xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision f700969ab27181023551541cb5293e922f1587eb)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2013 by Delphix. All rights reserved.
24  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25  * Copyright (c) 2014 RackTop Systems.
26  */
27 
28 #include <sys/dmu_objset.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_prop.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dmu_impl.h>
35 #include <sys/dmu_tx.h>
36 #include <sys/arc.h>
37 #include <sys/zio.h>
38 #include <sys/zap.h>
39 #include <sys/zfeature.h>
40 #include <sys/unique.h>
41 #include <sys/zfs_context.h>
42 #include <sys/zfs_ioctl.h>
43 #include <sys/spa.h>
44 #include <sys/zfs_znode.h>
45 #include <sys/zfs_onexit.h>
46 #include <sys/zvol.h>
47 #include <sys/dsl_scan.h>
48 #include <sys/dsl_deadlist.h>
49 #include <sys/dsl_destroy.h>
50 #include <sys/dsl_userhold.h>
51 #include <sys/dsl_bookmark.h>
52 
53 #define	SWITCH64(x, y) \
54 	{ \
55 		uint64_t __tmp = (x); \
56 		(x) = (y); \
57 		(y) = __tmp; \
58 	}
59 
60 #define	DS_REF_MAX	(1ULL << 62)
61 
62 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
63 
64 /*
65  * Figure out how much of this delta should be propogated to the dsl_dir
66  * layer.  If there's a refreservation, that space has already been
67  * partially accounted for in our ancestors.
68  */
69 static int64_t
70 parent_delta(dsl_dataset_t *ds, int64_t delta)
71 {
72 	uint64_t old_bytes, new_bytes;
73 
74 	if (ds->ds_reserved == 0)
75 		return (delta);
76 
77 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
78 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
79 
80 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
81 	return (new_bytes - old_bytes);
82 }
83 
84 void
85 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
86 {
87 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
88 	int compressed = BP_GET_PSIZE(bp);
89 	int uncompressed = BP_GET_UCSIZE(bp);
90 	int64_t delta;
91 
92 	dprintf_bp(bp, "ds=%p", ds);
93 
94 	ASSERT(dmu_tx_is_syncing(tx));
95 	/* It could have been compressed away to nothing */
96 	if (BP_IS_HOLE(bp))
97 		return;
98 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
99 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
100 	if (ds == NULL) {
101 		dsl_pool_mos_diduse_space(tx->tx_pool,
102 		    used, compressed, uncompressed);
103 		return;
104 	}
105 
106 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
107 	mutex_enter(&ds->ds_lock);
108 	delta = parent_delta(ds, used);
109 	ds->ds_phys->ds_referenced_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 }
119 
120 int
121 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
122     boolean_t async)
123 {
124 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
125 	int compressed = BP_GET_PSIZE(bp);
126 	int uncompressed = BP_GET_UCSIZE(bp);
127 
128 	if (BP_IS_HOLE(bp))
129 		return (0);
130 
131 	ASSERT(dmu_tx_is_syncing(tx));
132 	ASSERT(bp->blk_birth <= tx->tx_txg);
133 
134 	if (ds == NULL) {
135 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
136 		dsl_pool_mos_diduse_space(tx->tx_pool,
137 		    -used, -compressed, -uncompressed);
138 		return (used);
139 	}
140 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
141 
142 	ASSERT(!dsl_dataset_is_snapshot(ds));
143 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
144 
145 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
146 		int64_t delta;
147 
148 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
149 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
150 
151 		mutex_enter(&ds->ds_lock);
152 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
153 		    !DS_UNIQUE_IS_ACCURATE(ds));
154 		delta = parent_delta(ds, -used);
155 		ds->ds_phys->ds_unique_bytes -= used;
156 		mutex_exit(&ds->ds_lock);
157 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
158 		    delta, -compressed, -uncompressed, tx);
159 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
160 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
161 	} else {
162 		dprintf_bp(bp, "putting on dead list: %s", "");
163 		if (async) {
164 			/*
165 			 * We are here as part of zio's write done callback,
166 			 * which means we're a zio interrupt thread.  We can't
167 			 * call dsl_deadlist_insert() now because it may block
168 			 * waiting for I/O.  Instead, put bp on the deferred
169 			 * queue and let dsl_pool_sync() finish the job.
170 			 */
171 			bplist_append(&ds->ds_pending_deadlist, bp);
172 		} else {
173 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
174 		}
175 		ASSERT3U(ds->ds_prev->ds_object, ==,
176 		    ds->ds_phys->ds_prev_snap_obj);
177 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
178 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
179 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
180 		    ds->ds_object && bp->blk_birth >
181 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
182 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
183 			mutex_enter(&ds->ds_prev->ds_lock);
184 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
185 			mutex_exit(&ds->ds_prev->ds_lock);
186 		}
187 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
188 			dsl_dir_transfer_space(ds->ds_dir, used,
189 			    DD_USED_HEAD, DD_USED_SNAP, tx);
190 		}
191 	}
192 	mutex_enter(&ds->ds_lock);
193 	ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
194 	ds->ds_phys->ds_referenced_bytes -= used;
195 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
196 	ds->ds_phys->ds_compressed_bytes -= compressed;
197 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
198 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
199 	mutex_exit(&ds->ds_lock);
200 
201 	return (used);
202 }
203 
204 uint64_t
205 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
206 {
207 	uint64_t trysnap = 0;
208 
209 	if (ds == NULL)
210 		return (0);
211 	/*
212 	 * The snapshot creation could fail, but that would cause an
213 	 * incorrect FALSE return, which would only result in an
214 	 * overestimation of the amount of space that an operation would
215 	 * consume, which is OK.
216 	 *
217 	 * There's also a small window where we could miss a pending
218 	 * snapshot, because we could set the sync task in the quiescing
219 	 * phase.  So this should only be used as a guess.
220 	 */
221 	if (ds->ds_trysnap_txg >
222 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
223 		trysnap = ds->ds_trysnap_txg;
224 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
225 }
226 
227 boolean_t
228 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
229     uint64_t blk_birth)
230 {
231 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
232 	    (bp != NULL && BP_IS_HOLE(bp)))
233 		return (B_FALSE);
234 
235 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
236 
237 	return (B_TRUE);
238 }
239 
240 /* ARGSUSED */
241 static void
242 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
243 {
244 	dsl_dataset_t *ds = dsv;
245 
246 	ASSERT(ds->ds_owner == NULL);
247 
248 	unique_remove(ds->ds_fsid_guid);
249 
250 	if (ds->ds_objset != NULL)
251 		dmu_objset_evict(ds->ds_objset);
252 
253 	if (ds->ds_prev) {
254 		dsl_dataset_rele(ds->ds_prev, ds);
255 		ds->ds_prev = NULL;
256 	}
257 
258 	bplist_destroy(&ds->ds_pending_deadlist);
259 	if (ds->ds_phys->ds_deadlist_obj != 0)
260 		dsl_deadlist_close(&ds->ds_deadlist);
261 	if (ds->ds_dir)
262 		dsl_dir_rele(ds->ds_dir, ds);
263 
264 	ASSERT(!list_link_active(&ds->ds_synced_link));
265 
266 	mutex_destroy(&ds->ds_lock);
267 	mutex_destroy(&ds->ds_opening_lock);
268 	refcount_destroy(&ds->ds_longholds);
269 
270 	kmem_free(ds, sizeof (dsl_dataset_t));
271 }
272 
273 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 != 0)
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 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 int
319 dsl_dataset_snap_remove(dsl_dataset_t *ds, const 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 	dsl_dir_snap_cmtime_update(ds->ds_dir);
327 
328 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
329 		mt = MT_FIRST;
330 	else
331 		mt = MT_EXACT;
332 
333 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
334 	if (err == ENOTSUP && mt == MT_FIRST)
335 		err = zap_remove(mos, snapobj, name, tx);
336 	return (err);
337 }
338 
339 int
340 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
341     dsl_dataset_t **dsp)
342 {
343 	objset_t *mos = dp->dp_meta_objset;
344 	dmu_buf_t *dbuf;
345 	dsl_dataset_t *ds;
346 	int err;
347 	dmu_object_info_t doi;
348 
349 	ASSERT(dsl_pool_config_held(dp));
350 
351 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
352 	if (err != 0)
353 		return (err);
354 
355 	/* Make sure dsobj has the correct object type. */
356 	dmu_object_info_from_db(dbuf, &doi);
357 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
358 		dmu_buf_rele(dbuf, tag);
359 		return (SET_ERROR(EINVAL));
360 	}
361 
362 	ds = dmu_buf_get_user(dbuf);
363 	if (ds == NULL) {
364 		dsl_dataset_t *winner = NULL;
365 
366 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
367 		ds->ds_dbuf = dbuf;
368 		ds->ds_object = dsobj;
369 		ds->ds_phys = dbuf->db_data;
370 
371 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
372 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
373 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
374 		refcount_create(&ds->ds_longholds);
375 
376 		bplist_create(&ds->ds_pending_deadlist);
377 		dsl_deadlist_open(&ds->ds_deadlist,
378 		    mos, ds->ds_phys->ds_deadlist_obj);
379 
380 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
381 		    offsetof(dmu_sendarg_t, dsa_link));
382 
383 		if (err == 0) {
384 			err = dsl_dir_hold_obj(dp,
385 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
386 		}
387 		if (err != 0) {
388 			mutex_destroy(&ds->ds_lock);
389 			mutex_destroy(&ds->ds_opening_lock);
390 			refcount_destroy(&ds->ds_longholds);
391 			bplist_destroy(&ds->ds_pending_deadlist);
392 			dsl_deadlist_close(&ds->ds_deadlist);
393 			kmem_free(ds, sizeof (dsl_dataset_t));
394 			dmu_buf_rele(dbuf, tag);
395 			return (err);
396 		}
397 
398 		if (!dsl_dataset_is_snapshot(ds)) {
399 			ds->ds_snapname[0] = '\0';
400 			if (ds->ds_phys->ds_prev_snap_obj != 0) {
401 				err = dsl_dataset_hold_obj(dp,
402 				    ds->ds_phys->ds_prev_snap_obj,
403 				    ds, &ds->ds_prev);
404 			}
405 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
406 				int zaperr = zap_lookup(mos, ds->ds_object,
407 				    DS_FIELD_BOOKMARK_NAMES,
408 				    sizeof (ds->ds_bookmarks), 1,
409 				    &ds->ds_bookmarks);
410 				if (zaperr != ENOENT)
411 					VERIFY0(zaperr);
412 			}
413 		} else {
414 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
415 				err = dsl_dataset_get_snapname(ds);
416 			if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
417 				err = zap_count(
418 				    ds->ds_dir->dd_pool->dp_meta_objset,
419 				    ds->ds_phys->ds_userrefs_obj,
420 				    &ds->ds_userrefs);
421 			}
422 		}
423 
424 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
425 			err = dsl_prop_get_int_ds(ds,
426 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
427 			    &ds->ds_reserved);
428 			if (err == 0) {
429 				err = dsl_prop_get_int_ds(ds,
430 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
431 				    &ds->ds_quota);
432 			}
433 		} else {
434 			ds->ds_reserved = ds->ds_quota = 0;
435 		}
436 
437 		if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
438 		    &ds->ds_phys, dsl_dataset_evict)) != NULL) {
439 			bplist_destroy(&ds->ds_pending_deadlist);
440 			dsl_deadlist_close(&ds->ds_deadlist);
441 			if (ds->ds_prev)
442 				dsl_dataset_rele(ds->ds_prev, ds);
