xref: /freebsd/sys/contrib/openzfs/module/zfs/dsl_destroy.c (revision 7a7741af18d6c8a804cc643cb7ecda9d730c6aa6)
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 https://opensource.org/licenses/CDDL-1.0.
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) 2016 Actifio, Inc. All rights reserved.
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/dsl_bookmark.h>
35 #include <sys/dmu_tx.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/dsl_dir.h>
38 #include <sys/dmu_traverse.h>
39 #include <sys/dsl_scan.h>
40 #include <sys/dmu_objset.h>
41 #include <sys/zap.h>
42 #include <sys/zfeature.h>
43 #include <sys/zfs_ioctl.h>
44 #include <sys/dsl_deleg.h>
45 #include <sys/dmu_impl.h>
46 #include <sys/zvol.h>
47 #include <sys/zcp.h>
48 #include <sys/dsl_deadlist.h>
49 #include <sys/zthr.h>
50 #include <sys/spa_impl.h>
51 
52 extern int zfs_snapshot_history_enabled;
53 
54 int
dsl_destroy_snapshot_check_impl(dsl_dataset_t * ds,boolean_t defer)55 dsl_destroy_snapshot_check_impl(dsl_dataset_t *ds, boolean_t defer)
56 {
57 	if (!ds->ds_is_snapshot)
58 		return (SET_ERROR(EINVAL));
59 
60 	if (dsl_dataset_long_held(ds))
61 		return (SET_ERROR(EBUSY));
62 
63 	/*
64 	 * Only allow deferred destroy on pools that support it.
65 	 * NOTE: deferred destroy is only supported on snapshots.
66 	 */
67 	if (defer) {
68 		if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
69 		    SPA_VERSION_USERREFS)
70 			return (SET_ERROR(ENOTSUP));
71 		return (0);
72 	}
73 
74 	/*
75 	 * If this snapshot has an elevated user reference count,
76 	 * we can't destroy it yet.
77 	 */
78 	if (ds->ds_userrefs > 0)
79 		return (SET_ERROR(EBUSY));
80 
81 	/*
82 	 * Can't delete a branch point.
83 	 */
84 	if (dsl_dataset_phys(ds)->ds_num_children > 1)
85 		return (SET_ERROR(EEXIST));
86 
87 	return (0);
88 }
89 
90 int
dsl_destroy_snapshot_check(void * arg,dmu_tx_t * tx)91 dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx)
92 {
93 	dsl_destroy_snapshot_arg_t *ddsa = arg;
94 	const char *dsname = ddsa->ddsa_name;
95 	boolean_t defer = ddsa->ddsa_defer;
96 
97 	dsl_pool_t *dp = dmu_tx_pool(tx);
98 	int error = 0;
99 	dsl_dataset_t *ds;
100 
101 	error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
102 
103 	/*
104 	 * If the snapshot does not exist, silently ignore it, and
105 	 * dsl_destroy_snapshot_sync() will be a no-op
106 	 * (it's "already destroyed").
107 	 */
108 	if (error == ENOENT)
109 		return (0);
110 
111 	if (error == 0) {
112 		error = dsl_destroy_snapshot_check_impl(ds, defer);
113 		dsl_dataset_rele(ds, FTAG);
114 	}
115 
116 	return (error);
117 }
118 
119 struct process_old_arg {
120 	dsl_dataset_t *ds;
121 	dsl_dataset_t *ds_prev;
122 	boolean_t after_branch_point;
123 	zio_t *pio;
124 	uint64_t used, comp, uncomp;
125 };
126 
127 static int
process_old_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)128 process_old_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
129 {
130 	struct process_old_arg *poa = arg;
131 	dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
132 
133 	ASSERT(!BP_IS_HOLE(bp));
134 
135 	if (BP_GET_LOGICAL_BIRTH(bp) <=
136 	    dsl_dataset_phys(poa->ds)->ds_prev_snap_txg) {
137 		dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, bp_freed, tx);
138 		if (poa->ds_prev && !poa->after_branch_point &&
139 		    BP_GET_LOGICAL_BIRTH(bp) >
140 		    dsl_dataset_phys(poa->ds_prev)->ds_prev_snap_txg) {
141 			dsl_dataset_phys(poa->ds_prev)->ds_unique_bytes +=
142 			    bp_get_dsize_sync(dp->dp_spa, bp);
143 		}
144 	} else {
145 		poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
146 		poa->comp += BP_GET_PSIZE(bp);
147 		poa->uncomp += BP_GET_UCSIZE(bp);
148 		dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
149 	}
150 	return (0);
151 }
152 
153 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)154 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
155     dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
156 {
157 	struct process_old_arg poa = { 0 };
158 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
159 	objset_t *mos = dp->dp_meta_objset;
160 	uint64_t deadlist_obj;
161 
162 	ASSERT(ds->ds_deadlist.dl_oldfmt);
163 	ASSERT(ds_next->ds_deadlist.dl_oldfmt);
164 
165 	poa.ds = ds;
166 	poa.ds_prev = ds_prev;
167 	poa.after_branch_point = after_branch_point;
168 	poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
169 	VERIFY0(bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
170 	    process_old_cb, &poa, tx));
171 	VERIFY0(zio_wait(poa.pio));
172 	ASSERT3U(poa.used, ==, dsl_dataset_phys(ds)->ds_unique_bytes);
173 
174 	/* change snapused */
175 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
176 	    -poa.used, -poa.comp, -poa.uncomp, tx);
177 
178 	/* swap next's deadlist to our deadlist */
179 	dsl_deadlist_close(&ds->ds_deadlist);
180 	dsl_deadlist_close(&ds_next->ds_deadlist);
181 	deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
182 	dsl_dataset_phys(ds)->ds_deadlist_obj =
183 	    dsl_dataset_phys(ds_next)->ds_deadlist_obj;
184 	dsl_dataset_phys(ds_next)->ds_deadlist_obj = deadlist_obj;
185 	dsl_deadlist_open(&ds->ds_deadlist, mos,
186 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
187 	dsl_deadlist_open(&ds_next->ds_deadlist, mos,
188 	    dsl_dataset_phys(ds_next)->ds_deadlist_obj);
189 }
190 
191 typedef struct remaining_clones_key {
192 	dsl_dataset_t *rck_clone;
193 	list_node_t rck_node;
194 } remaining_clones_key_t;
195 
196 static remaining_clones_key_t *
rck_alloc(dsl_dataset_t * clone)197 rck_alloc(dsl_dataset_t *clone)
198 {
199 	remaining_clones_key_t *rck = kmem_alloc(sizeof (*rck), KM_SLEEP);
200 	rck->rck_clone = clone;
201 	return (rck);
202 }
203 
204 static void
dsl_dir_remove_clones_key_impl(dsl_dir_t * dd,uint64_t mintxg,dmu_tx_t * tx,list_t * stack,const void * tag)205 dsl_dir_remove_clones_key_impl(dsl_dir_t *dd, uint64_t mintxg, dmu_tx_t *tx,
206     list_t *stack, const void *tag)
207 {
208 	objset_t *mos = dd->dd_pool->dp_meta_objset;
209 
210 	/*
211 	 * If it is the old version, dd_clones doesn't exist so we can't
212 	 * find the clones, but dsl_deadlist_remove_key() is a no-op so it
213 	 * doesn't matter.
