xref: /illumos-gate/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 5422785d352a2bb398daceab3d1898a8aa64d006)
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) 2011, 2014 by Delphix. All rights reserved.
24  * Copyright (c) 2014, 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 /*
54  * The SPA supports block sizes up to 16MB.  However, very large blocks
55  * can have an impact on i/o latency (e.g. tying up a spinning disk for
56  * ~300ms), and also potentially on the memory allocator.  Therefore,
57  * we do not allow the recordsize to be set larger than zfs_max_recordsize
58  * (default 1MB).  Larger blocks can be created by changing this tunable,
59  * and pools with larger blocks can always be imported and used, regardless
60  * of this setting.
61  */
62 int zfs_max_recordsize = 1 * 1024 * 1024;
63 
64 #define	SWITCH64(x, y) \
65 	{ \
66 		uint64_t __tmp = (x); \
67 		(x) = (y); \
68 		(y) = __tmp; \
69 	}
70 
71 #define	DS_REF_MAX	(1ULL << 62)
72 
73 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
74 extern inline boolean_t dsl_dataset_is_snapshot(dsl_dataset_t *ds);
75 
76 /*
77  * Figure out how much of this delta should be propogated to the dsl_dir
78  * layer.  If there's a refreservation, that space has already been
79  * partially accounted for in our ancestors.
80  */
81 static int64_t
82 parent_delta(dsl_dataset_t *ds, int64_t delta)
83 {
84 	dsl_dataset_phys_t *ds_phys;
85 	uint64_t old_bytes, new_bytes;
86 
87 	if (ds->ds_reserved == 0)
88 		return (delta);
89 
90 	ds_phys = dsl_dataset_phys(ds);
91 	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
92 	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
93 
94 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
95 	return (new_bytes - old_bytes);
96 }
97 
98 void
99 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
100 {
101 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
102 	int compressed = BP_GET_PSIZE(bp);
103 	int uncompressed = BP_GET_UCSIZE(bp);
104 	int64_t delta;
105 
106 	dprintf_bp(bp, "ds=%p", ds);
107 
108 	ASSERT(dmu_tx_is_syncing(tx));
109 	/* It could have been compressed away to nothing */
110 	if (BP_IS_HOLE(bp))
111 		return;
112 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
113 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
114 	if (ds == NULL) {
115 		dsl_pool_mos_diduse_space(tx->tx_pool,
116 		    used, compressed, uncompressed);
117 		return;
118 	}
119 
120 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
121 	mutex_enter(&ds->ds_lock);
122 	delta = parent_delta(ds, used);
123 	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
124 	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
125 	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
126 	dsl_dataset_phys(ds)->ds_unique_bytes += used;
127 	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE)
128 		ds->ds_need_large_blocks = B_TRUE;
129 	mutex_exit(&ds->ds_lock);
130 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
131 	    compressed, uncompressed, tx);
132 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
133 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
134 }
135 
136 int
137 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
138     boolean_t async)
139 {
140 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
141 	int compressed = BP_GET_PSIZE(bp);
142 	int uncompressed = BP_GET_UCSIZE(bp);
143 
144 	if (BP_IS_HOLE(bp))
145 		return (0);
146 
147 	ASSERT(dmu_tx_is_syncing(tx));
148 	ASSERT(bp->blk_birth <= tx->tx_txg);
149 
150 	if (ds == NULL) {
151 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
152 		dsl_pool_mos_diduse_space(tx->tx_pool,
153 		    -used, -compressed, -uncompressed);
154 		return (used);
155 	}
156 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
157 
158 	ASSERT(!dsl_dataset_is_snapshot(ds));
159 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
160 
161 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
162 		int64_t delta;
163 
164 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
165 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
166 
167 		mutex_enter(&ds->ds_lock);
168 		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
169 		    !DS_UNIQUE_IS_ACCURATE(ds));
170 		delta = parent_delta(ds, -used);
171 		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
172 		mutex_exit(&ds->ds_lock);
173 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
174 		    delta, -compressed, -uncompressed, tx);
175 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
176 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
177 	} else {
178 		dprintf_bp(bp, "putting on dead list: %s", "");
179 		if (async) {
180 			/*
181 			 * We are here as part of zio's write done callback,
182 			 * which means we're a zio interrupt thread.  We can't
183 			 * call dsl_deadlist_insert() now because it may block
184 			 * waiting for I/O.  Instead, put bp on the deferred
185 			 * queue and let dsl_pool_sync() finish the job.
186 			 */
187 			bplist_append(&ds->ds_pending_deadlist, bp);
188 		} else {
189 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
190 		}
191 		ASSERT3U(ds->ds_prev->ds_object, ==,
192 		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
193 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
194 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
195 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
196 		    ds->ds_object && bp->blk_birth >
197 		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
198 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
199 			mutex_enter(&ds->ds_prev->ds_lock);
200 			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
201 			mutex_exit(&ds->ds_prev->ds_lock);
202 		}
203 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
204 			dsl_dir_transfer_space(ds->ds_dir, used,
205 			    DD_USED_HEAD, DD_USED_SNAP, tx);
206 		}
207 	}
208 	mutex_enter(&ds->ds_lock);
209 	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
210 	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
211 	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
212 	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
213 	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
214 	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
215 	mutex_exit(&ds->ds_lock);
216 
217 	return (used);
218 }
219 
220 uint64_t
221 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
222 {
223 	uint64_t trysnap = 0;
224 
225 	if (ds == NULL)
226 		return (0);
227 	/*
228 	 * The snapshot creation could fail, but that would cause an
229 	 * incorrect FALSE return, which would only result in an
230 	 * overestimation of the amount of space that an operation would
231 	 * consume, which is OK.
232 	 *
233 	 * There's also a small window where we could miss a pending
234 	 * snapshot, because we could set the sync task in the quiescing
235 	 * phase.  So this should only be used as a guess.
236 	 */
237 	if (ds->ds_trysnap_txg >
238 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
239 		trysnap = ds->ds_trysnap_txg;
240 	return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
241 }
242 
243 boolean_t
244 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
245     uint64_t blk_birth)
246 {
247 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
248 	    (bp != NULL && BP_IS_HOLE(bp)))
249 		return (B_FALSE);
250 
251 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
252 
253 	return (B_TRUE);
254 }
255 
256 /* ARGSUSED */
257 static void
258 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
259 {
260 	dsl_dataset_t *ds = dsv;
261 
262 	ASSERT(ds->ds_owner == NULL);
263 
264 	unique_remove(ds->ds_fsid_guid);
265 
266 	if (ds->ds_objset != NULL)
267 		dmu_objset_evict(ds->ds_objset);
268 
269 	if (ds->ds_prev) {
270 		dsl_dataset_rele(ds->ds_prev, ds);
271 		ds->ds_prev = NULL;
272 	}
273 
274 	bplist_destroy(&ds->ds_pending_deadlist);
275 	if (dsl_dataset_phys(ds)->ds_deadlist_obj != 0)
276 		dsl_deadlist_close(&ds->ds_deadlist);
277 	if (ds->ds_dir)
278 		dsl_dir_rele(ds->ds_dir, ds);
279 
280 	ASSERT(!list_link_active(&ds->ds_synced_link));
281 
282 	mutex_destroy(&ds->ds_lock);
283 	mutex_destroy(&ds->ds_opening_lock);
284 	mutex_destroy(&ds->ds_sendstream_lock);
285 	refcount_destroy(&ds->ds_longholds);
286 
287 	kmem_free(ds, sizeof (dsl_dataset_t));
288 }
289 
290 int
291 dsl_dataset_get_snapname(dsl_dataset_t *ds)
292 {
293 	dsl_dataset_phys_t *headphys;
294 	int err;
295 	dmu_buf_t *headdbuf;
296 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
297 	objset_t *mos = dp->dp_meta_objset;
298 
299 	if (ds->ds_snapname[0])
300 		return (0);
301 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
302 		return (0);
303 
304 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
305 	    FTAG, &headdbuf);
306 	if (err != 0)
307 		return (err);
308 	headphys = headdbuf->db_data;
309 	err = zap_value_search(dp->dp_meta_objset,
310 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
311 	dmu_buf_rele(headdbuf, FTAG);
312 	return (err);
313 }
314 
315 int
316 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
317 {