443 			dsl_dir_rele(ds->ds_dir, ds);
444 			mutex_destroy(&ds->ds_lock);
445 			mutex_destroy(&ds->ds_opening_lock);
446 			refcount_destroy(&ds->ds_longholds);
447 			kmem_free(ds, sizeof (dsl_dataset_t));
448 			if (err != 0) {
449 				dmu_buf_rele(dbuf, tag);
450 				return (err);
451 			}
452 			ds = winner;
453 		} else {
454 			ds->ds_fsid_guid =
455 			    unique_insert(ds->ds_phys->ds_fsid_guid);
456 		}
457 	}
458 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
459 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
460 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
461 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
462 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
463 	*dsp = ds;
464 	return (0);
465 }
466 
467 int
468 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
469     void *tag, dsl_dataset_t **dsp)
470 {
471 	dsl_dir_t *dd;
472 	const char *snapname;
473 	uint64_t obj;
474 	int err = 0;
475 
476 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
477 	if (err != 0)
478 		return (err);
479 
480 	ASSERT(dsl_pool_config_held(dp));
481 	obj = dd->dd_phys->dd_head_dataset_obj;
482 	if (obj != 0)
483 		err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
484 	else
485 		err = SET_ERROR(ENOENT);
486 
487 	/* we may be looking for a snapshot */
488 	if (err == 0 && snapname != NULL) {
489 		dsl_dataset_t *ds;
490 
491 		if (*snapname++ != '@') {
492 			dsl_dataset_rele(*dsp, tag);
493 			dsl_dir_rele(dd, FTAG);
494 			return (SET_ERROR(ENOENT));
495 		}
496 
497 		dprintf("looking for snapshot '%s'\n", snapname);
498 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
499 		if (err == 0)
500 			err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
501 		dsl_dataset_rele(*dsp, tag);
502 
503 		if (err == 0) {
504 			mutex_enter(&ds->ds_lock);
505 			if (ds->ds_snapname[0] == 0)
506 				(void) strlcpy(ds->ds_snapname, snapname,
507 				    sizeof (ds->ds_snapname));
508 			mutex_exit(&ds->ds_lock);
509 			*dsp = ds;
510 		}
511 	}
512 
513 	dsl_dir_rele(dd, FTAG);
514 	return (err);
515 }
516 
517 int
518 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
519     void *tag, dsl_dataset_t **dsp)
520 {
521 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
522 	if (err != 0)
523 		return (err);
524 	if (!dsl_dataset_tryown(*dsp, tag)) {
525 		dsl_dataset_rele(*dsp, tag);
526 		*dsp = NULL;
527 		return (SET_ERROR(EBUSY));
528 	}
529 	return (0);
530 }
531 
532 int
533 dsl_dataset_own(dsl_pool_t *dp, const char *name,
534     void *tag, dsl_dataset_t **dsp)
535 {
536 	int err = dsl_dataset_hold(dp, name, tag, dsp);
537 	if (err != 0)
538 		return (err);
539 	if (!dsl_dataset_tryown(*dsp, tag)) {
540 		dsl_dataset_rele(*dsp, tag);
541 		return (SET_ERROR(EBUSY));
542 	}
543 	return (0);
544 }
545 
546 /*
547  * See the comment above dsl_pool_hold() for details.  In summary, a long
548  * hold is used to prevent destruction of a dataset while the pool hold
549  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
550  *
551  * The dataset and pool must be held when this function is called.  After it
552  * is called, the pool hold may be released while the dataset is still held
553  * and accessed.
554  */
555 void
556 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
557 {
558 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
559 	(void) refcount_add(&ds->ds_longholds, tag);
560 }
561 
562 void
563 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
564 {
565 	(void) refcount_remove(&ds->ds_longholds, tag);
566 }
567 
568 /* Return B_TRUE if there are any long holds on this dataset. */
569 boolean_t
570 dsl_dataset_long_held(dsl_dataset_t *ds)
571 {
572 	return (!refcount_is_zero(&ds->ds_longholds));
573 }
574 
575 void
576 dsl_dataset_name(dsl_dataset_t *ds, char *name)
577 {
578 	if (ds == NULL) {
579 		(void) strcpy(name, "mos");
580 	} else {
581 		dsl_dir_name(ds->ds_dir, name);
582 		VERIFY0(dsl_dataset_get_snapname(ds));
583 		if (ds->ds_snapname[0]) {
584 			(void) strcat(name, "@");
585 			/*
586 			 * We use a "recursive" mutex so that we
587 			 * can call dprintf_ds() with ds_lock held.
588 			 */
589 			if (!MUTEX_HELD(&ds->ds_lock)) {
590 				mutex_enter(&ds->ds_lock);
591 				(void) strcat(name, ds->ds_snapname);
592 				mutex_exit(&ds->ds_lock);
593 			} else {
594 				(void) strcat(name, ds->ds_snapname);
595 			}
596 		}
597 	}
598 }
599 
600 void
601 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
602 {
603 	dmu_buf_rele(ds->ds_dbuf, tag);
604 }
605 
606 void
607 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
608 {
609 	ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL);
610 
611 	mutex_enter(&ds->ds_lock);
612 	ds->ds_owner = NULL;
613 	mutex_exit(&ds->ds_lock);
614 	dsl_dataset_long_rele(ds, tag);
615 	if (ds->ds_dbuf != NULL)
616 		dsl_dataset_rele(ds, tag);
617 	else
618 		dsl_dataset_evict(NULL, ds);
619 }
620 
621 boolean_t
622 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
623 {
624 	boolean_t gotit = FALSE;
625 
626 	mutex_enter(&ds->ds_lock);
627 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
628 		ds->ds_owner = tag;
629 		dsl_dataset_long_hold(ds, tag);
630 		gotit = TRUE;
631 	}
632 	mutex_exit(&ds->ds_lock);
633 	return (gotit);
634 }
635 
636 uint64_t
637 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
638     uint64_t flags, dmu_tx_t *tx)
639 {
640 	dsl_pool_t *dp = dd->dd_pool;
641 	dmu_buf_t *dbuf;
642 	dsl_dataset_phys_t *dsphys;
643 	uint64_t dsobj;
644 	objset_t *mos = dp->dp_meta_objset;
645 
646 	if (origin == NULL)
647 		origin = dp->dp_origin_snap;
648 
649 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
650 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
651 	ASSERT(dmu_tx_is_syncing(tx));
652 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
653 
654 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
655 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
656 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
657 	dmu_buf_will_dirty(dbuf, tx);
658 	dsphys = dbuf->db_data;
659 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
660 	dsphys->ds_dir_obj = dd->dd_object;
661 	dsphys->ds_flags = flags;
662 	dsphys->ds_fsid_guid = unique_create();
663 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
664 	    sizeof (dsphys->ds_guid));
665 	dsphys->ds_snapnames_zapobj =
666 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
667 	    DMU_OT_NONE, 0, tx);
668 	dsphys->ds_creation_time = gethrestime_sec();
669 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
670 
671 	if (origin == NULL) {
672 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
673 	} else {
674 		dsl_dataset_t *ohds; /* head of the origin snapshot */
675 
676 		dsphys->ds_prev_snap_obj = origin->ds_object;
677 		dsphys->ds_prev_snap_txg =
678 		    origin->ds_phys->ds_creation_txg;
679 		dsphys->ds_referenced_bytes =
680 		    origin->ds_phys->ds_referenced_bytes;
681 		dsphys->ds_compressed_bytes =
682 		    origin->ds_phys->ds_compressed_bytes;
683 		dsphys->ds_uncompressed_bytes =
684 		    origin->ds_phys->ds_uncompressed_bytes;
685 		dsphys->ds_bp = origin->ds_phys->ds_bp;
686 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
687 
688 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
689 		origin->ds_phys->ds_num_children++;
690 
691 		VERIFY0(dsl_dataset_hold_obj(dp,
692 		    origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
693 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
694 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
695 		dsl_dataset_rele(ohds, FTAG);
696 
697 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
698 			if (origin->ds_phys->ds_next_clones_obj == 0) {
699 				origin->ds_phys->ds_next_clones_obj =
700 				    zap_create(mos,
701 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
702 			}
703 			VERIFY0(zap_add_int(mos,
704 			    origin->ds_phys->ds_next_clones_obj, dsobj, tx));
705 		}
706 
707 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
708 		dd->dd_phys->dd_origin_obj = origin->ds_object;
709 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
710 			if (origin->ds_dir->dd_phys->dd_clones == 0) {
711 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
712 				origin->ds_dir->dd_phys->dd_clones =
713 				    zap_create(mos,
714 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
715 			}
716 			VERIFY0(zap_add_int(mos,
717 			    origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
718 		}
719 	}
720 
721 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
722 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
723 
724 	dmu_buf_rele(dbuf, FTAG);
725 
726 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
727 	dd->dd_phys->dd_head_dataset_obj = dsobj;
728 
729 	return (dsobj);
730 }
731 
732 static void
733 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
734 {
735 	objset_t *os;
736 
737 	VERIFY0(dmu_objset_from_ds(ds, &os));
738 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
739 	dsl_dataset_dirty(ds, tx);
740 }
741 
742 uint64_t
743 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
744     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
745 {
746 	dsl_pool_t *dp = pdd->dd_pool;
747 	uint64_t dsobj, ddobj;
748 	dsl_dir_t *dd;
749 
750 	ASSERT(dmu_tx_is_syncing(tx));
751 	ASSERT(lastname[0] != '@');
752 
753 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
754 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
755 
756 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
757 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
758 
759 	dsl_deleg_set_create_perms(dd, tx, cr);
760 
761 	dsl_dir_rele(dd, FTAG);
762 
763 	/*
764 	 * If we are creating a clone, make sure we zero out any stale
765 	 * data from the origin snapshots zil header.
766 	 */
767 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
768 		dsl_dataset_t *ds;
769 
770 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
771 		dsl_dataset_zero_zil(ds, tx);
772 		dsl_dataset_rele(ds, FTAG);
773 	}
774 
775 	return (dsobj);
776 }
777 
778 /*
779  * The unique space in the head dataset can be calculated by subtracting
780  * the space used in the most recent snapshot, that is still being used
781  * in this file system, from the space currently in use.  To figure out
782  * the space in the most recent snapshot still in use, we need to take
783  * the total space used in the snapshot and subtract out the space that
784  * has been freed up since the snapshot was taken.
785  */
786 void
787 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
788 {
789 	uint64_t mrs_used;
790 	uint64_t dlused, dlcomp, dluncomp;
791 
792 	ASSERT(!dsl_dataset_is_snapshot(ds));
793 
794 	if (ds->ds_phys->ds_prev_snap_obj != 0)
795 		mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
796 	else
797 		mrs_used = 0;
798 
799 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
800 
801 	ASSERT3U(dlused, <=, mrs_used);
802 	ds->ds_phys->ds_unique_bytes =
803 	    ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
804 
805 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
806 	    SPA_VERSION_UNIQUE_ACCURATE)
807 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
808 }
809 
810 void
811 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
812     dmu_tx_t *tx)
813 {
814 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
815 	uint64_t count;
816 	int err;
817 
818 	ASSERT(ds->ds_phys->ds_num_children >= 2);
819 	err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
820 	/*
821 	 * The err should not be ENOENT, but a bug in a previous version
822 	 * of the code could cause upgrade_clones_cb() to not set
823 	 * ds_next_snap_obj when it should, leading to a missing entry.
