level == ZB_ZIL_LEVEL) {
561 ASSERT(zilog != NULL);
562 /*
563 * It's a block in the intent log. It has no
564 * accounting, so just free it.
565 */
566 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
567 } else {
568 ASSERT(zilog == NULL);
569 ASSERT3U(bp->blk_birth, >,
570 dsl_dataset_phys(ka->ds)->ds_prev_snap_txg);
571 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
572 }
573
574 return (0);
575 }
576
577 static void
old_synchronous_dataset_destroy(dsl_dataset_t * ds,dmu_tx_t * tx)578 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
579 {
580 struct killarg ka;
581
582 /*
583 * Free everything that we point to (that's born after
584 * the previous snapshot, if we are a clone)
585 *
586 * NB: this should be very quick, because we already
587 * freed all the objects in open context.
588 */
589 ka.ds = ds;
590 ka.tx = tx;
591 VERIFY0(traverse_dataset(ds,
592 dsl_dataset_phys(ds)->ds_prev_snap_txg, TRAVERSE_POST,
593 kill_blkptr, &ka));
594 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
595 dsl_dataset_phys(ds)->ds_unique_bytes == 0);
596 }
597
598 typedef struct dsl_destroy_head_arg {
599 const char *ddha_name;
600 } dsl_destroy_head_arg_t;
601
602 int
dsl_destroy_head_check_impl(dsl_dataset_t * ds,int expected_holds)603 dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
604 {
605 int error;
606 uint64_t count;
607 objset_t *mos;
608
609 ASSERT(!ds->ds_is_snapshot);
610 if (ds->ds_is_snapshot)
611 return (SET_ERROR(EINVAL));
612
613 if (refcount_count(&ds->ds_longholds) != expected_holds)
614 return (SET_ERROR(EBUSY));
615
616 mos = ds->ds_dir->dd_pool->dp_meta_objset;
617
618 /*
619 * Can't delete a head dataset if there are snapshots of it.
620 * (Except if the only snapshots are from the branch we cloned
621 * from.)
622 */
623 if (ds->ds_prev != NULL &&
624 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object)
625 return (SET_ERROR(EBUSY));
626
627 /*
628 * Can't delete if there are children of this fs.
629 */
630 error = zap_count(mos,
631 dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, &count);
632 if (error != 0)
633 return (error);
634 if (count != 0)
635 return (SET_ERROR(EEXIST));
636
637 if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev) &&
638 dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
639 ds->ds_prev->ds_userrefs == 0) {
640 /* We need to remove the origin snapshot as well. */
641 if (!refcount_is_zero(&ds->ds_prev->ds_longholds))
642 return (SET_ERROR(EBUSY));
643 }
644 return (0);
645 }
646
647 static int
dsl_destroy_head_check(void * arg,dmu_tx_t * tx)648 dsl_destroy_head_check(void *arg, dmu_tx_t *tx)
649 {
650 dsl_destroy_head_arg_t *ddha = arg;
651 dsl_pool_t *dp = dmu_tx_pool(tx);
652 dsl_dataset_t *ds;
653 int error;
654
655 error = dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds);
656 if (error != 0)
657 return (error);
658
659 error = dsl_destroy_head_check_impl(ds, 0);
660 dsl_dataset_rele(ds, FTAG);
661 return (error);
662 }
663
664 static void
dsl_dir_destroy_sync(uint64_t ddobj,dmu_tx_t * tx)665 dsl_dir_destroy_sync(uint64_t ddobj, dmu_tx_t *tx)
666 {
667 dsl_dir_t *dd;
668 dsl_pool_t *dp = dmu_tx_pool(tx);
669 objset_t *mos = dp->dp_meta_objset;
670 dd_used_t t;
671
672 ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx)->dp_config_rwlock));
673
674 VERIFY0(dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd));
675
676 ASSERT0(dsl_dir_phys(dd)->dd_head_dataset_obj);
677
678 /*
679 * Decrement the filesystem count for all parent filesystems.
680 *
681 * When we receive an incremental stream into a filesystem that already
682 * exists, a temporary clone is created. We never count this temporary
683 * clone, whose name begins with a '%'.
684 */
685 if (dd->dd_myname[0] != '%' && dd->dd_parent != NULL)
686 dsl_fs_ss_count_adjust(dd->dd_parent, -1,
687 DD_FIELD_FILESYSTEM_COUNT, tx);
688
689 /*
690 * Remove our reservation. The impl() routine avoids setting the
691 * actual property, which would require the (already destroyed) ds.
