xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 56f9a274cc7ca7f2d6f19959b2db143d94a4e7e0)
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 &&
1215 	    ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL)
1216 		return (EINVAL);
1217 
1218 	/*
1219 	 * This must not be a snapshot.
1220 	 */
1221 	if (ds->ds_phys->ds_next_snap_obj != 0)
1222 		return (EINVAL);
1223 
1224 	/*
1225 	 * If we made changes this txg, traverse_dataset won't find
1226 	 * them.  Try again.
1227 	 */
1228 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1229 		return (EAGAIN);
1230 
1231 	return (0);
1232 }
1233 
1234 /* ARGSUSED */
1235 static void
1236 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1237 {
1238 	dsl_dataset_t *ds = arg1;
1239 	dmu_objset_type_t *ost = arg2;
1240 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1241 
1242 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1243 
1244 	if (ds->ds_user_ptr != NULL) {
1245 		/*
1246 		 * We need to make sure that the objset_impl_t is reopened after
1247 		 * we do the rollback, otherwise it will have the wrong
1248 		 * objset_phys_t.  Normally this would happen when this
1249 		 * dataset-open is closed, thus causing the
1250 		 * dataset to be immediately evicted.  But when doing "zfs recv
1251 		 * -F", we reopen the objset before that, so that there is no
1252 		 * window where the dataset is closed and inconsistent.
1253 		 */
1254 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1255 		ds->ds_user_ptr = NULL;
1256 	}
1257 
1258 	/* Transfer space that was freed since last snap back to the head. */
1259 	{
1260 		uint64_t used;
1261 
1262 		VERIFY(0 == bplist_space_birthrange(&ds->ds_deadlist,
1263 		    ds->ds_origin_txg, UINT64_MAX, &used));
1264 		dsl_dir_transfer_space(ds->ds_dir, used,
1265 		    DD_USED_SNAP, DD_USED_HEAD, tx);
1266 	}
1267 
1268 	/* Zero out the deadlist. */
1269 	bplist_close(&ds->ds_deadlist);
1270 	bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1271 	ds->ds_phys->ds_deadlist_obj =
1272 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1273 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1274 	    ds->ds_phys->ds_deadlist_obj));
1275 
1276 	{
1277 		/*
1278 		 * Free blkptrs that we gave birth to - this covers
1279 		 * claimed but not played log blocks too.
1280 		 */
1281 		zio_t *zio;
1282 		struct killarg ka;
1283 
1284 		zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
1285 		    ZIO_FLAG_MUSTSUCCEED);
1286 		ka.ds = ds;
1287 		ka.zio = zio;
1288 		ka.tx = tx;
1289 		(void) traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1290 		    TRAVERSE_POST, kill_blkptr, &ka);
1291 		(void) zio_wait(zio);
1292 	}
1293 
1294 	ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1295 
1296 	if (ds->ds_prev && ds->ds_prev != ds->ds_dir->dd_pool->dp_origin_snap) {
1297 		/* Change our contents to that of the prev snapshot */
1298 
1299 		ASSERT3U(ds->ds_prev->ds_object, ==,
1300 		    ds->ds_phys->ds_prev_snap_obj);
1301 		ASSERT3U(ds->ds_phys->ds_used_bytes, <=,
1302 		    ds->ds_prev->ds_phys->ds_used_bytes);
1303 
1304 		ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1305 		ds->ds_phys->ds_used_bytes =
1306 		    ds->ds_prev->ds_phys->ds_used_bytes;
1307 		ds->ds_phys->ds_compressed_bytes =
1308 		    ds->ds_prev->ds_phys->ds_compressed_bytes;
1309 		ds->ds_phys->ds_uncompressed_bytes =
1310 		    ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1311 		ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1312 
1313 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1314 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1315 			ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1316 		}
1317 	} else {
1318 		objset_impl_t *osi;
1319 
1320 		ASSERT(*ost != DMU_OST_ZVOL);
1321 		ASSERT3U(ds->ds_phys->ds_used_bytes, ==, 0);
1322 		ASSERT3U(ds->ds_phys->ds_compressed_bytes, ==, 0);
1323 		ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, ==, 0);
1324 
1325 		bzero(&ds->ds_phys->ds_bp, sizeof (blkptr_t));
1326 		ds->ds_phys->ds_flags = 0;
1327 		ds->ds_phys->ds_unique_bytes = 0;
1328 		if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1329 		    SPA_VERSION_UNIQUE_ACCURATE)
1330 			ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1331 
1332 		osi = dmu_objset_create_impl(ds->ds_dir->dd_pool->dp_spa, ds,
1333 		    &ds->ds_phys->ds_bp, *ost, tx);
1334 #ifdef _KERNEL
1335 		zfs_create_fs(&osi->os, kcred, NULL, tx);
1336 #endif
1337 	}
1338 
1339 	spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa,
1340 	    tx, cr, "dataset = %llu", ds->ds_object);
1341 }
1342 
1343 /* ARGSUSED */
1344 static int
1345 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1346 {
1347 	dsl_dataset_t *ds = arg1;
1348 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1349 	uint64_t count;
1350 	int err;
1351 
1352 	/*
1353 	 * Can't delete a head dataset if there are snapshots of it.
1354 	 * (Except if the only snapshots are from the branch we cloned
1355 	 * from.)
1356 	 */
1357 	if (ds->ds_prev != NULL &&
1358 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1359 		return (EINVAL);
1360 
1361 	/*
1362 	 * This is really a dsl_dir thing, but check it here so that
1363 	 * we'll be less likely to leave this dataset inconsistent &
1364 	 * nearly destroyed.
1365 	 */
1366 	err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1367 	if (err)
1368 		return (err);
1369 	if (count != 0)
1370 		return (EEXIST);
1371 
1372 	return (0);
1373 }
1374 
1375 /* ARGSUSED */
1376 static void
1377 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1378 {
1379 	dsl_dataset_t *ds = arg1;
1380 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1381 
1382 	/* Mark it as inconsistent on-disk, in case we crash */
1383 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1384 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1385 
1386 	spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1387 	    cr, "dataset = %llu", ds->ds_object);
1388 }
1389 
1390 /* ARGSUSED */
1391 int
1392 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1393 {
1394 	dsl_dataset_t *ds = arg1;
1395 
1396 	/* we have an owner hold, so noone else can destroy us */
1397 	ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1398 
1399 	/* Can't delete a branch point. */
1400 	if (ds->ds_phys->ds_num_children > 1)
1401 		return (EEXIST);
1402 
1403 	/*
1404 	 * Can't delete a head dataset if there are snapshots of it.
1405 	 * (Except if the only snapshots are from the branch we cloned
1406 	 * from.)
1407 	 */
1408 	if (ds->ds_prev != NULL &&
1409 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1410 		return (EINVAL);
1411 
1412 	/*
1413 	 * If we made changes this txg, traverse_dsl_dataset won't find
1414 	 * them.  Try again.