318 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
319 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
320 	matchtype_t mt;
321 	int err;
322 
323 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
324 		mt = MT_FIRST;
325 	else
326 		mt = MT_EXACT;
327 
328 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
329 	    value, mt, NULL, 0, NULL);
330 	if (err == ENOTSUP && mt == MT_FIRST)
331 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
332 	return (err);
333 }
334 
335 int
336 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
337     boolean_t adj_cnt)
338 {
339 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
340 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
341 	matchtype_t mt;
342 	int err;
343 
344 	dsl_dir_snap_cmtime_update(ds->ds_dir);
345 
346 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
347 		mt = MT_FIRST;
348 	else
349 		mt = MT_EXACT;
350 
351 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
352 	if (err == ENOTSUP && mt == MT_FIRST)
353 		err = zap_remove(mos, snapobj, name, tx);
354 
355 	if (err == 0 && adj_cnt)
356 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
357 		    DD_FIELD_SNAPSHOT_COUNT, tx);
358 
359 	return (err);
360 }
361 
362 int
363 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
364     dsl_dataset_t **dsp)
365 {
366 	objset_t *mos = dp->dp_meta_objset;
367 	dmu_buf_t *dbuf;
368 	dsl_dataset_t *ds;
369 	int err;
370 	dmu_object_info_t doi;
371 
372 	ASSERT(dsl_pool_config_held(dp));
373 
374 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
375 	if (err != 0)
376 		return (err);
377 
378 	/* Make sure dsobj has the correct object type. */
379 	dmu_object_info_from_db(dbuf, &doi);
380 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
381 		dmu_buf_rele(dbuf, tag);
382 		return (SET_ERROR(EINVAL));
383 	}
384 
385 	ds = dmu_buf_get_user(dbuf);
386 	if (ds == NULL) {
387 		dsl_dataset_t *winner = NULL;
388 
389 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
390 		ds->ds_dbuf = dbuf;
391 		ds->ds_object = dsobj;
392 
393 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
394 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
395 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
396 		refcount_create(&ds->ds_longholds);
397 
398 		bplist_create(&ds->ds_pending_deadlist);
399 		dsl_deadlist_open(&ds->ds_deadlist,
400 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
401 
402 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
403 		    offsetof(dmu_sendarg_t, dsa_link));
404 
405 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
406 			err = zap_contains(mos, dsobj, DS_FIELD_LARGE_BLOCKS);
407 			if (err == 0)
408 				ds->ds_large_blocks = B_TRUE;
409 			else
410 				ASSERT3U(err, ==, ENOENT);
411 		}
412 
413 		if (err == 0) {
414 			err = dsl_dir_hold_obj(dp,
415 			    dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds,
416 			    &ds->ds_dir);
417 		}
418 		if (err != 0) {
419 			mutex_destroy(&ds->ds_lock);
420 			mutex_destroy(&ds->ds_opening_lock);
421 			mutex_destroy(&ds->ds_sendstream_lock);
422 			refcount_destroy(&ds->ds_longholds);
423 			bplist_destroy(&ds->ds_pending_deadlist);
424 			dsl_deadlist_close(&ds->ds_deadlist);
425 			kmem_free(ds, sizeof (dsl_dataset_t));
426 			dmu_buf_rele(dbuf, tag);
427 			return (err);
428 		}
429 
430 		if (!dsl_dataset_is_snapshot(ds)) {
431 			ds->ds_snapname[0] = '\0';
432 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
433 				err = dsl_dataset_hold_obj(dp,
434 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
435 				    ds, &ds->ds_prev);
436 			}
437 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
438 				int zaperr = zap_lookup(mos, ds->ds_object,
439 				    DS_FIELD_BOOKMARK_NAMES,
440 				    sizeof (ds->ds_bookmarks), 1,
441 				    &ds->ds_bookmarks);
442 				if (zaperr != ENOENT)
443 					VERIFY0(zaperr);
444 			}
445 		} else {
446 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
447 				err = dsl_dataset_get_snapname(ds);
448 			if (err == 0 &&
449 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
450 				err = zap_count(
451 				    ds->ds_dir->dd_pool->dp_meta_objset,
452 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
453 				    &ds->ds_userrefs);
454 			}
455 		}
456 
457 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
458 			err = dsl_prop_get_int_ds(ds,
459 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
460 			    &ds->ds_reserved);
461 			if (err == 0) {
462 				err = dsl_prop_get_int_ds(ds,
463 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
464 				    &ds->ds_quota);
465 			}
466 		} else {
467 			ds->ds_reserved = ds->ds_quota = 0;
468 		}
469 
470 		if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
471 		    dsl_dataset_evict)) != NULL) {
472 			bplist_destroy(&ds->ds_pending_deadlist);
473 			dsl_deadlist_close(&ds->ds_deadlist);
474 			if (ds->ds_prev)
475 				dsl_dataset_rele(ds->ds_prev, ds);
476 			dsl_dir_rele(ds->ds_dir, ds);
477 			mutex_destroy(&ds->ds_lock);
478 			mutex_destroy(&ds->ds_opening_lock);
479 			mutex_destroy(&ds->ds_sendstream_lock);
480 			refcount_destroy(&ds->ds_longholds);
481 			kmem_free(ds, sizeof (dsl_dataset_t));
482 			if (err != 0) {
483 				dmu_buf_rele(dbuf, tag);
484 				return (err);
485 			}
486 			ds = winner;
487 		} else {
488 			ds->ds_fsid_guid =
489 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
490 		}
491 	}
492 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
493 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
494 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
495 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
496 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
497 	*dsp = ds;
498 	return (0);
499 }
500 
501 int
502 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
503     void *tag, dsl_dataset_t **dsp)
504 {
505 	dsl_dir_t *dd;
506 	const char *snapname;
507 	uint64_t obj;
508 	int err = 0;
509 
510 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
511 	if (err != 0)
512 		return (err);
513 
514 	ASSERT(dsl_pool_config_held(dp));
515 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
516 	if (obj != 0)
517 		err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
518 	else
519 		err = SET_ERROR(ENOENT);
520 
521 	/* we may be looking for a snapshot */
522 	if (err == 0 && snapname != NULL) {
523 		dsl_dataset_t *ds;
524 
525 		if (*snapname++ != '@') {
526 			dsl_dataset_rele(*dsp, tag);
527 			dsl_dir_rele(dd, FTAG);
528 			return (SET_ERROR(ENOENT));
529 		}
530 
531 		dprintf("looking for snapshot '%s'\n", snapname);
532 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
533 		if (err == 0)
534 			err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
535 		dsl_dataset_rele(*dsp, tag);
536 
537 		if (err == 0) {
538 			mutex_enter(&ds->ds_lock);
539 			if (ds->ds_snapname[0] == 0)
540 				(void) strlcpy(ds->ds_snapname, snapname,
541 				    sizeof (ds->ds_snapname));
542 			mutex_exit(&ds->ds_lock);
543 			*dsp = ds;
544 		}
545 	}
546 
547 	dsl_dir_rele(dd, FTAG);
548 	return (err);
549 }
550 
551 int
552 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
553     void *tag, dsl_dataset_t **dsp)
554 {
555 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
556 	if (err != 0)
557 		return (err);
558 	if (!dsl_dataset_tryown(*dsp, tag)) {
559 		dsl_dataset_rele(*dsp, tag);
560 		*dsp = NULL;
561 		return (SET_ERROR(EBUSY));
562 	}
563 	return (0);
564 }
565 
566 int
567 dsl_dataset_own(dsl_pool_t *dp, const char *name,
568     void *tag, dsl_dataset_t **dsp)
569 {
570 	int err = dsl_dataset_hold(dp, name, tag, dsp);
571 	if (err != 0)
572 		return (err);
573 	if (!dsl_dataset_tryown(*dsp, tag)) {
574 		dsl_dataset_rele(*dsp, tag);
575 		return (SET_ERROR(EBUSY));
576 	}
577 	return (0);
578 }
579 
580 /*
581  * See the comment above dsl_pool_hold() for details.  In summary, a long
582  * hold is used to prevent destruction of a dataset while the pool hold
583  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
584  *
585  * The dataset and pool must be held when this function is called.  After it
586  * is called, the pool hold may be released while the dataset is still held
587  * and accessed.
588  */
589 void
590 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
591 {
592 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
593 	(void) refcount_add(&ds->ds_longholds, tag);
594 }
595 
596 void
597 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
598 {
599 	(void) refcount_remove(&ds->ds_longholds, tag);
600 }
601 
602 /* Return B_TRUE if there are any long holds on this dataset. */
603 boolean_t
604 dsl_dataset_long_held(dsl_dataset_t *ds)
605 {
606 	return (!refcount_is_zero(&ds->ds_longholds));
607 }
608 
609 void
610 dsl_dataset_name(dsl_dataset_t *ds, char *name)
611 {
612 	if (ds == NULL) {
613 		(void) strcpy(name, "mos");
614 	} else {
615 		dsl_dir_name(ds->ds_dir, name);
616 		VERIFY0(dsl_dataset_get_snapname(ds));
617 		if (ds->ds_snapname[0]) {
618 			(void) strcat(name, "@");
619 			/*
620 			 * We use a "recursive" mutex so that we
621 			 * can call dprintf_ds() with ds_lock held.
