xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision b509e89b2befbaa42939abad9da1d7f5a8c6aaae)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/dmu_objset.h>
27 #include <sys/dsl_dataset.h>
28 #include <sys/dsl_dir.h>
29 #include <sys/dsl_prop.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dmu_traverse.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/arc.h>
34 #include <sys/zio.h>
35 #include <sys/zap.h>
36 #include <sys/unique.h>
37 #include <sys/zfs_context.h>
38 #include <sys/zfs_ioctl.h>
39 #include <sys/spa.h>
40 #include <sys/zfs_znode.h>
41 #include <sys/sunddi.h>
42 
43 static char *dsl_reaper = "the grim reaper";
44 
45 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
46 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
47 static dsl_checkfunc_t dsl_dataset_rollback_check;
48 static dsl_syncfunc_t dsl_dataset_rollback_sync;
49 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
50 
51 #define	DS_REF_MAX	(1ULL << 62)
52 
53 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
54 
55 #define	DSL_DATASET_IS_DESTROYED(ds)	((ds)->ds_owner == dsl_reaper)
56 
57 
58 /*
59  * Figure out how much of this delta should be propogated to the dsl_dir
60  * layer.  If there's a refreservation, that space has already been
61  * partially accounted for in our ancestors.
62  */
63 static int64_t
64 parent_delta(dsl_dataset_t *ds, int64_t delta)
65 {
66 	uint64_t old_bytes, new_bytes;
67 
68 	if (ds->ds_reserved == 0)
69 		return (delta);
70 
71 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
72 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
73 
74 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
75 	return (new_bytes - old_bytes);
76 }
77 
78 void
79 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
80 {
81 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
82 	int compressed = BP_GET_PSIZE(bp);
83 	int uncompressed = BP_GET_UCSIZE(bp);
84 	int64_t delta;
85 
86 	dprintf_bp(bp, "born, ds=%p\n", ds);
87 
88 	ASSERT(dmu_tx_is_syncing(tx));
89 	/* It could have been compressed away to nothing */
90 	if (BP_IS_HOLE(bp))
91 		return;
92 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
93 	ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
94 	if (ds == NULL) {
95 		/*
96 		 * Account for the meta-objset space in its placeholder
97 		 * dsl_dir.
98 		 */
99 		ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
100 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
101 		    used, compressed, uncompressed, tx);
102 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
103 		return;
104 	}
105 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
106 	mutex_enter(&ds->ds_dir->dd_lock);
107 	mutex_enter(&ds->ds_lock);
108 	delta = parent_delta(ds, used);
109 	ds->ds_phys->ds_used_bytes += used;
110 	ds->ds_phys->ds_compressed_bytes += compressed;
111 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
112 	ds->ds_phys->ds_unique_bytes += used;
113 	mutex_exit(&ds->ds_lock);
114 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
115 	    compressed, uncompressed, tx);
116 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
117 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
118 	mutex_exit(&ds->ds_dir->dd_lock);
119 }
120 
121 int
122 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
123     dmu_tx_t *tx)
124 {
125 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
126 	int compressed = BP_GET_PSIZE(bp);
127 	int uncompressed = BP_GET_UCSIZE(bp);
128 
129 	ASSERT(pio != NULL);
130 	ASSERT(dmu_tx_is_syncing(tx));
131 	/* No block pointer => nothing to free */
132 	if (BP_IS_HOLE(bp))
133 		return (0);
134 
135 	ASSERT(used > 0);
136 	if (ds == NULL) {
137 		int err;
138 		/*
139 		 * Account for the meta-objset space in its placeholder
140 		 * dataset.
141 		 */
142 		err = dsl_free(pio, tx->tx_pool,
143 		    tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT);
144 		ASSERT(err == 0);
145 
146 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
147 		    -used, -compressed, -uncompressed, tx);
148 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
149 		return (used);
150 	}
151 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
152 
153 	ASSERT(!dsl_dataset_is_snapshot(ds));
154 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
155 
156 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
157 		int err;
158 		int64_t delta;
159 
160 		dprintf_bp(bp, "freeing: %s", "");
161 		err = dsl_free(pio, tx->tx_pool,
162 		    tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT);
163 		ASSERT(err == 0);
164 
165 		mutex_enter(&ds->ds_dir->dd_lock);
166 		mutex_enter(&ds->ds_lock);
167 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
168 		    !DS_UNIQUE_IS_ACCURATE(ds));
169 		delta = parent_delta(ds, -used);
170 		ds->ds_phys->ds_unique_bytes -= used;
171 		mutex_exit(&ds->ds_lock);
172 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
173 		    delta, -compressed, -uncompressed, tx);
174 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
175 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
176 		mutex_exit(&ds->ds_dir->dd_lock);
177 	} else {
178 		dprintf_bp(bp, "putting on dead list: %s", "");
179 		VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
180 		ASSERT3U(ds->ds_prev->ds_object, ==,
181 		    ds->ds_phys->ds_prev_snap_obj);
182 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
183 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
184 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
185 		    ds->ds_object && bp->blk_birth >
186 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
187 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
188 			mutex_enter(&ds->ds_prev->ds_lock);
189 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
190 			mutex_exit(&ds->ds_prev->ds_lock);
191 		}
192 		if (bp->blk_birth > ds->ds_origin_txg) {
193 			dsl_dir_transfer_space(ds->ds_dir, used,
194 			    DD_USED_HEAD, DD_USED_SNAP, tx);
195 		}
196 	}
197 	mutex_enter(&ds->ds_lock);
198 	ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
199 	ds->ds_phys->ds_used_bytes -= used;
200 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
201 	ds->ds_phys->ds_compressed_bytes -= compressed;
202 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
203 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
204 	mutex_exit(&ds->ds_lock);
205 
206 	return (used);
207 }
208 
209 uint64_t
210 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
211 {
212 	uint64_t trysnap = 0;
213 
214 	if (ds == NULL)
215 		return (0);
216 	/*
217 	 * The snapshot creation could fail, but that would cause an
218 	 * incorrect FALSE return, which would only result in an
219 	 * overestimation of the amount of space that an operation would
220 	 * consume, which is OK.
221 	 *
222 	 * There's also a small window where we could miss a pending
223 	 * snapshot, because we could set the sync task in the quiescing
224 	 * phase.  So this should only be used as a guess.
225 	 */
226 	if (ds->ds_trysnap_txg >
227 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
228 		trysnap = ds->ds_trysnap_txg;
229 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
230 }
231 
232 boolean_t
233 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
234 {
235 	return (blk_birth > dsl_dataset_prev_snap_txg(ds));
236 }
237 
238 /* ARGSUSED */
239 static void
240 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
241 {
242 	dsl_dataset_t *ds = dsv;
243 
244 	ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
245 
246 	dprintf_ds(ds, "evicting %s\n", "");
247 
248 	unique_remove(ds->ds_fsid_guid);
249 
250 	if (ds->ds_user_ptr != NULL)
251 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
252 
253 	if (ds->ds_prev) {
254 		dsl_dataset_drop_ref(ds->ds_prev, ds);
255 		ds->ds_prev = NULL;
256 	}
257 
258 	bplist_close(&ds->ds_deadlist);
259 	if (ds->ds_dir)
260 		dsl_dir_close(ds->ds_dir, ds);
261 
262 	ASSERT(!list_link_active(&ds->ds_synced_link));
263 
264 	mutex_destroy(&ds->ds_lock);
265 	mutex_destroy(&ds->ds_opening_lock);
266 	mutex_destroy(&ds->ds_deadlist.bpl_lock);
267 	rw_destroy(&ds->ds_rwlock);
268 	cv_destroy(&ds->ds_exclusive_cv);
269 
270 	kmem_free(ds, sizeof (dsl_dataset_t));
271 }
272 
273 static int
274 dsl_dataset_get_snapname(dsl_dataset_t *ds)
275 {
276 	dsl_dataset_phys_t *headphys;
277 	int err;
278 	dmu_buf_t *headdbuf;
279 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
280 	objset_t *mos = dp->dp_meta_objset;
281 
282 	if (ds->ds_snapname[0])
283 		return (0);
284 	if (ds->ds_phys->ds_next_snap_obj == 0)
285 		return (0);
286 
287 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
288 	    FTAG, &headdbuf);
289 	if (err)
290 		return (err);
291 	headphys = headdbuf->db_data;
292 	err = zap_value_search(dp->dp_meta_objset,
293 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
294 	dmu_buf_rele(headdbuf, FTAG);
295 	return (err);
296 }
297 
298 static int
299 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
300 {
301 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
302 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
303 	matchtype_t mt;
304 	int err;
305 
306 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
307 		mt = MT_FIRST;
308 	else
309 		mt = MT_EXACT;
310 
311 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
312 	    value, mt, NULL, 0, NULL);
313 	if (err == ENOTSUP && mt == MT_FIRST)
314 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
315 	return (err);
316 }
317 
318 static int
319 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
320 {
321 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
322 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
323 	matchtype_t mt;
324 	int err;
325 
326 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
327 		mt = MT_FIRST;
328 	else
329 		mt = MT_EXACT;
330 
331 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
332 	if (err == ENOTSUP && mt == MT_FIRST)
333 		err = zap_remove(mos, snapobj, name, tx);
334 	return (err);
335 }
336 
337 static int
338 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
339     dsl_dataset_t **dsp)
340 {
341 	objset_t *mos = dp->dp_meta_objset;
342 	dmu_buf_t *dbuf;
343 	dsl_dataset_t *ds;
344 	int err;
345 
346 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
347 	    dsl_pool_sync_context(dp));
348 
349 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
350 	if (err)
351 		return (err);
352 	ds = dmu_buf_get_user(dbuf);
353 	if (ds == NULL) {
354 		dsl_dataset_t *winner;
355 
356 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
357 		ds->ds_dbuf = dbuf;
358 		ds->ds_object = dsobj;
359 		ds->ds_phys = dbuf->db_data;
360 
361 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
362 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
363 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
364 		    NULL);
365 		rw_init(&ds->ds_rwlock, 0, 0, 0);
366 		cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
367 
368 		err = bplist_open(&ds->ds_deadlist,
369 		    mos, ds->ds_phys->ds_deadlist_obj);
370 		if (err == 0) {
371 			err = dsl_dir_open_obj(dp,
372 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
373 		}
374 		if (err) {
375 			/*
376 			 * we don't really need to close the blist if we
377 			 * just opened it.
378 			 */
379 			mutex_destroy(&ds->ds_lock);
380 			mutex_destroy(&ds->ds_opening_lock);
381 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
382 			rw_destroy(&ds->ds_rwlock);
383 			cv_destroy(&ds->ds_exclusive_cv);
384 			kmem_free(ds, sizeof (dsl_dataset_t));
385 			dmu_buf_rele(dbuf, tag);
386 			return (err);
387 		}
388 
389 		if (!dsl_dataset_is_snapshot(ds)) {
390 			ds->ds_snapname[0] = '\0';
391 			if (ds->ds_phys->ds_prev_snap_obj) {
392 				err = dsl_dataset_get_ref(dp,
393 				    ds->ds_phys->ds_prev_snap_obj,
394 				    ds, &ds->ds_prev);
395 			}
396 
397 			if (err == 0 && dsl_dir_is_clone(ds->ds_dir)) {
398 				dsl_dataset_t *origin;
399 
400 				err = dsl_dataset_hold_obj(dp,
401 				    ds->ds_dir->dd_phys->dd_origin_obj,
402 				    FTAG, &origin);
403 				if (err == 0) {
404 					ds->ds_origin_txg =
405 					    origin->ds_phys->ds_creation_txg;
406 					dsl_dataset_rele(origin, FTAG);
407 				}
408 			}
409 		} else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
410 			err = dsl_dataset_get_snapname(ds);
411 		}
412 
413 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
414 			/*
415 			 * In sync context, we're called with either no lock
416 			 * or with the write lock.  If we're not syncing,
417 			 * we're always called with the read lock held.