1415 	 */
1416 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1417 		return (EAGAIN);
1418 
1419 	/* XXX we should do some i/o error checking... */
1420 	return (0);
1421 }
1422 
1423 struct refsarg {
1424 	kmutex_t lock;
1425 	boolean_t gone;
1426 	kcondvar_t cv;
1427 };
1428 
1429 /* ARGSUSED */
1430 static void
1431 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1432 {
1433 	struct refsarg *arg = argv;
1434 
1435 	mutex_enter(&arg->lock);
1436 	arg->gone = TRUE;
1437 	cv_signal(&arg->cv);
1438 	mutex_exit(&arg->lock);
1439 }
1440 
1441 static void
1442 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1443 {
1444 	struct refsarg arg;
1445 
1446 	mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1447 	cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1448 	arg.gone = FALSE;
1449 	(void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1450 	    dsl_dataset_refs_gone);
1451 	dmu_buf_rele(ds->ds_dbuf, tag);
1452 	mutex_enter(&arg.lock);
1453 	while (!arg.gone)
1454 		cv_wait(&arg.cv, &arg.lock);
1455 	ASSERT(arg.gone);
1456 	mutex_exit(&arg.lock);
1457 	ds->ds_dbuf = NULL;
1458 	ds->ds_phys = NULL;
1459 	mutex_destroy(&arg.lock);
1460 	cv_destroy(&arg.cv);
1461 }
1462 
1463 void
1464 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1465 {
1466 	dsl_dataset_t *ds = arg1;
1467 	zio_t *zio;
1468 	int err;
1469 	int after_branch_point = FALSE;
1470 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1471 	objset_t *mos = dp->dp_meta_objset;
1472 	dsl_dataset_t *ds_prev = NULL;
1473 	uint64_t obj;
1474 
1475 	ASSERT(ds->ds_owner);
1476 	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1477 	ASSERT(ds->ds_prev == NULL ||
1478 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1479 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1480 
1481 	/* signal any waiters that this dataset is going away */
1482 	mutex_enter(&ds->ds_lock);
1483 	ds->ds_owner = dsl_reaper;
1484 	cv_broadcast(&ds->ds_exclusive_cv);
1485 	mutex_exit(&ds->ds_lock);
1486 
1487 	/* Remove our reservation */
1488 	if (ds->ds_reserved != 0) {
1489 		uint64_t val = 0;
1490 		dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1491 		ASSERT3U(ds->ds_reserved, ==, 0);
1492 	}
1493 
1494 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1495 
1496 	dsl_pool_ds_destroyed(ds, tx);
1497 
1498 	obj = ds->ds_object;
1499 
1500 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1501 		if (ds->ds_prev) {
1502 			ds_prev = ds->ds_prev;
1503 		} else {
1504 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1505 			    ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1506 		}
1507 		after_branch_point =
1508 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1509 
1510 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1511 		if (after_branch_point &&
1512 		    ds_prev->ds_phys->ds_next_clones_obj != 0) {
1513 			VERIFY3U(0, ==, zap_remove_int(mos,
1514 			    ds_prev->ds_phys->ds_next_clones_obj, obj, tx));
1515 			if (ds->ds_phys->ds_next_snap_obj != 0) {
1516 				VERIFY(0 == zap_add_int(mos,
1517 				    ds_prev->ds_phys->ds_next_clones_obj,
1518 				    ds->ds_phys->ds_next_snap_obj, tx));
1519 			}
1520 		}
1521 		if (after_branch_point &&
1522 		    ds->ds_phys->ds_next_snap_obj == 0) {
1523 			/* This clone is toast. */
1524 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1525 			ds_prev->ds_phys->ds_num_children--;
1526 		} else if (!after_branch_point) {
1527 			ds_prev->ds_phys->ds_next_snap_obj =
1528 			    ds->ds_phys->ds_next_snap_obj;
1529 		}
1530 	}
1531 
1532 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1533 
1534 	if (ds->ds_phys->ds_next_snap_obj != 0) {
1535 		blkptr_t bp;
1536 		dsl_dataset_t *ds_next;
1537 		uint64_t itor = 0;
1538 		uint64_t old_unique;
1539 		int64_t used = 0, compressed = 0, uncompressed = 0;
1540 
1541 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1542 		    ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1543 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1544 
1545 		old_unique = dsl_dataset_unique(ds_next);
1546 
1547 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1548 		ds_next->ds_phys->ds_prev_snap_obj =
1549 		    ds->ds_phys->ds_prev_snap_obj;
1550 		ds_next->ds_phys->ds_prev_snap_txg =
1551 		    ds->ds_phys->ds_prev_snap_txg;
1552 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1553 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1554 
1555 		/*
1556 		 * Transfer to our deadlist (which will become next's
1557 		 * new deadlist) any entries from next's current
1558 		 * deadlist which were born before prev, and free the
1559 		 * other entries.
1560 		 *
1561 		 * XXX we're doing this long task with the config lock held
1562 		 */
1563 		while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) {
1564 			if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1565 				VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1566 				    &bp, tx));
1567 				if (ds_prev && !after_branch_point &&
1568 				    bp.blk_birth >
1569 				    ds_prev->ds_phys->ds_prev_snap_txg) {
1570 					ds_prev->ds_phys->ds_unique_bytes +=
1571 					    bp_get_dasize(dp->dp_spa, &bp);
1572 				}
1573 			} else {
1574 				used += bp_get_dasize(dp->dp_spa, &bp);
1575 				compressed += BP_GET_PSIZE(&bp);
1576 				uncompressed += BP_GET_UCSIZE(&bp);
1577 				/* XXX check return value? */
1578 				(void) dsl_free(zio, dp, tx->tx_txg,
1579 				    &bp, NULL, NULL, ARC_NOWAIT);
1580 			}
1581 		}
1582 
1583 		ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1584 
1585 		/* change snapused */
1586 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1587 		    -used, -compressed, -uncompressed, tx);
1588 
1589 		/* free next's deadlist */
1590 		bplist_close(&ds_next->ds_deadlist);
1591 		bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1592 
1593 		/* set next's deadlist to our deadlist */
1594 		bplist_close(&ds->ds_deadlist);
1595 		ds_next->ds_phys->ds_deadlist_obj =
1596 		    ds->ds_phys->ds_deadlist_obj;
1597 		VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1598 		    ds_next->ds_phys->ds_deadlist_obj));
1599 		ds->ds_phys->ds_deadlist_obj = 0;
1600 
1601 		if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1602 			/*
1603 			 * Update next's unique to include blocks which
1604 			 * were previously shared by only this snapshot
1605 			 * and it.  Those blocks will be born after the
1606 			 * prev snap and before this snap, and will have
1607 			 * died after the next snap and before the one
1608 			 * after that (ie. be on the snap after next's
1609 			 * deadlist).
1610 			 *
1611 			 * XXX we're doing this long task with the
1612 			 * config lock held
1613 			 */
1614 			dsl_dataset_t *ds_after_next;
1615 			uint64_t space;
1616 
1617 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1618 			    ds_next->ds_phys->ds_next_snap_obj,
1619 			    FTAG, &ds_after_next));
1620 
1621 			VERIFY(0 ==
1622 			    bplist_space_birthrange(&ds_after_next->ds_deadlist,
1623 			    ds->ds_phys->ds_prev_snap_txg,
1624 			    ds->ds_phys->ds_creation_txg, &space));
1625 			ds_next->ds_phys->ds_unique_bytes += space;
1626 
1627 			dsl_dataset_rele(ds_after_next, FTAG);
1628 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1629 		} else {
1630 			ASSERT3P(ds_next->ds_prev, ==, ds);
1631 			dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1632 			ds_next->ds_prev = NULL;
1633 			if (ds_prev) {
1634 				VERIFY(0 == dsl_dataset_get_ref(dp,
1635 				    ds->ds_phys->ds_prev_snap_obj,
1636 				    ds_next, &ds_next->ds_prev));
1637 			}
1638 
1639 			dsl_dataset_recalc_head_uniq(ds_next);
1640 
1641 			/*
1642 			 * Reduce the amount of our unconsmed refreservation
1643 			 * being charged to our parent by the amount of
1644 			 * new unique data we have gained.
1645 			 */
1646 			if (old_unique < ds_next->ds_reserved) {
1647 				int64_t mrsdelta;
1648 				uint64_t new_unique =
1649 				    ds_next->ds_phys->ds_unique_bytes;
1650 
1651 				ASSERT(old_unique <= new_unique);
1652 				mrsdelta = MIN(new_unique - old_unique,
1653 				    ds_next->ds_reserved - old_unique);
1654 				dsl_dir_diduse_space(ds->ds_dir,
1655 				    DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1656 			}
1657 		}
1658 		dsl_dataset_rele(ds_next, FTAG);
1659 	} else {
1660 		/*
1661 		 * There's no next snapshot, so this is a head dataset.