214 	 */
215 	if (dsl_dir_phys(dd)->dd_clones == 0)
216 		return;
217 
218 	zap_cursor_t *zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
219 	zap_attribute_t *za = zap_attribute_alloc();
220 
221 	for (zap_cursor_init(zc, mos, dsl_dir_phys(dd)->dd_clones);
222 	    zap_cursor_retrieve(zc, za) == 0;
223 	    zap_cursor_advance(zc)) {
224 		dsl_dataset_t *clone;
225 
226 		VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
227 		    za->za_first_integer, tag, &clone));
228 
229 		if (clone->ds_dir->dd_origin_txg > mintxg) {
230 			dsl_deadlist_remove_key(&clone->ds_deadlist,
231 			    mintxg, tx);
232 
233 			if (dsl_dataset_remap_deadlist_exists(clone)) {
234 				dsl_deadlist_remove_key(
235 				    &clone->ds_remap_deadlist, mintxg, tx);
236 			}
237 
238 			list_insert_head(stack, rck_alloc(clone));
239 		} else {
240 			dsl_dataset_rele(clone, tag);
241 		}
242 	}
243 	zap_cursor_fini(zc);
244 
245 	zap_attribute_free(za);
246 	kmem_free(zc, sizeof (zap_cursor_t));
247 }
248 
249 void
dsl_dir_remove_clones_key(dsl_dir_t * top_dd,uint64_t mintxg,dmu_tx_t * tx)250 dsl_dir_remove_clones_key(dsl_dir_t *top_dd, uint64_t mintxg, dmu_tx_t *tx)
251 {
252 	list_t stack;
253 
254 	list_create(&stack, sizeof (remaining_clones_key_t),
255 	    offsetof(remaining_clones_key_t, rck_node));
256 
257 	dsl_dir_remove_clones_key_impl(top_dd, mintxg, tx, &stack, FTAG);
258 	for (remaining_clones_key_t *rck = list_remove_head(&stack);
259 	    rck != NULL; rck = list_remove_head(&stack)) {
260 		dsl_dataset_t *clone = rck->rck_clone;
261 		dsl_dir_t *clone_dir = clone->ds_dir;
262 
263 		kmem_free(rck, sizeof (*rck));
264 
265 		dsl_dir_remove_clones_key_impl(clone_dir, mintxg, tx,
266 		    &stack, FTAG);
267 		dsl_dataset_rele(clone, FTAG);
268 	}
269 
270 	list_destroy(&stack);
271 }
272 
273 static void
dsl_destroy_snapshot_handle_remaps(dsl_dataset_t * ds,dsl_dataset_t * ds_next,dmu_tx_t * tx)274 dsl_destroy_snapshot_handle_remaps(dsl_dataset_t *ds, dsl_dataset_t *ds_next,
275     dmu_tx_t *tx)
276 {
277 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
278 
279 	/* Move blocks to be obsoleted to pool's obsolete list. */
280 	if (dsl_dataset_remap_deadlist_exists(ds_next)) {
281 		if (!bpobj_is_open(&dp->dp_obsolete_bpobj))
282 			dsl_pool_create_obsolete_bpobj(dp, tx);
283 
284 		dsl_deadlist_move_bpobj(&ds_next->ds_remap_deadlist,
285 		    &dp->dp_obsolete_bpobj,
286 		    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
287 	}
288 
289 	/* Merge our deadlist into next's and free it. */
290 	if (dsl_dataset_remap_deadlist_exists(ds)) {
291 		uint64_t remap_deadlist_object =
292 		    dsl_dataset_get_remap_deadlist_object(ds);
293 		ASSERT(remap_deadlist_object != 0);
294 
295 		mutex_enter(&ds_next->ds_remap_deadlist_lock);
296 		if (!dsl_dataset_remap_deadlist_exists(ds_next))
297 			dsl_dataset_create_remap_deadlist(ds_next, tx);
298 		mutex_exit(&ds_next->ds_remap_deadlist_lock);
299 
300 		dsl_deadlist_merge(&ds_next->ds_remap_deadlist,
301 		    remap_deadlist_object, tx);
302 		dsl_dataset_destroy_remap_deadlist(ds, tx);
303 	}
304 }
305 
306 void
dsl_destroy_snapshot_sync_impl(dsl_dataset_t * ds,boolean_t defer,dmu_tx_t * tx)307 dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
308 {
309 	int after_branch_point = FALSE;
310 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
311 	objset_t *mos = dp->dp_meta_objset;
312 	dsl_dataset_t *ds_prev = NULL;
313 	uint64_t obj;
314 
315 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
316 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
317 	ASSERT3U(BP_GET_LOGICAL_BIRTH(&dsl_dataset_phys(ds)->ds_bp), <=,
318 	    tx->tx_txg);
319 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
320 	ASSERT(zfs_refcount_is_zero(&ds->ds_longholds));
321 
322 	if (defer &&
323 	    (ds->ds_userrefs > 0 ||
324 	    dsl_dataset_phys(ds)->ds_num_children > 1)) {
325 		ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
326 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
327 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_DEFER_DESTROY;
328 		if (zfs_snapshot_history_enabled) {
329 			spa_history_log_internal_ds(ds, "defer_destroy", tx,
330 			    " ");
331 		}
332 		return;
333 	}
334 
335 	ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
336 
337 	if (zfs_snapshot_history_enabled) {
338 		/* We need to log before removing it from the namespace. */
339 		spa_history_log_internal_ds(ds, "destroy", tx, " ");
340 	}
341 
342 	dsl_scan_ds_destroyed(ds, tx);
343 
344 	obj = ds->ds_object;
345 
346 	boolean_t book_exists = dsl_bookmark_ds_destroyed(ds, tx);
347 
348 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
349 		if (dsl_dataset_feature_is_active(ds, f))
350 			dsl_dataset_deactivate_feature(ds, f, tx);
351 	}
352 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
353 		ASSERT3P(ds->ds_prev, ==, NULL);
354 		VERIFY0(dsl_dataset_hold_obj(dp,
355 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &ds_prev));
356 		after_branch_point =
357 		    (dsl_dataset_phys(ds_prev)->ds_next_snap_obj != obj);
358 
359 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
360 		if (after_branch_point &&
361 		    dsl_dataset_phys(ds_prev)->ds_next_clones_obj != 0) {
362 			dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
363 			if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
364 				VERIFY0(zap_add_int(mos,
365 				    dsl_dataset_phys(ds_prev)->
366 				    ds_next_clones_obj,
367 				    dsl_dataset_phys(ds)->ds_next_snap_obj,