418 			 */
419 			boolean_t need_lock =
420 			    !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
421 			    dsl_pool_sync_context(dp);
422 
423 			if (need_lock)
424 				rw_enter(&dp->dp_config_rwlock, RW_READER);
425 
426 			err = dsl_prop_get_ds(ds,
427 			    "refreservation", sizeof (uint64_t), 1,
428 			    &ds->ds_reserved, NULL);
429 			if (err == 0) {
430 				err = dsl_prop_get_ds(ds,
431 				    "refquota", sizeof (uint64_t), 1,
432 				    &ds->ds_quota, NULL);
433 			}
434 
435 			if (need_lock)
436 				rw_exit(&dp->dp_config_rwlock);
437 		} else {
438 			ds->ds_reserved = ds->ds_quota = 0;
439 		}
440 
441 		if (err == 0) {
442 			winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
443 			    dsl_dataset_evict);
444 		}
445 		if (err || winner) {
446 			bplist_close(&ds->ds_deadlist);
447 			if (ds->ds_prev)
448 				dsl_dataset_drop_ref(ds->ds_prev, ds);
449 			dsl_dir_close(ds->ds_dir, ds);
450 			mutex_destroy(&ds->ds_lock);
451 			mutex_destroy(&ds->ds_opening_lock);
452 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
453 			rw_destroy(&ds->ds_rwlock);
454 			cv_destroy(&ds->ds_exclusive_cv);
455 			kmem_free(ds, sizeof (dsl_dataset_t));
456 			if (err) {
457 				dmu_buf_rele(dbuf, tag);
458 				return (err);
459 			}
460 			ds = winner;
461 		} else {
462 			ds->ds_fsid_guid =
463 			    unique_insert(ds->ds_phys->ds_fsid_guid);
464 		}
465 	}
466 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
467 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
468 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
469 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
470 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
471 	mutex_enter(&ds->ds_lock);
472 	if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
473 		mutex_exit(&ds->ds_lock);
474 		dmu_buf_rele(ds->ds_dbuf, tag);
475 		return (ENOENT);
476 	}
477 	mutex_exit(&ds->ds_lock);
478 	*dsp = ds;
479 	return (0);
480 }
481 
482 static int
483 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
484 {
485 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
486 
487 	/*
488 	 * In syncing context we don't want the rwlock lock: there
489 	 * may be an existing writer waiting for sync phase to
490 	 * finish.  We don't need to worry about such writers, since
491 	 * sync phase is single-threaded, so the writer can't be
492 	 * doing anything while we are active.
493 	 */
494 	if (dsl_pool_sync_context(dp)) {
495 		ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
496 		return (0);
497 	}
498 
499 	/*
500 	 * Normal users will hold the ds_rwlock as a READER until they
501 	 * are finished (i.e., call dsl_dataset_rele()).  "Owners" will
502 	 * drop their READER lock after they set the ds_owner field.
503 	 *
504 	 * If the dataset is being destroyed, the destroy thread will
505 	 * obtain a WRITER lock for exclusive access after it's done its
506 	 * open-context work and then change the ds_owner to
507 	 * dsl_reaper once destruction is assured.  So threads
508 	 * may block here temporarily, until the "destructability" of
509 	 * the dataset is determined.
510 	 */
511 	ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
512 	mutex_enter(&ds->ds_lock);
513 	while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
514 		rw_exit(&dp->dp_config_rwlock);
515 		cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
516 		if (DSL_DATASET_IS_DESTROYED(ds)) {
517 			mutex_exit(&ds->ds_lock);
518 			dsl_dataset_drop_ref(ds, tag);
519 			rw_enter(&dp->dp_config_rwlock, RW_READER);
520 			return (ENOENT);
521 		}
522 		rw_enter(&dp->dp_config_rwlock, RW_READER);
523 	}
524 	mutex_exit(&ds->ds_lock);
525 	return (0);
526 }
527 
528 int
529 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
530     dsl_dataset_t **dsp)
531 {
532 	int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
533 
534 	if (err)
535 		return (err);
536 	return (dsl_dataset_hold_ref(*dsp, tag));
537 }
538 
539 int
540 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, int flags, void *owner,
541     dsl_dataset_t **dsp)
542 {
543 	int err = dsl_dataset_hold_obj(dp, dsobj, owner, dsp);
544 
545 	ASSERT(DS_MODE_TYPE(flags) != DS_MODE_USER);
546 
547 	if (err)
548 		return (err);
549 	if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
550 		dsl_dataset_rele(*dsp, owner);
551 		*dsp = NULL;
552 		return (EBUSY);
553 	}
554 	return (0);
555 }
556 
557 int
558 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
559 {
560 	dsl_dir_t *dd;
561 	dsl_pool_t *dp;
562 	const char *snapname;
563 	uint64_t obj;
564 	int err = 0;
565 
566 	err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
567 	if (err)
568 		return (err);
569 
570 	dp = dd->dd_pool;
571 	obj = dd->dd_phys->dd_head_dataset_obj;
572 	rw_enter(&dp->dp_config_rwlock, RW_READER);
573 	if (obj)
574 		err = dsl_dataset_get_ref(dp, obj, tag, dsp);
575 	else
576 		err = ENOENT;
577 	if (err)
578 		goto out;
579 
580 	err = dsl_dataset_hold_ref(*dsp, tag);
581 
582 	/* we may be looking for a snapshot */
583 	if (err == 0 && snapname != NULL) {
584 		dsl_dataset_t *ds = NULL;
585 
586 		if (*snapname++ != '@') {
587 			dsl_dataset_rele(*dsp, tag);
588 			err = ENOENT;
589 			goto out;
590 		}
591 
592 		dprintf("looking for snapshot '%s'\n", snapname);
593 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
594 		if (err == 0)
595 			err = dsl_dataset_get_ref(dp, obj, tag, &ds);
596 		dsl_dataset_rele(*dsp, tag);
597 
598 		ASSERT3U((err == 0), ==, (ds != NULL));
599 
600 		if (ds) {
601 			mutex_enter(&ds->ds_lock);
602 			if (ds->ds_snapname[0] == 0)
603 				(void) strlcpy(ds->ds_snapname, snapname,
604 				    sizeof (ds->ds_snapname));
605 			mutex_exit(&ds->ds_lock);
606 			err = dsl_dataset_hold_ref(ds, tag);
607 			*dsp = err ? NULL : ds;
608 		}
609 	}
610 out:
611 	rw_exit(&dp->dp_config_rwlock);
612 	dsl_dir_close(dd, FTAG);
613 	return (err);
614 }
615 
616 int
617 dsl_dataset_own(const char *name, int flags, void *owner, dsl_dataset_t **dsp)
618 {
619 	int err = dsl_dataset_hold(name, owner, dsp);
620 	if (err)
621 		return (err);
622 	if ((*dsp)->ds_phys->ds_num_children > 0 &&
623 	    !DS_MODE_IS_READONLY(flags)) {
624 		dsl_dataset_rele(*dsp, owner);
625 		return (EROFS);
626 	}
627 	if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
628 		dsl_dataset_rele(*dsp, owner);
629 		return (EBUSY);
630 	}
631 	return (0);
632 }
633 
634 void
635 dsl_dataset_name(dsl_dataset_t *ds, char *name)
636 {
637 	if (ds == NULL) {
638 		(void) strcpy(name, "mos");
639 	} else {
640 		dsl_dir_name(ds->ds_dir, name);
641 		VERIFY(0 == dsl_dataset_get_snapname(ds));
642 		if (ds->ds_snapname[0]) {
643 			(void) strcat(name, "@");
644 			/*
645 			 * We use a "recursive" mutex so that we
646 			 * can call dprintf_ds() with ds_lock held.
647 			 */
648 			if (!MUTEX_HELD(&ds->ds_lock)) {
649 				mutex_enter(&ds->ds_lock);
650 				(void) strcat(name, ds->ds_snapname);
651 				mutex_exit(&ds->ds_lock);
652 			} else {
653 				(void) strcat(name, ds->ds_snapname);
654 			}
655 		}
656 	}
657 }
658 
659 static int
660 dsl_dataset_namelen(dsl_dataset_t *ds)
661 {
662 	int result;
663 
664 	if (ds == NULL) {
665 		result = 3;	/* "mos" */
666 	} else {
667 		result = dsl_dir_namelen(ds->ds_dir);
668 		VERIFY(0 == dsl_dataset_get_snapname(ds));
669 		if (ds->ds_snapname[0]) {
670 			++result;	/* adding one for the @-sign */
671 			if (!MUTEX_HELD(&ds->ds_lock)) {
672 				mutex_enter(&ds->ds_lock);
673 				result += strlen(ds->ds_snapname);
674 				mutex_exit(&ds->ds_lock);
675 			} else {
676 				result += strlen(ds->ds_snapname);
677 			}
678 		}
679 	}
680 
681 	return (result);
682 }
683 
684 void
685 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
686 {
687 	dmu_buf_rele(ds->ds_dbuf, tag);
688 }
689 
690 void
691 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
692 {
693 	if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
694 		rw_exit(&ds->ds_rwlock);
695 	}
696 	dsl_dataset_drop_ref(ds, tag);
697 }
698 
699 void
700 dsl_dataset_disown(dsl_dataset_t *ds, void *owner)
701 {
702 	ASSERT((ds->ds_owner == owner && ds->ds_dbuf) ||
703 	    (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
704 
705 	mutex_enter(&ds->ds_lock);
706 	ds->ds_owner = NULL;
707 	if (RW_WRITE_HELD(&ds->ds_rwlock)) {
708 		rw_exit(&ds->ds_rwlock);
709 		cv_broadcast(&ds->ds_exclusive_cv);
710 	}
711 	mutex_exit(&ds->ds_lock);
712 	if (ds->ds_dbuf)
713 		dsl_dataset_drop_ref(ds, owner);
714 	else
715 		dsl_dataset_evict(ds->ds_dbuf, ds);
716 }
717 
718 boolean_t
719 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *owner)
720 {
721 	boolean_t gotit = FALSE;
722 
723 	mutex_enter(&ds->ds_lock);
724 	if (ds->ds_owner == NULL &&
725 	    (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
726 		ds->ds_owner = owner;
727 		if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
728 			rw_exit(&ds->ds_rwlock);
729 		gotit = TRUE;
730 	}
731 	mutex_exit(&ds->ds_lock);
732 	return (gotit);
733 }
734 
735 void
736 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
737 {
738 	ASSERT3P(owner, ==, ds->ds_owner);
739 	if (!RW_WRITE_HELD(&ds->ds_rwlock))
740 		rw_enter(&ds->ds_rwlock, RW_WRITER);
741 }
742 
743 uint64_t
744 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
745     uint64_t flags, dmu_tx_t *tx)
746 {
747 	dsl_pool_t *dp = dd->dd_pool;
748 	dmu_buf_t *dbuf;
749 	dsl_dataset_phys_t *dsphys;
750 	uint64_t dsobj;
751 	objset_t *mos = dp->dp_meta_objset;
752 
753 	if (origin == NULL)
754 		origin = dp->dp_origin_snap;
755 
756 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
757 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
758 	ASSERT(dmu_tx_is_syncing(tx));
759 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
760 
761 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
762 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
763 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
764 	dmu_buf_will_dirty(dbuf, tx);
765 	dsphys = dbuf->db_data;
766 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
767 	dsphys->ds_dir_obj = dd->dd_object;
768 	dsphys->ds_flags = flags;
769 	dsphys->ds_fsid_guid = unique_create();
770 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
771 	    sizeof (dsphys->ds_guid));
772 	dsphys->ds_snapnames_zapobj =
773 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
774 	    DMU_OT_NONE, 0, tx);
775 	dsphys->ds_creation_time = gethrestime_sec();
776 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
777 	dsphys->ds_deadlist_obj =
778 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
779 
780 	if (origin) {
781 		dsphys->ds_prev_snap_obj = origin->ds_object;
782 		dsphys->ds_prev_snap_txg =
783 		    origin->ds_phys->ds_creation_txg;
784 		dsphys->ds_used_bytes =
785 		    origin->ds_phys->ds_used_bytes;
786 		dsphys->ds_compressed_bytes =
787 		    origin->ds_phys->ds_compressed_bytes;
788 		dsphys->ds_uncompressed_bytes =
789 		    origin->ds_phys->ds_uncompressed_bytes;
790 		dsphys->ds_bp = origin->ds_phys->ds_bp;
791 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
792 
793 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
794 		origin->ds_phys->ds_num_children++;
795 
796 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
797 			if (origin->ds_phys->ds_next_clones_obj == 0) {
798 				origin->ds_phys->ds_next_clones_obj =
799 				    zap_create(mos,
800 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
801 			}
802 			VERIFY(0 == zap_add_int(mos,
803 			    origin->ds_phys->ds_next_clones_obj,
804 			    dsobj, tx));
805 		}
806 
807 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
808 		dd->dd_phys->dd_origin_obj = origin->ds_object;
809 	}
810 
811 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
812 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
813 
814 	dmu_buf_rele(dbuf, FTAG);
815 
816 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
817 	dd->dd_phys->dd_head_dataset_obj = dsobj;
818 
819 	return (dsobj);
820 }
821 
822 uint64_t
823 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
824     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
825 {
826 	dsl_pool_t *dp = pdd->dd_pool;
827 	uint64_t dsobj, ddobj;
828 	dsl_dir_t *dd;
829 
830 	ASSERT(lastname[0] != '@');
831 
832 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
833 	VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
834 
835 	dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
836 
837 	dsl_deleg_set_create_perms(dd, tx, cr);
838 
839 	dsl_dir_close(dd, FTAG);
840 
841 	return (dsobj);
842 }
843 
844 struct destroyarg {
845 	dsl_sync_task_group_t *dstg;
846 	char *snapname;
847 	char *failed;
848 };
849 
850 static int
851 dsl_snapshot_destroy_one(char *name, void *arg)
852 {
853 	struct destroyarg *da = arg;
854 	dsl_dataset_t *ds;
855 	char *cp;
856 	int err;
857 
858 	(void) strcat(name, "@");
859 	(void) strcat(name, da->snapname);
860 	err = dsl_dataset_own(name, DS_MODE_READONLY | DS_MODE_INCONSISTENT,
861 	    da->dstg, &ds);
862 	cp = strchr(name, '@');
863 	*cp = '\0';
864 	if (err == 0) {
865 		dsl_dataset_make_exclusive(ds, da->dstg);
866 		if (ds->ds_user_ptr) {
867 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
868 			ds->ds_user_ptr = NULL;
869 		}
870 		dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
871 		    dsl_dataset_destroy_sync, ds, da->dstg, 0);
872 	} else if (err == ENOENT) {
873 		err = 0;
874 	} else {
875 		(void) strcpy(da->failed, name);
876 	}
877 	return (err);
878 }
879 
880 /*
881  * Destroy 'snapname' in all descendants of 'fsname'.