824 	 * If we knew that the pool was created after
825 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
826 	 * ENOENT.  However, at least we can check that we don't have
827 	 * too many entries in the next_clones_obj even after failing to
828 	 * remove this one.
829 	 */
830 	if (err != ENOENT)
831 		VERIFY0(err);
832 	ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
833 	    &count));
834 	ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
835 }
836 
837 
838 blkptr_t *
839 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
840 {
841 	return (&ds->ds_phys->ds_bp);
842 }
843 
844 void
845 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
846 {
847 	ASSERT(dmu_tx_is_syncing(tx));
848 	/* If it's the meta-objset, set dp_meta_rootbp */
849 	if (ds == NULL) {
850 		tx->tx_pool->dp_meta_rootbp = *bp;
851 	} else {
852 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
853 		ds->ds_phys->ds_bp = *bp;
854 	}
855 }
856 
857 spa_t *
858 dsl_dataset_get_spa(dsl_dataset_t *ds)
859 {
860 	return (ds->ds_dir->dd_pool->dp_spa);
861 }
862 
863 void
864 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
865 {
866 	dsl_pool_t *dp;
867 
868 	if (ds == NULL) /* this is the meta-objset */
869 		return;
870 
871 	ASSERT(ds->ds_objset != NULL);
872 
873 	if (ds->ds_phys->ds_next_snap_obj != 0)
874 		panic("dirtying snapshot!");
875 
876 	dp = ds->ds_dir->dd_pool;
877 
878 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
879 		/* up the hold count until we can be written out */
880 		dmu_buf_add_ref(ds->ds_dbuf, ds);
881 	}
882 }
883 
884 boolean_t
885 dsl_dataset_is_dirty(dsl_dataset_t *ds)
886 {
887 	for (int t = 0; t < TXG_SIZE; t++) {
888 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
889 		    ds, t))
890 			return (B_TRUE);
891 	}
892 	return (B_FALSE);
893 }
894 
895 static int
896 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
897 {
898 	uint64_t asize;
899 
900 	if (!dmu_tx_is_syncing(tx))
901 		return (0);
902 
903 	/*
904 	 * If there's an fs-only reservation, any blocks that might become
905 	 * owned by the snapshot dataset must be accommodated by space
906 	 * outside of the reservation.
907 	 */
908 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
909 	asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
910 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
911 		return (SET_ERROR(ENOSPC));
912 
913 	/*
914 	 * Propagate any reserved space for this snapshot to other
915 	 * snapshot checks in this sync group.
916 	 */
917 	if (asize > 0)
918 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
919 
920 	return (0);
921 }
922 
923 typedef struct dsl_dataset_snapshot_arg {
924 	nvlist_t *ddsa_snaps;
925 	nvlist_t *ddsa_props;
926 	nvlist_t *ddsa_errors;
927 } dsl_dataset_snapshot_arg_t;
928 
929 int
930 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
931     dmu_tx_t *tx, boolean_t recv)
932 {
933 	int error;
934 	uint64_t value;
935 
936 	ds->ds_trysnap_txg = tx->tx_txg;
937 
938 	if (!dmu_tx_is_syncing(tx))
939 		return (0);
940 
941 	/*
942 	 * We don't allow multiple snapshots of the same txg.  If there
943 	 * is already one, try again.
944 	 */
945 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
946 		return (SET_ERROR(EAGAIN));
947 
948 	/*
949 	 * Check for conflicting snapshot name.
950 	 */
951 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
952 	if (error == 0)
953 		return (SET_ERROR(EEXIST));
954 	if (error != ENOENT)
955 		return (error);
956 
957 	/*
958 	 * We don't allow taking snapshots of inconsistent datasets, such as
959 	 * those into which we are currently receiving.  However, if we are
960 	 * creating this snapshot as part of a receive, this check will be
961 	 * executed atomically with respect to the completion of the receive
962 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
963 	 * case we ignore this, knowing it will be fixed up for us shortly in
964 	 * dmu_recv_end_sync().
965 	 */
966 	if (!recv && DS_IS_INCONSISTENT(ds))
967 		return (SET_ERROR(EBUSY));
968 
969 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
970 	if (error != 0)
971 		return (error);
972 
973 	return (0);
974 }
975 
976 static int
977 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
978 {
979 	dsl_dataset_snapshot_arg_t *ddsa = arg;
980 	dsl_pool_t *dp = dmu_tx_pool(tx);
981 	nvpair_t *pair;
982 	int rv = 0;
983 
984 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
985 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
986 		int error = 0;
987 		dsl_dataset_t *ds;
988 		char *name, *atp;
989 		char dsname[MAXNAMELEN];
990 
991 		name = nvpair_name(pair);
992 		if (strlen(name) >= MAXNAMELEN)
993 			error = SET_ERROR(ENAMETOOLONG);
994 		if (error == 0) {
995 			atp = strchr(name, '@');
996 			if (atp == NULL)
997 				error = SET_ERROR(EINVAL);
998 			if (error == 0)
999 				(void) strlcpy(dsname, name, atp - name + 1);
1000 		}
1001 		if (error == 0)
1002 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1003 		if (error == 0) {
1004 			error = dsl_dataset_snapshot_check_impl(ds,
1005 			    atp + 1, tx, B_FALSE);
1006 			dsl_dataset_rele(ds, FTAG);
1007 		}
1008 
1009 		if (error != 0) {
1010 			if (ddsa->ddsa_errors != NULL) {
1011 				fnvlist_add_int32(ddsa->ddsa_errors,
1012 				    name, error);
1013 			}
1014 			rv = error;
1015 		}
1016 	}
1017 	return (rv);
1018 }
1019 
1020 void
1021 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1022     dmu_tx_t *tx)
1023 {
1024 	static zil_header_t zero_zil;
1025 
1026 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1027 	dmu_buf_t *dbuf;
1028 	dsl_dataset_phys_t *dsphys;
1029 	uint64_t dsobj, crtxg;
1030 	objset_t *mos = dp->dp_meta_objset;
1031 	objset_t *os;
1032 
1033 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1034 
1035 	/*
1036 	 * If we are on an old pool, the zil must not be active, in which
1037 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1038 	 */
1039 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1040 	    dmu_objset_from_ds(ds, &os) != 0 ||
1041 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1042 	    sizeof (zero_zil)) == 0);
1043 
1044 
1045 	/*
1046 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1047 	 */
1048 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1049 		crtxg = 1;
1050 	else
1051 		crtxg = tx->tx_txg;
1052 
1053 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1054 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1055 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1056 	dmu_buf_will_dirty(dbuf, tx);
1057 	dsphys = dbuf->db_data;
1058 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1059 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1060 	dsphys->ds_fsid_guid = unique_create();
1061 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1062 	    sizeof (dsphys->ds_guid));
1063 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1064 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1065 	dsphys->ds_next_snap_obj = ds->ds_object;
1066 	dsphys->ds_num_children = 1;
1067 	dsphys->ds_creation_time = gethrestime_sec();
1068 	dsphys->ds_creation_txg = crtxg;
1069 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1070 	dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
1071 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1072 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1073 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1074 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1075 	dmu_buf_rele(dbuf, FTAG);
1076 
1077 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1078 	if (ds->ds_prev) {
1079 		uint64_t next_clones_obj =
1080 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1081 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1082 		    ds->ds_object ||
1083 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1084 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1085 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1086 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1087 			    ds->ds_prev->ds_phys->ds_creation_txg);
1088 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1089 		} else if (next_clones_obj != 0) {
1090 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1091 			    dsphys->ds_next_snap_obj, tx);
1092 			VERIFY0(zap_add_int(mos,
1093 			    next_clones_obj, dsobj, tx));
1094 		}
1095 	}
1096 
1097 	/*
1098 	 * If we have a reference-reservation on this dataset, we will
1099 	 * need to increase the amount of refreservation being charged
1100 	 * since our unique space is going to zero.
1101 	 */
1102 	if (ds->ds_reserved) {
1103 		int64_t delta;
1104 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1105 		delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1106 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1107 		    delta, 0, 0, tx);
1108 	}
1109 
1110 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1111 	ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
1112 	    UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
1113 	dsl_deadlist_close(&ds->ds_deadlist);
1114 	dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1115 	dsl_deadlist_add_key(&ds->ds_deadlist,
1116 	    ds->ds_phys->ds_prev_snap_txg, tx);
1117 
1118 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1119 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1120 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1121 	ds->ds_phys->ds_unique_bytes = 0;
1122 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1123 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1124 
1125 	VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1126 	    snapname, 8, 1, &dsobj, tx));
1127 
1128 	if (ds->ds_prev)
1129 		dsl_dataset_rele(ds->ds_prev, ds);
1130 	VERIFY0(dsl_dataset_hold_obj(dp,
1131 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1132 
1133 	dsl_scan_ds_snapshotted(ds, tx);
1134 
1135 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1136 
1137 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1138 }
1139 
1140 static void
1141 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1142 {
1143 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1144 	dsl_pool_t *dp = dmu_tx_pool(tx);
1145 	nvpair_t *pair;
1146 
1147 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1148 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1149 		dsl_dataset_t *ds;
1150 		char *name, *atp;
1151 		char dsname[MAXNAMELEN];
1152 
1153 		name = nvpair_name(pair);
1154 		atp = strchr(name, '@');
1155 		(void) strlcpy(dsname, name, atp - name + 1);
1156 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1157 
1158 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1159 		if (ddsa->ddsa_props != NULL) {
1160 			dsl_props_set_sync_impl(ds->ds_prev,
1161 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1162 		}
1163 		dsl_dataset_rele(ds, FTAG);
1164 	}
1165 }
1166 
1167 /*
1168  * The snapshots must all be in the same pool.
1169  * All-or-nothing: if there are any failures, nothing will be modified.