622 			 */
623 			if (!MUTEX_HELD(&ds->ds_lock)) {
624 				mutex_enter(&ds->ds_lock);
625 				(void) strcat(name, ds->ds_snapname);
626 				mutex_exit(&ds->ds_lock);
627 			} else {
628 				(void) strcat(name, ds->ds_snapname);
629 			}
630 		}
631 	}
632 }
633 
634 void
635 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
636 {
637 	dmu_buf_rele(ds->ds_dbuf, tag);
638 }
639 
640 void
641 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
642 {
643 	ASSERT3P(ds->ds_owner, ==, tag);
644 	ASSERT(ds->ds_dbuf != NULL);
645 
646 	mutex_enter(&ds->ds_lock);
647 	ds->ds_owner = NULL;
648 	mutex_exit(&ds->ds_lock);
649 	dsl_dataset_long_rele(ds, tag);
650 	dsl_dataset_rele(ds, tag);
651 }
652 
653 boolean_t
654 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
655 {
656 	boolean_t gotit = FALSE;
657 
658 	mutex_enter(&ds->ds_lock);
659 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
660 		ds->ds_owner = tag;
661 		dsl_dataset_long_hold(ds, tag);
662 		gotit = TRUE;
663 	}
664 	mutex_exit(&ds->ds_lock);
665 	return (gotit);
666 }
667 
668 uint64_t
669 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
670     uint64_t flags, dmu_tx_t *tx)
671 {
672 	dsl_pool_t *dp = dd->dd_pool;
673 	dmu_buf_t *dbuf;
674 	dsl_dataset_phys_t *dsphys;
675 	uint64_t dsobj;
676 	objset_t *mos = dp->dp_meta_objset;
677 
678 	if (origin == NULL)
679 		origin = dp->dp_origin_snap;
680 
681 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
682 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
683 	ASSERT(dmu_tx_is_syncing(tx));
684 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
685 
686 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
687 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
688 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
689 	dmu_buf_will_dirty(dbuf, tx);
690 	dsphys = dbuf->db_data;
691 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
692 	dsphys->ds_dir_obj = dd->dd_object;
693 	dsphys->ds_flags = flags;
694 	dsphys->ds_fsid_guid = unique_create();
695 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
696 	    sizeof (dsphys->ds_guid));
697 	dsphys->ds_snapnames_zapobj =
698 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
699 	    DMU_OT_NONE, 0, tx);
700 	dsphys->ds_creation_time = gethrestime_sec();
701 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
702 
703 	if (origin == NULL) {
704 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
705 	} else {
706 		dsl_dataset_t *ohds; /* head of the origin snapshot */
707 
708 		dsphys->ds_prev_snap_obj = origin->ds_object;
709 		dsphys->ds_prev_snap_txg =
710 		    dsl_dataset_phys(origin)->ds_creation_txg;
711 		dsphys->ds_referenced_bytes =
712 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
713 		dsphys->ds_compressed_bytes =
714 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
715 		dsphys->ds_uncompressed_bytes =
716 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
717 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
718 
719 		/*
720 		 * Inherit flags that describe the dataset's contents
721 		 * (INCONSISTENT) or properties (Case Insensitive).
722 		 */
723 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
724 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
725 
726 		if (origin->ds_large_blocks)
727 			dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
728 
729 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
730 		dsl_dataset_phys(origin)->ds_num_children++;
731 
732 		VERIFY0(dsl_dataset_hold_obj(dp,
733 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
734 		    FTAG, &ohds));
735 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
736 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
737 		dsl_dataset_rele(ohds, FTAG);
738 
739 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
740 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
741 				dsl_dataset_phys(origin)->ds_next_clones_obj =
742 				    zap_create(mos,
743 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
744 			}
745 			VERIFY0(zap_add_int(mos,
746 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
747 			    dsobj, tx));
748 		}
749 
750 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
751 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
752 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
753 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
754 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
755 				dsl_dir_phys(origin->ds_dir)->dd_clones =
756 				    zap_create(mos,
757 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
758 			}
759 			VERIFY0(zap_add_int(mos,
760 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
761 			    dsobj, tx));
762 		}
763 	}
764 
765 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
766 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
767 
768 	dmu_buf_rele(dbuf, FTAG);
769 
770 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
771 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
772 
773 	return (dsobj);
774 }
775 
776 static void
777 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
778 {
779 	objset_t *os;
780 
781 	VERIFY0(dmu_objset_from_ds(ds, &os));
782 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
783 	dsl_dataset_dirty(ds, tx);
784 }
785 
786 uint64_t
787 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
788     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
789 {
790 	dsl_pool_t *dp = pdd->dd_pool;
791 	uint64_t dsobj, ddobj;
792 	dsl_dir_t *dd;
793 
794 	ASSERT(dmu_tx_is_syncing(tx));
795 	ASSERT(lastname[0] != '@');
796 
797 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
798 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
799 
800 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
801 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
802 
803 	dsl_deleg_set_create_perms(dd, tx, cr);
804 
805 	/*
806 	 * Since we're creating a new node we know it's a leaf, so we can
807 	 * initialize the counts if the limit feature is active.
808 	 */
809 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
810 		uint64_t cnt = 0;
811 		objset_t *os = dd->dd_pool->dp_meta_objset;
812 
813 		dsl_dir_zapify(dd, tx);
814 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
815 		    sizeof (cnt), 1, &cnt, tx));
816 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
817 		    sizeof (cnt), 1, &cnt, tx));
818 	}
819 
820 	dsl_dir_rele(dd, FTAG);
821 
822 	/*
823 	 * If we are creating a clone, make sure we zero out any stale
824 	 * data from the origin snapshots zil header.
825 	 */
826 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
827 		dsl_dataset_t *ds;
828 
829 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
830 		dsl_dataset_zero_zil(ds, tx);
831 		dsl_dataset_rele(ds, FTAG);
832 	}
833 
834 	return (dsobj);
835 }
836 
837 /*
838  * The unique space in the head dataset can be calculated by subtracting
839  * the space used in the most recent snapshot, that is still being used
840  * in this file system, from the space currently in use.  To figure out
841  * the space in the most recent snapshot still in use, we need to take
842  * the total space used in the snapshot and subtract out the space that
843  * has been freed up since the snapshot was taken.
844  */
845 void
846 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
847 {
848 	uint64_t mrs_used;
849 	uint64_t dlused, dlcomp, dluncomp;
850 
851 	ASSERT(!dsl_dataset_is_snapshot(ds));
852 
853 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
854 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
855 	else
856 		mrs_used = 0;
857 
858 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
859 
860 	ASSERT3U(dlused, <=, mrs_used);
861 	dsl_dataset_phys(ds)->ds_unique_bytes =
862 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
863 
864 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
865 	    SPA_VERSION_UNIQUE_ACCURATE)
866 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
867 }
868 
869 void
870 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
871     dmu_tx_t *tx)
872 {
873 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
874 	uint64_t count;
875 	int err;
876 
877 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
878 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
879 	    obj, tx);
880 	/*
881 	 * The err should not be ENOENT, but a bug in a previous version
882 	 * of the code could cause upgrade_clones_cb() to not set
883 	 * ds_next_snap_obj when it should, leading to a missing entry.
884 	 * If we knew that the pool was created after
885 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
886 	 * ENOENT.  However, at least we can check that we don't have
887 	 * too many entries in the next_clones_obj even after failing to
888 	 * remove this one.
889 	 */
890 	if (err != ENOENT)
891 		VERIFY0(err);
892 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
893 	    &count));
894 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
895 }
896 
897 
898 blkptr_t *
899 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
900 {
901 	return (&dsl_dataset_phys(ds)->ds_bp);
902 }
903 
904 void
905 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
906 {
907 	ASSERT(dmu_tx_is_syncing(tx));
908 	/* If it's the meta-objset, set dp_meta_rootbp */
909 	if (ds == NULL) {
910 		tx->tx_pool->dp_meta_rootbp = *bp;
911 	} else {
912 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
913 		dsl_dataset_phys(ds)->ds_bp = *bp;
914 	}
915 }
916 
917 spa_t *
918 dsl_dataset_get_spa(dsl_dataset_t *ds)
919 {
920 	return (ds->ds_dir->dd_pool->dp_spa);
921 }
922 
923 void
924 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
925 {
926 	dsl_pool_t *dp;
927 
928 	if (ds == NULL) /* this is the meta-objset */
929 		return;
930 
931 	ASSERT(ds->ds_objset != NULL);
932 
933 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
934 		panic("dirtying snapshot!");
935 
936 	dp = ds->ds_dir->dd_pool;
937 
938 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
939 		/* up the hold count until we can be written out */
940 		dmu_buf_add_ref(ds->ds_dbuf, ds);
941 	}
942 }
943 
944 boolean_t
945 dsl_dataset_is_dirty(dsl_dataset_t *ds)
946 {
947 	for (int t = 0; t < TXG_SIZE; t++) {
948 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
949 		    ds, t))
950 			return (B_TRUE);
951 	}
952 	return (B_FALSE);
953 }
954 
955 static int
956 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
957 {
958 	uint64_t asize;
959 
960 	if (!dmu_tx_is_syncing(tx))
961 		return (0);
962 
963 	/*
964 	 * If there's an fs-only reservation, any blocks that might become
965 	 * owned by the snapshot dataset must be accommodated by space
966 	 * outside of the reservation.
967 	 */
968 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
969 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
970 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
971 		return (SET_ERROR(ENOSPC));
972 
973 	/*
974 	 * Propagate any reserved space for this snapshot to other
975 	 * snapshot checks in this sync group.