368 				    tx));
369 			}
370 		}
371 		if (!after_branch_point) {
372 			dsl_dataset_phys(ds_prev)->ds_next_snap_obj =
373 			    dsl_dataset_phys(ds)->ds_next_snap_obj;
374 		}
375 	}
376 
377 	dsl_dataset_t *ds_next;
378 	uint64_t old_unique;
379 	uint64_t used = 0, comp = 0, uncomp = 0;
380 
381 	VERIFY0(dsl_dataset_hold_obj(dp,
382 	    dsl_dataset_phys(ds)->ds_next_snap_obj, FTAG, &ds_next));
383 	ASSERT3U(dsl_dataset_phys(ds_next)->ds_prev_snap_obj, ==, obj);
384 
385 	old_unique = dsl_dataset_phys(ds_next)->ds_unique_bytes;
386 
387 	dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
388 	dsl_dataset_phys(ds_next)->ds_prev_snap_obj =
389 	    dsl_dataset_phys(ds)->ds_prev_snap_obj;
390 	dsl_dataset_phys(ds_next)->ds_prev_snap_txg =
391 	    dsl_dataset_phys(ds)->ds_prev_snap_txg;
392 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
393 	    ds_prev ? dsl_dataset_phys(ds_prev)->ds_creation_txg : 0);
394 
395 	if (ds_next->ds_deadlist.dl_oldfmt) {
396 		process_old_deadlist(ds, ds_prev, ds_next,
397 		    after_branch_point, tx);
398 	} else {
399 		/* Adjust prev's unique space. */
400 		if (ds_prev && !after_branch_point) {
401 			dsl_deadlist_space_range(&ds_next->ds_deadlist,
402 			    dsl_dataset_phys(ds_prev)->ds_prev_snap_txg,
403 			    dsl_dataset_phys(ds)->ds_prev_snap_txg,
404 			    &used, &comp, &uncomp);
405 			dsl_dataset_phys(ds_prev)->ds_unique_bytes += used;
406 		}
407 
408 		/* Adjust snapused. */
409 		dsl_deadlist_space_range(&ds_next->ds_deadlist,
410 		    dsl_dataset_phys(ds)->ds_prev_snap_txg, UINT64_MAX,
411 		    &used, &comp, &uncomp);
412 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
413 		    -used, -comp, -uncomp, tx);
414 
415 		/* Move blocks to be freed to pool's free list. */
416 		dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
417 		    &dp->dp_free_bpobj, dsl_dataset_phys(ds)->ds_prev_snap_txg,
418 		    tx);
419 		dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
420 		    DD_USED_HEAD, used, comp, uncomp, tx);
421 
422 		/* Merge our deadlist into next's and free it. */
423 		dsl_deadlist_merge(&ds_next->ds_deadlist,
424 		    dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
425 
426 		/*
427 		 * We are done with the deadlist tree (generated/used
428 		 * by dsl_deadlist_move_bpobj() and dsl_deadlist_merge()).
429 		 * Discard it to save memory.
430 		 */
431 		dsl_deadlist_discard_tree(&ds_next->ds_deadlist);
432 	}
433 
434 	dsl_deadlist_close(&ds->ds_deadlist);
435 	dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
436 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
437 	dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
438 
439 	dsl_destroy_snapshot_handle_remaps(ds, ds_next, tx);
440 
441 	if (!book_exists) {
442 		/* Collapse range in clone heads */
443 		dsl_dir_remove_clones_key(ds->ds_dir,
444 		    dsl_dataset_phys(ds)->ds_creation_txg, tx);
445 	}
446 
447 	if (ds_next->ds_is_snapshot) {
448 		dsl_dataset_t *ds_nextnext;
449 
450 		/*
451 		 * Update next's unique to include blocks which
452 		 * were previously shared by only this snapshot
453 		 * and it.  Those blocks will be born after the
454 		 * prev snap and before this snap, and will have
455 		 * died after the next snap and before the one
456 		 * after that (ie. be on the snap after next's
457 		 * deadlist).
458 		 */
459 		VERIFY0(dsl_dataset_hold_obj(dp,
460 		    dsl_dataset_phys(ds_next)->ds_next_snap_obj,
461 		    FTAG, &ds_nextnext));
462 		dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
463 		    dsl_dataset_phys(ds)->ds_prev_snap_txg,
464 		    dsl_dataset_phys(ds)->ds_creation_txg,
465 		    &used, &comp, &uncomp);
466 		dsl_dataset_phys(ds_next)->ds_unique_bytes += used;
467 		dsl_dataset_rele(ds_nextnext, FTAG);
468 		ASSERT3P(ds_next->ds_prev, ==, NULL);
469 
470 		/* Collapse range in this head. */
471 		dsl_dataset_t *hds;
472 		VERIFY0(dsl_dataset_hold_obj(dp,
473 		    dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
474 		    FTAG, &hds));
475 		if (!book_exists) {
476 			/* Collapse range in this head. */
477 			dsl_deadlist_remove_key(&hds->ds_deadlist,
478 			    dsl_dataset_phys(ds)->ds_creation_txg, tx);
479 		}
480 		if (dsl_dataset_remap_deadlist_exists(hds)) {
481 			dsl_deadlist_remove_key(&hds->ds_remap_deadlist,
482 			    dsl_dataset_phys(ds)->ds_creation_txg, tx);
483 		}
484 		dsl_dataset_rele(hds, FTAG);
485 
486 	} else {
487 		ASSERT3P(ds_next->ds_prev, ==, ds);
488 		dsl_dataset_rele(ds_next->ds_prev, ds_next);
489 		ds_next->ds_prev = NULL;
490 		if (ds_prev) {
491 			VERIFY0(dsl_dataset_hold_obj(dp,
492 			    dsl_dataset_phys(ds)->ds_prev_snap_obj,
493 			    ds_next, &ds_next->ds_prev));
494 		}
495 
496 		dsl_dataset_recalc_head_uniq(ds_next);
497 
498 		/*
499 		 * Reduce the amount of our unconsumed refreservation
500 		 * being charged to our parent by the amount of
501 		 * new unique data we have gained.
502 		 */
503 		if (old_unique < ds_next->ds_reserved) {
504 			int64_t mrsdelta;
505 			uint64_t new_unique =
506 			    dsl_dataset_phys(ds_next)->ds_unique_bytes;
507 
508 			ASSERT(old_unique <= new_unique);
509 			mrsdelta = MIN(new_unique - old_unique,
510 			    ds_next->ds_reserved - old_unique);
511 			dsl_dir_diduse_space(ds->ds_dir,
512 			    DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
513 		}
514 	}
515 	dsl_dataset_rele(ds_next, FTAG);
516 
517 	/*
518 	 * This must be done after the dsl_traverse(), because it will
519 	 * re-open the objset.