692 */
693 dsl_dir_set_reservation_sync_impl(dd, 0, tx);
694
695 ASSERT0(dsl_dir_phys(dd)->dd_used_bytes);
696 ASSERT0(dsl_dir_phys(dd)->dd_reserved);
697 for (t = 0; t < DD_USED_NUM; t++)
698 ASSERT0(dsl_dir_phys(dd)->dd_used_breakdown[t]);
699
700 VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_child_dir_zapobj, tx));
701 VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_props_zapobj, tx));
702 VERIFY0(dsl_deleg_destroy(mos, dsl_dir_phys(dd)->dd_deleg_zapobj, tx));
703 VERIFY0(zap_remove(mos,
704 dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
705 dd->dd_myname, tx));
706
707 dsl_dir_rele(dd, FTAG);
708 dmu_object_free_zapified(mos, ddobj, tx);
709 }
710
711 void
dsl_destroy_head_sync_impl(dsl_dataset_t * ds,dmu_tx_t * tx)712 dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
713 {
714 dsl_pool_t *dp = dmu_tx_pool(tx);
715 objset_t *mos = dp->dp_meta_objset;
716 uint64_t obj, ddobj, prevobj = 0;
717 boolean_t rmorigin;
718
719 ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
720 ASSERT(ds->ds_prev == NULL ||
721 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj != ds->ds_object);
722 ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
723 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
724
725 /* We need to log before removing it from the namespace. */
726 spa_history_log_internal_ds(ds, "destroy", tx, "");
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 obj = ds->ds_object;
742
743 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
744 if (ds->ds_feature_inuse[f]) {
745 dsl_dataset_deactivate_feature(obj, f, tx);
746 ds->ds_feature_inuse[f] = B_FALSE;
747 }
748 }
749
750 dsl_scan_ds_destroyed(ds, tx);
751
752 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
753 /* This is a clone */
754 ASSERT(ds->ds_prev != NULL);
755 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj, !=,
756 obj);
757 ASSERT0(dsl_dataset_phys(ds)->ds_next_snap_obj);
758
759 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
760 if (dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj != 0) {
761 dsl_dataset_remove_from_next_clones(ds->ds_prev,
762 obj, tx);
763 }
764
765 ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_num_children, >, 1);
766 dsl_dataset_phys(ds->ds_prev)->ds_num_children--;
767 }
768
769 /*
770 * Destroy the deadlist. Unless it's a clone, the
771 * deadlist should be empty. (If it's a clone, it's
772 * safe to ignore the deadlist contents.)
773 */
774 dsl_deadlist_close(&ds->ds_deadlist);
775 dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
776 dmu_buf_will_dirty(ds->ds_dbuf, tx);
777 dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
778
779 objset_t *os;
780 VERIFY0(dmu_objset_from_ds(ds, &os));
781
782 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) {
783 old_synchronous_dataset_destroy(ds, tx);
784 } else {
785 /*
786 * Move the bptree into the pool's list of trees to
787 * clean up and update space accounting information.
788 */
789 uint64_t used, comp, uncomp;
790
791 zil_destroy_sync(dmu_objset_zil(os), tx);
792
793 if (!spa_feature_is_active(dp->dp_spa,
794 SPA_FEATURE_ASYNC_DESTROY)) {
795 dsl_scan_t *scn = dp->dp_scan;
796 spa_feature_incr(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY,
797 tx);
798 dp->dp_bptree_obj = bptree_alloc(mos, tx);
799 VERIFY0(zap_add(mos,
800 DMU_POOL_DIRECTORY_OBJECT,
801 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
802 &dp->dp_bptree_obj, tx));
803 ASSERT(!scn->scn_async_destroying);
804 scn->scn_async_destroying = B_TRUE;
805 }
806
807 used = dsl_dir_phys(ds->ds_dir)->dd_used_bytes;
808 comp = dsl_dir_phys(ds->ds_dir)->dd_compressed_bytes;
809 uncomp = dsl_dir_phys(ds->ds_dir)->dd_uncompressed_bytes;
810
811 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
812 dsl_dataset_phys(ds)->ds_unique_bytes == used);
813
814 bptree_add(mos, dp->dp_bptree_obj,
815 &dsl_dataset_phys(ds)->ds_bp,
816 dsl_dataset_phys(ds)->ds_prev_snap_txg,
817 used, comp, uncomp, tx);
818 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
819 -used, -comp, -uncomp, tx);
820 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
821 used, comp, uncomp, tx);
822 }
823
824 if (ds->ds_prev != NULL) {
825 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
826 VERIFY0(zap_remove_int(mos,
827 dsl_dir_phys(ds->ds_prev->ds_dir)->dd_clones,
828 ds->ds_object, tx));
829 }
830 prevobj = ds->ds_prev->ds_object;
831 dsl_dataset_rele(ds->ds_prev, ds);
832 ds->ds_prev = NULL;
833 }
834
835 /*
836 * This must be done after the dsl_traverse(), because it will
837 * re-open the objset.