1662 		 * Destroy the deadlist.  Unless it's a clone, the
1663 		 * deadlist should be empty.  (If it's a clone, it's
1664 		 * safe to ignore the deadlist contents.)
1665 		 */
1666 		struct killarg ka;
1667 
1668 		ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1669 		bplist_close(&ds->ds_deadlist);
1670 		bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1671 		ds->ds_phys->ds_deadlist_obj = 0;
1672 
1673 		/*
1674 		 * Free everything that we point to (that's born after
1675 		 * the previous snapshot, if we are a clone)
1676 		 *
1677 		 * NB: this should be very quick, because we already
1678 		 * freed all the objects in open context.
1679 		 */
1680 		ka.ds = ds;
1681 		ka.zio = zio;
1682 		ka.tx = tx;
1683 		err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1684 		    TRAVERSE_POST, kill_blkptr, &ka);
1685 		ASSERT3U(err, ==, 0);
1686 		ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1687 		    ds->ds_phys->ds_unique_bytes == 0);
1688 	}
1689 
1690 	err = zio_wait(zio);
1691 	ASSERT3U(err, ==, 0);
1692 
1693 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1694 		/* Erase the link in the dir */
1695 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1696 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1697 		ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1698 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1699 		ASSERT(err == 0);
1700 	} else {
1701 		/* remove from snapshot namespace */
1702 		dsl_dataset_t *ds_head;
1703 		ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1704 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1705 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1706 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1707 #ifdef ZFS_DEBUG
1708 		{
1709 			uint64_t val;
1710 
1711 			err = dsl_dataset_snap_lookup(ds_head,
1712 			    ds->ds_snapname, &val);
1713 			ASSERT3U(err, ==, 0);
1714 			ASSERT3U(val, ==, obj);
1715 		}
1716 #endif
1717 		err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1718 		ASSERT(err == 0);
1719 		dsl_dataset_rele(ds_head, FTAG);
1720 	}
1721 
1722 	if (ds_prev && ds->ds_prev != ds_prev)
1723 		dsl_dataset_rele(ds_prev, FTAG);
1724 
1725 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1726 	spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1727 	    cr, "dataset = %llu", ds->ds_object);
1728 
1729 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1730 		uint64_t count;
1731 		ASSERT(0 == zap_count(mos,
1732 		    ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1733 		VERIFY(0 == dmu_object_free(mos,
1734 		    ds->ds_phys->ds_next_clones_obj, tx));
1735 	}
1736 	if (ds->ds_phys->ds_props_obj != 0)
1737 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1738 	dsl_dir_close(ds->ds_dir, ds);
1739 	ds->ds_dir = NULL;
1740 	dsl_dataset_drain_refs(ds, tag);
1741 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1742 }
1743 
1744 static int
1745 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1746 {
1747 	uint64_t asize;
1748 
1749 	if (!dmu_tx_is_syncing(tx))
1750 		return (0);
1751 
1752 	/*
1753 	 * If there's an fs-only reservation, any blocks that might become
1754 	 * owned by the snapshot dataset must be accommodated by space
1755 	 * outside of the reservation.
1756 	 */
1757 	asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1758 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1759 		return (ENOSPC);
1760 
1761 	/*
1762 	 * Propogate any reserved space for this snapshot to other
1763 	 * snapshot checks in this sync group.
1764 	 */
1765 	if (asize > 0)
1766 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1767 
1768 	return (0);
1769 }
1770 
1771 /* ARGSUSED */
1772 int
1773 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1774 {
1775 	dsl_dataset_t *ds = arg1;
1776 	const char *snapname = arg2;
1777 	int err;
1778 	uint64_t value;
1779 
1780 	/*
1781 	 * We don't allow multiple snapshots of the same txg.  If there
1782 	 * is already one, try again.
1783 	 */
1784 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1785 		return (EAGAIN);
1786 
1787 	/*
1788 	 * Check for conflicting name snapshot name.
1789 	 */
1790 	err = dsl_dataset_snap_lookup(ds, snapname, &value);
1791 	if (err == 0)
1792 		return (EEXIST);
1793 	if (err != ENOENT)
1794 		return (err);
1795 
1796 	/*
1797 	 * Check that the dataset's name is not too long.  Name consists
1798 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1799 	 */
1800 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1801 		return (ENAMETOOLONG);
1802 
1803 	err = dsl_dataset_snapshot_reserve_space(ds, tx);
1804 	if (err)
1805 		return (err);
1806 
1807 	ds->ds_trysnap_txg = tx->tx_txg;
1808 	return (0);
1809 }
1810 
1811 void
1812 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1813 {
1814 	dsl_dataset_t *ds = arg1;
1815 	const char *snapname = arg2;
1816 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1817 	dmu_buf_t *dbuf;
1818 	dsl_dataset_phys_t *dsphys;
1819 	uint64_t dsobj, crtxg;
1820 	objset_t *mos = dp->dp_meta_objset;
1821 	int err;
1822 
1823 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1824 
1825 	/*
1826 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1827 	 */
1828 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1829 		crtxg = 1;
1830 	else
1831 		crtxg = tx->tx_txg;
1832 
1833 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1834 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1835 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1836 	dmu_buf_will_dirty(dbuf, tx);
1837 	dsphys = dbuf->db_data;
1838 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1839 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1840 	dsphys->ds_fsid_guid = unique_create();
1841 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1842 	    sizeof (dsphys->ds_guid));
1843 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1844 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1845 	dsphys->ds_next_snap_obj = ds->ds_object;
1846 	dsphys->ds_num_children = 1;
1847 	dsphys->ds_creation_time = gethrestime_sec();
1848 	dsphys->ds_creation_txg = crtxg;
1849 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1850 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1851 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1852 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1853 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1854 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1855 	dmu_buf_rele(dbuf, FTAG);
1856 
1857 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1858 	if (ds->ds_prev) {
1859 		uint64_t next_clones_obj =
1860 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1861 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1862 		    ds->ds_object ||
1863 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1864 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1865 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1866 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1867 			    ds->ds_prev->ds_phys->ds_creation_txg);
1868 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1869 		} else if (next_clones_obj != 0) {
1870 			VERIFY3U(0, ==, zap_remove_int(mos,
1871 			    next_clones_obj, dsphys->ds_next_snap_obj, tx));
1872 			VERIFY3U(0, ==, zap_add_int(mos,
1873 			    next_clones_obj, dsobj, tx));
1874 		}
1875 	}
1876 
1877 	/*
1878 	 * If we have a reference-reservation on this dataset, we will
1879 	 * need to increase the amount of refreservation being charged
1880 	 * since our unique space is going to zero.
1881 	 */
1882 	if (ds->ds_reserved) {
1883 		int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1884 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1885 		    add, 0, 0, tx);
1886 	}
1887 
1888 	bplist_close(&ds->ds_deadlist);
1889 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1890 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1891 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1892 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1893 	ds->ds_phys->ds_unique_bytes = 0;
1894 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1895 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1896 	ds->ds_phys->ds_deadlist_obj =
1897 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1898 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1899 	    ds->ds_phys->ds_deadlist_obj));
1900 
1901 	dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1902 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1903 	    snapname, 8, 1, &dsobj, tx);
1904 	ASSERT(err == 0);
1905 
1906 	if (ds->ds_prev)
1907 		dsl_dataset_drop_ref(ds->ds_prev, ds);
1908 	VERIFY(0 == dsl_dataset_get_ref(dp,
1909 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1910 
1911 	dsl_pool_ds_snapshotted(ds, tx);
1912 
1913 	spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1914 	    "dataset = %llu", dsobj);
1915 }
1916 
1917 void
1918 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1919 {
1920 	ASSERT(dmu_tx_is_syncing(tx));
1921 	ASSERT(ds->ds_user_ptr != NULL);
1922 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1923 
1924 	/*
1925 	 * in case we had to change ds_fsid_guid when we opened it,
1926 	 * sync it out now.