520 	 */
521 	if (ds->ds_objset) {
522 		dmu_objset_evict(ds->ds_objset);
523 		ds->ds_objset = NULL;
524 	}
525 
526 	/* remove from snapshot namespace */
527 	dsl_dataset_t *ds_head;
528 	ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0);
529 	VERIFY0(dsl_dataset_hold_obj(dp,
530 	    dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj, FTAG, &ds_head));
531 	VERIFY0(dsl_dataset_get_snapname(ds));
532 #ifdef ZFS_DEBUG
533 	{
534 		uint64_t val;
535 		int err;
536 
537 		err = dsl_dataset_snap_lookup(ds_head,
538 		    ds->ds_snapname, &val);
539 		ASSERT0(err);
540 		ASSERT3U(val, ==, obj);
541 	}
542 #endif
543 	VERIFY0(dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx, B_TRUE));
544 	dsl_dataset_rele(ds_head, FTAG);
545 
546 	if (ds_prev != NULL)
547 		dsl_dataset_rele(ds_prev, FTAG);
548 
549 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
550 
551 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
552 		uint64_t count __maybe_unused;
553 		ASSERT0(zap_count(mos,
554 		    dsl_dataset_phys(ds)->ds_next_clones_obj, &count) &&
555 		    count == 0);
556 		VERIFY0(dmu_object_free(mos,
557 		    dsl_dataset_phys(ds)->ds_next_clones_obj, tx));
558 	}
559 	if (dsl_dataset_phys(ds)->ds_props_obj != 0)
560 		VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_props_obj,
561 		    tx));
562 	if (dsl_dataset_phys(ds)->ds_userrefs_obj != 0)
563 		VERIFY0(zap_destroy(mos, dsl_dataset_phys(ds)->ds_userrefs_obj,
564 		    tx));
565 	dsl_dir_rele(ds->ds_dir, ds);
566 	ds->ds_dir = NULL;
567 	dmu_object_free_zapified(mos, obj, tx);
568 }
569 
570 void
dsl_destroy_snapshot_sync(void * arg,dmu_tx_t * tx)571 dsl_destroy_snapshot_sync(void *arg, dmu_tx_t *tx)
572 {
573 	dsl_destroy_snapshot_arg_t *ddsa = arg;
574 	const char *dsname = ddsa->ddsa_name;
575 	boolean_t defer = ddsa->ddsa_defer;
576 
577 	dsl_pool_t *dp = dmu_tx_pool(tx);
578 	dsl_dataset_t *ds;
579 
580 	int error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
581 	if (error == ENOENT)
582 		return;
583 	ASSERT0(error);
584 	dsl_destroy_snapshot_sync_impl(ds, defer, tx);
585 	zvol_remove_minors(dp->dp_spa, dsname, B_TRUE);
586 	dsl_dataset_rele(ds, FTAG);
587 }
588 
589 /*
590  * The semantics of this function are described in the comment above
591  * lzc_destroy_snaps().  To summarize:
592  *
593  * The snapshots must all be in the same pool.
594  *
595  * Snapshots that don't exist will be silently ignored (considered to be
596  * "already deleted").
597  *
598  * On success, all snaps will be destroyed and this will return 0.
599  * On failure, no snaps will be destroyed, the errlist will be filled in,
600  * and this will return an errno.
601  */
602 int
dsl_destroy_snapshots_nvl(nvlist_t * snaps,boolean_t defer,nvlist_t * errlist)603 dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t defer,
604     nvlist_t *errlist)
605 {
606 	if (nvlist_next_nvpair(snaps, NULL) == NULL)
607 		return (0);
608 
609 	/*
610 	 * lzc_destroy_snaps() is documented to take an nvlist whose
611 	 * values "don't matter".  We need to convert that nvlist to
612 	 * one that we know can be converted to LUA.
613 	 */
614 	nvlist_t *snaps_normalized = fnvlist_alloc();
615 	for (nvpair_t *pair = nvlist_next_nvpair(snaps, NULL);
616 	    pair != NULL; pair = nvlist_next_nvpair(snaps, pair)) {
617 		fnvlist_add_boolean_value(snaps_normalized,
618 		    nvpair_name(pair), B_TRUE);
619 	}
620 
621 	nvlist_t *arg = fnvlist_alloc();
622 	fnvlist_add_nvlist(arg, "snaps", snaps_normalized);
623 	fnvlist_free(snaps_normalized);
624 	fnvlist_add_boolean_value(arg, "defer", defer);
625 
626 	nvlist_t *wrapper = fnvlist_alloc();
627 	fnvlist_add_nvlist(wrapper, ZCP_ARG_ARGLIST, arg);
628 	fnvlist_free(arg);
629 
630 	const char *program =
631 	    "arg = ...\n"
632 	    "snaps = arg['snaps']\n"
633 	    "defer = arg['defer']\n"
634 	    "errors = { }\n"
635 	    "has_errors = false\n"
636 	    "for snap, v in pairs(snaps) do\n"
637 	    "    errno = zfs.check.destroy{snap, defer=defer}\n"
638 	    "    zfs.debug('snap: ' .. snap .. ' errno: ' .. errno)\n"
639 	    "    if errno == ENOENT then\n"
640 	    "        snaps[snap] = nil\n"
641 	    "    elseif errno ~= 0 then\n"
642 	    "        errors[snap] = errno\n"
643 	    "        has_errors = true\n"
644 	    "    end\n"
645 	    "end\n"
646 	    "if has_errors then\n"
647 	    "    return errors\n"
648 	    "end\n"
649 	    "for snap, v in pairs(snaps) do\n"
650 	    "    errno = zfs.sync.destroy{snap, defer=defer}\n"
651 	    "    assert(errno == 0)\n"
652 	    "end\n"
653 	    "return { }\n";
654 
655 	nvlist_t *result = fnvlist_alloc();
656 	int error = zcp_eval(nvpair_name(nvlist_next_nvpair(snaps, NULL)),
657 	    program,
658 	    B_TRUE,
659 	    0,
660 	    zfs_lua_max_memlimit,
661 	    fnvlist_lookup_nvpair(wrapper, ZCP_ARG_ARGLIST), result);
662 	if (error != 0) {
663 		const char *errorstr = NULL;
664 		(void) nvlist_lookup_string(result, ZCP_RET_ERROR, &errorstr);
665 		if (errorstr != NULL) {
666 			zfs_dbgmsg("%s", errorstr);
667 		}
668 		fnvlist_free(wrapper);
669 		fnvlist_free(result);
670 		return (error);
671 	}
672 	fnvlist_free(wrapper);
673 
674 	/*
675 	 * lzc_destroy_snaps() is documented to fill the errlist with
676 	 * int32 values, so we need to convert the int64 values that are
677 	 * returned from LUA.