882  */
883 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
884 int
885 dsl_snapshots_destroy(char *fsname, char *snapname)
886 {
887 	int err;
888 	struct destroyarg da;
889 	dsl_sync_task_t *dst;
890 	spa_t *spa;
891 
892 	err = spa_open(fsname, &spa, FTAG);
893 	if (err)
894 		return (err);
895 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
896 	da.snapname = snapname;
897 	da.failed = fsname;
898 
899 	err = dmu_objset_find(fsname,
900 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
901 
902 	if (err == 0)
903 		err = dsl_sync_task_group_wait(da.dstg);
904 
905 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
906 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
907 		dsl_dataset_t *ds = dst->dst_arg1;
908 		/*
909 		 * Return the file system name that triggered the error
910 		 */
911 		if (dst->dst_err) {
912 			dsl_dataset_name(ds, fsname);
913 			*strchr(fsname, '@') = '\0';
914 		}
915 		dsl_dataset_disown(ds, da.dstg);
916 	}
917 
918 	dsl_sync_task_group_destroy(da.dstg);
919 	spa_close(spa, FTAG);
920 	return (err);
921 }
922 
923 /*
924  * ds must be opened as OWNER.  On return (whether successful or not),
925  * ds will be closed and caller can no longer dereference it.
926  */
927 int
928 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
929 {
930 	int err;
931 	dsl_sync_task_group_t *dstg;
932 	objset_t *os;
933 	dsl_dir_t *dd;
934 	uint64_t obj;
935 
936 	if (dsl_dataset_is_snapshot(ds)) {
937 		/* Destroying a snapshot is simpler */
938 		dsl_dataset_make_exclusive(ds, tag);
939 
940 		if (ds->ds_user_ptr) {
941 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
942 			ds->ds_user_ptr = NULL;
943 		}
944 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
945 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
946 		    ds, tag, 0);
947 		goto out;
948 	}
949 
950 	dd = ds->ds_dir;
951 
952 	/*
953 	 * Check for errors and mark this ds as inconsistent, in
954 	 * case we crash while freeing the objects.
955 	 */
956 	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
957 	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
958 	if (err)
959 		goto out;
960 
961 	err = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
962 	if (err)
963 		goto out;
964 
965 	/*
966 	 * remove the objects in open context, so that we won't
967 	 * have too much to do in syncing context.
968 	 */
969 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
970 	    ds->ds_phys->ds_prev_snap_txg)) {
971 		/*
972 		 * Ignore errors, if there is not enough disk space
973 		 * we will deal with it in dsl_dataset_destroy_sync().
974 		 */
975 		(void) dmu_free_object(os, obj);
976 	}
977 
978 	/*
979 	 * We need to sync out all in-flight IO before we try to evict
980 	 * (the dataset evict func is trying to clear the cached entries
981 	 * for this dataset in the ARC).
982 	 */
983 	txg_wait_synced(dd->dd_pool, 0);
984 
985 	/*
986 	 * If we managed to free all the objects in open
987 	 * context, the user space accounting should be zero.
988 	 */
989 	if (ds->ds_phys->ds_bp.blk_fill == 0 &&
990 	    dmu_objset_userused_enabled(os->os)) {
991 		uint64_t count;
992 
993 		ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
994 		    count == 0);
995 		ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
996 		    count == 0);
997 	}
998 
999 	dmu_objset_close(os);
1000 	if (err != ESRCH)
1001 		goto out;
1002 
1003 	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1004 	err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1005 	rw_exit(&dd->dd_pool->dp_config_rwlock);
1006 
1007 	if (err)
1008 		goto out;
1009 
1010 	if (ds->ds_user_ptr) {
1011 		/*
1012 		 * We need to sync out all in-flight IO before we try
1013 		 * to evict (the dataset evict func is trying to clear
1014 		 * the cached entries for this dataset in the ARC).
1015 		 */
1016 		txg_wait_synced(dd->dd_pool, 0);
1017 	}
1018 
1019 	/*
1020 	 * Blow away the dsl_dir + head dataset.
1021 	 */
1022 	dsl_dataset_make_exclusive(ds, tag);
1023 	if (ds->ds_user_ptr) {
1024 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1025 		ds->ds_user_ptr = NULL;
1026 	}
1027 	dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1028 	dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1029 	    dsl_dataset_destroy_sync, ds, tag, 0);
1030 	dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1031 	    dsl_dir_destroy_sync, dd, FTAG, 0);
1032 	err = dsl_sync_task_group_wait(dstg);
1033 	dsl_sync_task_group_destroy(dstg);
1034 	/* if it is successful, dsl_dir_destroy_sync will close the dd */
1035 	if (err)
1036 		dsl_dir_close(dd, FTAG);
1037 out:
1038 	dsl_dataset_disown(ds, tag);
1039 	return (err);
1040 }
1041 
1042 int
1043 dsl_dataset_rollback(dsl_dataset_t *ds, dmu_objset_type_t ost)
1044 {
1045 	int err;
1046 
1047 	ASSERT(ds->ds_owner);
1048 
1049 	dsl_dataset_make_exclusive(ds, ds->ds_owner);
1050 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1051 	    dsl_dataset_rollback_check, dsl_dataset_rollback_sync,
1052 	    ds, &ost, 0);
1053 	/* drop exclusive access */
1054 	mutex_enter(&ds->ds_lock);
1055 	rw_exit(&ds->ds_rwlock);
1056 	cv_broadcast(&ds->ds_exclusive_cv);
1057 	mutex_exit(&ds->ds_lock);
1058 	return (err);
1059 }
1060 
1061 void *
1062 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
1063     void *p, dsl_dataset_evict_func_t func)
1064 {
1065 	void *old;
1066 
1067 	mutex_enter(&ds->ds_lock);
1068 	old = ds->ds_user_ptr;
1069 	if (old == NULL) {
1070 		ds->ds_user_ptr = p;
1071 		ds->ds_user_evict_func = func;
1072 	}
1073 	mutex_exit(&ds->ds_lock);
1074 	return (old);
1075 }
1076 
1077 void *
1078 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
1079 {
1080 	return (ds->ds_user_ptr);
1081 }
1082 
1083 blkptr_t *
1084 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1085 {
1086 	return (&ds->ds_phys->ds_bp);
1087 }
1088 
1089 void
1090 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1091 {
1092 	ASSERT(dmu_tx_is_syncing(tx));
1093 	/* If it's the meta-objset, set dp_meta_rootbp */
1094 	if (ds == NULL) {
1095 		tx->tx_pool->dp_meta_rootbp = *bp;
1096 	} else {
1097 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1098 		ds->ds_phys->ds_bp = *bp;
1099 	}
1100 }
1101 
1102 spa_t *
1103 dsl_dataset_get_spa(dsl_dataset_t *ds)
1104 {
1105 	return (ds->ds_dir->dd_pool->dp_spa);
1106 }
1107 
1108 void
1109 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1110 {
1111 	dsl_pool_t *dp;
1112 
1113 	if (ds == NULL) /* this is the meta-objset */
1114 		return;
1115 
1116 	ASSERT(ds->ds_user_ptr != NULL);
1117 
1118 	if (ds->ds_phys->ds_next_snap_obj != 0)
1119 		panic("dirtying snapshot!");
1120 
1121 	dp = ds->ds_dir->dd_pool;
1122 
1123 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1124 		/* up the hold count until we can be written out */
1125 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1126 	}
1127 }
1128 
1129 /*
1130  * The unique space in the head dataset can be calculated by subtracting
1131  * the space used in the most recent snapshot, that is still being used
1132  * in this file system, from the space currently in use.  To figure out
1133  * the space in the most recent snapshot still in use, we need to take
1134  * the total space used in the snapshot and subtract out the space that
1135  * has been freed up since the snapshot was taken.
1136  */
1137 static void
1138 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1139 {
1140 	uint64_t mrs_used;
1141 	uint64_t dlused, dlcomp, dluncomp;
1142 
1143 	ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj);
1144 
1145 	if (ds->ds_phys->ds_prev_snap_obj != 0)
1146 		mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1147 	else
1148 		mrs_used = 0;
1149 
1150 	VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp,
1151 	    &dluncomp));
1152 
1153 	ASSERT3U(dlused, <=, mrs_used);
1154 	ds->ds_phys->ds_unique_bytes =
1155 	    ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1156 
1157 	if (!DS_UNIQUE_IS_ACCURATE(ds) &&
1158 	    spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1159 	    SPA_VERSION_UNIQUE_ACCURATE)
1160 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1161 }
1162 
1163 static uint64_t
1164 dsl_dataset_unique(dsl_dataset_t *ds)
1165 {
1166 	if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds))
1167 		dsl_dataset_recalc_head_uniq(ds);
1168 
1169 	return (ds->ds_phys->ds_unique_bytes);
1170 }
1171 
1172 struct killarg {
1173 	dsl_dataset_t *ds;
1174 	zio_t *zio;
1175 	dmu_tx_t *tx;
1176 };
1177 
1178 /* ARGSUSED */
1179 static int
1180 kill_blkptr(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb,
1181     const dnode_phys_t *dnp, void *arg)
1182 {
1183 	struct killarg *ka = arg;
1184 
1185 	if (bp == NULL)
1186 		return (0);
1187 
1188 	if ((zb->zb_level == -1ULL && zb->zb_blkid != 0) ||
1189 	    (zb->zb_object != 0 && dnp == NULL)) {
1190 		/*
1191 		 * It's a block in the intent log.  It has no
1192 		 * accounting, so just free it.
1193 		 */
1194 		VERIFY3U(0, ==, dsl_free(ka->zio, ka->tx->tx_pool,
1195 		    ka->tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT));
1196 	} else {
1197 		ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1198 		(void) dsl_dataset_block_kill(ka->ds, bp, ka->zio, ka->tx);
1199 	}
1200 
1201 	return (0);
1202 }
1203 
1204 /* ARGSUSED */
1205 static int
1206 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
1207 {
1208 	dsl_dataset_t *ds = arg1;
1209 	dmu_objset_type_t *ost = arg2;
1210 
1211 	/*
1212 	 * We can only roll back to emptyness if it is a ZPL objset.