976 	 */
977 	if (asize > 0)
978 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
979 
980 	return (0);
981 }
982 
983 typedef struct dsl_dataset_snapshot_arg {
984 	nvlist_t *ddsa_snaps;
985 	nvlist_t *ddsa_props;
986 	nvlist_t *ddsa_errors;
987 	cred_t *ddsa_cr;
988 } dsl_dataset_snapshot_arg_t;
989 
990 int
991 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
992     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
993 {
994 	int error;
995 	uint64_t value;
996 
997 	ds->ds_trysnap_txg = tx->tx_txg;
998 
999 	if (!dmu_tx_is_syncing(tx))
1000 		return (0);
1001 
1002 	/*
1003 	 * We don't allow multiple snapshots of the same txg.  If there
1004 	 * is already one, try again.
1005 	 */
1006 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1007 		return (SET_ERROR(EAGAIN));
1008 
1009 	/*
1010 	 * Check for conflicting snapshot name.
1011 	 */
1012 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1013 	if (error == 0)
1014 		return (SET_ERROR(EEXIST));
1015 	if (error != ENOENT)
1016 		return (error);
1017 
1018 	/*
1019 	 * We don't allow taking snapshots of inconsistent datasets, such as
1020 	 * those into which we are currently receiving.  However, if we are
1021 	 * creating this snapshot as part of a receive, this check will be
1022 	 * executed atomically with respect to the completion of the receive
1023 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1024 	 * case we ignore this, knowing it will be fixed up for us shortly in
1025 	 * dmu_recv_end_sync().
1026 	 */
1027 	if (!recv && DS_IS_INCONSISTENT(ds))
1028 		return (SET_ERROR(EBUSY));
1029 
1030 	/*
1031 	 * Skip the check for temporary snapshots or if we have already checked
1032 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1033 	 * check the count here when we're receiving a stream.
1034 	 */
1035 	if (cnt != 0 && cr != NULL) {
1036 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1037 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1038 		if (error != 0)
1039 			return (error);
1040 	}
1041 
1042 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1043 	if (error != 0)
1044 		return (error);
1045 
1046 	return (0);
1047 }
1048 
1049 static int
1050 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1051 {
1052 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1053 	dsl_pool_t *dp = dmu_tx_pool(tx);
1054 	nvpair_t *pair;
1055 	int rv = 0;
1056 
1057 	/*
1058 	 * Pre-compute how many total new snapshots will be created for each
1059 	 * level in the tree and below. This is needed for validating the
1060 	 * snapshot limit when either taking a recursive snapshot or when
1061 	 * taking multiple snapshots.
1062 	 *
1063 	 * The problem is that the counts are not actually adjusted when
1064 	 * we are checking, only when we finally sync. For a single snapshot,
1065 	 * this is easy, the count will increase by 1 at each node up the tree,
1066 	 * but its more complicated for the recursive/multiple snapshot case.
1067 	 *
1068 	 * The dsl_fs_ss_limit_check function does recursively check the count
1069 	 * at each level up the tree but since it is validating each snapshot
1070 	 * independently we need to be sure that we are validating the complete
1071 	 * count for the entire set of snapshots. We do this by rolling up the
1072 	 * counts for each component of the name into an nvlist and then
1073 	 * checking each of those cases with the aggregated count.
1074 	 *
1075 	 * This approach properly handles not only the recursive snapshot
1076 	 * case (where we get all of those on the ddsa_snaps list) but also
1077 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1078 	 * validate the limit on 'a' using a count of 2).
1079 	 *
1080 	 * We validate the snapshot names in the third loop and only report
1081 	 * name errors once.
1082 	 */
1083 	if (dmu_tx_is_syncing(tx)) {
1084 		nvlist_t *cnt_track = NULL;
1085 		cnt_track = fnvlist_alloc();
1086 
1087 		/* Rollup aggregated counts into the cnt_track list */
1088 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1089 		    pair != NULL;
1090 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1091 			char *pdelim;
1092 			uint64_t val;
1093 			char nm[MAXPATHLEN];
1094 
1095 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1096 			pdelim = strchr(nm, '@');
1097 			if (pdelim == NULL)
1098 				continue;
1099 			*pdelim = '\0';
1100 
1101 			do {
1102 				if (nvlist_lookup_uint64(cnt_track, nm,
1103 				    &val) == 0) {
1104 					/* update existing entry */
1105 					fnvlist_add_uint64(cnt_track, nm,
1106 					    val + 1);
1107 				} else {
1108 					/* add to list */
1109 					fnvlist_add_uint64(cnt_track, nm, 1);
1110 				}
1111 
1112 				pdelim = strrchr(nm, '/');
1113 				if (pdelim != NULL)
1114 					*pdelim = '\0';
1115 			} while (pdelim != NULL);
1116 		}
1117 
1118 		/* Check aggregated counts at each level */
1119 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1120 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1121 			int error = 0;
1122 			char *name;
1123 			uint64_t cnt = 0;
1124 			dsl_dataset_t *ds;
1125 
1126 			name = nvpair_name(pair);
1127 			cnt = fnvpair_value_uint64(pair);
1128 			ASSERT(cnt > 0);
1129 
1130 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1131 			if (error == 0) {
1132 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1133 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1134 				    ddsa->ddsa_cr);
1135 				dsl_dataset_rele(ds, FTAG);
1136 			}
1137 
1138 			if (error != 0) {
1139 				if (ddsa->ddsa_errors != NULL)
1140 					fnvlist_add_int32(ddsa->ddsa_errors,
1141 					    name, error);
1142 				rv = error;
1143 				/* only report one error for this check */
1144 				break;
1145 			}
1146 		}
1147 		nvlist_free(cnt_track);
1148 	}
1149 
1150 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1151 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1152 		int error = 0;
1153 		dsl_dataset_t *ds;
1154 		char *name, *atp;
1155 		char dsname[MAXNAMELEN];
1156 
1157 		name = nvpair_name(pair);
1158 		if (strlen(name) >= MAXNAMELEN)
1159 			error = SET_ERROR(ENAMETOOLONG);
1160 		if (error == 0) {
1161 			atp = strchr(name, '@');
1162 			if (atp == NULL)
1163 				error = SET_ERROR(EINVAL);
1164 			if (error == 0)
1165 				(void) strlcpy(dsname, name, atp - name + 1);
1166 		}
1167 		if (error == 0)
1168 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1169 		if (error == 0) {
1170 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1171 			error = dsl_dataset_snapshot_check_impl(ds,
1172 			    atp + 1, tx, B_FALSE, 0, NULL);
1173 			dsl_dataset_rele(ds, FTAG);
1174 		}
1175 
1176 		if (error != 0) {
1177 			if (ddsa->ddsa_errors != NULL) {
1178 				fnvlist_add_int32(ddsa->ddsa_errors,
1179 				    name, error);
1180 			}
1181 			rv = error;
1182 		}
1183 	}
1184 
1185 	return (rv);
1186 }
1187 
1188 void
1189 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1190     dmu_tx_t *tx)
1191 {
1192 	static zil_header_t zero_zil;
1193 
1194 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1195 	dmu_buf_t *dbuf;
1196 	dsl_dataset_phys_t *dsphys;
1197 	uint64_t dsobj, crtxg;
1198 	objset_t *mos = dp->dp_meta_objset;
1199 	objset_t *os;
1200 
1201 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1202 
1203 	/*
1204 	 * If we are on an old pool, the zil must not be active, in which
1205 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1206 	 */
1207 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1208 	    dmu_objset_from_ds(ds, &os) != 0 ||
1209 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1210 	    sizeof (zero_zil)) == 0);
1211 
1212 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1213 
1214 	/*
1215 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1216 	 */
1217 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1218 		crtxg = 1;
1219 	else
1220 		crtxg = tx->tx_txg;
1221 
1222 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1223 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1224 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1225 	dmu_buf_will_dirty(dbuf, tx);
1226 	dsphys = dbuf->db_data;
1227 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1228 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1229 	dsphys->ds_fsid_guid = unique_create();
1230 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1231 	    sizeof (dsphys->ds_guid));
1232 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1233 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1234 	dsphys->ds_next_snap_obj = ds->ds_object;
1235 	dsphys->ds_num_children = 1;
1236 	dsphys->ds_creation_time = gethrestime_sec();
1237 	dsphys->ds_creation_txg = crtxg;
1238 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1239 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1240 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1241 	dsphys->ds_uncompressed_bytes =
1242 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1243 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1244 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1245 	dmu_buf_rele(dbuf, FTAG);
1246 
1247 	if (ds->ds_large_blocks)
1248 		dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1249 
1250 	ASSERT3U(ds->ds_prev != 0, ==,
1251 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1252 	if (ds->ds_prev) {
1253 		uint64_t next_clones_obj =
1254 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1255 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1256 		    ds->ds_object ||
1257 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1258 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1259 		    ds->ds_object) {
1260 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1261 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1262 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1263 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1264 		} else if (next_clones_obj != 0) {
1265 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1266 			    dsphys->ds_next_snap_obj, tx);
1267 			VERIFY0(zap_add_int(mos,
1268 			    next_clones_obj, dsobj, tx));
1269 		}
1270 	}
1271 
1272 	/*
1273 	 * If we have a reference-reservation on this dataset, we will
1274 	 * need to increase the amount of refreservation being charged
1275 	 * since our unique space is going to zero.