1170  */
1171 int
1172 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1173 {
1174 	dsl_dataset_snapshot_arg_t ddsa;
1175 	nvpair_t *pair;
1176 	boolean_t needsuspend;
1177 	int error;
1178 	spa_t *spa;
1179 	char *firstname;
1180 	nvlist_t *suspended = NULL;
1181 
1182 	pair = nvlist_next_nvpair(snaps, NULL);
1183 	if (pair == NULL)
1184 		return (0);
1185 	firstname = nvpair_name(pair);
1186 
1187 	error = spa_open(firstname, &spa, FTAG);
1188 	if (error != 0)
1189 		return (error);
1190 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1191 	spa_close(spa, FTAG);
1192 
1193 	if (needsuspend) {
1194 		suspended = fnvlist_alloc();
1195 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1196 		    pair = nvlist_next_nvpair(snaps, pair)) {
1197 			char fsname[MAXNAMELEN];
1198 			char *snapname = nvpair_name(pair);
1199 			char *atp;
1200 			void *cookie;
1201 
1202 			atp = strchr(snapname, '@');
1203 			if (atp == NULL) {
1204 				error = SET_ERROR(EINVAL);
1205 				break;
1206 			}
1207 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1208 
1209 			error = zil_suspend(fsname, &cookie);
1210 			if (error != 0)
1211 				break;
1212 			fnvlist_add_uint64(suspended, fsname,
1213 			    (uintptr_t)cookie);
1214 		}
1215 	}
1216 
1217 	ddsa.ddsa_snaps = snaps;
1218 	ddsa.ddsa_props = props;
1219 	ddsa.ddsa_errors = errors;
1220 
1221 	if (error == 0) {
1222 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1223 		    dsl_dataset_snapshot_sync, &ddsa,
1224 		    fnvlist_num_pairs(snaps) * 3);
1225 	}
1226 
1227 	if (suspended != NULL) {
1228 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1229 		    pair = nvlist_next_nvpair(suspended, pair)) {
1230 			zil_resume((void *)(uintptr_t)
1231 			    fnvpair_value_uint64(pair));
1232 		}
1233 		fnvlist_free(suspended);
1234 	}
1235 
1236 	return (error);
1237 }
1238 
1239 typedef struct dsl_dataset_snapshot_tmp_arg {
1240 	const char *ddsta_fsname;
1241 	const char *ddsta_snapname;
1242 	minor_t ddsta_cleanup_minor;
1243 	const char *ddsta_htag;
1244 } dsl_dataset_snapshot_tmp_arg_t;
1245 
1246 static int
1247 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1248 {
1249 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1250 	dsl_pool_t *dp = dmu_tx_pool(tx);
1251 	dsl_dataset_t *ds;
1252 	int error;
1253 
1254 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1255 	if (error != 0)
1256 		return (error);
1257 
1258 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1259 	    tx, B_FALSE);
1260 	if (error != 0) {
1261 		dsl_dataset_rele(ds, FTAG);
1262 		return (error);
1263 	}
1264 
1265 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1266 		dsl_dataset_rele(ds, FTAG);
1267 		return (SET_ERROR(ENOTSUP));
1268 	}
1269 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1270 	    B_TRUE, tx);
1271 	if (error != 0) {
1272 		dsl_dataset_rele(ds, FTAG);
1273 		return (error);
1274 	}
1275 
1276 	dsl_dataset_rele(ds, FTAG);
1277 	return (0);
1278 }
1279 
1280 static void
1281 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1282 {
1283 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1284 	dsl_pool_t *dp = dmu_tx_pool(tx);
1285 	dsl_dataset_t *ds;
1286 
1287 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1288 
1289 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1290 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1291 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1292 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1293 
1294 	dsl_dataset_rele(ds, FTAG);
1295 }
1296 
1297 int
1298 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1299     minor_t cleanup_minor, const char *htag)
1300 {
1301 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1302 	int error;
1303 	spa_t *spa;
1304 	boolean_t needsuspend;
1305 	void *cookie;
1306 
1307 	ddsta.ddsta_fsname = fsname;
1308 	ddsta.ddsta_snapname = snapname;
1309 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1310 	ddsta.ddsta_htag = htag;
1311 
1312 	error = spa_open(fsname, &spa, FTAG);
1313 	if (error != 0)
1314 		return (error);
1315 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1316 	spa_close(spa, FTAG);
1317 
1318 	if (needsuspend) {
1319 		error = zil_suspend(fsname, &cookie);
1320 		if (error != 0)
1321 			return (error);
1322 	}
1323 
1324 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1325 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3);
1326 
1327 	if (needsuspend)
1328 		zil_resume(cookie);
1329 	return (error);
1330 }
1331 
1332 
1333 void
1334 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1335 {
1336 	ASSERT(dmu_tx_is_syncing(tx));
1337 	ASSERT(ds->ds_objset != NULL);
1338 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1339 
1340 	/*
1341 	 * in case we had to change ds_fsid_guid when we opened it,
1342 	 * sync it out now.
1343 	 */
1344 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1345 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1346 
1347 	dmu_objset_sync(ds->ds_objset, zio, tx);
1348 }
1349 
1350 static void
1351 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1352 {
1353 	uint64_t count = 0;
1354 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1355 	zap_cursor_t zc;
1356 	zap_attribute_t za;
1357 	nvlist_t *propval = fnvlist_alloc();
1358 	nvlist_t *val = fnvlist_alloc();
1359 
1360 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1361 
1362 	/*
1363 	 * There may be missing entries in ds_next_clones_obj
1364 	 * due to a bug in a previous version of the code.
1365 	 * Only trust it if it has the right number of entries.
1366 	 */
1367 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1368 		VERIFY0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1369 		    &count));
1370 	}
1371 	if (count != ds->ds_phys->ds_num_children - 1)
1372 		goto fail;
1373 	for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
1374 	    zap_cursor_retrieve(&zc, &za) == 0;
1375 	    zap_cursor_advance(&zc)) {
1376 		dsl_dataset_t *clone;
1377 		char buf[ZFS_MAXNAMELEN];
1378 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1379 		    za.za_first_integer, FTAG, &clone));
1380 		dsl_dir_name(clone->ds_dir, buf);
1381 		fnvlist_add_boolean(val, buf);
1382 		dsl_dataset_rele(clone, FTAG);
1383 	}
1384 	zap_cursor_fini(&zc);
1385 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1386 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1387 fail:
1388 	nvlist_free(val);
1389 	nvlist_free(propval);
1390 }
1391 
1392 void
1393 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1394 {
1395 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1396 	uint64_t refd, avail, uobjs, aobjs, ratio;
1397 
1398 	ASSERT(dsl_pool_config_held(dp));
1399 
1400 	ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1401 	    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1402 	    ds->ds_phys->ds_compressed_bytes);
1403 
1404 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1405 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1406 	    ds->ds_phys->ds_uncompressed_bytes);
1407 
1408 	if (dsl_dataset_is_snapshot(ds)) {
1409 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1410 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1411 		    ds->ds_phys->ds_unique_bytes);
1412 		get_clones_stat(ds, nv);
1413 	} else {
1414 		dsl_dir_stats(ds->ds_dir, nv);
1415 	}
1416 
1417 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1418 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1419 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1420 
1421 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1422 	    ds->ds_phys->ds_creation_time);
1423 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1424 	    ds->ds_phys->ds_creation_txg);
1425 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1426 	    ds->ds_quota);
1427 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1428 	    ds->ds_reserved);
1429 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1430 	    ds->ds_phys->ds_guid);
1431 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1432 	    ds->ds_phys->ds_unique_bytes);
1433 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1434 	    ds->ds_object);
1435 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1436 	    ds->ds_userrefs);
1437 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1438 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1439 
1440 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1441 		uint64_t written, comp, uncomp;
1442 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1443 		dsl_dataset_t *prev;
1444 
1445 		int err = dsl_dataset_hold_obj(dp,
1446 		    ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
1447 		if (err == 0) {
1448 			err = dsl_dataset_space_written(prev, ds, &written,
1449 			    &comp, &uncomp);
1450 			dsl_dataset_rele(prev, FTAG);
1451 			if (err == 0) {
1452 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1453 				    written);
1454 			}
1455 		}
1456 	}
1457 }
1458 
1459 void
1460 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1461 {
1462 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1463 	ASSERT(dsl_pool_config_held(dp));
1464 
1465 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1466 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1467 	stat->dds_guid = ds->ds_phys->ds_guid;
1468 	stat->dds_origin[0] = '\0';
1469 	if (dsl_dataset_is_snapshot(ds)) {
1470 		stat->dds_is_snapshot = B_TRUE;
1471 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1472 	} else {
1473 		stat->dds_is_snapshot = B_FALSE;
1474 		stat->dds_num_clones = 0;
1475 
1476 		if (dsl_dir_is_clone(ds->ds_dir)) {
1477 			dsl_dataset_t *ods;
1478 
1479 			VERIFY0(dsl_dataset_hold_obj(dp,
1480 			    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1481 			dsl_dataset_name(ods, stat->dds_origin);
1482 			dsl_dataset_rele(ods, FTAG);
1483 		}
1484 	}
1485 }
1486 
1487 uint64_t
1488 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1489 {
1490 	return (ds->ds_fsid_guid);
1491 }
1492 
1493 void
1494 dsl_dataset_space(dsl_dataset_t *ds,
1495     uint64_t *refdbytesp, uint64_t *availbytesp,
1496     uint64_t *usedobjsp, uint64_t *availobjsp)
1497 {
1498 	*refdbytesp = ds->ds_phys->ds_referenced_bytes;
1499 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1500 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1501 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1502 	if (ds->ds_quota != 0) {
1503 		/*
1504 		 * Adjust available bytes according to refquota
1505 		 */
1506 		if (*refdbytesp < ds->ds_quota)
1507 			*availbytesp = MIN(*availbytesp,
1508 			    ds->ds_quota - *refdbytesp);
1509 		else
1510 			*availbytesp = 0;
1511 	}
1512 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1513 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1514 }
1515 
1516 boolean_t
1517 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1518 {
1519 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1520 
1521 	ASSERT(dsl_pool_config_held(dp));
1522 	if (snap == NULL)
1523 		return (B_FALSE);
1524 	if (ds->ds_phys->ds_bp.blk_birth >
1525 	    snap->ds_phys->ds_creation_txg) {
1526 		objset_t *os, *os_snap;
1527 		/*
1528 		 * It may be that only the ZIL differs, because it was
1529 		 * reset in the head.  Don't count that as being
1530 		 * modified.
1531 		 */
1532 		if (dmu_objset_from_ds(ds, &os) != 0)
1533 			return (B_TRUE);
1534 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1535 			return (B_TRUE);
1536 		return (bcmp(&os->os_phys->os_meta_dnode,
1537 		    &os_snap->os_phys->os_meta_dnode,
1538 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1539 	}
1540 	return (B_FALSE);
1541 }
1542 
1543 typedef struct dsl_dataset_rename_snapshot_arg {
1544 	const char *ddrsa_fsname;
1545 	const char *ddrsa_oldsnapname;
1546 	const char *ddrsa_newsnapname;
1547 	boolean_t ddrsa_recursive;
1548 	dmu_tx_t *ddrsa_tx;
1549 } dsl_dataset_rename_snapshot_arg_t;
1550 
1551 /* ARGSUSED */
1552 static int
1553 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1554     dsl_dataset_t *hds, void *arg)
1555 {
1556 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1557 	int error;
1558 	uint64_t val;
1559 
1560 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1561 	if (error != 0) {
1562 		/* ignore nonexistent snapshots */
1563 		return (error == ENOENT ? 0 : error);
1564 	}
1565 
1566 	/* new name should not exist */
1567 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1568 	if (error == 0)
1569 		error = SET_ERROR(EEXIST);
1570 	else if (error == ENOENT)
1571 		error = 0;
1572 
1573 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1574 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1575 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1576 		error = SET_ERROR(ENAMETOOLONG);
1577 
1578 	return (error);
1579 }
1580 
1581 static int
1582 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1583 {
1584 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1585 	dsl_pool_t *dp = dmu_tx_pool(tx);
1586 	dsl_dataset_t *hds;
1587 	int error;
1588 
1589 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1590 	if (error != 0)
1591 		return (error);
1592 
1593 	if (ddrsa->ddrsa_recursive) {
1594 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1595 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1596 		    DS_FIND_CHILDREN);
1597 	} else {
1598 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1599 	}
1600 	dsl_dataset_rele(hds, FTAG);
1601 	return (error);
1602 }
1603 
1604 static int
1605 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1606     dsl_dataset_t *hds, void *arg)
1607 {
1608 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1609 	dsl_dataset_t *ds;
1610 	uint64_t val;
1611 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1612 	int error;
1613 
1614 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1615 	ASSERT(error == 0 || error == ENOENT);
1616 	if (error == ENOENT) {
1617 		/* ignore nonexistent snapshots */
1618 		return (0);
1619 	}
1620 
1621 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1622 
1623 	/* log before we change the name */
1624 	spa_history_log_internal_ds(ds, "rename", tx,
1625 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1626 
1627 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx));
1628 	mutex_enter(&ds->ds_lock);
1629 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1630 	mutex_exit(&ds->ds_lock);
1631 	VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj,
1632 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1633 
1634 	dsl_dataset_rele(ds, FTAG);
1635 	return (0);
1636 }
1637 
1638 static void
1639 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1640 {
1641 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1642 	dsl_pool_t *dp = dmu_tx_pool(tx);
1643 	dsl_dataset_t *hds;
1644 
1645 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1646 	ddrsa->ddrsa_tx = tx;
1647 	if (ddrsa->ddrsa_recursive) {
1648 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1649 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1650 		    DS_FIND_CHILDREN));
1651 	} else {
1652 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1653 	}
1654 	dsl_dataset_rele(hds, FTAG);
1655 }
1656 
1657 int
1658 dsl_dataset_rename_snapshot(const char *fsname,
1659     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1660 {
1661 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1662 
1663 	ddrsa.ddrsa_fsname = fsname;
1664 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1665 	ddrsa.ddrsa_newsnapname = newsnapname;
1666 	ddrsa.ddrsa_recursive = recursive;
1667 
1668 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1669 	    dsl_dataset_rename_snapshot_sync, &ddrsa, 1));
1670 }
1671 
1672 /*
1673  * If we're doing an ownership handoff, we need to make sure that there is
1674  * only one long hold on the dataset.  We're not allowed to change anything here
1675  * so we don't permanently release the long hold or regular hold here.  We want
1676  * to do this only when syncing to avoid the dataset unexpectedly going away
1677  * when we release the long hold.