838 */
839 if (ds->ds_objset) {
840 dmu_objset_evict(ds->ds_objset);
841 ds->ds_objset = NULL;
842 }
843
844 /* Erase the link in the dir */
845 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
846 dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj = 0;
847 ddobj = ds->ds_dir->dd_object;
848 ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0);
849 VERIFY0(zap_destroy(mos,
850 dsl_dataset_phys(ds)->ds_snapnames_zapobj, tx));
851
852 if (ds->ds_bookmarks != 0) {
853 VERIFY0(zap_destroy(mos, ds->ds_bookmarks, tx));
854 spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
855 }
856
857 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
858
859 ASSERT0(dsl_dataset_phys(ds)->ds_next_clones_obj);
860 ASSERT0(dsl_dataset_phys(ds)->ds_props_obj);
861 ASSERT0(dsl_dataset_phys(ds)->ds_userrefs_obj);
862 dsl_dir_rele(ds->ds_dir, ds);
863 ds->ds_dir = NULL;
864 dmu_object_free_zapified(mos, obj, tx);
865
866 dsl_dir_destroy_sync(ddobj, tx);
867
868 if (rmorigin) {
869 dsl_dataset_t *prev;
870 VERIFY0(dsl_dataset_hold_obj(dp, prevobj, FTAG, &prev));
871 dsl_destroy_snapshot_sync_impl(prev, B_FALSE, tx);
872 dsl_dataset_rele(prev, FTAG);
873 }
874 }
875
876 static void
dsl_destroy_head_sync(void * arg,dmu_tx_t * tx)877 dsl_destroy_head_sync(void *arg, dmu_tx_t *tx)
878 {
879 dsl_destroy_head_arg_t *ddha = arg;
880 dsl_pool_t *dp = dmu_tx_pool(tx);
881 dsl_dataset_t *ds;
882
883 VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
884 dsl_destroy_head_sync_impl(ds, tx);
885 dsl_dataset_rele(ds, FTAG);
886 }
887
888 static void
dsl_destroy_head_begin_sync(void * arg,dmu_tx_t * tx)889 dsl_destroy_head_begin_sync(void *arg, dmu_tx_t *tx)
890 {
891 dsl_destroy_head_arg_t *ddha = arg;
892 dsl_pool_t *dp = dmu_tx_pool(tx);
893 dsl_dataset_t *ds;
894
895 VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
896
897 /* Mark it as inconsistent on-disk, in case we crash */
898 dmu_buf_will_dirty(ds->ds_dbuf, tx);
899 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT;
900
901 spa_history_log_internal_ds(ds, "destroy begin", tx, "");
902 dsl_dataset_rele(ds, FTAG);
903 }
904
905 int
dsl_destroy_head(const char * name)906 dsl_destroy_head(const char *name)
907 {
908 dsl_destroy_head_arg_t ddha;
909 int error;
910 spa_t *spa;
911 boolean_t isenabled;
912
913 #ifdef _KERNEL
914 zfs_destroy_unmount_origin(name);
915 #endif
916
917 error = spa_open(name, &spa, FTAG);
918 if (error != 0)
919 return (error);
920 isenabled = spa_feature_is_enabled(spa, SPA_FEATURE_ASYNC_DESTROY);
921 spa_close(spa, FTAG);
922
923 ddha.ddha_name = name;
924
925 if (!isenabled) {
926 objset_t *os;
927
928 error = dsl_sync_task(name, dsl_destroy_head_check,
929 dsl_destroy_head_begin_sync, &ddha,
930 0, ZFS_SPACE_CHECK_NONE);
931 if (error != 0)
932 return (error);
933
934 /*
935 * Head deletion is processed in one txg on old pools;
936 * remove the objects from open context so that the txg sync
937 * is not too long.
938 */
939 error = dmu_objset_own(name, DMU_OST_ANY, B_FALSE, FTAG, &os);
940 if (error == 0) {
941 uint64_t prev_snap_txg =
942 dsl_dataset_phys(dmu_objset_ds(os))->
943 ds_prev_snap_txg;
944 for (uint64_t obj = 0; error == 0;
945 error = dmu_object_next(os, &obj, FALSE,
946 prev_snap_txg))
947 (void) dmu_free_long_object(os, obj);
948 /* sync out all frees */
949 txg_wait_synced(dmu_objset_pool(os), 0);
950 dmu_objset_disown(os, FTAG);
951 }
952 }
953
954 return (dsl_sync_task(name, dsl_destroy_head_check,
955 dsl_destroy_head_sync, &ddha, 0, ZFS_SPACE_CHECK_NONE));
956 }
957
958 /*
959 * Note, this function is used as the callback for dmu_objset_find(). We
960 * always return 0 so that we will continue to find and process
961 * inconsistent datasets, even if we encounter an error trying to
962 * process one of them.
963 */
964 /* ARGSUSED */
965 int
dsl_destroy_inconsistent(const char * dsname,void * arg)966 dsl_destroy_inconsistent(const char *dsname, void *arg)
967 {
968 objset_t *os;
969
970 if (dmu_objset_hold(dsname, FTAG, &os) == 0) {
971 boolean_t need_destroy = DS_IS_INCONSISTENT(dmu_objset_ds(os));
972
973 /*
974 * If the dataset is inconsistent because a resumable receive
975 * has failed, then do not destroy it.
976 */
977 if (dsl_dataset_has_resume_receive_state(dmu_objset_ds(os)))
978 need_destroy = B_FALSE;
979
980 dmu_objset_rele(os, FTAG);
981 if (need_destroy)
982 (void) dsl_destroy_head(dsname);
983 }
984 return (0);
985 }
986