1927 	 */
1928 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1929 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1930 
1931 	dsl_dir_dirty(ds->ds_dir, tx);
1932 	dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1933 }
1934 
1935 void
1936 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1937 {
1938 	uint64_t refd, avail, uobjs, aobjs;
1939 
1940 	dsl_dir_stats(ds->ds_dir, nv);
1941 
1942 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1943 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1944 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1945 
1946 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1947 	    ds->ds_phys->ds_creation_time);
1948 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1949 	    ds->ds_phys->ds_creation_txg);
1950 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1951 	    ds->ds_quota);
1952 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1953 	    ds->ds_reserved);
1954 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1955 	    ds->ds_phys->ds_guid);
1956 
1957 	if (ds->ds_phys->ds_next_snap_obj) {
1958 		/*
1959 		 * This is a snapshot; override the dd's space used with
1960 		 * our unique space and compression ratio.
1961 		 */
1962 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1963 		    ds->ds_phys->ds_unique_bytes);
1964 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1965 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1966 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1967 		    ds->ds_phys->ds_compressed_bytes));
1968 	}
1969 }
1970 
1971 void
1972 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1973 {
1974 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1975 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1976 	stat->dds_guid = ds->ds_phys->ds_guid;
1977 	if (ds->ds_phys->ds_next_snap_obj) {
1978 		stat->dds_is_snapshot = B_TRUE;
1979 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1980 	} else {
1981 		stat->dds_is_snapshot = B_FALSE;
1982 		stat->dds_num_clones = 0;
1983 	}
1984 
1985 	/* clone origin is really a dsl_dir thing... */
1986 	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1987 	if (dsl_dir_is_clone(ds->ds_dir)) {
1988 		dsl_dataset_t *ods;
1989 
1990 		VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
1991 		    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1992 		dsl_dataset_name(ods, stat->dds_origin);
1993 		dsl_dataset_drop_ref(ods, FTAG);
1994 	} else {
1995 		stat->dds_origin[0] = '\0';
1996 	}
1997 	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1998 }
1999 
2000 uint64_t
2001 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2002 {
2003 	return (ds->ds_fsid_guid);
2004 }
2005 
2006 void
2007 dsl_dataset_space(dsl_dataset_t *ds,
2008     uint64_t *refdbytesp, uint64_t *availbytesp,
2009     uint64_t *usedobjsp, uint64_t *availobjsp)
2010 {
2011 	*refdbytesp = ds->ds_phys->ds_used_bytes;
2012 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2013 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2014 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2015 	if (ds->ds_quota != 0) {
2016 		/*
2017 		 * Adjust available bytes according to refquota
2018 		 */
2019 		if (*refdbytesp < ds->ds_quota)
2020 			*availbytesp = MIN(*availbytesp,
2021 			    ds->ds_quota - *refdbytesp);
2022 		else
2023 			*availbytesp = 0;
2024 	}
2025 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2026 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
2027 }
2028 
2029 boolean_t
2030 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2031 {
2032 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2033 
2034 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2035 	    dsl_pool_sync_context(dp));
2036 	if (ds->ds_prev == NULL)
2037 		return (B_FALSE);
2038 	if (ds->ds_phys->ds_bp.blk_birth >
2039 	    ds->ds_prev->ds_phys->ds_creation_txg)
2040 		return (B_TRUE);
2041 	return (B_FALSE);
2042 }
2043 
2044 /* ARGSUSED */
2045 static int
2046 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2047 {
2048 	dsl_dataset_t *ds = arg1;
2049 	char *newsnapname = arg2;
2050 	dsl_dir_t *dd = ds->ds_dir;
2051 	dsl_dataset_t *hds;
2052 	uint64_t val;
2053 	int err;
2054 
2055 	err = dsl_dataset_hold_obj(dd->dd_pool,
2056 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2057 	if (err)
2058 		return (err);
2059 
2060 	/* new name better not be in use */
2061 	err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2062 	dsl_dataset_rele(hds, FTAG);
2063 
2064 	if (err == 0)
2065 		err = EEXIST;
2066 	else if (err == ENOENT)
2067 		err = 0;
2068 
2069 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2070 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2071 		err = ENAMETOOLONG;
2072 
2073 	return (err);
2074 }
2075 
2076 static void
2077 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
2078     cred_t *cr, dmu_tx_t *tx)
2079 {
2080 	dsl_dataset_t *ds = arg1;
2081 	const char *newsnapname = arg2;
2082 	dsl_dir_t *dd = ds->ds_dir;
2083 	objset_t *mos = dd->dd_pool->dp_meta_objset;
2084 	dsl_dataset_t *hds;
2085 	int err;
2086 
2087 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2088 
2089 	VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2090 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2091 
2092 	VERIFY(0 == dsl_dataset_get_snapname(ds));
2093 	err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2094 	ASSERT3U(err, ==, 0);
2095 	mutex_enter(&ds->ds_lock);
2096 	(void) strcpy(ds->ds_snapname, newsnapname);
2097 	mutex_exit(&ds->ds_lock);
2098 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2099 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2100 	ASSERT3U(err, ==, 0);
2101 
2102 	spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2103 	    cr, "dataset = %llu", ds->ds_object);
2104 	dsl_dataset_rele(hds, FTAG);
2105 }
2106 
2107 struct renamesnaparg {
2108 	dsl_sync_task_group_t *dstg;
2109 	char failed[MAXPATHLEN];
2110 	char *oldsnap;
2111 	char *newsnap;
2112 };
2113 
2114 static int
2115 dsl_snapshot_rename_one(char *name, void *arg)
2116 {
2117 	struct renamesnaparg *ra = arg;
2118 	dsl_dataset_t *ds = NULL;
2119 	char *cp;
2120 	int err;
2121 
2122 	cp = name + strlen(name);
2123 	*cp = '@';
2124 	(void) strcpy(cp + 1, ra->oldsnap);
2125 
2126 	/*
2127 	 * For recursive snapshot renames the parent won't be changing
2128 	 * so we just pass name for both the to/from argument.
2129 	 */
2130 	err = zfs_secpolicy_rename_perms(name, name, CRED());
2131 	if (err == ENOENT) {
2132 		return (0);
2133 	} else if (err) {
2134 		(void) strcpy(ra->failed, name);
2135 		return (err);
2136 	}
2137 
2138 #ifdef _KERNEL
2139 	/*
2140 	 * For all filesystems undergoing rename, we'll need to unmount it.