678 	 */
679 	int rv = 0;
680 	nvlist_t *errlist_raw = fnvlist_lookup_nvlist(result, ZCP_RET_RETURN);
681 	for (nvpair_t *pair = nvlist_next_nvpair(errlist_raw, NULL);
682 	    pair != NULL; pair = nvlist_next_nvpair(errlist_raw, pair)) {
683 		int32_t val = (int32_t)fnvpair_value_int64(pair);
684 		if (rv == 0)
685 			rv = val;
686 		fnvlist_add_int32(errlist, nvpair_name(pair), val);
687 	}
688 	fnvlist_free(result);
689 	return (rv);
690 }
691 
692 int
dsl_destroy_snapshot(const char * name,boolean_t defer)693 dsl_destroy_snapshot(const char *name, boolean_t defer)
694 {
695 	int error;
696 	nvlist_t *nvl = fnvlist_alloc();
697 	nvlist_t *errlist = fnvlist_alloc();
698 
699 	fnvlist_add_boolean(nvl, name);
700 	error = dsl_destroy_snapshots_nvl(nvl, defer, errlist);
701 	fnvlist_free(errlist);
702 	fnvlist_free(nvl);
703 	return (error);
704 }
705 
706 struct killarg {
707 	dsl_dataset_t *ds;
708 	dmu_tx_t *tx;
709 };
710 
711 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)712 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
713     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
714 {
715 	(void) spa, (void) dnp;
716 	struct killarg *ka = arg;
717 	dmu_tx_t *tx = ka->tx;
718 
719 	if (zb->zb_level == ZB_DNODE_LEVEL || BP_IS_HOLE(bp) ||
720 	    BP_IS_EMBEDDED(bp))
721 		return (0);
722 
723 	if (zb->zb_level == ZB_ZIL_LEVEL) {
724 		ASSERT(zilog != NULL);
725 		/*
726 		 * It's a block in the intent log.  It has no
727 		 * accounting, so just free it.
728 		 */
729 		dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
730 	} else {
731 		ASSERT(zilog == NULL);
732 		ASSERT3U(BP_GET_LOGICAL_BIRTH(bp), >,
733 		    dsl_dataset_phys(ka->ds)->ds_prev_snap_txg);
734 		(void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
735 	}
736 
737 	return (0);
738 }
739 
740 static void
old_synchronous_dataset_destroy(dsl_dataset_t * ds,dmu_tx_t * tx)741 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
742 {
743 	struct killarg ka;
744 
745 	spa_history_log_internal_ds(ds, "destroy", tx,
746 	    "(synchronous, mintxg=%llu)",
747 	    (long long)dsl_dataset_phys(ds)->ds_prev_snap_txg);
748 
749 	/*
750 	 * Free everything that we point to (that's born after
751 	 * the previous snapshot, if we are a clone)
752 	 *
753 	 * NB: this should be very quick, because we already
754 	 * freed all the objects in open context.
755 	 */
756 	ka.ds = ds;
757 	ka.tx = tx;
758 	VERIFY0(traverse_dataset(ds,
759 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, TRAVERSE_POST |
760 	    TRAVERSE_NO_DECRYPT, kill_blkptr, &ka));
761 	ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
762 	    dsl_dataset_phys(ds)->ds_unique_bytes == 0);
763 }
764 
765 int
dsl_destroy_head_check_impl(dsl_dataset_t * ds,int expected_holds)766 dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
767 {
768 	int error;
769 	uint64_t count;
770 	objset_t *mos;
771 
772 	ASSERT(!ds->ds_is_snapshot);
773 	if (ds->ds_is_snapshot)
774 		return (SET_ERROR(EINVAL));
775 
776 	if (zfs_refcount_count(&ds->ds_longholds) != expected_holds)
777 		return (SET_ERROR(EBUSY));
778 
779 	ASSERT0(ds->ds_dir->dd_activity_waiters);
780 
781 	mos = ds->ds_dir->dd_pool->dp_meta_objset;
782 
783 	/*
784 	 * Can't delete a head dataset if there are snapshots of it.
785 	 * (Except if the only snapshots are from the branch we cloned
786 	 * from.)
787 	 */
788 	if (ds->ds_prev != NULL &&
789 	    dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj == ds->ds_object)
790 		return (SET_ERROR(EBUSY));
791 
792 	/*
793 	 * Can't delete if there are children of this fs.
794 	 */
795 	error = zap_count(mos,
796 	    dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, &count);
797 	if (error != 0)
798 		return (error);
799 	if (count != 0)
800 		return (SET_ERROR(EEXIST));
801 
802 	if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev) &&
803 	    dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
804 	    ds->ds_prev->ds_userrefs == 0) {
805 		/* We need to remove the origin snapshot as well. */
806 		if (!zfs_refcount_is_zero(&ds->ds_prev->ds_longholds))
807 			return (SET_ERROR(EBUSY));
808 	}
809 	return (0);
810 }
811 
812 int
dsl_destroy_head_check(void * arg,dmu_tx_t * tx)813 dsl_destroy_head_check(void *arg, dmu_tx_t *tx)
814 {
815 	dsl_destroy_head_arg_t *ddha = arg;
816 	dsl_pool_t *dp = dmu_tx_pool(tx);
817 	dsl_dataset_t *ds;
818 	int error;
819 
820 	error = dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds);
821 	if (error != 0)
822 		return (error);
823 
824 	error = dsl_destroy_head_check_impl(ds, 0);
825 	dsl_dataset_rele(ds, FTAG);
826 	return (error);
827 }
828 
829 static void
dsl_dir_destroy_sync(uint64_t ddobj,dmu_tx_t * tx)830 dsl_dir_destroy_sync(uint64_t ddobj, dmu_tx_t *tx)
831 {
832 	dsl_dir_t *dd;
833 	dsl_pool_t *dp = dmu_tx_pool(tx);
834 	objset_t *mos = dp->dp_meta_objset;
835 	dd_used_t t;
836 
837 	ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx)->dp_config_rwlock));
838 
839 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd));
840 
841 	ASSERT0(dsl_dir_phys(dd)->dd_head_dataset_obj);
842 
843 	/* Decrement the filesystem count for all parent filesystems. */
844 	if (dd->dd_parent != NULL)
845 		dsl_fs_ss_count_adjust(dd->dd_parent, -1,
846 		    DD_FIELD_FILESYSTEM_COUNT, tx);
847 
848 	/*
849 	 * Remove our reservation. The impl() routine avoids setting the
850 	 * actual property, which would require the (already destroyed) ds.