1213 	 */
1214 	if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0)
1215 		return (EINVAL);
1216 
1217 	/*
1218 	 * This must not be a snapshot.
1219 	 */
1220 	if (ds->ds_phys->ds_next_snap_obj != 0)
1221 		return (EINVAL);
1222 
1223 	/*
1224 	 * If we made changes this txg, traverse_dataset won't find
1225 	 * them.  Try again.
1226 	 */
1227 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1228 		return (EAGAIN);
1229 
1230 	return (0);
1231 }
1232 
1233 /* ARGSUSED */
1234 static void
1235 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1236 {
1237 	dsl_dataset_t *ds = arg1;
1238 	dmu_objset_type_t *ost = arg2;
1239 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1240 
1241 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1242 
1243 	if (ds->ds_user_ptr != NULL) {
1244 		/*
1245 		 * We need to make sure that the objset_impl_t is reopened after
1246 		 * we do the rollback, otherwise it will have the wrong
1247 		 * objset_phys_t.  Normally this would happen when this
1248 		 * dataset-open is closed, thus causing the
1249 		 * dataset to be immediately evicted.  But when doing "zfs recv
1250 		 * -F", we reopen the objset before that, so that there is no
1251 		 * window where the dataset is closed and inconsistent.
1252 		 */
1253 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1254 		ds->ds_user_ptr = NULL;
1255 	}
1256 
1257 	/* Transfer space that was freed since last snap back to the head. */
1258 	{
1259 		uint64_t used;
1260 
1261 		VERIFY(0 == bplist_space_birthrange(&ds->ds_deadlist,
1262 		    ds->ds_origin_txg, UINT64_MAX, &used));
1263 		dsl_dir_transfer_space(ds->ds_dir, used,
1264 		    DD_USED_SNAP, DD_USED_HEAD, tx);
1265 	}
1266 
1267 	/* Zero out the deadlist. */
1268 	bplist_close(&ds->ds_deadlist);
1269 	bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1270 	ds->ds_phys->ds_deadlist_obj =
1271 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1272 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1273 	    ds->ds_phys->ds_deadlist_obj));
1274 
1275 	{
1276 		/*
1277 		 * Free blkptrs that we gave birth to - this covers
1278 		 * claimed but not played log blocks too.
1279 		 */
1280 		zio_t *zio;
1281 		struct killarg ka;
1282 
1283 		zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
1284 		    ZIO_FLAG_MUSTSUCCEED);
1285 		ka.ds = ds;
1286 		ka.zio = zio;
1287 		ka.tx = tx;
1288 		(void) traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1289 		    TRAVERSE_POST, kill_blkptr, &ka);
1290 		(void) zio_wait(zio);
1291 	}
1292 
1293 	ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1294 
1295 	if (ds->ds_prev && ds->ds_prev != ds->ds_dir->dd_pool->dp_origin_snap) {
1296 		/* Change our contents to that of the prev snapshot */
1297 
1298 		ASSERT3U(ds->ds_prev->ds_object, ==,
1299 		    ds->ds_phys->ds_prev_snap_obj);
1300 		ASSERT3U(ds->ds_phys->ds_used_bytes, <=,
1301 		    ds->ds_prev->ds_phys->ds_used_bytes);
1302 
1303 		ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1304 		ds->ds_phys->ds_used_bytes =
1305 		    ds->ds_prev->ds_phys->ds_used_bytes;
1306 		ds->ds_phys->ds_compressed_bytes =
1307 		    ds->ds_prev->ds_phys->ds_compressed_bytes;
1308 		ds->ds_phys->ds_uncompressed_bytes =
1309 		    ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1310 		ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1311 
1312 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1313 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1314 			ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1315 		}
1316 	} else {
1317 		objset_impl_t *osi;
1318 
1319 		ASSERT3U(ds->ds_phys->ds_used_bytes, ==, 0);
1320 		ASSERT3U(ds->ds_phys->ds_compressed_bytes, ==, 0);
1321 		ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, ==, 0);
1322 
1323 		bzero(&ds->ds_phys->ds_bp, sizeof (blkptr_t));
1324 		ds->ds_phys->ds_flags = 0;
1325 		ds->ds_phys->ds_unique_bytes = 0;
1326 		if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1327 		    SPA_VERSION_UNIQUE_ACCURATE)
1328 			ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1329 
1330 		osi = dmu_objset_create_impl(ds->ds_dir->dd_pool->dp_spa, ds,
1331 		    &ds->ds_phys->ds_bp, *ost, tx);
1332 #ifdef _KERNEL
1333 		zfs_create_fs(&osi->os, kcred, NULL, tx);
1334 #endif
1335 	}
1336 
1337 	spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa,
1338 	    tx, cr, "dataset = %llu", ds->ds_object);
1339 }
1340 
1341 /* ARGSUSED */
1342 static int
1343 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1344 {
1345 	dsl_dataset_t *ds = arg1;
1346 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1347 	uint64_t count;
1348 	int err;
1349 
1350 	/*
1351 	 * Can't delete a head dataset if there are snapshots of it.
1352 	 * (Except if the only snapshots are from the branch we cloned
1353 	 * from.)
1354 	 */
1355 	if (ds->ds_prev != NULL &&
1356 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1357 		return (EINVAL);
1358 
1359 	/*
1360 	 * This is really a dsl_dir thing, but check it here so that
1361 	 * we'll be less likely to leave this dataset inconsistent &
1362 	 * nearly destroyed.
1363 	 */
1364 	err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1365 	if (err)
1366 		return (err);
1367 	if (count != 0)
1368 		return (EEXIST);
1369 
1370 	return (0);
1371 }
1372 
1373 /* ARGSUSED */
1374 static void
1375 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1376 {
1377 	dsl_dataset_t *ds = arg1;
1378 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1379 
1380 	/* Mark it as inconsistent on-disk, in case we crash */
1381 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1382 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1383 
1384 	spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1385 	    cr, "dataset = %llu", ds->ds_object);
1386 }
1387 
1388 /* ARGSUSED */
1389 int
1390 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1391 {
1392 	dsl_dataset_t *ds = arg1;
1393 
1394 	/* we have an owner hold, so noone else can destroy us */
1395 	ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1396 
1397 	/* Can't delete a branch point. */
1398 	if (ds->ds_phys->ds_num_children > 1)
1399 		return (EEXIST);
1400 
1401 	/*
1402 	 * Can't delete a head dataset if there are snapshots of it.
1403 	 * (Except if the only snapshots are from the branch we cloned
1404 	 * from.)
1405 	 */
1406 	if (ds->ds_prev != NULL &&
1407 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1408 		return (EINVAL);
1409 
1410 	/*
1411 	 * If we made changes this txg, traverse_dsl_dataset won't find
1412 	 * them.  Try again.
1413 	 */
1414 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1415 		return (EAGAIN);
1416 
1417 	/* XXX we should do some i/o error checking... */
1418 	return (0);
1419 }
1420 
1421 struct refsarg {
1422 	kmutex_t lock;
1423 	boolean_t gone;
1424 	kcondvar_t cv;
1425 };
1426 
1427 /* ARGSUSED */
1428 static void
1429 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1430 {
1431 	struct refsarg *arg = argv;
1432 
1433 	mutex_enter(&arg->lock);
1434 	arg->gone = TRUE;
1435 	cv_signal(&arg->cv);
1436 	mutex_exit(&arg->lock);
1437 }
1438 
1439 static void
1440 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1441 {
1442 	struct refsarg arg;
1443 
1444 	mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1445 	cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1446 	arg.gone = FALSE;
1447 	(void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1448 	    dsl_dataset_refs_gone);
1449 	dmu_buf_rele(ds->ds_dbuf, tag);
1450 	mutex_enter(&arg.lock);
1451 	while (!arg.gone)
1452 		cv_wait(&arg.cv, &arg.lock);
1453 	ASSERT(arg.gone);
1454 	mutex_exit(&arg.lock);
1455 	ds->ds_dbuf = NULL;
1456 	ds->ds_phys = NULL;
1457 	mutex_destroy(&arg.lock);
1458 	cv_destroy(&arg.cv);
1459 }
1460 
1461 void
1462 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1463 {
1464 	dsl_dataset_t *ds = arg1;
1465 	zio_t *zio;
1466 	int err;
1467 	int after_branch_point = FALSE;
1468 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1469 	objset_t *mos = dp->dp_meta_objset;
1470 	dsl_dataset_t *ds_prev = NULL;
1471 	uint64_t obj;
1472 
1473 	ASSERT(ds->ds_owner);
1474 	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1475 	ASSERT(ds->ds_prev == NULL ||
1476 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1477 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1478 
1479 	/* signal any waiters that this dataset is going away */
1480 	mutex_enter(&ds->ds_lock);
1481 	ds->ds_owner = dsl_reaper;
1482 	cv_broadcast(&ds->ds_exclusive_cv);
1483 	mutex_exit(&ds->ds_lock);
1484 
1485 	/* Remove our reservation */
1486 	if (ds->ds_reserved != 0) {
1487 		uint64_t val = 0;
1488 		dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1489 		ASSERT3U(ds->ds_reserved, ==, 0);
1490 	}
1491 
1492 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1493 
1494 	dsl_pool_ds_destroyed(ds, tx);
1495 
1496 	obj = ds->ds_object;
1497 
1498 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1499 		if (ds->ds_prev) {
1500 			ds_prev = ds->ds_prev;
1501 		} else {
1502 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1503 			    ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1504 		}
1505 		after_branch_point =
1506 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1507 
1508 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1509 		if (after_branch_point &&
1510 		    ds_prev->ds_phys->ds_next_clones_obj != 0) {
1511 			VERIFY3U(0, ==, zap_remove_int(mos,
1512 			    ds_prev->ds_phys->ds_next_clones_obj, obj, tx));
1513 			if (ds->ds_phys->ds_next_snap_obj != 0) {
1514 				VERIFY(0 == zap_add_int(mos,
1515 				    ds_prev->ds_phys->ds_next_clones_obj,
1516 				    ds->ds_phys->ds_next_snap_obj, tx));
1517 			}
1518 		}
1519 		if (after_branch_point &&
1520 		    ds->ds_phys->ds_next_snap_obj == 0) {
1521 			/* This clone is toast. */
1522 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1523 			ds_prev->ds_phys->ds_num_children--;
1524 		} else if (!after_branch_point) {
1525 			ds_prev->ds_phys->ds_next_snap_obj =
1526 			    ds->ds_phys->ds_next_snap_obj;
1527 		}
1528 	}
1529 
1530 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1531 
1532 	if (ds->ds_phys->ds_next_snap_obj != 0) {
1533 		blkptr_t bp;
1534 		dsl_dataset_t *ds_next;
1535 		uint64_t itor = 0;
1536 		uint64_t old_unique;
1537 		int64_t used = 0, compressed = 0, uncompressed = 0;
1538 
1539 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1540 		    ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1541 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1542 
1543 		old_unique = dsl_dataset_unique(ds_next);
1544 
1545 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1546 		ds_next->ds_phys->ds_prev_snap_obj =
1547 		    ds->ds_phys->ds_prev_snap_obj;
1548 		ds_next->ds_phys->ds_prev_snap_txg =
1549 		    ds->ds_phys->ds_prev_snap_txg;
1550 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1551 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1552 
1553 		/*
1554 		 * Transfer to our deadlist (which will become next's
1555 		 * new deadlist) any entries from next's current
1556 		 * deadlist which were born before prev, and free the
1557 		 * other entries.