2141 	 */
2142 	(void) zfs_unmount_snap(name, NULL);
2143 #endif
2144 	err = dsl_dataset_hold(name, ra->dstg, &ds);
2145 	*cp = '\0';
2146 	if (err == ENOENT) {
2147 		return (0);
2148 	} else if (err) {
2149 		(void) strcpy(ra->failed, name);
2150 		return (err);
2151 	}
2152 
2153 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2154 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2155 
2156 	return (0);
2157 }
2158 
2159 static int
2160 dsl_recursive_rename(char *oldname, const char *newname)
2161 {
2162 	int err;
2163 	struct renamesnaparg *ra;
2164 	dsl_sync_task_t *dst;
2165 	spa_t *spa;
2166 	char *cp, *fsname = spa_strdup(oldname);
2167 	int len = strlen(oldname);
2168 
2169 	/* truncate the snapshot name to get the fsname */
2170 	cp = strchr(fsname, '@');
2171 	*cp = '\0';
2172 
2173 	err = spa_open(fsname, &spa, FTAG);
2174 	if (err) {
2175 		kmem_free(fsname, len + 1);
2176 		return (err);
2177 	}
2178 	ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2179 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2180 
2181 	ra->oldsnap = strchr(oldname, '@') + 1;
2182 	ra->newsnap = strchr(newname, '@') + 1;
2183 	*ra->failed = '\0';
2184 
2185 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2186 	    DS_FIND_CHILDREN);
2187 	kmem_free(fsname, len + 1);
2188 
2189 	if (err == 0) {
2190 		err = dsl_sync_task_group_wait(ra->dstg);
2191 	}
2192 
2193 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2194 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2195 		dsl_dataset_t *ds = dst->dst_arg1;
2196 		if (dst->dst_err) {
2197 			dsl_dir_name(ds->ds_dir, ra->failed);
2198 			(void) strcat(ra->failed, "@");
2199 			(void) strcat(ra->failed, ra->newsnap);
2200 		}
2201 		dsl_dataset_rele(ds, ra->dstg);
2202 	}
2203 
2204 	if (err)
2205 		(void) strcpy(oldname, ra->failed);
2206 
2207 	dsl_sync_task_group_destroy(ra->dstg);
2208 	kmem_free(ra, sizeof (struct renamesnaparg));
2209 	spa_close(spa, FTAG);
2210 	return (err);
2211 }
2212 
2213 static int
2214 dsl_valid_rename(char *oldname, void *arg)
2215 {
2216 	int delta = *(int *)arg;
2217 
2218 	if (strlen(oldname) + delta >= MAXNAMELEN)
2219 		return (ENAMETOOLONG);
2220 
2221 	return (0);
2222 }
2223 
2224 #pragma weak dmu_objset_rename = dsl_dataset_rename
2225 int
2226 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2227 {
2228 	dsl_dir_t *dd;
2229 	dsl_dataset_t *ds;
2230 	const char *tail;
2231 	int err;
2232 
2233 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2234 	if (err)
2235 		return (err);
2236 	/*
2237 	 * If there are more than 2 references there may be holds
2238 	 * hanging around that haven't been cleared out yet.
2239 	 */
2240 	if (dmu_buf_refcount(dd->dd_dbuf) > 2)
2241 		txg_wait_synced(dd->dd_pool, 0);
2242 	if (tail == NULL) {
2243 		int delta = strlen(newname) - strlen(oldname);
2244 
2245 		/* if we're growing, validate child name lengths */
2246 		if (delta > 0)
2247 			err = dmu_objset_find(oldname, dsl_valid_rename,
2248 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2249 
2250 		if (!err)
2251 			err = dsl_dir_rename(dd, newname);
2252 		dsl_dir_close(dd, FTAG);
2253 		return (err);
2254 	}
2255 	if (tail[0] != '@') {
2256 		/* the name ended in a nonexistant component */
2257 		dsl_dir_close(dd, FTAG);
2258 		return (ENOENT);
2259 	}
2260 
2261 	dsl_dir_close(dd, FTAG);
2262 
2263 	/* new name must be snapshot in same filesystem */
2264 	tail = strchr(newname, '@');
2265 	if (tail == NULL)
2266 		return (EINVAL);
2267 	tail++;
2268 	if (strncmp(oldname, newname, tail - newname) != 0)
2269 		return (EXDEV);
2270 
2271 	if (recursive) {
2272 		err = dsl_recursive_rename(oldname, newname);
2273 	} else {
2274 		err = dsl_dataset_hold(oldname, FTAG, &ds);
2275 		if (err)
2276 			return (err);
2277 
2278 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2279 		    dsl_dataset_snapshot_rename_check,
2280 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2281 
2282 		dsl_dataset_rele(ds, FTAG);
2283 	}
2284 
2285 	return (err);
2286 }
2287 
2288 struct promotenode {
2289 	list_node_t link;
2290 	dsl_dataset_t *ds;
2291 };
2292 
2293 struct promotearg {
2294 	list_t shared_snaps, origin_snaps, clone_snaps;
2295 	dsl_dataset_t *origin_origin, *origin_head;
2296 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2297 };
2298 
2299 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2300 
2301 /* ARGSUSED */
2302 static int
2303 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2304 {
2305 	dsl_dataset_t *hds = arg1;
2306 	struct promotearg *pa = arg2;
2307 	struct promotenode *snap = list_head(&pa->shared_snaps);
2308 	dsl_dataset_t *origin_ds = snap->ds;
2309 	int err;
2310 
2311 	/* Check that it is a real clone */
2312 	if (!dsl_dir_is_clone(hds->ds_dir))
2313 		return (EINVAL);
2314 
2315 	/* Since this is so expensive, don't do the preliminary check */
2316 	if (!dmu_tx_is_syncing(tx))
2317 		return (0);
2318 
2319 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2320 		return (EXDEV);
2321 
2322 	/* compute origin's new unique space */
2323 	snap = list_tail(&pa->clone_snaps);
2324 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2325 	err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2326 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, &pa->unique);
2327 	if (err)
2328 		return (err);
2329 
2330 	/*
2331 	 * Walk the snapshots that we are moving
2332 	 *
2333 	 * Compute space to transfer.  Consider the incremental changes
2334 	 * to used for each snapshot:
2335 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2336 	 * So each snapshot gave birth to:
2337 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2338 	 * So a sequence would look like:
2339 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2340 	 * Which simplifies to:
2341 	 * uN + kN + kN-1 + ... + k1 + k0
2342 	 * Note however, if we stop before we reach the ORIGIN we get:
2343 	 * uN + kN + kN-1 + ... + kM - uM-1
2344 	 */
2345 	pa->used = origin_ds->ds_phys->ds_used_bytes;
2346 	pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2347 	pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2348 	for (snap = list_head(&pa->shared_snaps); snap;
2349 	    snap = list_next(&pa->shared_snaps, snap)) {
2350 		uint64_t val, dlused, dlcomp, dluncomp;
2351 		dsl_dataset_t *ds = snap->ds;
2352 
2353 		/* Check that the snapshot name does not conflict */
2354 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2355 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2356 		if (err == 0)
2357 			return (EEXIST);
2358 		if (err != ENOENT)
2359 			return (err);
2360 
2361 		/* The very first snapshot does not have a deadlist */
2362 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2363 			continue;
2364 
2365 		if (err = bplist_space(&ds->ds_deadlist,
2366 		    &dlused, &dlcomp, &dluncomp))
2367 			return (err);
2368 		pa->used += dlused;
2369 		pa->comp += dlcomp;
2370 		pa->uncomp += dluncomp;
2371 	}
2372 
2373 	/*
2374 	 * If we are a clone of a clone then we never reached ORIGIN,
2375 	 * so we need to subtract out the clone origin's used space.
2376 	 */
2377 	if (pa->origin_origin) {
2378 		pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2379 		pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2380 		pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2381 	}
2382 
2383 	/* Check that there is enough space here */
2384 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2385 	    pa->used);
2386 	if (err)
2387 		return (err);
2388 
2389 	/*
2390 	 * Compute the amounts of space that will be used by snapshots
2391 	 * after the promotion (for both origin and clone).  For each,
2392 	 * it is the amount of space that will be on all of their
2393 	 * deadlists (that was not born before their new origin).
2394 	 */
2395 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2396 		uint64_t space;
2397 
2398 		/*
2399 		 * Note, typically this will not be a clone of a clone,
2400 		 * so snap->ds->ds_origin_txg will be < TXG_INITIAL, so
2401 		 * these snaplist_space() -> bplist_space_birthrange()
2402 		 * calls will be fast because they do not have to
2403 		 * iterate over all bps.