851 	 */
852 	dsl_dir_set_reservation_sync_impl(dd, 0, tx);
853 
854 	ASSERT0(dsl_dir_phys(dd)->dd_used_bytes);
855 	ASSERT0(dsl_dir_phys(dd)->dd_reserved);
856 	for (t = 0; t < DD_USED_NUM; t++)
857 		ASSERT0(dsl_dir_phys(dd)->dd_used_breakdown[t]);
858 
859 	if (dd->dd_crypto_obj != 0) {
860 		dsl_crypto_key_destroy_sync(dd->dd_crypto_obj, tx);
861 		(void) spa_keystore_unload_wkey_impl(dp->dp_spa, dd->dd_object);
862 	}
863 
864 	VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_child_dir_zapobj, tx));
865 	VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_props_zapobj, tx));
866 	if (dsl_dir_phys(dd)->dd_clones != 0)
867 		VERIFY0(zap_destroy(mos, dsl_dir_phys(dd)->dd_clones, tx));
868 	VERIFY0(dsl_deleg_destroy(mos, dsl_dir_phys(dd)->dd_deleg_zapobj, tx));
869 	VERIFY0(zap_remove(mos,
870 	    dsl_dir_phys(dd->dd_parent)->dd_child_dir_zapobj,
871 	    dd->dd_myname, tx));
872 
873 	dsl_dir_rele(dd, FTAG);
874 	dmu_object_free_zapified(mos, ddobj, tx);
875 }
876 
877 static void
dsl_clone_destroy_assert(dsl_dir_t * dd)878 dsl_clone_destroy_assert(dsl_dir_t *dd)
879 {
880 	uint64_t used, comp, uncomp;
881 
882 	ASSERT(dsl_dir_is_clone(dd));
883 	dsl_deadlist_space(&dd->dd_livelist, &used, &comp, &uncomp);
884 
885 	ASSERT3U(dsl_dir_phys(dd)->dd_used_bytes, ==, used);
886 	ASSERT3U(dsl_dir_phys(dd)->dd_compressed_bytes, ==, comp);
887 	/*
888 	 * Greater than because we do not track embedded block pointers in
889 	 * the livelist
890 	 */
891 	ASSERT3U(dsl_dir_phys(dd)->dd_uncompressed_bytes, >=, uncomp);
892 
893 	ASSERT(list_is_empty(&dd->dd_pending_allocs.bpl_list));
894 	ASSERT(list_is_empty(&dd->dd_pending_frees.bpl_list));
895 }
896 
897 /*
898  * Start the delete process for a clone. Free its zil, verify the space usage
899  * and queue the blkptrs for deletion by adding the livelist to the pool-wide
900  * delete queue.
901  */
902 static void
dsl_async_clone_destroy(dsl_dataset_t * ds,dmu_tx_t * tx)903 dsl_async_clone_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
904 {
905 	uint64_t zap_obj, to_delete, used, comp, uncomp;
906 	objset_t *os;
907 	dsl_dir_t *dd = ds->ds_dir;
908 	dsl_pool_t *dp = dmu_tx_pool(tx);
909 	objset_t *mos = dp->dp_meta_objset;
910 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
911 	VERIFY0(dmu_objset_from_ds(ds, &os));
912 
913 	uint64_t mintxg = 0;
914 	dsl_deadlist_entry_t *dle = dsl_deadlist_first(&dd->dd_livelist);
915 	if (dle != NULL)
916 		mintxg = dle->dle_mintxg;
917 
918 	spa_history_log_internal_ds(ds, "destroy", tx,
919 	    "(livelist, mintxg=%llu)", (long long)mintxg);
920 
921 	/* Check that the clone is in a correct state to be deleted */
922 	dsl_clone_destroy_assert(dd);
923 
924 	/* Destroy the zil */
925 	zil_destroy_sync(dmu_objset_zil(os), tx);
926 
927 	VERIFY0(zap_lookup(mos, dd->dd_object,
928 	    DD_FIELD_LIVELIST, sizeof (uint64_t), 1, &to_delete));
929 	/* Initialize deleted_clones entry to track livelists to cleanup */
930 	int error = zap_lookup(mos, DMU_POOL_DIRECTORY_OBJECT,
931 	    DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1, &zap_obj);
932 	if (error == ENOENT) {
933 		zap_obj = zap_create(mos, DMU_OTN_ZAP_METADATA,
934 		    DMU_OT_NONE, 0, tx);
935 		VERIFY0(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT,
936 		    DMU_POOL_DELETED_CLONES, sizeof (uint64_t), 1,
937 		    &(zap_obj), tx));
938 		spa->spa_livelists_to_delete = zap_obj;
939 	} else if (error != 0) {
940 		zfs_panic_recover("zfs: error %d was returned while looking "
941 		    "up DMU_POOL_DELETED_CLONES in the zap", error);
942 		return;
943 	}
944 	VERIFY0(zap_add_int(mos, zap_obj, to_delete, tx));
945 
946 	/* Clone is no longer using space, now tracked by dp_free_dir */
947 	dsl_deadlist_space(&dd->dd_livelist, &used, &comp, &uncomp);
948 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
949 	    -used, -comp, -dsl_dir_phys(dd)->dd_uncompressed_bytes,
950 	    tx);
951 	dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
952 	    used, comp, uncomp, tx);
953 	dsl_dir_remove_livelist(dd, tx, B_FALSE);
954 	zthr_wakeup(spa->spa_livelist_delete_zthr);
955 }
956 
957 /*
958  * Move the bptree into the pool's list of trees to clean up, update space
959  * accounting information and destroy the zil.