1558 		 *
1559 		 * XXX we're doing this long task with the config lock held
1560 		 */
1561 		while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) {
1562 			if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1563 				VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1564 				    &bp, tx));
1565 				if (ds_prev && !after_branch_point &&
1566 				    bp.blk_birth >
1567 				    ds_prev->ds_phys->ds_prev_snap_txg) {
1568 					ds_prev->ds_phys->ds_unique_bytes +=
1569 					    bp_get_dasize(dp->dp_spa, &bp);
1570 				}
1571 			} else {
1572 				used += bp_get_dasize(dp->dp_spa, &bp);
1573 				compressed += BP_GET_PSIZE(&bp);
1574 				uncompressed += BP_GET_UCSIZE(&bp);
1575 				/* XXX check return value? */
1576 				(void) dsl_free(zio, dp, tx->tx_txg,
1577 				    &bp, NULL, NULL, ARC_NOWAIT);
1578 			}
1579 		}
1580 
1581 		ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1582 
1583 		/* change snapused */
1584 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1585 		    -used, -compressed, -uncompressed, tx);
1586 
1587 		/* free next's deadlist */
1588 		bplist_close(&ds_next->ds_deadlist);
1589 		bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1590 
1591 		/* set next's deadlist to our deadlist */
1592 		bplist_close(&ds->ds_deadlist);
1593 		ds_next->ds_phys->ds_deadlist_obj =
1594 		    ds->ds_phys->ds_deadlist_obj;
1595 		VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1596 		    ds_next->ds_phys->ds_deadlist_obj));
1597 		ds->ds_phys->ds_deadlist_obj = 0;
1598 
1599 		if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1600 			/*
1601 			 * Update next's unique to include blocks which
1602 			 * were previously shared by only this snapshot
1603 			 * and it.  Those blocks will be born after the
1604 			 * prev snap and before this snap, and will have
1605 			 * died after the next snap and before the one
1606 			 * after that (ie. be on the snap after next's
1607 			 * deadlist).
1608 			 *
1609 			 * XXX we're doing this long task with the
1610 			 * config lock held
1611 			 */
1612 			dsl_dataset_t *ds_after_next;
1613 			uint64_t space;
1614 
1615 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1616 			    ds_next->ds_phys->ds_next_snap_obj,
1617 			    FTAG, &ds_after_next));
1618 
1619 			VERIFY(0 ==
1620 			    bplist_space_birthrange(&ds_after_next->ds_deadlist,
1621 			    ds->ds_phys->ds_prev_snap_txg,
1622 			    ds->ds_phys->ds_creation_txg, &space));
1623 			ds_next->ds_phys->ds_unique_bytes += space;
1624 
1625 			dsl_dataset_rele(ds_after_next, FTAG);
1626 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1627 		} else {
1628 			ASSERT3P(ds_next->ds_prev, ==, ds);
1629 			dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1630 			ds_next->ds_prev = NULL;
1631 			if (ds_prev) {
1632 				VERIFY(0 == dsl_dataset_get_ref(dp,
1633 				    ds->ds_phys->ds_prev_snap_obj,
1634 				    ds_next, &ds_next->ds_prev));
1635 			}
1636 
1637 			dsl_dataset_recalc_head_uniq(ds_next);
1638 
1639 			/*
1640 			 * Reduce the amount of our unconsmed refreservation
1641 			 * being charged to our parent by the amount of
1642 			 * new unique data we have gained.
1643 			 */
1644 			if (old_unique < ds_next->ds_reserved) {
1645 				int64_t mrsdelta;
1646 				uint64_t new_unique =
1647 				    ds_next->ds_phys->ds_unique_bytes;
1648 
1649 				ASSERT(old_unique <= new_unique);
1650 				mrsdelta = MIN(new_unique - old_unique,
1651 				    ds_next->ds_reserved - old_unique);
1652 				dsl_dir_diduse_space(ds->ds_dir,
1653 				    DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1654 			}
1655 		}
1656 		dsl_dataset_rele(ds_next, FTAG);
1657 	} else {
1658 		/*
1659 		 * There's no next snapshot, so this is a head dataset.
1660 		 * Destroy the deadlist.  Unless it's a clone, the
1661 		 * deadlist should be empty.  (If it's a clone, it's
1662 		 * safe to ignore the deadlist contents.)
1663 		 */
1664 		struct killarg ka;
1665 
1666 		ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1667 		bplist_close(&ds->ds_deadlist);
1668 		bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1669 		ds->ds_phys->ds_deadlist_obj = 0;
1670 
1671 		/*
1672 		 * Free everything that we point to (that's born after
1673 		 * the previous snapshot, if we are a clone)
1674 		 *
1675 		 * NB: this should be very quick, because we already
1676 		 * freed all the objects in open context.
1677 		 */
1678 		ka.ds = ds;
1679 		ka.zio = zio;
1680 		ka.tx = tx;
1681 		err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1682 		    TRAVERSE_POST, kill_blkptr, &ka);
1683 		ASSERT3U(err, ==, 0);
1684 		ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1685 		    ds->ds_phys->ds_unique_bytes == 0);
1686 	}
1687 
1688 	err = zio_wait(zio);
1689 	ASSERT3U(err, ==, 0);
1690 
1691 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1692 		/* Erase the link in the dir */
1693 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1694 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1695 		ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1696 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1697 		ASSERT(err == 0);
1698 	} else {
1699 		/* remove from snapshot namespace */
1700 		dsl_dataset_t *ds_head;
1701 		ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1702 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1703 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1704 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1705 #ifdef ZFS_DEBUG
1706 		{
1707 			uint64_t val;
1708 
1709 			err = dsl_dataset_snap_lookup(ds_head,
1710 			    ds->ds_snapname, &val);
1711 			ASSERT3U(err, ==, 0);
1712 			ASSERT3U(val, ==, obj);
1713 		}
1714 #endif
1715 		err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1716 		ASSERT(err == 0);
1717 		dsl_dataset_rele(ds_head, FTAG);
1718 	}
1719 
1720 	if (ds_prev && ds->ds_prev != ds_prev)
1721 		dsl_dataset_rele(ds_prev, FTAG);
1722 
1723 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1724 	spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1725 	    cr, "dataset = %llu", ds->ds_object);
1726 
1727 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1728 		uint64_t count;
1729 		ASSERT(0 == zap_count(mos,
1730 		    ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1731 		VERIFY(0 == dmu_object_free(mos,
1732 		    ds->ds_phys->ds_next_clones_obj, tx));
1733 	}
1734 	if (ds->ds_phys->ds_props_obj != 0)
1735 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1736 	dsl_dir_close(ds->ds_dir, ds);
1737 	ds->ds_dir = NULL;
1738 	dsl_dataset_drain_refs(ds, tag);
1739 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1740 }
1741 
1742 static int
1743 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1744 {
1745 	uint64_t asize;
1746 
1747 	if (!dmu_tx_is_syncing(tx))
1748 		return (0);
1749 
1750 	/*
1751 	 * If there's an fs-only reservation, any blocks that might become
1752 	 * owned by the snapshot dataset must be accommodated by space
1753 	 * outside of the reservation.
1754 	 */
1755 	asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1756 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1757 		return (ENOSPC);
1758 
1759 	/*
1760 	 * Propogate any reserved space for this snapshot to other
1761 	 * snapshot checks in this sync group.
1762 	 */
1763 	if (asize > 0)
1764 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1765 
1766 	return (0);
1767 }
1768 
1769 /* ARGSUSED */
1770 int
1771 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1772 {
1773 	dsl_dataset_t *ds = arg1;
1774 	const char *snapname = arg2;
1775 	int err;
1776 	uint64_t value;
1777 
1778 	/*
1779 	 * We don't allow multiple snapshots of the same txg.  If there
1780 	 * is already one, try again.
1781 	 */
1782 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1783 		return (EAGAIN);
1784 
1785 	/*
1786 	 * Check for conflicting name snapshot name.
1787 	 */
1788 	err = dsl_dataset_snap_lookup(ds, snapname, &value);
1789 	if (err == 0)
1790 		return (EEXIST);
1791 	if (err != ENOENT)
1792 		return (err);
1793 
1794 	/*
1795 	 * Check that the dataset's name is not too long.  Name consists
1796 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1797 	 */
1798 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1799 		return (ENAMETOOLONG);
1800 
1801 	err = dsl_dataset_snapshot_reserve_space(ds, tx);
1802 	if (err)
1803 		return (err);
1804 
1805 	ds->ds_trysnap_txg = tx->tx_txg;
1806 	return (0);
1807 }
1808 
1809 void
1810 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1811 {
1812 	dsl_dataset_t *ds = arg1;
1813 	const char *snapname = arg2;
1814 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1815 	dmu_buf_t *dbuf;
1816 	dsl_dataset_phys_t *dsphys;
1817 	uint64_t dsobj, crtxg;
1818 	objset_t *mos = dp->dp_meta_objset;
1819 	int err;
1820 
1821 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1822 
1823 	/*
1824 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1825 	 */
1826 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1827 		crtxg = 1;
1828 	else
1829 		crtxg = tx->tx_txg;
1830 
1831 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1832 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1833 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1834 	dmu_buf_will_dirty(dbuf, tx);
1835 	dsphys = dbuf->db_data;
1836 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1837 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1838 	dsphys->ds_fsid_guid = unique_create();
1839 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1840 	    sizeof (dsphys->ds_guid));
1841 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1842 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1843 	dsphys->ds_next_snap_obj = ds->ds_object;
1844 	dsphys->ds_num_children = 1;
1845 	dsphys->ds_creation_time = gethrestime_sec();
1846 	dsphys->ds_creation_txg = crtxg;
1847 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1848 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1849 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1850 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1851 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1852 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1853 	dmu_buf_rele(dbuf, FTAG);
1854 
1855 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1856 	if (ds->ds_prev) {
1857 		uint64_t next_clones_obj =
1858 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1859 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1860 		    ds->ds_object ||
1861 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1862 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1863 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1864 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1865 			    ds->ds_prev->ds_phys->ds_creation_txg);
1866 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1867 		} else if (next_clones_obj != 0) {
1868 			VERIFY3U(0, ==, zap_remove_int(mos,
1869 			    next_clones_obj, dsphys->ds_next_snap_obj, tx));
1870 			VERIFY3U(0, ==, zap_add_int(mos,
1871 			    next_clones_obj, dsobj, tx));
1872 		}
1873 	}
1874 
1875 	/*
1876 	 * If we have a reference-reservation on this dataset, we will
1877 	 * need to increase the amount of refreservation being charged
1878 	 * since our unique space is going to zero.
1879 	 */
1880 	if (ds->ds_reserved) {
1881 		int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1882 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1883 		    add, 0, 0, tx);
1884 	}
1885 
1886 	bplist_close(&ds->ds_deadlist);
1887 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1888 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1889 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1890 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1891 	ds->ds_phys->ds_unique_bytes = 0;
1892 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1893 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1894 	ds->ds_phys->ds_deadlist_obj =
1895 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1896 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1897 	    ds->ds_phys->ds_deadlist_obj));
1898 
1899 	dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1900 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1901 	    snapname, 8, 1, &dsobj, tx);
1902 	ASSERT(err == 0);
1903 
1904 	if (ds->ds_prev)
1905 		dsl_dataset_drop_ref(ds->ds_prev, ds);
1906 	VERIFY(0 == dsl_dataset_get_ref(dp,
1907 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1908 
1909 	dsl_pool_ds_snapshotted(ds, tx);
1910 
1911 	spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1912 	    "dataset = %llu", dsobj);
1913 }
1914 
1915 void
1916 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1917 {
1918 	ASSERT(dmu_tx_is_syncing(tx));
1919 	ASSERT(ds->ds_user_ptr != NULL);
1920 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1921 
1922 	/*
1923 	 * in case we had to change ds_fsid_guid when we opened it,
1924 	 * sync it out now.
1925 	 */
1926 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1927 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1928 
1929 	dsl_dir_dirty(ds->ds_dir, tx);
1930 	dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1931 }
1932 
1933 void
1934 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1935 {
1936 	uint64_t refd, avail, uobjs, aobjs;
1937 
1938 	dsl_dir_stats(ds->ds_dir, nv);
1939 
1940 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1941 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1942 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1943 
1944 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1945 	    ds->ds_phys->ds_creation_time);
1946 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1947 	    ds->ds_phys->ds_creation_txg);
1948 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1949 	    ds->ds_quota);
1950 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1951 	    ds->ds_reserved);
1952 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1953 	    ds->ds_phys->ds_guid);
1954 
1955 	if (ds->ds_phys->ds_next_snap_obj) {
1956 		/*
1957 		 * This is a snapshot; override the dd's space used with
1958 		 * our unique space and compression ratio.