1276 	 */
1277 	if (ds->ds_reserved) {
1278 		int64_t delta;
1279 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1280 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1281 		    ds->ds_reserved);
1282 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1283 		    delta, 0, 0, tx);
1284 	}
1285 
1286 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1287 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1288 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1289 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1290 	dsl_deadlist_close(&ds->ds_deadlist);
1291 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1292 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1293 	dsl_deadlist_add_key(&ds->ds_deadlist,
1294 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1295 
1296 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1297 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1298 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1299 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1300 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1301 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1302 
1303 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1304 	    snapname, 8, 1, &dsobj, tx));
1305 
1306 	if (ds->ds_prev)
1307 		dsl_dataset_rele(ds->ds_prev, ds);
1308 	VERIFY0(dsl_dataset_hold_obj(dp,
1309 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1310 
1311 	dsl_scan_ds_snapshotted(ds, tx);
1312 
1313 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1314 
1315 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1316 }
1317 
1318 static void
1319 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1320 {
1321 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1322 	dsl_pool_t *dp = dmu_tx_pool(tx);
1323 	nvpair_t *pair;
1324 
1325 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1326 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1327 		dsl_dataset_t *ds;
1328 		char *name, *atp;
1329 		char dsname[MAXNAMELEN];
1330 
1331 		name = nvpair_name(pair);
1332 		atp = strchr(name, '@');
1333 		(void) strlcpy(dsname, name, atp - name + 1);
1334 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1335 
1336 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1337 		if (ddsa->ddsa_props != NULL) {
1338 			dsl_props_set_sync_impl(ds->ds_prev,
1339 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1340 		}
1341 		dsl_dataset_rele(ds, FTAG);
1342 	}
1343 }
1344 
1345 /*
1346  * The snapshots must all be in the same pool.
1347  * All-or-nothing: if there are any failures, nothing will be modified.
1348  */
1349 int
1350 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1351 {
1352 	dsl_dataset_snapshot_arg_t ddsa;
1353 	nvpair_t *pair;
1354 	boolean_t needsuspend;
1355 	int error;
1356 	spa_t *spa;
1357 	char *firstname;
1358 	nvlist_t *suspended = NULL;
1359 
1360 	pair = nvlist_next_nvpair(snaps, NULL);
1361 	if (pair == NULL)
1362 		return (0);
1363 	firstname = nvpair_name(pair);
1364 
1365 	error = spa_open(firstname, &spa, FTAG);
1366 	if (error != 0)
1367 		return (error);
1368 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1369 	spa_close(spa, FTAG);
1370 
1371 	if (needsuspend) {
1372 		suspended = fnvlist_alloc();
1373 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1374 		    pair = nvlist_next_nvpair(snaps, pair)) {
1375 			char fsname[MAXNAMELEN];
1376 			char *snapname = nvpair_name(pair);
1377 			char *atp;
1378 			void *cookie;
1379 
1380 			atp = strchr(snapname, '@');
1381 			if (atp == NULL) {
1382 				error = SET_ERROR(EINVAL);
1383 				break;
1384 			}
1385 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1386 
1387 			error = zil_suspend(fsname, &cookie);
1388 			if (error != 0)
1389 				break;
1390 			fnvlist_add_uint64(suspended, fsname,
1391 			    (uintptr_t)cookie);
1392 		}
1393 	}
1394 
1395 	ddsa.ddsa_snaps = snaps;
1396 	ddsa.ddsa_props = props;
1397 	ddsa.ddsa_errors = errors;
1398 	ddsa.ddsa_cr = CRED();
1399 
1400 	if (error == 0) {
1401 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1402 		    dsl_dataset_snapshot_sync, &ddsa,
1403 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1404 	}
1405 
1406 	if (suspended != NULL) {
1407 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1408 		    pair = nvlist_next_nvpair(suspended, pair)) {
1409 			zil_resume((void *)(uintptr_t)
1410 			    fnvpair_value_uint64(pair));
1411 		}
1412 		fnvlist_free(suspended);
1413 	}
1414 
1415 	return (error);
1416 }
1417 
1418 typedef struct dsl_dataset_snapshot_tmp_arg {
1419 	const char *ddsta_fsname;
1420 	const char *ddsta_snapname;
1421 	minor_t ddsta_cleanup_minor;
1422 	const char *ddsta_htag;
1423 } dsl_dataset_snapshot_tmp_arg_t;
1424 
1425 static int
1426 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1427 {
1428 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1429 	dsl_pool_t *dp = dmu_tx_pool(tx);
1430 	dsl_dataset_t *ds;
1431 	int error;
1432 
1433 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1434 	if (error != 0)
1435 		return (error);
1436 
1437 	/* NULL cred means no limit check for tmp snapshot */
1438 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1439 	    tx, B_FALSE, 0, NULL);
1440 	if (error != 0) {
1441 		dsl_dataset_rele(ds, FTAG);
1442 		return (error);
1443 	}
1444 
1445 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1446 		dsl_dataset_rele(ds, FTAG);
1447 		return (SET_ERROR(ENOTSUP));
1448 	}
1449 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1450 	    B_TRUE, tx);
1451 	if (error != 0) {
1452 		dsl_dataset_rele(ds, FTAG);
1453 		return (error);
1454 	}
1455 
1456 	dsl_dataset_rele(ds, FTAG);
1457 	return (0);
1458 }
1459 
1460 static void
1461 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1462 {
1463 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1464 	dsl_pool_t *dp = dmu_tx_pool(tx);
1465 	dsl_dataset_t *ds;
1466 
1467 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1468 
1469 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1470 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1471 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1472 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1473 
1474 	dsl_dataset_rele(ds, FTAG);
1475 }
1476 
1477 int
1478 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1479     minor_t cleanup_minor, const char *htag)
1480 {
1481 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1482 	int error;
1483 	spa_t *spa;
1484 	boolean_t needsuspend;
1485 	void *cookie;
1486 
1487 	ddsta.ddsta_fsname = fsname;
1488 	ddsta.ddsta_snapname = snapname;
1489 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1490 	ddsta.ddsta_htag = htag;
1491 
1492 	error = spa_open(fsname, &spa, FTAG);
1493 	if (error != 0)
1494 		return (error);
1495 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1496 	spa_close(spa, FTAG);
1497 
1498 	if (needsuspend) {
1499 		error = zil_suspend(fsname, &cookie);
1500 		if (error != 0)
1501 			return (error);
1502 	}
1503 
1504 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1505 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1506 
1507 	if (needsuspend)
1508 		zil_resume(cookie);
1509 	return (error);
1510 }
1511 
1512 
1513 void
1514 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1515 {
1516 	ASSERT(dmu_tx_is_syncing(tx));
1517 	ASSERT(ds->ds_objset != NULL);
1518 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1519 
1520 	/*
1521 	 * in case we had to change ds_fsid_guid when we opened it,
1522 	 * sync it out now.
1523 	 */
1524 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1525 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1526 
1527 	dmu_objset_sync(ds->ds_objset, zio, tx);
1528 
1529 	if (ds->ds_need_large_blocks && !ds->ds_large_blocks) {
1530 		dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
1531 		ds->ds_large_blocks = B_TRUE;
1532 	}
1533 }
1534 
1535 static void
1536 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1537 {
1538 	uint64_t count = 0;
1539 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1540 	zap_cursor_t zc;
1541 	zap_attribute_t za;
1542 	nvlist_t *propval = fnvlist_alloc();
1543 	nvlist_t *val = fnvlist_alloc();
1544 
1545 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1546 
1547 	/*
1548 	 * There may be missing entries in ds_next_clones_obj
1549 	 * due to a bug in a previous version of the code.
1550 	 * Only trust it if it has the right number of entries.