1678  */
1679 static int
1680 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1681 {
1682 	boolean_t held;
1683 
1684 	if (!dmu_tx_is_syncing(tx))
1685 		return (0);
1686 
1687 	if (owner != NULL) {
1688 		VERIFY3P(ds->ds_owner, ==, owner);
1689 		dsl_dataset_long_rele(ds, owner);
1690 	}
1691 
1692 	held = dsl_dataset_long_held(ds);
1693 
1694 	if (owner != NULL)
1695 		dsl_dataset_long_hold(ds, owner);
1696 
1697 	if (held)
1698 		return (SET_ERROR(EBUSY));
1699 
1700 	return (0);
1701 }
1702 
1703 typedef struct dsl_dataset_rollback_arg {
1704 	const char *ddra_fsname;
1705 	void *ddra_owner;
1706 	nvlist_t *ddra_result;
1707 } dsl_dataset_rollback_arg_t;
1708 
1709 static int
1710 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1711 {
1712 	dsl_dataset_rollback_arg_t *ddra = arg;
1713 	dsl_pool_t *dp = dmu_tx_pool(tx);
1714 	dsl_dataset_t *ds;
1715 	int64_t unused_refres_delta;
1716 	int error;
1717 
1718 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1719 	if (error != 0)
1720 		return (error);
1721 
1722 	/* must not be a snapshot */
1723 	if (dsl_dataset_is_snapshot(ds)) {
1724 		dsl_dataset_rele(ds, FTAG);
1725 		return (SET_ERROR(EINVAL));
1726 	}
1727 
1728 	/* must have a most recent snapshot */
1729 	if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) {
1730 		dsl_dataset_rele(ds, FTAG);
1731 		return (SET_ERROR(EINVAL));
1732 	}
1733 
1734 	/* must not have any bookmarks after the most recent snapshot */
1735 	nvlist_t *proprequest = fnvlist_alloc();
1736 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
1737 	nvlist_t *bookmarks = fnvlist_alloc();
1738 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
1739 	fnvlist_free(proprequest);
1740 	if (error != 0)
1741 		return (error);
1742 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
1743 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
1744 		nvlist_t *valuenv =
1745 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
1746 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
1747 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
1748 		if (createtxg > ds->ds_phys->ds_prev_snap_txg) {
1749 			fnvlist_free(bookmarks);
1750 			dsl_dataset_rele(ds, FTAG);
1751 			return (SET_ERROR(EEXIST));
1752 		}
1753 	}
1754 	fnvlist_free(bookmarks);
1755 
1756 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
1757 	if (error != 0) {
1758 		dsl_dataset_rele(ds, FTAG);
1759 		return (error);
1760 	}
1761 
1762 	/*
1763 	 * Check if the snap we are rolling back to uses more than
1764 	 * the refquota.
1765 	 */
1766 	if (ds->ds_quota != 0 &&
1767 	    ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) {
1768 		dsl_dataset_rele(ds, FTAG);
1769 		return (SET_ERROR(EDQUOT));
1770 	}
1771 
1772 	/*
1773 	 * When we do the clone swap, we will temporarily use more space
1774 	 * due to the refreservation (the head will no longer have any
1775 	 * unique space, so the entire amount of the refreservation will need
1776 	 * to be free).  We will immediately destroy the clone, freeing
1777 	 * this space, but the freeing happens over many txg's.
1778 	 */
1779 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1780 	    ds->ds_phys->ds_unique_bytes);
1781 
1782 	if (unused_refres_delta > 0 &&
1783 	    unused_refres_delta >
1784 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1785 		dsl_dataset_rele(ds, FTAG);
1786 		return (SET_ERROR(ENOSPC));
1787 	}
1788 
1789 	dsl_dataset_rele(ds, FTAG);
1790 	return (0);
1791 }
1792 
1793 static void
1794 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1795 {
1796 	dsl_dataset_rollback_arg_t *ddra = arg;
1797 	dsl_pool_t *dp = dmu_tx_pool(tx);
1798 	dsl_dataset_t *ds, *clone;
1799 	uint64_t cloneobj;
1800 	char namebuf[ZFS_MAXNAMELEN];
1801 
1802 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
1803 
1804 	dsl_dataset_name(ds->ds_prev, namebuf);
1805 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
1806 
1807 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
1808 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
1809 
1810 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
1811 
1812 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
1813 	dsl_dataset_zero_zil(ds, tx);
1814 
1815 	dsl_destroy_head_sync_impl(clone, tx);
1816 
1817 	dsl_dataset_rele(clone, FTAG);
1818 	dsl_dataset_rele(ds, FTAG);
1819 }
1820 
1821 /*
1822  * Rolls back the given filesystem or volume to the most recent snapshot.
1823  * The name of the most recent snapshot will be returned under key "target"
1824  * in the result nvlist.
1825  *
1826  * If owner != NULL:
1827  * - The existing dataset MUST be owned by the specified owner at entry
1828  * - Upon return, dataset will still be held by the same owner, whether we
1829  *   succeed or not.
1830  *
1831  * This mode is required any time the existing filesystem is mounted.  See
1832  * notes above zfs_suspend_fs() for further details.
1833  */
1834 int
1835 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
1836 {
1837 	dsl_dataset_rollback_arg_t ddra;
1838 
1839 	ddra.ddra_fsname = fsname;
1840 	ddra.ddra_owner = owner;
1841 	ddra.ddra_result = result;
1842 
1843 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
1844 	    dsl_dataset_rollback_sync, &ddra, 1));
1845 }
1846 
1847 struct promotenode {
1848 	list_node_t link;
1849 	dsl_dataset_t *ds;
1850 };
1851 
1852 typedef struct dsl_dataset_promote_arg {
1853 	const char *ddpa_clonename;
1854 	dsl_dataset_t *ddpa_clone;
1855 	list_t shared_snaps, origin_snaps, clone_snaps;
1856 	dsl_dataset_t *origin_origin; /* origin of the origin */
1857 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
1858 	char *err_ds;
1859 } dsl_dataset_promote_arg_t;
1860 
1861 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
1862 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
1863     void *tag);
1864 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
1865 
1866 static int
1867 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
1868 {
1869 	dsl_dataset_promote_arg_t *ddpa = arg;
1870 	dsl_pool_t *dp = dmu_tx_pool(tx);
1871 	dsl_dataset_t *hds;
1872 	struct promotenode *snap;
1873 	dsl_dataset_t *origin_ds;
1874 	int err;
1875 	uint64_t unused;
1876 
1877 	err = promote_hold(ddpa, dp, FTAG);
1878 	if (err != 0)
1879 		return (err);
1880 
1881 	hds = ddpa->ddpa_clone;
1882 
1883 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
1884 		promote_rele(ddpa, FTAG);
1885 		return (SET_ERROR(EXDEV));
1886 	}
1887 
1888 	/*
1889 	 * Compute and check the amount of space to transfer.  Since this is
1890 	 * so expensive, don't do the preliminary check.
1891 	 */
1892 	if (!dmu_tx_is_syncing(tx)) {
1893 		promote_rele(ddpa, FTAG);
1894 		return (0);
1895 	}
1896 
1897 	snap = list_head(&ddpa->shared_snaps);
1898 	origin_ds = snap->ds;
1899 
1900 	/* compute origin's new unique space */
1901 	snap = list_tail(&ddpa->clone_snaps);
1902 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1903 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
1904 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1905 	    &ddpa->unique, &unused, &unused);
1906 
1907 	/*
1908 	 * Walk the snapshots that we are moving
1909 	 *
1910 	 * Compute space to transfer.  Consider the incremental changes
1911 	 * to used by each snapshot:
1912 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
1913 	 * So each snapshot gave birth to:
1914 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
1915 	 * So a sequence would look like:
1916 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
1917 	 * Which simplifies to:
1918 	 * uN + kN + kN-1 + ... + k1 + k0
1919 	 * Note however, if we stop before we reach the ORIGIN we get:
1920 	 * uN + kN + kN-1 + ... + kM - uM-1
1921 	 */
1922 	ddpa->used = origin_ds->ds_phys->ds_referenced_bytes;
1923 	ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes;
1924 	ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
1925 	for (snap = list_head(&ddpa->shared_snaps); snap;
1926 	    snap = list_next(&ddpa->shared_snaps, snap)) {
1927 		uint64_t val, dlused, dlcomp, dluncomp;
1928 		dsl_dataset_t *ds = snap->ds;
1929 
1930 		/*
1931 		 * If there are long holds, we won't be able to evict
1932 		 * the objset.
1933 		 */
1934 		if (dsl_dataset_long_held(ds)) {
1935 			err = SET_ERROR(EBUSY);
1936 			goto out;
1937 		}
1938 
1939 		/* Check that the snapshot name does not conflict */
1940 		VERIFY0(dsl_dataset_get_snapname(ds));
1941 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
1942 		if (err == 0) {
1943 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
1944 			err = SET_ERROR(EEXIST);
1945 			goto out;
1946 		}
1947 		if (err != ENOENT)
1948 			goto out;
1949 
1950 		/* The very first snapshot does not have a deadlist */
1951 		if (ds->ds_phys->ds_prev_snap_obj == 0)
1952 			continue;
1953 
1954 		dsl_deadlist_space(&ds->ds_deadlist,
1955 		    &dlused, &dlcomp, &dluncomp);
1956 		ddpa->used += dlused;
1957 		ddpa->comp += dlcomp;
1958 		ddpa->uncomp += dluncomp;
1959 	}
1960 
1961 	/*
1962 	 * If we are a clone of a clone then we never reached ORIGIN,
1963 	 * so we need to subtract out the clone origin's used space.