960  */
961 static void
dsl_async_dataset_destroy(dsl_dataset_t * ds,dmu_tx_t * tx)962 dsl_async_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
963 {
964 	uint64_t used, comp, uncomp;
965 	objset_t *os;
966 
967 	VERIFY0(dmu_objset_from_ds(ds, &os));
968 	dsl_pool_t *dp = dmu_tx_pool(tx);
969 	objset_t *mos = dp->dp_meta_objset;
970 
971 	spa_history_log_internal_ds(ds, "destroy", tx,
972 	    "(bptree, mintxg=%llu)",
973 	    (long long)dsl_dataset_phys(ds)->ds_prev_snap_txg);
974 
975 	zil_destroy_sync(dmu_objset_zil(os), tx);
976 
977 	if (!spa_feature_is_active(dp->dp_spa,
978 	    SPA_FEATURE_ASYNC_DESTROY)) {
979 		dsl_scan_t *scn = dp->dp_scan;
980 		spa_feature_incr(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY,
981 		    tx);
982 		dp->dp_bptree_obj = bptree_alloc(mos, tx);
983 		VERIFY0(zap_add(mos,
984 		    DMU_POOL_DIRECTORY_OBJECT,
985 		    DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
986 		    &dp->dp_bptree_obj, tx));
987 		ASSERT(!scn->scn_async_destroying);
988 		scn->scn_async_destroying = B_TRUE;
989 	}
990 
991 	used = dsl_dir_phys(ds->ds_dir)->dd_used_bytes;
992 	comp = dsl_dir_phys(ds->ds_dir)->dd_compressed_bytes;
993 	uncomp = dsl_dir_phys(ds->ds_dir)->dd_uncompressed_bytes;
994 
995 	ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
996 	    dsl_dataset_phys(ds)->ds_unique_bytes == used);
997 
998 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
999 	bptree_add(mos, dp->dp_bptree_obj,
1000 	    &dsl_dataset_phys(ds)->ds_bp,
1001 	    dsl_dataset_phys(ds)->ds_prev_snap_txg,
1002 	    used, comp, uncomp, tx);
1003 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
1004 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
1005 	    -used, -comp, -uncomp, tx);
1006 	dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1007 	    used, comp, uncomp, tx);
1008 }
1009 
1010 void
dsl_destroy_head_sync_impl(dsl_dataset_t * ds,dmu_tx_t * tx)1011 dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
1012 {
1013 	dsl_pool_t *dp = dmu_tx_pool(tx);
1014 	objset_t *mos = dp->dp_meta_objset;
1015 	uint64_t obj, ddobj, prevobj = 0;
1016 	boolean_t rmorigin;
1017 
1018 	ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
1019 	ASSERT(ds->ds_prev == NULL ||
1020 	    dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj != ds->ds_object);
1021 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1022 	ASSERT3U(BP_GET_LOGICAL_BIRTH(&dsl_dataset_phys(ds)->ds_bp), <=,
1023 	    tx->tx_txg);
1024 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
1025 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1026 
1027 	dsl_dir_cancel_waiters(ds->ds_dir);
1028 
1029 	rmorigin = (dsl_dir_is_clone(ds->ds_dir) &&
1030 	    DS_IS_DEFER_DESTROY(ds->ds_prev) &&
1031 	    dsl_dataset_phys(ds->ds_prev)->ds_num_children == 2 &&
1032 	    ds->ds_prev->ds_userrefs == 0);
1033 
1034 	/* Remove our reservation. */
1035 	if (ds->ds_reserved != 0) {
1036 		dsl_dataset_set_refreservation_sync_impl(ds,
1037 		    (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1038 		    0, tx);
1039 		ASSERT0(ds->ds_reserved);
1040 	}
1041 
1042 	obj = ds->ds_object;
1043 
1044 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1045 		if (dsl_dataset_feature_is_active(ds, f))
1046 			dsl_dataset_deactivate_feature(ds, f, tx);
1047 	}
1048 
1049 	dsl_scan_ds_destroyed(ds, tx);
1050 
1051 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1052 		/* This is a clone */
1053 		ASSERT(ds->ds_prev != NULL);
1054 		ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj, !=,
1055 		    obj);
1056 		ASSERT0(dsl_dataset_phys(ds)->ds_next_snap_obj);
1057 
1058 		dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1059 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj != 0) {
1060 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1061 			    obj, tx);
1062 		}
1063 
1064 		ASSERT3U(dsl_dataset_phys(ds->ds_prev)->ds_num_children, >, 1);
1065 		dsl_dataset_phys(ds->ds_prev)->ds_num_children--;
1066 	}
1067 
1068 	/*
1069 	 * Destroy the deadlist. Unless it's a clone, the
1070 	 * deadlist should be empty since the dataset has no snapshots.
1071 	 * (If it's a clone, it's safe to ignore the deadlist contents
1072 	 * since they are still referenced by the origin snapshot.)
1073 	 */
1074 	dsl_deadlist_close(&ds->ds_deadlist);
1075 	dsl_deadlist_free(mos, dsl_dataset_phys(ds)->ds_deadlist_obj, tx);
1076 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1077 	dsl_dataset_phys(ds)->ds_deadlist_obj = 0;
1078 
1079 	if (dsl_dataset_remap_deadlist_exists(ds))
1080 		dsl_dataset_destroy_remap_deadlist(ds, tx);
1081 
1082 	/*
1083 	 * Each destroy is responsible for both destroying (enqueuing
1084 	 * to be destroyed) the blkptrs comprising the dataset as well as
1085 	 * those belonging to the zil.
1086 	 */
1087 	if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist)) {
1088 		dsl_async_clone_destroy(ds, tx);
1089 	} else if (spa_feature_is_enabled(dp->dp_spa,
1090 	    SPA_FEATURE_ASYNC_DESTROY)) {
1091 		dsl_async_dataset_destroy(ds, tx);
1092 	} else {
1093 		old_synchronous_dataset_destroy(ds, tx);
1094 	}
1095 
1096 	if (ds->ds_prev != NULL) {
1097 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1098 			VERIFY0(zap_remove_int(mos,
1099 			    dsl_dir_phys(ds->ds_prev->ds_dir)->dd_clones,
1100 			    ds->ds_object, tx));
1101 		}
1102 		prevobj = ds->ds_prev->ds_object;
1103 		dsl_dataset_rele(ds->ds_prev, ds);
1104 		ds->ds_prev = NULL;
1105 	}
1106 
1107 	/*
1108 	 * This must be done after the dsl_traverse(), because it will
1109 	 * re-open the objset.
1110 	 */
1111 	if (ds->ds_objset) {
1112 		dmu_objset_evict(ds->ds_objset);
1113 		ds->ds_objset = NULL;
1114 	}
1115 
1116 	/* Erase the link in the dir */
1117 	dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1118 	dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj = 0;
1119 	ddobj = ds->ds_dir->dd_object;
1120 	ASSERT(dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0);
1121 	VERIFY0(zap_destroy(mos,
1122 	    dsl_dataset_phys(ds)->ds_snapnames_zapobj, tx));
1123 
1124 	if (ds->ds_bookmarks_obj != 0) {