1959 		 */
1960 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1961 		    ds->ds_phys->ds_unique_bytes);
1962 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1963 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1964 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1965 		    ds->ds_phys->ds_compressed_bytes));
1966 	}
1967 }
1968 
1969 void
1970 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1971 {
1972 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1973 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1974 	stat->dds_guid = ds->ds_phys->ds_guid;
1975 	if (ds->ds_phys->ds_next_snap_obj) {
1976 		stat->dds_is_snapshot = B_TRUE;
1977 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1978 	} else {
1979 		stat->dds_is_snapshot = B_FALSE;
1980 		stat->dds_num_clones = 0;
1981 	}
1982 
1983 	/* clone origin is really a dsl_dir thing... */
1984 	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1985 	if (dsl_dir_is_clone(ds->ds_dir)) {
1986 		dsl_dataset_t *ods;
1987 
1988 		VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
1989 		    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1990 		dsl_dataset_name(ods, stat->dds_origin);
1991 		dsl_dataset_drop_ref(ods, FTAG);
1992 	} else {
1993 		stat->dds_origin[0] = '\0';
1994 	}
1995 	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1996 }
1997 
1998 uint64_t
1999 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2000 {
2001 	return (ds->ds_fsid_guid);
2002 }
2003 
2004 void
2005 dsl_dataset_space(dsl_dataset_t *ds,
2006     uint64_t *refdbytesp, uint64_t *availbytesp,
2007     uint64_t *usedobjsp, uint64_t *availobjsp)
2008 {
2009 	*refdbytesp = ds->ds_phys->ds_used_bytes;
2010 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2011 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2012 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2013 	if (ds->ds_quota != 0) {
2014 		/*
2015 		 * Adjust available bytes according to refquota
2016 		 */
2017 		if (*refdbytesp < ds->ds_quota)
2018 			*availbytesp = MIN(*availbytesp,
2019 			    ds->ds_quota - *refdbytesp);
2020 		else
2021 			*availbytesp = 0;
2022 	}
2023 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2024 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
2025 }
2026 
2027 boolean_t
2028 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2029 {
2030 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2031 
2032 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2033 	    dsl_pool_sync_context(dp));
2034 	if (ds->ds_prev == NULL)
2035 		return (B_FALSE);
2036 	if (ds->ds_phys->ds_bp.blk_birth >
2037 	    ds->ds_prev->ds_phys->ds_creation_txg)
2038 		return (B_TRUE);
2039 	return (B_FALSE);
2040 }
2041 
2042 /* ARGSUSED */
2043 static int
2044 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2045 {
2046 	dsl_dataset_t *ds = arg1;
2047 	char *newsnapname = arg2;
2048 	dsl_dir_t *dd = ds->ds_dir;
2049 	dsl_dataset_t *hds;
2050 	uint64_t val;
2051 	int err;
2052 
2053 	err = dsl_dataset_hold_obj(dd->dd_pool,
2054 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2055 	if (err)
2056 		return (err);
2057 
2058 	/* new name better not be in use */
2059 	err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2060 	dsl_dataset_rele(hds, FTAG);
2061 
2062 	if (err == 0)
2063 		err = EEXIST;
2064 	else if (err == ENOENT)
2065 		err = 0;
2066 
2067 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2068 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2069 		err = ENAMETOOLONG;
2070 
2071 	return (err);
2072 }
2073 
2074 static void
2075 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
2076     cred_t *cr, dmu_tx_t *tx)
2077 {
2078 	dsl_dataset_t *ds = arg1;
2079 	const char *newsnapname = arg2;
2080 	dsl_dir_t *dd = ds->ds_dir;
2081 	objset_t *mos = dd->dd_pool->dp_meta_objset;
2082 	dsl_dataset_t *hds;
2083 	int err;
2084 
2085 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2086 
2087 	VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2088 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2089 
2090 	VERIFY(0 == dsl_dataset_get_snapname(ds));
2091 	err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2092 	ASSERT3U(err, ==, 0);
2093 	mutex_enter(&ds->ds_lock);
2094 	(void) strcpy(ds->ds_snapname, newsnapname);
2095 	mutex_exit(&ds->ds_lock);
2096 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2097 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2098 	ASSERT3U(err, ==, 0);
2099 
2100 	spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2101 	    cr, "dataset = %llu", ds->ds_object);
2102 	dsl_dataset_rele(hds, FTAG);
2103 }
2104 
2105 struct renamesnaparg {
2106 	dsl_sync_task_group_t *dstg;
2107 	char failed[MAXPATHLEN];
2108 	char *oldsnap;
2109 	char *newsnap;
2110 };
2111 
2112 static int
2113 dsl_snapshot_rename_one(char *name, void *arg)
2114 {
2115 	struct renamesnaparg *ra = arg;
2116 	dsl_dataset_t *ds = NULL;
2117 	char *cp;
2118 	int err;
2119 
2120 	cp = name + strlen(name);
2121 	*cp = '@';
2122 	(void) strcpy(cp + 1, ra->oldsnap);
2123 
2124 	/*
2125 	 * For recursive snapshot renames the parent won't be changing
2126 	 * so we just pass name for both the to/from argument.
2127 	 */
2128 	err = zfs_secpolicy_rename_perms(name, name, CRED());
2129 	if (err == ENOENT) {
2130 		return (0);
2131 	} else if (err) {
2132 		(void) strcpy(ra->failed, name);
2133 		return (err);
2134 	}
2135 
2136 #ifdef _KERNEL
2137 	/*
2138 	 * For all filesystems undergoing rename, we'll need to unmount it.
2139 	 */
2140 	(void) zfs_unmount_snap(name, NULL);
2141 #endif
2142 	err = dsl_dataset_hold(name, ra->dstg, &ds);
2143 	*cp = '\0';
2144 	if (err == ENOENT) {
2145 		return (0);
2146 	} else if (err) {
2147 		(void) strcpy(ra->failed, name);
2148 		return (err);
2149 	}
2150 
2151 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2152 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2153 
2154 	return (0);
2155 }
2156 
2157 static int
2158 dsl_recursive_rename(char *oldname, const char *newname)
2159 {
2160 	int err;
2161 	struct renamesnaparg *ra;
2162 	dsl_sync_task_t *dst;
2163 	spa_t *spa;
2164 	char *cp, *fsname = spa_strdup(oldname);
2165 	int len = strlen(oldname);
2166 
2167 	/* truncate the snapshot name to get the fsname */
2168 	cp = strchr(fsname, '@');
2169 	*cp = '\0';
2170 
2171 	err = spa_open(fsname, &spa, FTAG);
2172 	if (err) {
2173 		kmem_free(fsname, len + 1);
2174 		return (err);
2175 	}
2176 	ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2177 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2178 
2179 	ra->oldsnap = strchr(oldname, '@') + 1;
2180 	ra->newsnap = strchr(newname, '@') + 1;
2181 	*ra->failed = '\0';
2182 
2183 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2184 	    DS_FIND_CHILDREN);
2185 	kmem_free(fsname, len + 1);
2186 
2187 	if (err == 0) {
2188 		err = dsl_sync_task_group_wait(ra->dstg);
2189 	}
2190 
2191 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2192 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2193 		dsl_dataset_t *ds = dst->dst_arg1;
2194 		if (dst->dst_err) {
2195 			dsl_dir_name(ds->ds_dir, ra->failed);
2196 			(void) strcat(ra->failed, "@");
2197 			(void) strcat(ra->failed, ra->newsnap);
2198 		}
2199 		dsl_dataset_rele(ds, ra->dstg);
2200 	}
2201 
2202 	if (err)
2203 		(void) strcpy(oldname, ra->failed);
2204 
2205 	dsl_sync_task_group_destroy(ra->dstg);
2206 	kmem_free(ra, sizeof (struct renamesnaparg));
2207 	spa_close(spa, FTAG);
2208 	return (err);
2209 }
2210 
2211 static int
2212 dsl_valid_rename(char *oldname, void *arg)
2213 {
2214 	int delta = *(int *)arg;
2215 
2216 	if (strlen(oldname) + delta >= MAXNAMELEN)
2217 		return (ENAMETOOLONG);
2218 
2219 	return (0);
2220 }
2221 
2222 #pragma weak dmu_objset_rename = dsl_dataset_rename
2223 int
2224 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2225 {
2226 	dsl_dir_t *dd;
2227 	dsl_dataset_t *ds;
2228 	const char *tail;
2229 	int err;
2230 
2231 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2232 	if (err)
2233 		return (err);
2234 	/*
2235 	 * If there are more than 2 references there may be holds
2236 	 * hanging around that haven't been cleared out yet.
2237 	 */
2238 	if (dmu_buf_refcount(dd->dd_dbuf) > 2)
2239 		txg_wait_synced(dd->dd_pool, 0);
2240 	if (tail == NULL) {
2241 		int delta = strlen(newname) - strlen(oldname);
2242 
2243 		/* if we're growing, validate child name lengths */
2244 		if (delta > 0)
2245 			err = dmu_objset_find(oldname, dsl_valid_rename,
2246 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2247 
2248 		if (!err)
2249 			err = dsl_dir_rename(dd, newname);
2250 		dsl_dir_close(dd, FTAG);
2251 		return (err);
2252 	}
2253 	if (tail[0] != '@') {
2254 		/* the name ended in a nonexistant component */
2255 		dsl_dir_close(dd, FTAG);
2256 		return (ENOENT);
2257 	}
2258 
2259 	dsl_dir_close(dd, FTAG);
2260 
2261 	/* new name must be snapshot in same filesystem */
2262 	tail = strchr(newname, '@');
2263 	if (tail == NULL)
2264 		return (EINVAL);
2265 	tail++;
2266 	if (strncmp(oldname, newname, tail - newname) != 0)
2267 		return (EXDEV);
2268 
2269 	if (recursive) {
2270 		err = dsl_recursive_rename(oldname, newname);
2271 	} else {
2272 		err = dsl_dataset_hold(oldname, FTAG, &ds);
2273 		if (err)
2274 			return (err);
2275 
2276 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2277 		    dsl_dataset_snapshot_rename_check,
2278 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2279 
2280 		dsl_dataset_rele(ds, FTAG);
2281 	}
2282 
2283 	return (err);
2284 }
2285 
2286 struct promotenode {
2287 	list_node_t link;
2288 	dsl_dataset_t *ds;
2289 };
2290 
2291 struct promotearg {
2292 	list_t shared_snaps, origin_snaps, clone_snaps;
2293 	dsl_dataset_t *origin_origin, *origin_head;
2294 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2295 };
2296 
2297 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2298 
2299 /* ARGSUSED */
2300 static int
2301 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2302 {
2303 	dsl_dataset_t *hds = arg1;
2304 	struct promotearg *pa = arg2;
2305 	struct promotenode *snap = list_head(&pa->shared_snaps);
2306 	dsl_dataset_t *origin_ds = snap->ds;
2307 	int err;
2308 
2309 	/* Check that it is a real clone */
2310 	if (!dsl_dir_is_clone(hds->ds_dir))
2311 		return (EINVAL);
2312 
2313 	/* Since this is so expensive, don't do the preliminary check */
2314 	if (!dmu_tx_is_syncing(tx))
2315 		return (0);
2316 
2317 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2318 		return (EXDEV);
2319 
2320 	/* compute origin's new unique space */
2321 	snap = list_tail(&pa->clone_snaps);
2322 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2323 	err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2324 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, &pa->unique);
2325 	if (err)
2326 		return (err);
2327 
2328 	/*
2329 	 * Walk the snapshots that we are moving
2330 	 *
2331 	 * Compute space to transfer.  Consider the incremental changes
2332 	 * to used for each snapshot:
2333 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2334 	 * So each snapshot gave birth to:
2335 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2336 	 * So a sequence would look like:
2337 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2338 	 * Which simplifies to:
2339 	 * uN + kN + kN-1 + ... + k1 + k0
2340 	 * Note however, if we stop before we reach the ORIGIN we get:
2341 	 * uN + kN + kN-1 + ... + kM - uM-1
2342 	 */
2343 	pa->used = origin_ds->ds_phys->ds_used_bytes;
2344 	pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2345 	pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2346 	for (snap = list_head(&pa->shared_snaps); snap;
2347 	    snap = list_next(&pa->shared_snaps, snap)) {
2348 		uint64_t val, dlused, dlcomp, dluncomp;
2349 		dsl_dataset_t *ds = snap->ds;
2350 
2351 		/* Check that the snapshot name does not conflict */
2352 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2353 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2354 		if (err == 0)
2355 			return (EEXIST);
2356 		if (err != ENOENT)
2357 			return (err);
2358 
2359 		/* The very first snapshot does not have a deadlist */
2360 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2361 			continue;
2362 
2363 		if (err = bplist_space(&ds->ds_deadlist,
2364 		    &dlused, &dlcomp, &dluncomp))
2365 			return (err);
2366 		pa->used += dlused;
2367 		pa->comp += dlcomp;
2368 		pa->uncomp += dluncomp;
2369 	}
2370 
2371 	/*
2372 	 * If we are a clone of a clone then we never reached ORIGIN,
2373 	 * so we need to subtract out the clone origin's used space.