2404 		 */
2405 		snap = list_head(&pa->origin_snaps);
2406 		err = snaplist_space(&pa->shared_snaps,
2407 		    snap->ds->ds_origin_txg, &pa->cloneusedsnap);
2408 		if (err)
2409 			return (err);
2410 
2411 		err = snaplist_space(&pa->clone_snaps,
2412 		    snap->ds->ds_origin_txg, &space);
2413 		if (err)
2414 			return (err);
2415 		pa->cloneusedsnap += space;
2416 	}
2417 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2418 		err = snaplist_space(&pa->origin_snaps,
2419 		    origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2420 		if (err)
2421 			return (err);
2422 	}
2423 
2424 	return (0);
2425 }
2426 
2427 static void
2428 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2429 {
2430 	dsl_dataset_t *hds = arg1;
2431 	struct promotearg *pa = arg2;
2432 	struct promotenode *snap = list_head(&pa->shared_snaps);
2433 	dsl_dataset_t *origin_ds = snap->ds;
2434 	dsl_dataset_t *origin_head;
2435 	dsl_dir_t *dd = hds->ds_dir;
2436 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2437 	dsl_dir_t *odd = NULL;
2438 	uint64_t oldnext_obj;
2439 	int64_t delta;
2440 
2441 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2442 
2443 	snap = list_head(&pa->origin_snaps);
2444 	origin_head = snap->ds;
2445 
2446 	/*
2447 	 * We need to explicitly open odd, since origin_ds's dd will be
2448 	 * changing.
2449 	 */
2450 	VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2451 	    NULL, FTAG, &odd));
2452 
2453 	/* change origin's next snap */
2454 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2455 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2456 	snap = list_tail(&pa->clone_snaps);
2457 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2458 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2459 
2460 	/* change the origin's next clone */
2461 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2462 		VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2463 		    origin_ds->ds_phys->ds_next_clones_obj,
2464 		    origin_ds->ds_phys->ds_next_snap_obj, tx));
2465 		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2466 		    origin_ds->ds_phys->ds_next_clones_obj,
2467 		    oldnext_obj, tx));
2468 	}
2469 
2470 	/* change origin */
2471 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2472 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2473 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2474 	hds->ds_origin_txg = origin_head->ds_origin_txg;
2475 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2476 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2477 	origin_head->ds_origin_txg = origin_ds->ds_phys->ds_creation_txg;
2478 
2479 	/* move snapshots to this dir */
2480 	for (snap = list_head(&pa->shared_snaps); snap;
2481 	    snap = list_next(&pa->shared_snaps, snap)) {
2482 		dsl_dataset_t *ds = snap->ds;
2483 
2484 		/* unregister props as dsl_dir is changing */
2485 		if (ds->ds_user_ptr) {
2486 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
2487 			ds->ds_user_ptr = NULL;
2488 		}
2489 		/* move snap name entry */
2490 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2491 		VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2492 		    ds->ds_snapname, tx));
2493 		VERIFY(0 == zap_add(dp->dp_meta_objset,
2494 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2495 		    8, 1, &ds->ds_object, tx));
2496 		/* change containing dsl_dir */
2497 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2498 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2499 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2500 		ASSERT3P(ds->ds_dir, ==, odd);
2501 		dsl_dir_close(ds->ds_dir, ds);
2502 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2503 		    NULL, ds, &ds->ds_dir));
2504 
2505 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2506 	}
2507 
2508 	/*
2509 	 * Change space accounting.
2510 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2511 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2512 	 * is true for each of {clone,origin} independently.
2513 	 */
2514 
2515 	delta = pa->cloneusedsnap -
2516 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2517 	ASSERT3S(delta, >=, 0);
2518 	ASSERT3U(pa->used, >=, delta);
2519 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2520 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2521 	    pa->used - delta, pa->comp, pa->uncomp, tx);
2522 
2523 	delta = pa->originusedsnap -
2524 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2525 	ASSERT3S(delta, <=, 0);
2526 	ASSERT3U(pa->used, >=, -delta);
2527 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2528 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2529 	    -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2530 
2531 	origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2532 
2533 	/* log history record */
2534 	spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2535 	    cr, "dataset = %llu", hds->ds_object);
2536 
2537 	dsl_dir_close(odd, FTAG);
2538 }
2539 
2540 static char *snaplist_tag = "snaplist";
2541 /*
2542  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2543  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2544  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2545  * snapshots back to this dataset's origin.
2546  */
2547 static int
2548 snaplist_make(dsl_pool_t *dp, boolean_t own,
2549     uint64_t first_obj, uint64_t last_obj, list_t *l)
2550 {
2551 	uint64_t obj = last_obj;
2552 
2553 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2554 
2555 	list_create(l, sizeof (struct promotenode),
2556 	    offsetof(struct promotenode, link));
2557 
2558 	while (obj != first_obj) {
2559 		dsl_dataset_t *ds;
2560 		struct promotenode *snap;
2561 		int err;
2562 
2563 		if (own) {
2564 			err = dsl_dataset_own_obj(dp, obj,
2565 			    0, snaplist_tag, &ds);
2566 			if (err == 0)
2567 				dsl_dataset_make_exclusive(ds, snaplist_tag);
2568 		} else {
2569 			err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2570 		}
2571 		if (err == ENOENT) {
2572 			/* lost race with snapshot destroy */
2573 			struct promotenode *last = list_tail(l);
2574 			ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2575 			obj = last->ds->ds_phys->ds_prev_snap_obj;
2576 			continue;
2577 		} else if (err) {
2578 			return (err);
2579 		}
2580 
2581 		if (first_obj == 0)
2582 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2583 
2584 		snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2585 		snap->ds = ds;
2586 		list_insert_tail(l, snap);
2587 		obj = ds->ds_phys->ds_prev_snap_obj;
2588 	}
2589 
2590 	return (0);
2591 }
2592 
2593 static int
2594 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2595 {
2596 	struct promotenode *snap;
2597 
2598 	*spacep = 0;
2599 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2600 		uint64_t used;
2601 		int err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2602 		    mintxg, UINT64_MAX, &used);
2603 		if (err)
2604 			return (err);
2605 		*spacep += used;
2606 	}
2607 	return (0);
2608 }
2609 
2610 static void
2611 snaplist_destroy(list_t *l, boolean_t own)
2612 {
2613 	struct promotenode *snap;
2614 
2615 	if (!l || !list_link_active(&l->list_head))
2616 		return;
2617 
2618 	while ((snap = list_tail(l)) != NULL) {
2619 		list_remove(l, snap);
2620 		if (own)
2621 			dsl_dataset_disown(snap->ds, snaplist_tag);
2622 		else
2623 			dsl_dataset_rele(snap->ds, snaplist_tag);
2624 		kmem_free(snap, sizeof (struct promotenode));
2625 	}
2626 	list_destroy(l);
2627 }
2628 
2629 /*
2630  * Promote a clone.  Nomenclature note:
2631  * "clone" or "cds": the original clone which is being promoted
2632  * "origin" or "ods": the snapshot which is originally clone's origin
2633  * "origin head" or "ohds": the dataset which is the head
2634  * (filesystem/volume) for the origin
2635  * "origin origin": the origin of the origin's filesystem (typically
2636  * NULL, indicating that the clone is not a clone of a clone).
2637  */
2638 int
2639 dsl_dataset_promote(const char *name)
2640 {
2641 	dsl_dataset_t *ds;
2642 	dsl_dir_t *dd;
2643 	dsl_pool_t *dp;
2644 	dmu_object_info_t doi;
2645 	struct promotearg pa = { 0 };
2646 	struct promotenode *snap;
2647 	int err;
2648 
2649 	err = dsl_dataset_hold(name, FTAG, &ds);
2650 	if (err)
2651 		return (err);
2652 	dd = ds->ds_dir;
2653 	dp = dd->dd_pool;
2654 
2655 	err = dmu_object_info(dp->dp_meta_objset,
2656 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2657 	if (err) {
2658 		dsl_dataset_rele(ds, FTAG);
2659 		return (err);
2660 	}
2661 
2662 	if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2663 		dsl_dataset_rele(ds, FTAG);
2664 		return (EINVAL);
2665 	}
2666 
2667 	/*
2668 	 * We are going to inherit all the snapshots taken before our
2669 	 * origin (i.e., our new origin will be our parent's origin).