1125 		void *cookie = NULL;
1126 		dsl_bookmark_node_t *dbn;
1127 
1128 		while ((dbn = avl_destroy_nodes(&ds->ds_bookmarks, &cookie)) !=
1129 		    NULL) {
1130 			if (dbn->dbn_phys.zbm_redaction_obj != 0) {
1131 				dnode_t *rl;
1132 				VERIFY0(dnode_hold(mos,
1133 				    dbn->dbn_phys.zbm_redaction_obj, FTAG,
1134 				    &rl));
1135 				if (rl->dn_have_spill) {
1136 					spa_feature_decr(dmu_objset_spa(mos),
1137 					    SPA_FEATURE_REDACTION_LIST_SPILL,
1138 					    tx);
1139 				}
1140 				dnode_rele(rl, FTAG);
1141 				VERIFY0(dmu_object_free(mos,
1142 				    dbn->dbn_phys.zbm_redaction_obj, tx));
1143 				spa_feature_decr(dmu_objset_spa(mos),
1144 				    SPA_FEATURE_REDACTION_BOOKMARKS, tx);
1145 			}
1146 			if (dbn->dbn_phys.zbm_flags & ZBM_FLAG_HAS_FBN) {
1147 				spa_feature_decr(dmu_objset_spa(mos),
1148 				    SPA_FEATURE_BOOKMARK_WRITTEN, tx);
1149 			}
1150 			spa_strfree(dbn->dbn_name);
1151 			mutex_destroy(&dbn->dbn_lock);
1152 			kmem_free(dbn, sizeof (*dbn));
1153 		}
1154 		avl_destroy(&ds->ds_bookmarks);
1155 		VERIFY0(zap_destroy(mos, ds->ds_bookmarks_obj, tx));
1156 		spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
1157 	}
1158 
1159 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1160 
1161 	ASSERT0(dsl_dataset_phys(ds)->ds_next_clones_obj);
1162 	ASSERT0(dsl_dataset_phys(ds)->ds_props_obj);
1163 	ASSERT0(dsl_dataset_phys(ds)->ds_userrefs_obj);
1164 	dsl_dir_rele(ds->ds_dir, ds);
1165 	ds->ds_dir = NULL;
1166 	dmu_object_free_zapified(mos, obj, tx);
1167 
1168 	dsl_dir_destroy_sync(ddobj, tx);
1169 
1170 	if (rmorigin) {
1171 		dsl_dataset_t *prev;
1172 		VERIFY0(dsl_dataset_hold_obj(dp, prevobj, FTAG, &prev));
1173 		dsl_destroy_snapshot_sync_impl(prev, B_FALSE, tx);
1174 		dsl_dataset_rele(prev, FTAG);
1175 	}
1176 	/* Delete errlog. */
1177 	if (spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_HEAD_ERRLOG))
1178 		spa_delete_dataset_errlog(dp->dp_spa, ds->ds_object, tx);
1179 }
1180 
1181 void
dsl_destroy_head_sync(void * arg,dmu_tx_t * tx)1182 dsl_destroy_head_sync(void *arg, dmu_tx_t *tx)
1183 {
1184 	dsl_destroy_head_arg_t *ddha = arg;
1185 	dsl_pool_t *dp = dmu_tx_pool(tx);
1186 	dsl_dataset_t *ds;
1187 
1188 	VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
1189 	dsl_destroy_head_sync_impl(ds, tx);
1190 	zvol_remove_minors(dp->dp_spa, ddha->ddha_name, B_TRUE);
1191 	dsl_dataset_rele(ds, FTAG);
1192 }
1193 
1194 static void
dsl_destroy_head_begin_sync(void * arg,dmu_tx_t * tx)1195 dsl_destroy_head_begin_sync(void *arg, dmu_tx_t *tx)
1196 {
1197 	dsl_destroy_head_arg_t *ddha = arg;
1198 	dsl_pool_t *dp = dmu_tx_pool(tx);
1199 	dsl_dataset_t *ds;
1200 
1201 	VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
1202 
1203 	/* Mark it as inconsistent on-disk, in case we crash */
1204 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1205 	dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_INCONSISTENT;
1206 
1207 	spa_history_log_internal_ds(ds, "destroy begin", tx, " ");
1208 	dsl_dataset_rele(ds, FTAG);
1209 }
1210 
1211 int
dsl_destroy_head(const char * name)1212 dsl_destroy_head(const char *name)
1213 {
1214 	dsl_destroy_head_arg_t ddha;
1215 	int error;
1216 	spa_t *spa;
1217 	boolean_t isenabled;
1218 
1219 #ifdef _KERNEL
1220 	zfs_destroy_unmount_origin(name);
1221 #endif
1222 
1223 	error = spa_open(name, &spa, FTAG);
1224 	if (error != 0)
1225 		return (error);
1226 	isenabled = spa_feature_is_enabled(spa, SPA_FEATURE_ASYNC_DESTROY);
1227 	spa_close(spa, FTAG);
1228 
1229 	ddha.ddha_name = name;
1230 
1231 	if (!isenabled) {
1232 		objset_t *os;
1233 
1234 		error = dsl_sync_task(name, dsl_destroy_head_check,
1235 		    dsl_destroy_head_begin_sync, &ddha,
1236 		    0, ZFS_SPACE_CHECK_DESTROY);
1237 		if (error != 0)
1238 			return (error);
1239 
1240 		/*
1241 		 * Head deletion is processed in one txg on old pools;
1242 		 * remove the objects from open context so that the txg sync
1243 		 * is not too long. This optimization can only work for
1244 		 * encrypted datasets if the wrapping key is loaded.
1245 		 */
1246 		error = dmu_objset_own(name, DMU_OST_ANY, B_FALSE, B_TRUE,
1247 		    FTAG, &os);
1248 		if (error == 0) {
1249 			uint64_t prev_snap_txg =
1250 			    dsl_dataset_phys(dmu_objset_ds(os))->
1251 			    ds_prev_snap_txg;
1252 			for (uint64_t obj = 0; error == 0;
1253 			    error = dmu_object_next(os, &obj, FALSE,
1254 			    prev_snap_txg))
1255 				(void) dmu_free_long_object(os, obj);
1256 			/* sync out all frees */
1257 			txg_wait_synced(dmu_objset_pool(os), 0);
1258 			dmu_objset_disown(os, B_TRUE, FTAG);
1259 		}
1260 	}
1261 
1262 	return (dsl_sync_task(name, dsl_destroy_head_check,
1263 	    dsl_destroy_head_sync, &ddha, 0, ZFS_SPACE_CHECK_DESTROY));
1264 }
1265 
1266 /*
1267  * Note, this function is used as the callback for dmu_objset_find().  We
1268  * always return 0 so that we will continue to find and process
1269  * inconsistent datasets, even if we encounter an error trying to
1270  * process one of them.
1271  */
1272 int
dsl_destroy_inconsistent(const char * dsname,void * arg)1273 dsl_destroy_inconsistent(const char *dsname, void *arg)
1274 {
1275 	(void) arg;
1276 	objset_t *os;
1277 
1278 	if (dmu_objset_hold(dsname, FTAG, &os) == 0) {
1279 		boolean_t need_destroy = DS_IS_INCONSISTENT(dmu_objset_ds(os));
1280 
1281 		/*
1282 		 * If the dataset is inconsistent because a resumable receive
1283 		 * has failed, then do not destroy it.
1284 		 */
1285 		if (dsl_dataset_has_resume_receive_state(dmu_objset_ds(os)))
1286 			need_destroy = B_FALSE;
1287 
1288 		dmu_objset_rele(os, FTAG);
1289 		if (need_destroy)
1290 			(void) dsl_destroy_head(dsname);
1291 	}
1292 	return (0);
1293 }
1294 
1295 
1296 #if defined(_KERNEL)
1297 EXPORT_SYMBOL(dsl_destroy_head);
1298 EXPORT_SYMBOL(dsl_destroy_head_sync_impl);
1299 EXPORT_SYMBOL(dsl_dataset_user_hold_check_one);
1300 EXPORT_SYMBOL(dsl_destroy_snapshot_sync_impl);
1301 EXPORT_SYMBOL(dsl_destroy_inconsistent);
1302 EXPORT_SYMBOL(dsl_dataset_user_release_tmp);
1303 EXPORT_SYMBOL(dsl_destroy_head_check_impl);
1304 #endif
1305