2374 	 */
2375 	if (pa->origin_origin) {
2376 		pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2377 		pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2378 		pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2379 	}
2380 
2381 	/* Check that there is enough space here */
2382 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2383 	    pa->used);
2384 	if (err)
2385 		return (err);
2386 
2387 	/*
2388 	 * Compute the amounts of space that will be used by snapshots
2389 	 * after the promotion (for both origin and clone).  For each,
2390 	 * it is the amount of space that will be on all of their
2391 	 * deadlists (that was not born before their new origin).
2392 	 */
2393 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2394 		uint64_t space;
2395 
2396 		/*
2397 		 * Note, typically this will not be a clone of a clone,
2398 		 * so snap->ds->ds_origin_txg will be < TXG_INITIAL, so
2399 		 * these snaplist_space() -> bplist_space_birthrange()
2400 		 * calls will be fast because they do not have to
2401 		 * iterate over all bps.
2402 		 */
2403 		snap = list_head(&pa->origin_snaps);
2404 		err = snaplist_space(&pa->shared_snaps,
2405 		    snap->ds->ds_origin_txg, &pa->cloneusedsnap);
2406 		if (err)
2407 			return (err);
2408 
2409 		err = snaplist_space(&pa->clone_snaps,
2410 		    snap->ds->ds_origin_txg, &space);
2411 		if (err)
2412 			return (err);
2413 		pa->cloneusedsnap += space;
2414 	}
2415 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2416 		err = snaplist_space(&pa->origin_snaps,
2417 		    origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2418 		if (err)
2419 			return (err);
2420 	}
2421 
2422 	return (0);
2423 }
2424 
2425 static void
2426 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2427 {
2428 	dsl_dataset_t *hds = arg1;
2429 	struct promotearg *pa = arg2;
2430 	struct promotenode *snap = list_head(&pa->shared_snaps);
2431 	dsl_dataset_t *origin_ds = snap->ds;
2432 	dsl_dataset_t *origin_head;
2433 	dsl_dir_t *dd = hds->ds_dir;
2434 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2435 	dsl_dir_t *odd = NULL;
2436 	uint64_t oldnext_obj;
2437 	int64_t delta;
2438 
2439 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2440 
2441 	snap = list_head(&pa->origin_snaps);
2442 	origin_head = snap->ds;
2443 
2444 	/*
2445 	 * We need to explicitly open odd, since origin_ds's dd will be
2446 	 * changing.
2447 	 */
2448 	VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2449 	    NULL, FTAG, &odd));
2450 
2451 	/* change origin's next snap */
2452 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2453 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2454 	snap = list_tail(&pa->clone_snaps);
2455 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2456 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2457 
2458 	/* change the origin's next clone */
2459 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2460 		VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2461 		    origin_ds->ds_phys->ds_next_clones_obj,
2462 		    origin_ds->ds_phys->ds_next_snap_obj, tx));
2463 		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2464 		    origin_ds->ds_phys->ds_next_clones_obj,
2465 		    oldnext_obj, tx));
2466 	}
2467 
2468 	/* change origin */
2469 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2470 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2471 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2472 	hds->ds_origin_txg = origin_head->ds_origin_txg;
2473 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2474 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2475 	origin_head->ds_origin_txg = origin_ds->ds_phys->ds_creation_txg;
2476 
2477 	/* move snapshots to this dir */
2478 	for (snap = list_head(&pa->shared_snaps); snap;
2479 	    snap = list_next(&pa->shared_snaps, snap)) {
2480 		dsl_dataset_t *ds = snap->ds;
2481 
2482 		/* unregister props as dsl_dir is changing */
2483 		if (ds->ds_user_ptr) {
2484 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
2485 			ds->ds_user_ptr = NULL;
2486 		}
2487 		/* move snap name entry */
2488 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2489 		VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2490 		    ds->ds_snapname, tx));
2491 		VERIFY(0 == zap_add(dp->dp_meta_objset,
2492 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2493 		    8, 1, &ds->ds_object, tx));
2494 		/* change containing dsl_dir */
2495 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2496 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2497 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2498 		ASSERT3P(ds->ds_dir, ==, odd);
2499 		dsl_dir_close(ds->ds_dir, ds);
2500 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2501 		    NULL, ds, &ds->ds_dir));
2502 
2503 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2504 	}
2505 
2506 	/*
2507 	 * Change space accounting.
2508 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2509 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2510 	 * is true for each of {clone,origin} independently.
2511 	 */
2512 
2513 	delta = pa->cloneusedsnap -
2514 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2515 	ASSERT3S(delta, >=, 0);
2516 	ASSERT3U(pa->used, >=, delta);
2517 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2518 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2519 	    pa->used - delta, pa->comp, pa->uncomp, tx);
2520 
2521 	delta = pa->originusedsnap -
2522 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2523 	ASSERT3S(delta, <=, 0);
2524 	ASSERT3U(pa->used, >=, -delta);
2525 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2526 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2527 	    -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2528 
2529 	origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2530 
2531 	/* log history record */
2532 	spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2533 	    cr, "dataset = %llu", hds->ds_object);
2534 
2535 	dsl_dir_close(odd, FTAG);
2536 }
2537 
2538 static char *snaplist_tag = "snaplist";
2539 /*
2540  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2541  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2542  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2543  * snapshots back to this dataset's origin.
2544  */
2545 static int
2546 snaplist_make(dsl_pool_t *dp, boolean_t own,
2547     uint64_t first_obj, uint64_t last_obj, list_t *l)
2548 {
2549 	uint64_t obj = last_obj;
2550 
2551 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2552 
2553 	list_create(l, sizeof (struct promotenode),
2554 	    offsetof(struct promotenode, link));
2555 
2556 	while (obj != first_obj) {
2557 		dsl_dataset_t *ds;
2558 		struct promotenode *snap;
2559 		int err;
2560 
2561 		if (own) {
2562 			err = dsl_dataset_own_obj(dp, obj,
2563 			    0, snaplist_tag, &ds);
2564 			if (err == 0)
2565 				dsl_dataset_make_exclusive(ds, snaplist_tag);
2566 		} else {
2567 			err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2568 		}
2569 		if (err == ENOENT) {
2570 			/* lost race with snapshot destroy */
2571 			struct promotenode *last = list_tail(l);
2572 			ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2573 			obj = last->ds->ds_phys->ds_prev_snap_obj;
2574 			continue;
2575 		} else if (err) {
2576 			return (err);
2577 		}
2578 
2579 		if (first_obj == 0)
2580 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2581 
2582 		snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2583 		snap->ds = ds;
2584 		list_insert_tail(l, snap);
2585 		obj = ds->ds_phys->ds_prev_snap_obj;
2586 	}
2587 
2588 	return (0);
2589 }
2590 
2591 static int
2592 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2593 {
2594 	struct promotenode *snap;
2595 
2596 	*spacep = 0;
2597 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2598 		uint64_t used;
2599 		int err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2600 		    mintxg, UINT64_MAX, &used);
2601 		if (err)
2602 			return (err);
2603 		*spacep += used;
2604 	}
2605 	return (0);
2606 }
2607 
2608 static void
2609 snaplist_destroy(list_t *l, boolean_t own)
2610 {
2611 	struct promotenode *snap;
2612 
2613 	if (!l || !list_link_active(&l->list_head))
2614 		return;
2615 
2616 	while ((snap = list_tail(l)) != NULL) {
2617 		list_remove(l, snap);
2618 		if (own)
2619 			dsl_dataset_disown(snap->ds, snaplist_tag);
2620 		else
2621 			dsl_dataset_rele(snap->ds, snaplist_tag);
2622 		kmem_free(snap, sizeof (struct promotenode));
2623 	}
2624 	list_destroy(l);
2625 }
2626 
2627 /*
2628  * Promote a clone.  Nomenclature note:
2629  * "clone" or "cds": the original clone which is being promoted
2630  * "origin" or "ods": the snapshot which is originally clone's origin
2631  * "origin head" or "ohds": the dataset which is the head
2632  * (filesystem/volume) for the origin
2633  * "origin origin": the origin of the origin's filesystem (typically
2634  * NULL, indicating that the clone is not a clone of a clone).
2635  */
2636 int
2637 dsl_dataset_promote(const char *name)
2638 {
2639 	dsl_dataset_t *ds;
2640 	dsl_dir_t *dd;
2641 	dsl_pool_t *dp;
2642 	dmu_object_info_t doi;
2643 	struct promotearg pa = { 0 };
2644 	struct promotenode *snap;
2645 	int err;
2646 
2647 	err = dsl_dataset_hold(name, FTAG, &ds);
2648 	if (err)
2649 		return (err);
2650 	dd = ds->ds_dir;
2651 	dp = dd->dd_pool;
2652 
2653 	err = dmu_object_info(dp->dp_meta_objset,
2654 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2655 	if (err) {
2656 		dsl_dataset_rele(ds, FTAG);
2657 		return (err);
2658 	}
2659 
2660 	if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2661 		dsl_dataset_rele(ds, FTAG);
2662 		return (EINVAL);
2663 	}
2664 
2665 	/*
2666 	 * We are going to inherit all the snapshots taken before our
2667 	 * origin (i.e., our new origin will be our parent's origin).
2668 	 * Take ownership of them so that we can rename them into our
2669 	 * namespace.
2670 	 */
2671 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2672 
2673 	err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2674 	    &pa.shared_snaps);
2675 	if (err != 0)
2676 		goto out;
2677 
2678 	err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2679 	if (err != 0)
2680 		goto out;
2681 
2682 	snap = list_head(&pa.shared_snaps);
2683 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2684 	err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2685 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2686 	if (err != 0)
2687 		goto out;
2688 
2689 	if (dsl_dir_is_clone(snap->ds->ds_dir)) {
2690 		err = dsl_dataset_own_obj(dp,
2691 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2692 		    0, FTAG, &pa.origin_origin);
2693 		if (err != 0)
2694 			goto out;
2695 	}
2696 
2697 out:
2698 	rw_exit(&dp->dp_config_rwlock);
2699 
2700 	/*
2701 	 * Add in 128x the snapnames zapobj size, since we will be moving
2702 	 * a bunch of snapnames to the promoted ds, and dirtying their
2703 	 * bonus buffers.