1964 	 */
1965 	if (ddpa->origin_origin) {
1966 		ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes;
1967 		ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes;
1968 		ddpa->uncomp -=
1969 		    ddpa->origin_origin->ds_phys->ds_uncompressed_bytes;
1970 	}
1971 
1972 	/* Check that there is enough space here */
1973 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
1974 	    ddpa->used);
1975 	if (err != 0)
1976 		goto out;
1977 
1978 	/*
1979 	 * Compute the amounts of space that will be used by snapshots
1980 	 * after the promotion (for both origin and clone).  For each,
1981 	 * it is the amount of space that will be on all of their
1982 	 * deadlists (that was not born before their new origin).
1983 	 */
1984 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1985 		uint64_t space;
1986 
1987 		/*
1988 		 * Note, typically this will not be a clone of a clone,
1989 		 * so dd_origin_txg will be < TXG_INITIAL, so
1990 		 * these snaplist_space() -> dsl_deadlist_space_range()
1991 		 * calls will be fast because they do not have to
1992 		 * iterate over all bps.
1993 		 */
1994 		snap = list_head(&ddpa->origin_snaps);
1995 		err = snaplist_space(&ddpa->shared_snaps,
1996 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
1997 		if (err != 0)
1998 			goto out;
1999 
2000 		err = snaplist_space(&ddpa->clone_snaps,
2001 		    snap->ds->ds_dir->dd_origin_txg, &space);
2002 		if (err != 0)
2003 			goto out;
2004 		ddpa->cloneusedsnap += space;
2005 	}
2006 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2007 		err = snaplist_space(&ddpa->origin_snaps,
2008 		    origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap);
2009 		if (err != 0)
2010 			goto out;
2011 	}
2012 
2013 out:
2014 	promote_rele(ddpa, FTAG);
2015 	return (err);
2016 }
2017 
2018 static void
2019 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2020 {
2021 	dsl_dataset_promote_arg_t *ddpa = arg;
2022 	dsl_pool_t *dp = dmu_tx_pool(tx);
2023 	dsl_dataset_t *hds;
2024 	struct promotenode *snap;
2025 	dsl_dataset_t *origin_ds;
2026 	dsl_dataset_t *origin_head;
2027 	dsl_dir_t *dd;
2028 	dsl_dir_t *odd = NULL;
2029 	uint64_t oldnext_obj;
2030 	int64_t delta;
2031 
2032 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2033 	hds = ddpa->ddpa_clone;
2034 
2035 	ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE);
2036 
2037 	snap = list_head(&ddpa->shared_snaps);
2038 	origin_ds = snap->ds;
2039 	dd = hds->ds_dir;
2040 
2041 	snap = list_head(&ddpa->origin_snaps);
2042 	origin_head = snap->ds;
2043 
2044 	/*
2045 	 * We need to explicitly open odd, since origin_ds's dd will be
2046 	 * changing.
2047 	 */
2048 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2049 	    NULL, FTAG, &odd));
2050 
2051 	/* change origin's next snap */
2052 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2053 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2054 	snap = list_tail(&ddpa->clone_snaps);
2055 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2056 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2057 
2058 	/* change the origin's next clone */
2059 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2060 		dsl_dataset_remove_from_next_clones(origin_ds,
2061 		    snap->ds->ds_object, tx);
2062 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2063 		    origin_ds->ds_phys->ds_next_clones_obj,
2064 		    oldnext_obj, tx));
2065 	}
2066 
2067 	/* change origin */
2068 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2069 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2070 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2071 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2072 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2073 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2074 	origin_head->ds_dir->dd_origin_txg =
2075 	    origin_ds->ds_phys->ds_creation_txg;
2076 
2077 	/* change dd_clone entries */
2078 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2079 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2080 		    odd->dd_phys->dd_clones, hds->ds_object, tx));
2081 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2082 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2083 		    hds->ds_object, tx));
2084 
2085 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2086 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2087 		    origin_head->ds_object, tx));
2088 		if (dd->dd_phys->dd_clones == 0) {
2089 			dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2090 			    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2091 		}
2092 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2093 		    dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2094 	}
2095 
2096 	/* move snapshots to this dir */
2097 	for (snap = list_head(&ddpa->shared_snaps); snap;
2098 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2099 		dsl_dataset_t *ds = snap->ds;
2100 
2101 		/*
2102 		 * Property callbacks are registered to a particular
2103 		 * dsl_dir.  Since ours is changing, evict the objset
2104 		 * so that they will be unregistered from the old dsl_dir.
2105 		 */
2106 		if (ds->ds_objset) {
2107 			dmu_objset_evict(ds->ds_objset);
2108 			ds->ds_objset = NULL;
2109 		}
2110 
2111 		/* move snap name entry */
2112 		VERIFY0(dsl_dataset_get_snapname(ds));
2113 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2114 		    ds->ds_snapname, tx));
2115 		VERIFY0(zap_add(dp->dp_meta_objset,
2116 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2117 		    8, 1, &ds->ds_object, tx));
2118 
2119 		/* change containing dsl_dir */
2120 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2121 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2122 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2123 		ASSERT3P(ds->ds_dir, ==, odd);
2124 		dsl_dir_rele(ds->ds_dir, ds);
2125 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2126 		    NULL, ds, &ds->ds_dir));
2127 
2128 		/* move any clone references */
2129 		if (ds->ds_phys->ds_next_clones_obj &&
2130 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2131 			zap_cursor_t zc;
2132 			zap_attribute_t za;
2133 
2134 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2135 			    ds->ds_phys->ds_next_clones_obj);
2136 			    zap_cursor_retrieve(&zc, &za) == 0;
2137 			    zap_cursor_advance(&zc)) {
2138 				dsl_dataset_t *cnds;
2139 				uint64_t o;
2140 
2141 				if (za.za_first_integer == oldnext_obj) {
2142 					/*
2143 					 * We've already moved the
2144 					 * origin's reference.
2145 					 */
2146 					continue;
2147 				}
2148 
2149 				VERIFY0(dsl_dataset_hold_obj(dp,
2150 				    za.za_first_integer, FTAG, &cnds));
2151 				o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2152 
2153 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2154 				    odd->dd_phys->dd_clones, o, tx));
2155 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2156 				    dd->dd_phys->dd_clones, o, tx));
2157 				dsl_dataset_rele(cnds, FTAG);
2158 			}
2159 			zap_cursor_fini(&zc);
2160 		}
2161 
2162 		ASSERT(!dsl_prop_hascb(ds));
2163 	}
2164 
2165 	/*
2166 	 * Change space accounting.
2167 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2168 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2169 	 * is true for each of {clone,origin} independently.
2170 	 */
2171 
2172 	delta = ddpa->cloneusedsnap -
2173 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2174 	ASSERT3S(delta, >=, 0);
2175 	ASSERT3U(ddpa->used, >=, delta);
2176 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2177 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2178 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2179 
2180 	delta = ddpa->originusedsnap -
2181 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2182 	ASSERT3S(delta, <=, 0);
2183 	ASSERT3U(ddpa->used, >=, -delta);
2184 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2185 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2186 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2187 
2188 	origin_ds->ds_phys->ds_unique_bytes = ddpa->unique;
2189 
2190 	/* log history record */
2191 	spa_history_log_internal_ds(hds, "promote", tx, "");
2192 
2193 	dsl_dir_rele(odd, FTAG);
2194 	promote_rele(ddpa, FTAG);
2195 }
2196 
2197 /*
2198  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2199  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2200  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2201  * snapshots back to this dataset's origin.
2202  */
2203 static int
2204 snaplist_make(dsl_pool_t *dp,
2205     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2206 {
2207 	uint64_t obj = last_obj;
2208 
2209 	list_create(l, sizeof (struct promotenode),
2210 	    offsetof(struct promotenode, link));
2211 
2212 	while (obj != first_obj) {
2213 		dsl_dataset_t *ds;
2214 		struct promotenode *snap;
2215 		int err;
2216 
2217 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2218 		ASSERT(err != ENOENT);
2219 		if (err != 0)
2220 			return (err);
2221 
2222 		if (first_obj == 0)
2223 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2224 
2225 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2226 		snap->ds = ds;
2227 		list_insert_tail(l, snap);
2228 		obj = ds->ds_phys->ds_prev_snap_obj;
2229 	}
2230 
2231 	return (0);
2232 }
2233 
2234 static int
2235 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2236 {
2237 	struct promotenode *snap;
2238 
2239 	*spacep = 0;
2240 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2241 		uint64_t used, comp, uncomp;
2242 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2243 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2244 		*spacep += used;
2245 	}
2246 	return (0);
2247 }
2248 
2249 static void
2250 snaplist_destroy(list_t *l, void *tag)
2251 {
2252 	struct promotenode *snap;
2253 
2254 	if (l == NULL || !list_link_active(&l->list_head))
2255 		return;
2256 
2257 	while ((snap = list_tail(l)) != NULL) {
2258 		list_remove(l, snap);
2259 		dsl_dataset_rele(snap->ds, tag);
2260 		kmem_free(snap, sizeof (*snap));
2261 	}
2262 	list_destroy(l);
2263 }
2264 
2265 static int
2266 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2267 {
2268 	int error;
2269 	dsl_dir_t *dd;
2270 	struct promotenode *snap;
2271 
2272 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2273 	    &ddpa->ddpa_clone);
2274 	if (error != 0)
2275 		return (error);
2276 	dd = ddpa->ddpa_clone->ds_dir;
2277 
2278 	if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2279 	    !dsl_dir_is_clone(dd)) {
2280 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2281 		return (SET_ERROR(EINVAL));
2282 	}
2283 
2284 	error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj,
2285 	    &ddpa->shared_snaps, tag);
2286 	if (error != 0)
2287 		goto out;
2288 
2289 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2290 	    &ddpa->clone_snaps, tag);
2291 	if (error != 0)
2292 		goto out;
2293 
2294 	snap = list_head(&ddpa->shared_snaps);
2295 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2296 	error = snaplist_make(dp, dd->dd_phys->dd_origin_obj,
2297 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj,
2298 	    &ddpa->origin_snaps, tag);
2299 	if (error != 0)
2300 		goto out;
2301 
2302 	if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2303 		error = dsl_dataset_hold_obj(dp,
2304 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2305 		    tag, &ddpa->origin_origin);
2306 		if (error != 0)
2307 			goto out;
2308 	}
2309 out:
2310 	if (error != 0)
2311 		promote_rele(ddpa, tag);
2312 	return (error);
2313 }
2314 
2315 static void
2316 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2317 {
2318 	snaplist_destroy(&ddpa->shared_snaps, tag);
2319 	snaplist_destroy(&ddpa->clone_snaps, tag);
2320 	snaplist_destroy(&ddpa->origin_snaps, tag);
2321 	if (ddpa->origin_origin != NULL)
2322 		dsl_dataset_rele(ddpa->origin_origin, tag);
2323 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2324 }
2325 
2326 /*
2327  * Promote a clone.