2670 	 * Take ownership of them so that we can rename them into our
2671 	 * namespace.
2672 	 */
2673 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2674 
2675 	err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2676 	    &pa.shared_snaps);
2677 	if (err != 0)
2678 		goto out;
2679 
2680 	err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2681 	if (err != 0)
2682 		goto out;
2683 
2684 	snap = list_head(&pa.shared_snaps);
2685 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2686 	err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2687 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2688 	if (err != 0)
2689 		goto out;
2690 
2691 	if (dsl_dir_is_clone(snap->ds->ds_dir)) {
2692 		err = dsl_dataset_own_obj(dp,
2693 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2694 		    0, FTAG, &pa.origin_origin);
2695 		if (err != 0)
2696 			goto out;
2697 	}
2698 
2699 out:
2700 	rw_exit(&dp->dp_config_rwlock);
2701 
2702 	/*
2703 	 * Add in 128x the snapnames zapobj size, since we will be moving
2704 	 * a bunch of snapnames to the promoted ds, and dirtying their
2705 	 * bonus buffers.
2706 	 */
2707 	if (err == 0) {
2708 		err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2709 		    dsl_dataset_promote_sync, ds, &pa,
2710 		    2 + 2 * doi.doi_physical_blks);
2711 	}
2712 
2713 	snaplist_destroy(&pa.shared_snaps, B_TRUE);
2714 	snaplist_destroy(&pa.clone_snaps, B_FALSE);
2715 	snaplist_destroy(&pa.origin_snaps, B_FALSE);
2716 	if (pa.origin_origin)
2717 		dsl_dataset_disown(pa.origin_origin, FTAG);
2718 	dsl_dataset_rele(ds, FTAG);
2719 	return (err);
2720 }
2721 
2722 struct cloneswaparg {
2723 	dsl_dataset_t *cds; /* clone dataset */
2724 	dsl_dataset_t *ohds; /* origin's head dataset */
2725 	boolean_t force;
2726 	int64_t unused_refres_delta; /* change in unconsumed refreservation */
2727 };
2728 
2729 /* ARGSUSED */
2730 static int
2731 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2732 {
2733 	struct cloneswaparg *csa = arg1;
2734 
2735 	/* they should both be heads */
2736 	if (dsl_dataset_is_snapshot(csa->cds) ||
2737 	    dsl_dataset_is_snapshot(csa->ohds))
2738 		return (EINVAL);
2739 
2740 	/* the branch point should be just before them */
2741 	if (csa->cds->ds_prev != csa->ohds->ds_prev)
2742 		return (EINVAL);
2743 
2744 	/* cds should be the clone */
2745 	if (csa->cds->ds_prev->ds_phys->ds_next_snap_obj !=
2746 	    csa->ohds->ds_object)
2747 		return (EINVAL);
2748 
2749 	/* the clone should be a child of the origin */
2750 	if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2751 		return (EINVAL);
2752 
2753 	/* ohds shouldn't be modified unless 'force' */
2754 	if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2755 		return (ETXTBSY);
2756 
2757 	/* adjust amount of any unconsumed refreservation */
2758 	csa->unused_refres_delta =
2759 	    (int64_t)MIN(csa->ohds->ds_reserved,
2760 	    csa->ohds->ds_phys->ds_unique_bytes) -
2761 	    (int64_t)MIN(csa->ohds->ds_reserved,
2762 	    csa->cds->ds_phys->ds_unique_bytes);
2763 
2764 	if (csa->unused_refres_delta > 0 &&
2765 	    csa->unused_refres_delta >
2766 	    dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
2767 		return (ENOSPC);
2768 
2769 	return (0);
2770 }
2771 
2772 /* ARGSUSED */
2773 static void
2774 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2775 {
2776 	struct cloneswaparg *csa = arg1;
2777 	dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2778 
2779 	ASSERT(csa->cds->ds_reserved == 0);
2780 	ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota);
2781 
2782 	dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2783 	dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2784 	dmu_buf_will_dirty(csa->cds->ds_prev->ds_dbuf, tx);
2785 
2786 	if (csa->cds->ds_user_ptr != NULL) {
2787 		csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr);
2788 		csa->cds->ds_user_ptr = NULL;
2789 	}
2790 
2791 	if (csa->ohds->ds_user_ptr != NULL) {
2792 		csa->ohds->ds_user_evict_func(csa->ohds,
2793 		    csa->ohds->ds_user_ptr);
2794 		csa->ohds->ds_user_ptr = NULL;
2795 	}
2796 
2797 	/* reset origin's unique bytes */
2798 	VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2799 	    csa->cds->ds_prev->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2800 	    &csa->cds->ds_prev->ds_phys->ds_unique_bytes));
2801 
2802 	/* swap blkptrs */
2803 	{
2804 		blkptr_t tmp;
2805 		tmp = csa->ohds->ds_phys->ds_bp;
2806 		csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2807 		csa->cds->ds_phys->ds_bp = tmp;
2808 	}
2809 
2810 	/* set dd_*_bytes */
2811 	{
2812 		int64_t dused, dcomp, duncomp;
2813 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2814 		uint64_t odl_used, odl_comp, odl_uncomp;
2815 
2816 		ASSERT3U(csa->cds->ds_dir->dd_phys->
2817 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2818 
2819 		VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2820 		    &cdl_comp, &cdl_uncomp));
2821 		VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2822 		    &odl_comp, &odl_uncomp));
2823 
2824 		dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2825 		    (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2826 		dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2827 		    (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2828 		duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2829 		    cdl_uncomp -
2830 		    (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2831 
2832 		dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
2833 		    dused, dcomp, duncomp, tx);
2834 		dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
2835 		    -dused, -dcomp, -duncomp, tx);
2836 
2837 		/*
2838 		 * The difference in the space used by snapshots is the
2839 		 * difference in snapshot space due to the head's
2840 		 * deadlist (since that's the only thing that's
2841 		 * changing that affects the snapused).
2842 		 */
2843 		VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2844 		    csa->ohds->ds_origin_txg, UINT64_MAX, &cdl_used));
2845 		VERIFY(0 == bplist_space_birthrange(&csa->ohds->ds_deadlist,
2846 		    csa->ohds->ds_origin_txg, UINT64_MAX, &odl_used));
2847 		dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
2848 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2849 	}
2850 
2851 #define	SWITCH64(x, y) \
2852 	{ \
2853 		uint64_t __tmp = (x); \
2854 		(x) = (y); \
2855 		(y) = __tmp; \
2856 	}
2857 
2858 	/* swap ds_*_bytes */
2859 	SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2860 	    csa->cds->ds_phys->ds_used_bytes);
2861 	SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2862 	    csa->cds->ds_phys->ds_compressed_bytes);
2863 	SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2864 	    csa->cds->ds_phys->ds_uncompressed_bytes);
2865 	SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
2866 	    csa->cds->ds_phys->ds_unique_bytes);
2867 
2868 	/* apply any parent delta for change in unconsumed refreservation */
2869 	dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
2870 	    csa->unused_refres_delta, 0, 0, tx);
2871 
2872 	/* swap deadlists */
2873 	bplist_close(&csa->cds->ds_deadlist);
2874 	bplist_close(&csa->ohds->ds_deadlist);
2875 	SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2876 	    csa->cds->ds_phys->ds_deadlist_obj);
2877 	VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2878 	    csa->cds->ds_phys->ds_deadlist_obj));
2879 	VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2880 	    csa->ohds->ds_phys->ds_deadlist_obj));
2881 
2882 	dsl_pool_ds_clone_swapped(csa->ohds, csa->cds, tx);
2883 }
2884 
2885 /*
2886  * Swap 'clone' with its origin head file system.  Used at the end
2887  * of "online recv" to swizzle the file system to the new version.