2704 	 */
2705 	if (err == 0) {
2706 		err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2707 		    dsl_dataset_promote_sync, ds, &pa,
2708 		    2 + 2 * doi.doi_physical_blks);
2709 	}
2710 
2711 	snaplist_destroy(&pa.shared_snaps, B_TRUE);
2712 	snaplist_destroy(&pa.clone_snaps, B_FALSE);
2713 	snaplist_destroy(&pa.origin_snaps, B_FALSE);
2714 	if (pa.origin_origin)
2715 		dsl_dataset_disown(pa.origin_origin, FTAG);
2716 	dsl_dataset_rele(ds, FTAG);
2717 	return (err);
2718 }
2719 
2720 struct cloneswaparg {
2721 	dsl_dataset_t *cds; /* clone dataset */
2722 	dsl_dataset_t *ohds; /* origin's head dataset */
2723 	boolean_t force;
2724 	int64_t unused_refres_delta; /* change in unconsumed refreservation */
2725 };
2726 
2727 /* ARGSUSED */
2728 static int
2729 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2730 {
2731 	struct cloneswaparg *csa = arg1;
2732 
2733 	/* they should both be heads */
2734 	if (dsl_dataset_is_snapshot(csa->cds) ||
2735 	    dsl_dataset_is_snapshot(csa->ohds))
2736 		return (EINVAL);
2737 
2738 	/* the branch point should be just before them */
2739 	if (csa->cds->ds_prev != csa->ohds->ds_prev)
2740 		return (EINVAL);
2741 
2742 	/* cds should be the clone */
2743 	if (csa->cds->ds_prev->ds_phys->ds_next_snap_obj !=
2744 	    csa->ohds->ds_object)
2745 		return (EINVAL);
2746 
2747 	/* the clone should be a child of the origin */
2748 	if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2749 		return (EINVAL);
2750 
2751 	/* ohds shouldn't be modified unless 'force' */
2752 	if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2753 		return (ETXTBSY);
2754 
2755 	/* adjust amount of any unconsumed refreservation */
2756 	csa->unused_refres_delta =
2757 	    (int64_t)MIN(csa->ohds->ds_reserved,
2758 	    csa->ohds->ds_phys->ds_unique_bytes) -
2759 	    (int64_t)MIN(csa->ohds->ds_reserved,
2760 	    csa->cds->ds_phys->ds_unique_bytes);
2761 
2762 	if (csa->unused_refres_delta > 0 &&
2763 	    csa->unused_refres_delta >
2764 	    dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
2765 		return (ENOSPC);
2766 
2767 	return (0);
2768 }
2769 
2770 /* ARGSUSED */
2771 static void
2772 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2773 {
2774 	struct cloneswaparg *csa = arg1;
2775 	dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2776 
2777 	ASSERT(csa->cds->ds_reserved == 0);
2778 	ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota);
2779 
2780 	dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2781 	dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2782 	dmu_buf_will_dirty(csa->cds->ds_prev->ds_dbuf, tx);
2783 
2784 	if (csa->cds->ds_user_ptr != NULL) {
2785 		csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr);
2786 		csa->cds->ds_user_ptr = NULL;
2787 	}
2788 
2789 	if (csa->ohds->ds_user_ptr != NULL) {
2790 		csa->ohds->ds_user_evict_func(csa->ohds,
2791 		    csa->ohds->ds_user_ptr);
2792 		csa->ohds->ds_user_ptr = NULL;
2793 	}
2794 
2795 	/* reset origin's unique bytes */
2796 	VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2797 	    csa->cds->ds_prev->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2798 	    &csa->cds->ds_prev->ds_phys->ds_unique_bytes));
2799 
2800 	/* swap blkptrs */
2801 	{
2802 		blkptr_t tmp;
2803 		tmp = csa->ohds->ds_phys->ds_bp;
2804 		csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2805 		csa->cds->ds_phys->ds_bp = tmp;
2806 	}
2807 
2808 	/* set dd_*_bytes */
2809 	{
2810 		int64_t dused, dcomp, duncomp;
2811 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2812 		uint64_t odl_used, odl_comp, odl_uncomp;
2813 
2814 		ASSERT3U(csa->cds->ds_dir->dd_phys->
2815 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2816 
2817 		VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2818 		    &cdl_comp, &cdl_uncomp));
2819 		VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2820 		    &odl_comp, &odl_uncomp));
2821 
2822 		dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2823 		    (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2824 		dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2825 		    (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2826 		duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2827 		    cdl_uncomp -
2828 		    (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2829 
2830 		dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
2831 		    dused, dcomp, duncomp, tx);
2832 		dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
2833 		    -dused, -dcomp, -duncomp, tx);
2834 
2835 		/*
2836 		 * The difference in the space used by snapshots is the
2837 		 * difference in snapshot space due to the head's
2838 		 * deadlist (since that's the only thing that's
2839 		 * changing that affects the snapused).
2840 		 */
2841 		VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2842 		    csa->ohds->ds_origin_txg, UINT64_MAX, &cdl_used));
2843 		VERIFY(0 == bplist_space_birthrange(&csa->ohds->ds_deadlist,
2844 		    csa->ohds->ds_origin_txg, UINT64_MAX, &odl_used));
2845 		dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
2846 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2847 	}
2848 
2849 #define	SWITCH64(x, y) \
2850 	{ \
2851 		uint64_t __tmp = (x); \
2852 		(x) = (y); \
2853 		(y) = __tmp; \
2854 	}
2855 
2856 	/* swap ds_*_bytes */
2857 	SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2858 	    csa->cds->ds_phys->ds_used_bytes);
2859 	SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2860 	    csa->cds->ds_phys->ds_compressed_bytes);
2861 	SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2862 	    csa->cds->ds_phys->ds_uncompressed_bytes);
2863 	SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
2864 	    csa->cds->ds_phys->ds_unique_bytes);
2865 
2866 	/* apply any parent delta for change in unconsumed refreservation */
2867 	dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
2868 	    csa->unused_refres_delta, 0, 0, tx);
2869 
2870 	/* swap deadlists */
2871 	bplist_close(&csa->cds->ds_deadlist);
2872 	bplist_close(&csa->ohds->ds_deadlist);
2873 	SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2874 	    csa->cds->ds_phys->ds_deadlist_obj);
2875 	VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2876 	    csa->cds->ds_phys->ds_deadlist_obj));
2877 	VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2878 	    csa->ohds->ds_phys->ds_deadlist_obj));
2879 
2880 	dsl_pool_ds_clone_swapped(csa->ohds, csa->cds, tx);
2881 }
2882 
2883 /*
2884  * Swap 'clone' with its origin head file system.  Used at the end
2885  * of "online recv" to swizzle the file system to the new version.
2886  */
2887 int
2888 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2889     boolean_t force)
2890 {
2891 	struct cloneswaparg csa;
2892 	int error;
2893 
2894 	ASSERT(clone->ds_owner);
2895 	ASSERT(origin_head->ds_owner);
2896 retry:
2897 	/* Need exclusive access for the swap */
2898 	rw_enter(&clone->ds_rwlock, RW_WRITER);
2899 	if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
2900 		rw_exit(&clone->ds_rwlock);
2901 		rw_enter(&origin_head->ds_rwlock, RW_WRITER);
2902 		if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
2903 			rw_exit(&origin_head->ds_rwlock);
2904 			goto retry;
2905 		}
2906 	}
2907 	csa.cds = clone;
2908 	csa.ohds = origin_head;
2909 	csa.force = force;
2910 	error = dsl_sync_task_do(clone->ds_dir->dd_pool,
2911 	    dsl_dataset_clone_swap_check,
2912 	    dsl_dataset_clone_swap_sync, &csa, NULL, 9);
2913 	return (error);
2914 }
2915 
2916 /*
2917  * Given a pool name and a dataset object number in that pool,
2918  * return the name of that dataset.
2919  */
2920 int
2921 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2922 {
2923 	spa_t *spa;
2924 	dsl_pool_t *dp;
2925 	dsl_dataset_t *ds;
2926 	int error;
2927 
2928 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
2929 		return (error);
2930 	dp = spa_get_dsl(spa);
2931 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2932 	if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
2933 		dsl_dataset_name(ds, buf);
2934 		dsl_dataset_rele(ds, FTAG);
2935 	}
2936 	rw_exit(&dp->dp_config_rwlock);
2937 	spa_close(spa, FTAG);
2938 
2939 	return (error);
2940 }
2941 
2942 int
2943 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2944     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2945 {
2946 	int error = 0;
2947 
2948 	ASSERT3S(asize, >, 0);
2949 
2950 	/*
2951 	 * *ref_rsrv is the portion of asize that will come from any
2952 	 * unconsumed refreservation space.
2953 	 */
2954 	*ref_rsrv = 0;
2955 
2956 	mutex_enter(&ds->ds_lock);
2957 	/*
2958 	 * Make a space adjustment for reserved bytes.
2959 	 */
2960 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2961 		ASSERT3U(*used, >=,
2962 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2963 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2964 		*ref_rsrv =
2965 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2966 	}
2967 
2968 	if (!check_quota || ds->ds_quota == 0) {
2969 		mutex_exit(&ds->ds_lock);
2970 		return (0);
2971 	}
2972 	/*
2973 	 * If they are requesting more space, and our current estimate
2974 	 * is over quota, they get to try again unless the actual
2975 	 * on-disk is over quota and there are no pending changes (which
2976 	 * may free up space for us).
2977 	 */
2978 	if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
2979 		if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
2980 			error = ERESTART;
2981 		else
2982 			error = EDQUOT;
2983 	}
2984 	mutex_exit(&ds->ds_lock);
2985 
2986 	return (error);
2987 }
2988 
2989 /* ARGSUSED */
2990 static int
2991 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
2992 {
2993 	dsl_dataset_t *ds = arg1;
2994 	uint64_t *quotap = arg2;
2995 	uint64_t new_quota = *quotap;
2996 
2997 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
2998 		return (ENOTSUP);
2999 
3000 	if (new_quota == 0)
3001 		return (0);
3002 
3003 	if (new_quota < ds->ds_phys->ds_used_bytes ||
3004 	    new_quota < ds->ds_reserved)
3005 		return (ENOSPC);
3006 
3007 	return (0);
3008 }
3009 
3010 /* ARGSUSED */
3011 void
3012 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3013 {
3014 	dsl_dataset_t *ds = arg1;
3015 	uint64_t *quotap = arg2;
3016 	uint64_t new_quota = *quotap;
3017 
3018 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3019 
3020 	ds->ds_quota = new_quota;
3021 
3022 	dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
3023 
3024 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
3025 	    tx, cr, "%lld dataset = %llu ",
3026 	    (longlong_t)new_quota, ds->ds_object);
3027 }
3028 
3029 int
3030 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
3031 {
3032 	dsl_dataset_t *ds;
3033 	int err;
3034 
3035 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3036 	if (err)
3037 		return (err);
3038 
3039 	if (quota != ds->ds_quota) {
3040 		/*
3041 		 * If someone removes a file, then tries to set the quota, we
3042 		 * want to make sure the file freeing takes effect.
3043 		 */
3044 		txg_wait_open(ds->ds_dir->dd_pool, 0);
3045 
3046 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3047 		    dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3048 		    ds, &quota, 0);
3049 	}
3050 	dsl_dataset_rele(ds, FTAG);
3051 	return (err);
3052 }
3053 
3054 static int
3055 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3056 {
3057 	dsl_dataset_t *ds = arg1;
3058 	uint64_t *reservationp = arg2;
3059 	uint64_t new_reservation = *reservationp;
3060 	uint64_t unique;
3061 
3062 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3063 	    SPA_VERSION_REFRESERVATION)
3064 		return (ENOTSUP);
3065 
3066 	if (dsl_dataset_is_snapshot(ds))
3067 		return (EINVAL);
3068 
3069 	/*
3070 	 * If we are doing the preliminary check in open context, the
3071 	 * space estimates may be inaccurate.
3072 	 */
3073 	if (!dmu_tx_is_syncing(tx))
3074 		return (0);
3075 
3076 	mutex_enter(&ds->ds_lock);
3077 	unique = dsl_dataset_unique(ds);
3078 	mutex_exit(&ds->ds_lock);
3079 
3080 	if (MAX(unique, new_reservation) > MAX(unique, ds->ds_reserved)) {
3081 		uint64_t delta = MAX(unique, new_reservation) -
3082 		    MAX(unique, ds->ds_reserved);
3083 
3084 		if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3085 			return (ENOSPC);
3086 		if (ds->ds_quota > 0 &&
3087 		    new_reservation > ds->ds_quota)
3088 			return (ENOSPC);
3089 	}
3090 
3091 	return (0);
3092 }
3093 
3094 /* ARGSUSED */
3095 static void
3096 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
3097     dmu_tx_t *tx)
3098 {
3099 	dsl_dataset_t *ds = arg1;
3100 	uint64_t *reservationp = arg2;
3101 	uint64_t new_reservation = *reservationp;
3102 	uint64_t unique;
3103 	int64_t delta;
3104 
3105 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3106 
3107 	mutex_enter(&ds->ds_dir->dd_lock);
3108 	mutex_enter(&ds->ds_lock);
3109 	unique = dsl_dataset_unique(ds);
3110 	delta = MAX(0, (int64_t)(new_reservation - unique)) -
3111 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3112 	ds->ds_reserved = new_reservation;
3113 	mutex_exit(&ds->ds_lock);
3114 
3115 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3116 	mutex_exit(&ds->ds_dir->dd_lock);
3117 	dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
3118 	    new_reservation, cr, tx);
3119 
3120 	spa_history_internal_log(LOG_DS_REFRESERV,
3121 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
3122 	    (longlong_t)new_reservation, ds->ds_object);
3123 }
3124 
3125 int
3126 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
3127 {
3128 	dsl_dataset_t *ds;
3129 	int err;
3130 
3131 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3132 	if (err)
3133 		return (err);
3134 
3135 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3136 	    dsl_dataset_set_reservation_check,
3137 	    dsl_dataset_set_reservation_sync, ds, &reservation, 0);
3138 	dsl_dataset_rele(ds, FTAG);
3139 	return (err);
3140 }
3141