1551 	 */
1552 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1553 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1554 		    &count));
1555 	}
1556 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1557 		goto fail;
1558 	for (zap_cursor_init(&zc, mos,
1559 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1560 	    zap_cursor_retrieve(&zc, &za) == 0;
1561 	    zap_cursor_advance(&zc)) {
1562 		dsl_dataset_t *clone;
1563 		char buf[ZFS_MAXNAMELEN];
1564 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1565 		    za.za_first_integer, FTAG, &clone));
1566 		dsl_dir_name(clone->ds_dir, buf);
1567 		fnvlist_add_boolean(val, buf);
1568 		dsl_dataset_rele(clone, FTAG);
1569 	}
1570 	zap_cursor_fini(&zc);
1571 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1572 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1573 fail:
1574 	nvlist_free(val);
1575 	nvlist_free(propval);
1576 }
1577 
1578 void
1579 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1580 {
1581 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1582 	uint64_t refd, avail, uobjs, aobjs, ratio;
1583 
1584 	ASSERT(dsl_pool_config_held(dp));
1585 
1586 	ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1587 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1588 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1589 
1590 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1591 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1592 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1593 
1594 	if (dsl_dataset_is_snapshot(ds)) {
1595 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1596 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1597 		    dsl_dataset_phys(ds)->ds_unique_bytes);
1598 		get_clones_stat(ds, nv);
1599 	} else {
1600 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1601 			char buf[MAXNAMELEN];
1602 			dsl_dataset_name(ds->ds_prev, buf);
1603 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1604 		}
1605 
1606 		dsl_dir_stats(ds->ds_dir, nv);
1607 	}
1608 
1609 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1610 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1611 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1612 
1613 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1614 	    dsl_dataset_phys(ds)->ds_creation_time);
1615 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1616 	    dsl_dataset_phys(ds)->ds_creation_txg);
1617 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1618 	    ds->ds_quota);
1619 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1620 	    ds->ds_reserved);
1621 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1622 	    dsl_dataset_phys(ds)->ds_guid);
1623 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1624 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1625 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1626 	    ds->ds_object);
1627 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1628 	    ds->ds_userrefs);
1629 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1630 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1631 
1632 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1633 		uint64_t written, comp, uncomp;
1634 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1635 		dsl_dataset_t *prev;
1636 
1637 		int err = dsl_dataset_hold_obj(dp,
1638 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1639 		if (err == 0) {
1640 			err = dsl_dataset_space_written(prev, ds, &written,
1641 			    &comp, &uncomp);
1642 			dsl_dataset_rele(prev, FTAG);
1643 			if (err == 0) {
1644 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1645 				    written);
1646 			}
1647 		}
1648 	}
1649 }
1650 
1651 void
1652 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1653 {
1654 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1655 	ASSERT(dsl_pool_config_held(dp));
1656 
1657 	stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1658 	stat->dds_inconsistent =
1659 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1660 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1661 	stat->dds_origin[0] = '\0';
1662 	if (dsl_dataset_is_snapshot(ds)) {
1663 		stat->dds_is_snapshot = B_TRUE;
1664 		stat->dds_num_clones =
1665 		    dsl_dataset_phys(ds)->ds_num_children - 1;
1666 	} else {
1667 		stat->dds_is_snapshot = B_FALSE;
1668 		stat->dds_num_clones = 0;
1669 
1670 		if (dsl_dir_is_clone(ds->ds_dir)) {
1671 			dsl_dataset_t *ods;
1672 
1673 			VERIFY0(dsl_dataset_hold_obj(dp,
1674 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1675 			    FTAG, &ods));
1676 			dsl_dataset_name(ods, stat->dds_origin);
1677 			dsl_dataset_rele(ods, FTAG);
1678 		}
1679 	}
1680 }
1681 
1682 uint64_t
1683 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1684 {
1685 	return (ds->ds_fsid_guid);
1686 }
1687 
1688 void
1689 dsl_dataset_space(dsl_dataset_t *ds,
1690     uint64_t *refdbytesp, uint64_t *availbytesp,
1691     uint64_t *usedobjsp, uint64_t *availobjsp)
1692 {
1693 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1694 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1695 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1696 		*availbytesp +=
1697 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1698 	if (ds->ds_quota != 0) {
1699 		/*
1700 		 * Adjust available bytes according to refquota
1701 		 */
1702 		if (*refdbytesp < ds->ds_quota)
1703 			*availbytesp = MIN(*availbytesp,
1704 			    ds->ds_quota - *refdbytesp);
1705 		else
1706 			*availbytesp = 0;
1707 	}
1708 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1709 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1710 }
1711 
1712 boolean_t
1713 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1714 {
1715 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1716 
1717 	ASSERT(dsl_pool_config_held(dp));
1718 	if (snap == NULL)
1719 		return (B_FALSE);
1720 	if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1721 	    dsl_dataset_phys(snap)->ds_creation_txg) {
1722 		objset_t *os, *os_snap;
1723 		/*
1724 		 * It may be that only the ZIL differs, because it was
1725 		 * reset in the head.  Don't count that as being
1726 		 * modified.
1727 		 */
1728 		if (dmu_objset_from_ds(ds, &os) != 0)
1729 			return (B_TRUE);
1730 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1731 			return (B_TRUE);
1732 		return (bcmp(&os->os_phys->os_meta_dnode,
1733 		    &os_snap->os_phys->os_meta_dnode,
1734 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1735 	}
1736 	return (B_FALSE);
1737 }
1738 
1739 typedef struct dsl_dataset_rename_snapshot_arg {
1740 	const char *ddrsa_fsname;
1741 	const char *ddrsa_oldsnapname;
1742 	const char *ddrsa_newsnapname;
1743 	boolean_t ddrsa_recursive;
1744 	dmu_tx_t *ddrsa_tx;
1745 } dsl_dataset_rename_snapshot_arg_t;
1746 
1747 /* ARGSUSED */
1748 static int
1749 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1750     dsl_dataset_t *hds, void *arg)
1751 {
1752 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1753 	int error;
1754 	uint64_t val;
1755 
1756 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1757 	if (error != 0) {
1758 		/* ignore nonexistent snapshots */
1759 		return (error == ENOENT ? 0 : error);
1760 	}
1761 
1762 	/* new name should not exist */
1763 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1764 	if (error == 0)
1765 		error = SET_ERROR(EEXIST);
1766 	else if (error == ENOENT)
1767 		error = 0;
1768 
1769 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1770 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1771 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1772 		error = SET_ERROR(ENAMETOOLONG);
1773 
1774 	return (error);
1775 }
1776 
1777 static int
1778 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1779 {
1780 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1781 	dsl_pool_t *dp = dmu_tx_pool(tx);
1782 	dsl_dataset_t *hds;
1783 	int error;
1784 
1785 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1786 	if (error != 0)
1787 		return (error);
1788 
1789 	if (ddrsa->ddrsa_recursive) {
1790 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1791 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1792 		    DS_FIND_CHILDREN);
1793 	} else {
1794 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1795 	}
1796 	dsl_dataset_rele(hds, FTAG);
1797 	return (error);
1798 }
1799 
1800 static int
1801 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1802     dsl_dataset_t *hds, void *arg)
1803 {
1804 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1805 	dsl_dataset_t *ds;
1806 	uint64_t val;
1807 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1808 	int error;
1809 
1810 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1811 	ASSERT(error == 0 || error == ENOENT);
1812 	if (error == ENOENT) {
1813 		/* ignore nonexistent snapshots */
1814 		return (0);
1815 	}
1816 
1817 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1818 
1819 	/* log before we change the name */
1820 	spa_history_log_internal_ds(ds, "rename", tx,
1821 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1822 
1823 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1824 	    B_FALSE));
1825 	mutex_enter(&ds->ds_lock);
1826 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1827 	mutex_exit(&ds->ds_lock);
1828 	VERIFY0(zap_add(dp->dp_meta_objset,
1829 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
1830 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1831 
1832 	dsl_dataset_rele(ds, FTAG);
1833 	return (0);
1834 }
1835 
1836 static void
1837 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1838 {
1839 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1840 	dsl_pool_t *dp = dmu_tx_pool(tx);
1841 	dsl_dataset_t *hds;
1842 
1843 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1844 	ddrsa->ddrsa_tx = tx;
1845 	if (ddrsa->ddrsa_recursive) {
1846 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1847 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1848 		    DS_FIND_CHILDREN));
1849 	} else {
1850 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1851 	}
1852 	dsl_dataset_rele(hds, FTAG);
1853 }
1854 
1855 int
1856 dsl_dataset_rename_snapshot(const char *fsname,
1857     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1858 {
1859 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1860 
1861 	ddrsa.ddrsa_fsname = fsname;
1862 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1863 	ddrsa.ddrsa_newsnapname = newsnapname;
1864 	ddrsa.ddrsa_recursive = recursive;
1865 
1866 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1867 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
1868 	    1, ZFS_SPACE_CHECK_RESERVED));
1869 }
1870 
1871 /*
1872  * If we're doing an ownership handoff, we need to make sure that there is
1873  * only one long hold on the dataset.  We're not allowed to change anything here
1874  * so we don't permanently release the long hold or regular hold here.  We want
1875  * to do this only when syncing to avoid the dataset unexpectedly going away
1876  * when we release the long hold.