2328  *
2329  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2330  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2331  */
2332 int
2333 dsl_dataset_promote(const char *name, char *conflsnap)
2334 {
2335 	dsl_dataset_promote_arg_t ddpa = { 0 };
2336 	uint64_t numsnaps;
2337 	int error;
2338 	objset_t *os;
2339 
2340 	/*
2341 	 * We will modify space proportional to the number of
2342 	 * snapshots.  Compute numsnaps.
2343 	 */
2344 	error = dmu_objset_hold(name, FTAG, &os);
2345 	if (error != 0)
2346 		return (error);
2347 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2348 	    dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps);
2349 	dmu_objset_rele(os, FTAG);
2350 	if (error != 0)
2351 		return (error);
2352 
2353 	ddpa.ddpa_clonename = name;
2354 	ddpa.err_ds = conflsnap;
2355 
2356 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2357 	    dsl_dataset_promote_sync, &ddpa, 2 + numsnaps));
2358 }
2359 
2360 int
2361 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2362     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2363 {
2364 	int64_t unused_refres_delta;
2365 
2366 	/* they should both be heads */
2367 	if (dsl_dataset_is_snapshot(clone) ||
2368 	    dsl_dataset_is_snapshot(origin_head))
2369 		return (SET_ERROR(EINVAL));
2370 
2371 	/* if we are not forcing, the branch point should be just before them */
2372 	if (!force && clone->ds_prev != origin_head->ds_prev)
2373 		return (SET_ERROR(EINVAL));
2374 
2375 	/* clone should be the clone (unless they are unrelated) */
2376 	if (clone->ds_prev != NULL &&
2377 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2378 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2379 		return (SET_ERROR(EINVAL));
2380 
2381 	/* the clone should be a child of the origin */
2382 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2383 		return (SET_ERROR(EINVAL));
2384 
2385 	/* origin_head shouldn't be modified unless 'force' */
2386 	if (!force &&
2387 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2388 		return (SET_ERROR(ETXTBSY));
2389 
2390 	/* origin_head should have no long holds (e.g. is not mounted) */
2391 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2392 		return (SET_ERROR(EBUSY));
2393 
2394 	/* check amount of any unconsumed refreservation */
2395 	unused_refres_delta =
2396 	    (int64_t)MIN(origin_head->ds_reserved,
2397 	    origin_head->ds_phys->ds_unique_bytes) -
2398 	    (int64_t)MIN(origin_head->ds_reserved,
2399 	    clone->ds_phys->ds_unique_bytes);
2400 
2401 	if (unused_refres_delta > 0 &&
2402 	    unused_refres_delta >
2403 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2404 		return (SET_ERROR(ENOSPC));
2405 
2406 	/* clone can't be over the head's refquota */
2407 	if (origin_head->ds_quota != 0 &&
2408 	    clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota)
2409 		return (SET_ERROR(EDQUOT));
2410 
2411 	return (0);
2412 }
2413 
2414 void
2415 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2416     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2417 {
2418 	dsl_pool_t *dp = dmu_tx_pool(tx);
2419 	int64_t unused_refres_delta;
2420 
2421 	ASSERT(clone->ds_reserved == 0);
2422 	ASSERT(origin_head->ds_quota == 0 ||
2423 	    clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota);
2424 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2425 
2426 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2427 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2428 
2429 	if (clone->ds_objset != NULL) {
2430 		dmu_objset_evict(clone->ds_objset);
2431 		clone->ds_objset = NULL;
2432 	}
2433 
2434 	if (origin_head->ds_objset != NULL) {
2435 		dmu_objset_evict(origin_head->ds_objset);
2436 		origin_head->ds_objset = NULL;
2437 	}
2438 
2439 	unused_refres_delta =
2440 	    (int64_t)MIN(origin_head->ds_reserved,
2441 	    origin_head->ds_phys->ds_unique_bytes) -
2442 	    (int64_t)MIN(origin_head->ds_reserved,
2443 	    clone->ds_phys->ds_unique_bytes);
2444 
2445 	/*
2446 	 * Reset origin's unique bytes, if it exists.
2447 	 */
2448 	if (clone->ds_prev) {
2449 		dsl_dataset_t *origin = clone->ds_prev;
2450 		uint64_t comp, uncomp;
2451 
2452 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2453 		dsl_deadlist_space_range(&clone->ds_deadlist,
2454 		    origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2455 		    &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
2456 	}
2457 
2458 	/* swap blkptrs */
2459 	{
2460 		blkptr_t tmp;
2461 		tmp = origin_head->ds_phys->ds_bp;
2462 		origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp;
2463 		clone->ds_phys->ds_bp = tmp;
2464 	}
2465 
2466 	/* set dd_*_bytes */
2467 	{
2468 		int64_t dused, dcomp, duncomp;
2469 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2470 		uint64_t odl_used, odl_comp, odl_uncomp;
2471 
2472 		ASSERT3U(clone->ds_dir->dd_phys->
2473 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2474 
2475 		dsl_deadlist_space(&clone->ds_deadlist,
2476 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2477 		dsl_deadlist_space(&origin_head->ds_deadlist,
2478 		    &odl_used, &odl_comp, &odl_uncomp);
2479 
2480 		dused = clone->ds_phys->ds_referenced_bytes + cdl_used -
2481 		    (origin_head->ds_phys->ds_referenced_bytes + odl_used);
2482 		dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp -
2483 		    (origin_head->ds_phys->ds_compressed_bytes + odl_comp);
2484 		duncomp = clone->ds_phys->ds_uncompressed_bytes +
2485 		    cdl_uncomp -
2486 		    (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2487 
2488 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2489 		    dused, dcomp, duncomp, tx);
2490 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2491 		    -dused, -dcomp, -duncomp, tx);
2492 
2493 		/*
2494 		 * The difference in the space used by snapshots is the
2495 		 * difference in snapshot space due to the head's
2496 		 * deadlist (since that's the only thing that's
2497 		 * changing that affects the snapused).
2498 		 */
2499 		dsl_deadlist_space_range(&clone->ds_deadlist,
2500 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2501 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2502 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2503 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2504 		    &odl_used, &odl_comp, &odl_uncomp);
2505 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2506 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2507 	}
2508 
2509 	/* swap ds_*_bytes */
2510 	SWITCH64(origin_head->ds_phys->ds_referenced_bytes,
2511 	    clone->ds_phys->ds_referenced_bytes);
2512 	SWITCH64(origin_head->ds_phys->ds_compressed_bytes,
2513 	    clone->ds_phys->ds_compressed_bytes);
2514 	SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes,
2515 	    clone->ds_phys->ds_uncompressed_bytes);
2516 	SWITCH64(origin_head->ds_phys->ds_unique_bytes,
2517 	    clone->ds_phys->ds_unique_bytes);
2518 
2519 	/* apply any parent delta for change in unconsumed refreservation */
2520 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2521 	    unused_refres_delta, 0, 0, tx);
2522 
2523 	/*
2524 	 * Swap deadlists.
2525 	 */
2526 	dsl_deadlist_close(&clone->ds_deadlist);
2527 	dsl_deadlist_close(&origin_head->ds_deadlist);
2528 	SWITCH64(origin_head->ds_phys->ds_deadlist_obj,
2529 	    clone->ds_phys->ds_deadlist_obj);
2530 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2531 	    clone->ds_phys->ds_deadlist_obj);
2532 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2533 	    origin_head->ds_phys->ds_deadlist_obj);
2534 
2535 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2536 
2537 	spa_history_log_internal_ds(clone, "clone swap", tx,
2538 	    "parent=%s", origin_head->ds_dir->dd_myname);
2539 }
2540 
2541 /*
2542  * Given a pool name and a dataset object number in that pool,
2543  * return the name of that dataset.
2544  */
2545 int
2546 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2547 {
2548 	dsl_pool_t *dp;
2549 	dsl_dataset_t *ds;
2550 	int error;
2551 
2552 	error = dsl_pool_hold(pname, FTAG, &dp);
2553 	if (error != 0)
2554 		return (error);
2555 
2556 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2557 	if (error == 0) {
2558 		dsl_dataset_name(ds, buf);
2559 		dsl_dataset_rele(ds, FTAG);
2560 	}
2561 	dsl_pool_rele(dp, FTAG);
2562 
2563 	return (error);
2564 }
2565 
2566 int
2567 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2568     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2569 {
2570 	int error = 0;
2571 
2572 	ASSERT3S(asize, >, 0);
2573 
2574 	/*
2575 	 * *ref_rsrv is the portion of asize that will come from any
2576 	 * unconsumed refreservation space.
2577 	 */
2578 	*ref_rsrv = 0;
2579 
2580 	mutex_enter(&ds->ds_lock);
2581 	/*
2582 	 * Make a space adjustment for reserved bytes.
2583 	 */
2584 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2585 		ASSERT3U(*used, >=,
2586 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2587 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2588 		*ref_rsrv =
2589 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2590 	}
2591 
2592 	if (!check_quota || ds->ds_quota == 0) {
2593 		mutex_exit(&ds->ds_lock);
2594 		return (0);
2595 	}
2596 	/*
2597 	 * If they are requesting more space, and our current estimate
2598 	 * is over quota, they get to try again unless the actual
2599 	 * on-disk is over quota and there are no pending changes (which
2600 	 * may free up space for us).
2601 	 */
2602 	if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
2603 		if (inflight > 0 ||
2604 		    ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
2605 			error = SET_ERROR(ERESTART);
2606 		else
2607 			error = SET_ERROR(EDQUOT);
2608 	}
2609 	mutex_exit(&ds->ds_lock);
2610 
2611 	return (error);
2612 }
2613 
2614 typedef struct dsl_dataset_set_qr_arg {
2615 	const char *ddsqra_name;
2616 	zprop_source_t ddsqra_source;
2617 	uint64_t ddsqra_value;
2618 } dsl_dataset_set_qr_arg_t;
2619 
2620 
2621 /* ARGSUSED */
2622 static int
2623 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2624 {
2625 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2626 	dsl_pool_t *dp = dmu_tx_pool(tx);
2627 	dsl_dataset_t *ds;
2628 	int error;
2629 	uint64_t newval;
2630 
2631 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2632 		return (SET_ERROR(ENOTSUP));
2633 
2634 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2635 	if (error != 0)
2636 		return (error);
2637 
2638 	if (dsl_dataset_is_snapshot(ds)) {
2639 		dsl_dataset_rele(ds, FTAG);
2640 		return (SET_ERROR(EINVAL));
2641 	}
2642 
2643 	error = dsl_prop_predict(ds->ds_dir,
2644 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2645 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2646 	if (error != 0) {
2647 		dsl_dataset_rele(ds, FTAG);
2648 		return (error);
2649 	}
2650 
2651 	if (newval == 0) {
2652 		dsl_dataset_rele(ds, FTAG);
2653 		return (0);
2654 	}
2655 
2656 	if (newval < ds->ds_phys->ds_referenced_bytes ||
2657 	    newval < ds->ds_reserved) {
2658 		dsl_dataset_rele(ds, FTAG);
2659 		return (SET_ERROR(ENOSPC));
2660 	}
2661 
2662 	dsl_dataset_rele(ds, FTAG);
2663 	return (0);
2664 }
2665 
2666 static void
2667 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2668 {
2669 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2670 	dsl_pool_t *dp = dmu_tx_pool(tx);
2671 	dsl_dataset_t *ds;
2672 	uint64_t newval;
2673 
2674 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2675 
2676 	dsl_prop_set_sync_impl(ds,
2677 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2678 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2679 	    &ddsqra->ddsqra_value, tx);
2680 
2681 	VERIFY0(dsl_prop_get_int_ds(ds,
2682 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2683 
2684 	if (ds->ds_quota != newval) {
2685 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2686 		ds->ds_quota = newval;
2687 	}
2688 	dsl_dataset_rele(ds, FTAG);
2689 }
2690 
2691 int
2692 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2693     uint64_t refquota)
2694 {
2695 	dsl_dataset_set_qr_arg_t ddsqra;
2696 
2697 	ddsqra.ddsqra_name = dsname;
2698 	ddsqra.ddsqra_source = source;
2699 	ddsqra.ddsqra_value = refquota;
2700 
2701 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2702 	    dsl_dataset_set_refquota_sync, &ddsqra, 0));
2703 }
2704 
2705 static int
2706 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2707 {
2708 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2709 	dsl_pool_t *dp = dmu_tx_pool(tx);
2710 	dsl_dataset_t *ds;
2711 	int error;
2712 	uint64_t newval, unique;
2713 
2714 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2715 		return (SET_ERROR(ENOTSUP));
2716 
2717 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2718 	if (error != 0)
2719 		return (error);
2720 
2721 	if (dsl_dataset_is_snapshot(ds)) {
2722 		dsl_dataset_rele(ds, FTAG);
2723 		return (SET_ERROR(EINVAL));
2724 	}
2725 
2726 	error = dsl_prop_predict(ds->ds_dir,
2727 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2728 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2729 	if (error != 0) {
2730 		dsl_dataset_rele(ds, FTAG);
2731 		return (error);
2732 	}
2733 
2734 	/*
2735 	 * If we are doing the preliminary check in open context, the
2736 	 * space estimates may be inaccurate.