2888  */
2889 int
2890 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2891     boolean_t force)
2892 {
2893 	struct cloneswaparg csa;
2894 	int error;
2895 
2896 	ASSERT(clone->ds_owner);
2897 	ASSERT(origin_head->ds_owner);
2898 retry:
2899 	/* Need exclusive access for the swap */
2900 	rw_enter(&clone->ds_rwlock, RW_WRITER);
2901 	if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
2902 		rw_exit(&clone->ds_rwlock);
2903 		rw_enter(&origin_head->ds_rwlock, RW_WRITER);
2904 		if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
2905 			rw_exit(&origin_head->ds_rwlock);
2906 			goto retry;
2907 		}
2908 	}
2909 	csa.cds = clone;
2910 	csa.ohds = origin_head;
2911 	csa.force = force;
2912 	error = dsl_sync_task_do(clone->ds_dir->dd_pool,
2913 	    dsl_dataset_clone_swap_check,
2914 	    dsl_dataset_clone_swap_sync, &csa, NULL, 9);
2915 	return (error);
2916 }
2917 
2918 /*
2919  * Given a pool name and a dataset object number in that pool,
2920  * return the name of that dataset.
2921  */
2922 int
2923 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2924 {
2925 	spa_t *spa;
2926 	dsl_pool_t *dp;
2927 	dsl_dataset_t *ds;
2928 	int error;
2929 
2930 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
2931 		return (error);
2932 	dp = spa_get_dsl(spa);
2933 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2934 	if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
2935 		dsl_dataset_name(ds, buf);
2936 		dsl_dataset_rele(ds, FTAG);
2937 	}
2938 	rw_exit(&dp->dp_config_rwlock);
2939 	spa_close(spa, FTAG);
2940 
2941 	return (error);
2942 }
2943 
2944 int
2945 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2946     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2947 {
2948 	int error = 0;
2949 
2950 	ASSERT3S(asize, >, 0);
2951 
2952 	/*
2953 	 * *ref_rsrv is the portion of asize that will come from any
2954 	 * unconsumed refreservation space.
2955 	 */
2956 	*ref_rsrv = 0;
2957 
2958 	mutex_enter(&ds->ds_lock);
2959 	/*
2960 	 * Make a space adjustment for reserved bytes.
2961 	 */
2962 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2963 		ASSERT3U(*used, >=,
2964 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2965 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2966 		*ref_rsrv =
2967 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2968 	}
2969 
2970 	if (!check_quota || ds->ds_quota == 0) {
2971 		mutex_exit(&ds->ds_lock);
2972 		return (0);
2973 	}
2974 	/*
2975 	 * If they are requesting more space, and our current estimate
2976 	 * is over quota, they get to try again unless the actual
2977 	 * on-disk is over quota and there are no pending changes (which
2978 	 * may free up space for us).
2979 	 */
2980 	if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
2981 		if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
2982 			error = ERESTART;
2983 		else
2984 			error = EDQUOT;
2985 	}
2986 	mutex_exit(&ds->ds_lock);
2987 
2988 	return (error);
2989 }
2990 
2991 /* ARGSUSED */
2992 static int
2993 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
2994 {
2995 	dsl_dataset_t *ds = arg1;
2996 	uint64_t *quotap = arg2;
2997 	uint64_t new_quota = *quotap;
2998 
2999 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3000 		return (ENOTSUP);
3001 
3002 	if (new_quota == 0)
3003 		return (0);
3004 
3005 	if (new_quota < ds->ds_phys->ds_used_bytes ||
3006 	    new_quota < ds->ds_reserved)
3007 		return (ENOSPC);
3008 
3009 	return (0);
3010 }
3011 
3012 /* ARGSUSED */
3013 void
3014 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3015 {
3016 	dsl_dataset_t *ds = arg1;
3017 	uint64_t *quotap = arg2;
3018 	uint64_t new_quota = *quotap;
3019 
3020 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3021 
3022 	ds->ds_quota = new_quota;
3023 
3024 	dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
3025 
3026 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
3027 	    tx, cr, "%lld dataset = %llu ",
3028 	    (longlong_t)new_quota, ds->ds_object);
3029 }
3030 
3031 int
3032 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
3033 {
3034 	dsl_dataset_t *ds;
3035 	int err;
3036 
3037 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3038 	if (err)
3039 		return (err);
3040 
3041 	if (quota != ds->ds_quota) {
3042 		/*
3043 		 * If someone removes a file, then tries to set the quota, we
3044 		 * want to make sure the file freeing takes effect.
3045 		 */
3046 		txg_wait_open(ds->ds_dir->dd_pool, 0);
3047 
3048 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3049 		    dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3050 		    ds, &quota, 0);
3051 	}
3052 	dsl_dataset_rele(ds, FTAG);
3053 	return (err);
3054 }
3055 
3056 static int
3057 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3058 {
3059 	dsl_dataset_t *ds = arg1;
3060 	uint64_t *reservationp = arg2;
3061 	uint64_t new_reservation = *reservationp;
3062 	uint64_t unique;
3063 
3064 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3065 	    SPA_VERSION_REFRESERVATION)
3066 		return (ENOTSUP);
3067 
3068 	if (dsl_dataset_is_snapshot(ds))
3069 		return (EINVAL);
3070 
3071 	/*
3072 	 * If we are doing the preliminary check in open context, the
3073 	 * space estimates may be inaccurate.
3074 	 */
3075 	if (!dmu_tx_is_syncing(tx))
3076 		return (0);
3077 
3078 	mutex_enter(&ds->ds_lock);
3079 	unique = dsl_dataset_unique(ds);
3080 	mutex_exit(&ds->ds_lock);
3081 
3082 	if (MAX(unique, new_reservation) > MAX(unique, ds->ds_reserved)) {
3083 		uint64_t delta = MAX(unique, new_reservation) -
3084 		    MAX(unique, ds->ds_reserved);
3085 
3086 		if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3087 			return (ENOSPC);
3088 		if (ds->ds_quota > 0 &&
3089 		    new_reservation > ds->ds_quota)
3090 			return (ENOSPC);
3091 	}
3092 
3093 	return (0);
3094 }
3095 
3096 /* ARGSUSED */
3097 static void
3098 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
3099     dmu_tx_t *tx)
3100 {
3101 	dsl_dataset_t *ds = arg1;
3102 	uint64_t *reservationp = arg2;
3103 	uint64_t new_reservation = *reservationp;
3104 	uint64_t unique;
3105 	int64_t delta;
3106 
3107 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3108 
3109 	mutex_enter(&ds->ds_dir->dd_lock);
3110 	mutex_enter(&ds->ds_lock);
3111 	unique = dsl_dataset_unique(ds);
3112 	delta = MAX(0, (int64_t)(new_reservation - unique)) -
3113 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3114 	ds->ds_reserved = new_reservation;
3115 	mutex_exit(&ds->ds_lock);
3116 
3117 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3118 	mutex_exit(&ds->ds_dir->dd_lock);
3119 	dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
3120 	    new_reservation, cr, tx);
3121 
3122 	spa_history_internal_log(LOG_DS_REFRESERV,
3123 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
3124 	    (longlong_t)new_reservation, ds->ds_object);
3125 }
3126 
3127 int
3128 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
3129 {
3130 	dsl_dataset_t *ds;
3131 	int err;
3132 
3133 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3134 	if (err)
3135 		return (err);
3136 
3137 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3138 	    dsl_dataset_set_reservation_check,
3139 	    dsl_dataset_set_reservation_sync, ds, &reservation, 0);
3140 	dsl_dataset_rele(ds, FTAG);
3141 	return (err);
3142 }
3143