1877  */
1878 static int
1879 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1880 {
1881 	boolean_t held;
1882 
1883 	if (!dmu_tx_is_syncing(tx))
1884 		return (0);
1885 
1886 	if (owner != NULL) {
1887 		VERIFY3P(ds->ds_owner, ==, owner);
1888 		dsl_dataset_long_rele(ds, owner);
1889 	}
1890 
1891 	held = dsl_dataset_long_held(ds);
1892 
1893 	if (owner != NULL)
1894 		dsl_dataset_long_hold(ds, owner);
1895 
1896 	if (held)
1897 		return (SET_ERROR(EBUSY));
1898 
1899 	return (0);
1900 }
1901 
1902 typedef struct dsl_dataset_rollback_arg {
1903 	const char *ddra_fsname;
1904 	void *ddra_owner;
1905 	nvlist_t *ddra_result;
1906 } dsl_dataset_rollback_arg_t;
1907 
1908 static int
1909 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1910 {
1911 	dsl_dataset_rollback_arg_t *ddra = arg;
1912 	dsl_pool_t *dp = dmu_tx_pool(tx);
1913 	dsl_dataset_t *ds;
1914 	int64_t unused_refres_delta;
1915 	int error;
1916 
1917 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1918 	if (error != 0)
1919 		return (error);
1920 
1921 	/* must not be a snapshot */
1922 	if (dsl_dataset_is_snapshot(ds)) {
1923 		dsl_dataset_rele(ds, FTAG);
1924 		return (SET_ERROR(EINVAL));
1925 	}
1926 
1927 	/* must have a most recent snapshot */
1928 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
1929 		dsl_dataset_rele(ds, FTAG);
1930 		return (SET_ERROR(EINVAL));
1931 	}
1932 
1933 	/* must not have any bookmarks after the most recent snapshot */
1934 	nvlist_t *proprequest = fnvlist_alloc();
1935 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
1936 	nvlist_t *bookmarks = fnvlist_alloc();
1937 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
1938 	fnvlist_free(proprequest);
1939 	if (error != 0)
1940 		return (error);
1941 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
1942 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
1943 		nvlist_t *valuenv =
1944 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
1945 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
1946 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
1947 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
1948 			fnvlist_free(bookmarks);
1949 			dsl_dataset_rele(ds, FTAG);
1950 			return (SET_ERROR(EEXIST));
1951 		}
1952 	}
1953 	fnvlist_free(bookmarks);
1954 
1955 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
1956 	if (error != 0) {
1957 		dsl_dataset_rele(ds, FTAG);
1958 		return (error);
1959 	}
1960 
1961 	/*
1962 	 * Check if the snap we are rolling back to uses more than
1963 	 * the refquota.
1964 	 */
1965 	if (ds->ds_quota != 0 &&
1966 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
1967 		dsl_dataset_rele(ds, FTAG);
1968 		return (SET_ERROR(EDQUOT));
1969 	}
1970 
1971 	/*
1972 	 * When we do the clone swap, we will temporarily use more space
1973 	 * due to the refreservation (the head will no longer have any
1974 	 * unique space, so the entire amount of the refreservation will need
1975 	 * to be free).  We will immediately destroy the clone, freeing
1976 	 * this space, but the freeing happens over many txg's.
1977 	 */
1978 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1979 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1980 
1981 	if (unused_refres_delta > 0 &&
1982 	    unused_refres_delta >
1983 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1984 		dsl_dataset_rele(ds, FTAG);
1985 		return (SET_ERROR(ENOSPC));
1986 	}
1987 
1988 	dsl_dataset_rele(ds, FTAG);
1989 	return (0);
1990 }
1991 
1992 static void
1993 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1994 {
1995 	dsl_dataset_rollback_arg_t *ddra = arg;
1996 	dsl_pool_t *dp = dmu_tx_pool(tx);
1997 	dsl_dataset_t *ds, *clone;
1998 	uint64_t cloneobj;
1999 	char namebuf[ZFS_MAXNAMELEN];
2000 
2001 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2002 
2003 	dsl_dataset_name(ds->ds_prev, namebuf);
2004 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2005 
2006 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2007 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2008 
2009 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2010 
2011 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2012 	dsl_dataset_zero_zil(ds, tx);
2013 
2014 	dsl_destroy_head_sync_impl(clone, tx);
2015 
2016 	dsl_dataset_rele(clone, FTAG);
2017 	dsl_dataset_rele(ds, FTAG);
2018 }
2019 
2020 /*
2021  * Rolls back the given filesystem or volume to the most recent snapshot.
2022  * The name of the most recent snapshot will be returned under key "target"
2023  * in the result nvlist.
2024  *
2025  * If owner != NULL:
2026  * - The existing dataset MUST be owned by the specified owner at entry
2027  * - Upon return, dataset will still be held by the same owner, whether we
2028  *   succeed or not.
2029  *
2030  * This mode is required any time the existing filesystem is mounted.  See
2031  * notes above zfs_suspend_fs() for further details.
2032  */
2033 int
2034 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2035 {
2036 	dsl_dataset_rollback_arg_t ddra;
2037 
2038 	ddra.ddra_fsname = fsname;
2039 	ddra.ddra_owner = owner;
2040 	ddra.ddra_result = result;
2041 
2042 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2043 	    dsl_dataset_rollback_sync, &ddra,
2044 	    1, ZFS_SPACE_CHECK_RESERVED));
2045 }
2046 
2047 struct promotenode {
2048 	list_node_t link;
2049 	dsl_dataset_t *ds;
2050 };
2051 
2052 typedef struct dsl_dataset_promote_arg {
2053 	const char *ddpa_clonename;
2054 	dsl_dataset_t *ddpa_clone;
2055 	list_t shared_snaps, origin_snaps, clone_snaps;
2056 	dsl_dataset_t *origin_origin; /* origin of the origin */
2057 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2058 	char *err_ds;
2059 	cred_t *cr;
2060 } dsl_dataset_promote_arg_t;
2061 
2062 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2063 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2064     void *tag);
2065 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2066 
2067 static int
2068 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2069 {
2070 	dsl_dataset_promote_arg_t *ddpa = arg;
2071 	dsl_pool_t *dp = dmu_tx_pool(tx);
2072 	dsl_dataset_t *hds;
2073 	struct promotenode *snap;
2074 	dsl_dataset_t *origin_ds;
2075 	int err;
2076 	uint64_t unused;
2077 	uint64_t ss_mv_cnt;
2078 
2079 	err = promote_hold(ddpa, dp, FTAG);
2080 	if (err != 0)
2081 		return (err);
2082 
2083 	hds = ddpa->ddpa_clone;
2084 
2085 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2086 		promote_rele(ddpa, FTAG);
2087 		return (SET_ERROR(EXDEV));
2088 	}
2089 
2090 	/*
2091 	 * Compute and check the amount of space to transfer.  Since this is
2092 	 * so expensive, don't do the preliminary check.
2093 	 */
2094 	if (!dmu_tx_is_syncing(tx)) {
2095 		promote_rele(ddpa, FTAG);
2096 		return (0);
2097 	}
2098 
2099 	snap = list_head(&ddpa->shared_snaps);
2100 	origin_ds = snap->ds;
2101 
2102 	/* compute origin's new unique space */
2103 	snap = list_tail(&ddpa->clone_snaps);
2104 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2105 	    origin_ds->ds_object);
2106 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2107 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2108 	    &ddpa->unique, &unused, &unused);
2109 
2110 	/*
2111 	 * Walk the snapshots that we are moving
2112 	 *
2113 	 * Compute space to transfer.  Consider the incremental changes
2114 	 * to used by each snapshot:
2115 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2116 	 * So each snapshot gave birth to:
2117 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2118 	 * So a sequence would look like:
2119 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2120 	 * Which simplifies to:
2121 	 * uN + kN + kN-1 + ... + k1 + k0
2122 	 * Note however, if we stop before we reach the ORIGIN we get:
2123 	 * uN + kN + kN-1 + ... + kM - uM-1
2124 	 */
2125 	ss_mv_cnt = 0;
2126 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2127 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2128 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2129 	for (snap = list_head(&ddpa->shared_snaps); snap;
2130 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2131 		uint64_t val, dlused, dlcomp, dluncomp;
2132 		dsl_dataset_t *ds = snap->ds;
2133 
2134 		ss_mv_cnt++;
2135 
2136 		/*
2137 		 * If there are long holds, we won't be able to evict
2138 		 * the objset.
2139 		 */
2140 		if (dsl_dataset_long_held(ds)) {
2141 			err = SET_ERROR(EBUSY);
2142 			goto out;
2143 		}
2144 
2145 		/* Check that the snapshot name does not conflict */
2146 		VERIFY0(dsl_dataset_get_snapname(ds));
2147 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2148 		if (err == 0) {
2149 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2150 			err = SET_ERROR(EEXIST);
2151 			goto out;
2152 		}
2153 		if (err != ENOENT)
2154 			goto out;
2155 
2156 		/* The very first snapshot does not have a deadlist */
2157 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2158 			continue;
2159 
2160 		dsl_deadlist_space(&ds->ds_deadlist,
2161 		    &dlused, &dlcomp, &dluncomp);
2162 		ddpa->used += dlused;
2163 		ddpa->comp += dlcomp;
2164 		ddpa->uncomp += dluncomp;
2165 	}
2166 
2167 	/*
2168 	 * If we are a clone of a clone then we never reached ORIGIN,
2169 	 * so we need to subtract out the clone origin's used space.