2737 	 */
2738 	if (!dmu_tx_is_syncing(tx)) {
2739 		dsl_dataset_rele(ds, FTAG);
2740 		return (0);
2741 	}
2742 
2743 	mutex_enter(&ds->ds_lock);
2744 	if (!DS_UNIQUE_IS_ACCURATE(ds))
2745 		dsl_dataset_recalc_head_uniq(ds);
2746 	unique = ds->ds_phys->ds_unique_bytes;
2747 	mutex_exit(&ds->ds_lock);
2748 
2749 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2750 		uint64_t delta = MAX(unique, newval) -
2751 		    MAX(unique, ds->ds_reserved);
2752 
2753 		if (delta >
2754 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2755 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2756 			dsl_dataset_rele(ds, FTAG);
2757 			return (SET_ERROR(ENOSPC));
2758 		}
2759 	}
2760 
2761 	dsl_dataset_rele(ds, FTAG);
2762 	return (0);
2763 }
2764 
2765 void
2766 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2767     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2768 {
2769 	uint64_t newval;
2770 	uint64_t unique;
2771 	int64_t delta;
2772 
2773 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2774 	    source, sizeof (value), 1, &value, tx);
2775 
2776 	VERIFY0(dsl_prop_get_int_ds(ds,
2777 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
2778 
2779 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2780 	mutex_enter(&ds->ds_dir->dd_lock);
2781 	mutex_enter(&ds->ds_lock);
2782 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2783 	unique = ds->ds_phys->ds_unique_bytes;
2784 	delta = MAX(0, (int64_t)(newval - unique)) -
2785 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
2786 	ds->ds_reserved = newval;
2787 	mutex_exit(&ds->ds_lock);
2788 
2789 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
2790 	mutex_exit(&ds->ds_dir->dd_lock);
2791 }
2792 
2793 static void
2794 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
2795 {
2796 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2797 	dsl_pool_t *dp = dmu_tx_pool(tx);
2798 	dsl_dataset_t *ds;
2799 
2800 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2801 	dsl_dataset_set_refreservation_sync_impl(ds,
2802 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
2803 	dsl_dataset_rele(ds, FTAG);
2804 }
2805 
2806 int
2807 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
2808     uint64_t refreservation)
2809 {
2810 	dsl_dataset_set_qr_arg_t ddsqra;
2811 
2812 	ddsqra.ddsqra_name = dsname;
2813 	ddsqra.ddsqra_source = source;
2814 	ddsqra.ddsqra_value = refreservation;
2815 
2816 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
2817 	    dsl_dataset_set_refreservation_sync, &ddsqra, 0));
2818 }
2819 
2820 /*
2821  * Return (in *usedp) the amount of space written in new that is not
2822  * present in oldsnap.  New may be a snapshot or the head.  Old must be
2823  * a snapshot before new, in new's filesystem (or its origin).  If not then
2824  * fail and return EINVAL.
2825  *
2826  * The written space is calculated by considering two components:  First, we
2827  * ignore any freed space, and calculate the written as new's used space
2828  * minus old's used space.  Next, we add in the amount of space that was freed
2829  * between the two snapshots, thus reducing new's used space relative to old's.
2830  * Specifically, this is the space that was born before old->ds_creation_txg,
2831  * and freed before new (ie. on new's deadlist or a previous deadlist).
2832  *
2833  * space freed                         [---------------------]
2834  * snapshots                       ---O-------O--------O-------O------
2835  *                                         oldsnap            new
2836  */
2837 int
2838 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
2839     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2840 {
2841 	int err = 0;
2842 	uint64_t snapobj;
2843 	dsl_pool_t *dp = new->ds_dir->dd_pool;
2844 
2845 	ASSERT(dsl_pool_config_held(dp));
2846 
2847 	*usedp = 0;
2848 	*usedp += new->ds_phys->ds_referenced_bytes;
2849 	*usedp -= oldsnap->ds_phys->ds_referenced_bytes;
2850 
2851 	*compp = 0;
2852 	*compp += new->ds_phys->ds_compressed_bytes;
2853 	*compp -= oldsnap->ds_phys->ds_compressed_bytes;
2854 
2855 	*uncompp = 0;
2856 	*uncompp += new->ds_phys->ds_uncompressed_bytes;
2857 	*uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
2858 
2859 	snapobj = new->ds_object;
2860 	while (snapobj != oldsnap->ds_object) {
2861 		dsl_dataset_t *snap;
2862 		uint64_t used, comp, uncomp;
2863 
2864 		if (snapobj == new->ds_object) {
2865 			snap = new;
2866 		} else {
2867 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
2868 			if (err != 0)
2869 				break;
2870 		}
2871 
2872 		if (snap->ds_phys->ds_prev_snap_txg ==
2873 		    oldsnap->ds_phys->ds_creation_txg) {
2874 			/*
2875 			 * The blocks in the deadlist can not be born after
2876 			 * ds_prev_snap_txg, so get the whole deadlist space,
2877 			 * which is more efficient (especially for old-format
2878 			 * deadlists).  Unfortunately the deadlist code
2879 			 * doesn't have enough information to make this
2880 			 * optimization itself.
2881 			 */
2882 			dsl_deadlist_space(&snap->ds_deadlist,
2883 			    &used, &comp, &uncomp);
2884 		} else {
2885 			dsl_deadlist_space_range(&snap->ds_deadlist,
2886 			    0, oldsnap->ds_phys->ds_creation_txg,
2887 			    &used, &comp, &uncomp);
2888 		}
2889 		*usedp += used;
2890 		*compp += comp;
2891 		*uncompp += uncomp;
2892 
2893 		/*
2894 		 * If we get to the beginning of the chain of snapshots
2895 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
2896 		 * was not a snapshot of/before new.
2897 		 */
2898 		snapobj = snap->ds_phys->ds_prev_snap_obj;
2899 		if (snap != new)
2900 			dsl_dataset_rele(snap, FTAG);
2901 		if (snapobj == 0) {
2902 			err = SET_ERROR(EINVAL);
2903 			break;
2904 		}
2905 
2906 	}
2907 	return (err);
2908 }
2909 
2910 /*
2911  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
2912  * lastsnap, and all snapshots in between are deleted.
2913  *
2914  * blocks that would be freed            [---------------------------]
2915  * snapshots                       ---O-------O--------O-------O--------O
2916  *                                        firstsnap        lastsnap
2917  *
2918  * This is the set of blocks that were born after the snap before firstsnap,
2919  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
2920  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
2921  * We calculate this by iterating over the relevant deadlists (from the snap
2922  * after lastsnap, backward to the snap after firstsnap), summing up the
2923  * space on the deadlist that was born after the snap before firstsnap.
2924  */
2925 int
2926 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
2927     dsl_dataset_t *lastsnap,
2928     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2929 {
2930 	int err = 0;
2931 	uint64_t snapobj;
2932 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
2933 
2934 	ASSERT(dsl_dataset_is_snapshot(firstsnap));
2935 	ASSERT(dsl_dataset_is_snapshot(lastsnap));
2936 
2937 	/*
2938 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
2939 	 * is before lastsnap.
2940 	 */
2941 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
2942 	    firstsnap->ds_phys->ds_creation_txg >
2943 	    lastsnap->ds_phys->ds_creation_txg)
2944 		return (SET_ERROR(EINVAL));
2945 
2946 	*usedp = *compp = *uncompp = 0;
2947 
2948 	snapobj = lastsnap->ds_phys->ds_next_snap_obj;
2949 	while (snapobj != firstsnap->ds_object) {
2950 		dsl_dataset_t *ds;
2951 		uint64_t used, comp, uncomp;
2952 
2953 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
2954 		if (err != 0)
2955 			break;
2956 
2957 		dsl_deadlist_space_range(&ds->ds_deadlist,
2958 		    firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2959 		    &used, &comp, &uncomp);
2960 		*usedp += used;
2961 		*compp += comp;
2962 		*uncompp += uncomp;
2963 
2964 		snapobj = ds->ds_phys->ds_prev_snap_obj;
2965 		ASSERT3U(snapobj, !=, 0);
2966 		dsl_dataset_rele(ds, FTAG);
2967 	}
2968 	return (err);
2969 }
2970 
2971 /*
2972  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
2973  * For example, they could both be snapshots of the same filesystem, and
2974  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
2975  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
2976  * filesystem.  Or 'earlier' could be the origin's origin.
2977  *
2978  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
2979  */
2980 boolean_t
2981 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
2982 	uint64_t earlier_txg)
2983 {
2984 	dsl_pool_t *dp = later->ds_dir->dd_pool;
2985 	int error;
2986 	boolean_t ret;
2987 
2988 	ASSERT(dsl_pool_config_held(dp));
2989 	ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
2990 
2991 	if (earlier_txg == 0)
2992 		earlier_txg = earlier->ds_phys->ds_creation_txg;
2993 
2994 	if (dsl_dataset_is_snapshot(later) &&
2995 	    earlier_txg >= later->ds_phys->ds_creation_txg)
2996 		return (B_FALSE);
2997 
2998 	if (later->ds_dir == earlier->ds_dir)
2999 		return (B_TRUE);
3000 	if (!dsl_dir_is_clone(later->ds_dir))
3001 		return (B_FALSE);
3002 
3003 	if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object)
3004 		return (B_TRUE);
3005 	dsl_dataset_t *origin;
3006 	error = dsl_dataset_hold_obj(dp,
3007 	    later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
3008 	if (error != 0)
3009 		return (B_FALSE);
3010 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3011 	dsl_dataset_rele(origin, FTAG);
3012 	return (ret);
3013 }
3014 
3015 
3016 void
3017 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3018 {
3019 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3020 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3021 }
3022