2170 	 */
2171 	if (ddpa->origin_origin) {
2172 		ddpa->used -=
2173 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2174 		ddpa->comp -=
2175 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2176 		ddpa->uncomp -=
2177 		    dsl_dataset_phys(ddpa->origin_origin)->
2178 		    ds_uncompressed_bytes;
2179 	}
2180 
2181 	/* Check that there is enough space and limit headroom here */
2182 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2183 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2184 	if (err != 0)
2185 		goto out;
2186 
2187 	/*
2188 	 * Compute the amounts of space that will be used by snapshots
2189 	 * after the promotion (for both origin and clone).  For each,
2190 	 * it is the amount of space that will be on all of their
2191 	 * deadlists (that was not born before their new origin).
2192 	 */
2193 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2194 		uint64_t space;
2195 
2196 		/*
2197 		 * Note, typically this will not be a clone of a clone,
2198 		 * so dd_origin_txg will be < TXG_INITIAL, so
2199 		 * these snaplist_space() -> dsl_deadlist_space_range()
2200 		 * calls will be fast because they do not have to
2201 		 * iterate over all bps.
2202 		 */
2203 		snap = list_head(&ddpa->origin_snaps);
2204 		err = snaplist_space(&ddpa->shared_snaps,
2205 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2206 		if (err != 0)
2207 			goto out;
2208 
2209 		err = snaplist_space(&ddpa->clone_snaps,
2210 		    snap->ds->ds_dir->dd_origin_txg, &space);
2211 		if (err != 0)
2212 			goto out;
2213 		ddpa->cloneusedsnap += space;
2214 	}
2215 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2216 	    DD_FLAG_USED_BREAKDOWN) {
2217 		err = snaplist_space(&ddpa->origin_snaps,
2218 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2219 		    &ddpa->originusedsnap);
2220 		if (err != 0)
2221 			goto out;
2222 	}
2223 
2224 out:
2225 	promote_rele(ddpa, FTAG);
2226 	return (err);
2227 }
2228 
2229 static void
2230 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2231 {
2232 	dsl_dataset_promote_arg_t *ddpa = arg;
2233 	dsl_pool_t *dp = dmu_tx_pool(tx);
2234 	dsl_dataset_t *hds;
2235 	struct promotenode *snap;
2236 	dsl_dataset_t *origin_ds;
2237 	dsl_dataset_t *origin_head;
2238 	dsl_dir_t *dd;
2239 	dsl_dir_t *odd = NULL;
2240 	uint64_t oldnext_obj;
2241 	int64_t delta;
2242 
2243 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2244 	hds = ddpa->ddpa_clone;
2245 
2246 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2247 
2248 	snap = list_head(&ddpa->shared_snaps);
2249 	origin_ds = snap->ds;
2250 	dd = hds->ds_dir;
2251 
2252 	snap = list_head(&ddpa->origin_snaps);
2253 	origin_head = snap->ds;
2254 
2255 	/*
2256 	 * We need to explicitly open odd, since origin_ds's dd will be
2257 	 * changing.
2258 	 */
2259 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2260 	    NULL, FTAG, &odd));
2261 
2262 	/* change origin's next snap */
2263 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2264 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2265 	snap = list_tail(&ddpa->clone_snaps);
2266 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2267 	    origin_ds->ds_object);
2268 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2269 
2270 	/* change the origin's next clone */
2271 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2272 		dsl_dataset_remove_from_next_clones(origin_ds,
2273 		    snap->ds->ds_object, tx);
2274 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2275 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2276 		    oldnext_obj, tx));
2277 	}
2278 
2279 	/* change origin */
2280 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2281 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2282 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2283 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2284 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2285 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2286 	origin_head->ds_dir->dd_origin_txg =
2287 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2288 
2289 	/* change dd_clone entries */
2290 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2291 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2292 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2293 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2294 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2295 		    hds->ds_object, tx));
2296 
2297 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2298 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2299 		    origin_head->ds_object, tx));
2300 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2301 			dsl_dir_phys(dd)->dd_clones =
2302 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2303 			    DMU_OT_NONE, 0, tx);
2304 		}
2305 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2306 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2307 	}
2308 
2309 	/* move snapshots to this dir */
2310 	for (snap = list_head(&ddpa->shared_snaps); snap;
2311 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2312 		dsl_dataset_t *ds = snap->ds;
2313 
2314 		/*
2315 		 * Property callbacks are registered to a particular
2316 		 * dsl_dir.  Since ours is changing, evict the objset
2317 		 * so that they will be unregistered from the old dsl_dir.
2318 		 */
2319 		if (ds->ds_objset) {
2320 			dmu_objset_evict(ds->ds_objset);
2321 			ds->ds_objset = NULL;
2322 		}
2323 
2324 		/* move snap name entry */
2325 		VERIFY0(dsl_dataset_get_snapname(ds));
2326 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2327 		    ds->ds_snapname, tx, B_TRUE));
2328 		VERIFY0(zap_add(dp->dp_meta_objset,
2329 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2330 		    8, 1, &ds->ds_object, tx));
2331 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2332 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2333 
2334 		/* change containing dsl_dir */
2335 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2336 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2337 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2338 		ASSERT3P(ds->ds_dir, ==, odd);
2339 		dsl_dir_rele(ds->ds_dir, ds);
2340 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2341 		    NULL, ds, &ds->ds_dir));
2342 
2343 		/* move any clone references */
2344 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2345 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2346 			zap_cursor_t zc;
2347 			zap_attribute_t za;
2348 
2349 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2350 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
2351 			    zap_cursor_retrieve(&zc, &za) == 0;
2352 			    zap_cursor_advance(&zc)) {
2353 				dsl_dataset_t *cnds;
2354 				uint64_t o;
2355 
2356 				if (za.za_first_integer == oldnext_obj) {
2357 					/*
2358 					 * We've already moved the
2359 					 * origin's reference.
2360 					 */
2361 					continue;
2362 				}
2363 
2364 				VERIFY0(dsl_dataset_hold_obj(dp,
2365 				    za.za_first_integer, FTAG, &cnds));
2366 				o = dsl_dir_phys(cnds->ds_dir)->
2367 				    dd_head_dataset_obj;
2368 
2369 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2370 				    dsl_dir_phys(odd)->dd_clones, o, tx));
2371 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2372 				    dsl_dir_phys(dd)->dd_clones, o, tx));
2373 				dsl_dataset_rele(cnds, FTAG);
2374 			}
2375 			zap_cursor_fini(&zc);
2376 		}
2377 
2378 		ASSERT(!dsl_prop_hascb(ds));
2379 	}
2380 
2381 	/*
2382 	 * Change space accounting.
2383 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2384 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2385 	 * is true for each of {clone,origin} independently.
2386 	 */
2387 
2388 	delta = ddpa->cloneusedsnap -
2389 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2390 	ASSERT3S(delta, >=, 0);
2391 	ASSERT3U(ddpa->used, >=, delta);
2392 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2393 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2394 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2395 
2396 	delta = ddpa->originusedsnap -
2397 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2398 	ASSERT3S(delta, <=, 0);
2399 	ASSERT3U(ddpa->used, >=, -delta);
2400 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2401 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2402 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2403 
2404 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2405 
2406 	/* log history record */
2407 	spa_history_log_internal_ds(hds, "promote", tx, "");
2408 
2409 	dsl_dir_rele(odd, FTAG);
2410 	promote_rele(ddpa, FTAG);
2411 }
2412 
2413 /*
2414  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2415  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2416  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2417  * snapshots back to this dataset's origin.
2418  */
2419 static int
2420 snaplist_make(dsl_pool_t *dp,
2421     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2422 {
2423 	uint64_t obj = last_obj;
2424 
2425 	list_create(l, sizeof (struct promotenode),
2426 	    offsetof(struct promotenode, link));
2427 
2428 	while (obj != first_obj) {
2429 		dsl_dataset_t *ds;
2430 		struct promotenode *snap;
2431 		int err;
2432 
2433 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2434 		ASSERT(err != ENOENT);
2435 		if (err != 0)
2436 			return (err);
2437 
2438 		if (first_obj == 0)
2439 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2440 
2441 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2442 		snap->ds = ds;
2443 		list_insert_tail(l, snap);
2444 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2445 	}
2446 
2447 	return (0);
2448 }
2449 
2450 static int
2451 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2452 {
2453 	struct promotenode *snap;
2454 
2455 	*spacep = 0;
2456 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2457 		uint64_t used, comp, uncomp;
2458 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2459 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2460 		*spacep += used;
2461 	}
2462 	return (0);
2463 }
2464 
2465 static void
2466 snaplist_destroy(list_t *l, void *tag)
2467 {
2468 	struct promotenode *snap;
2469 
2470 	if (l == NULL || !list_link_active(&l->list_head))
2471 		return;
2472 
2473 	while ((snap = list_tail(l)) != NULL) {
2474 		list_remove(l, snap);
2475 		dsl_dataset_rele(snap->ds, tag);
2476 		kmem_free(snap, sizeof (*snap));
2477 	}
2478 	list_destroy(l);
2479 }
2480 
2481 static int
2482 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2483 {
2484 	int error;
2485 	dsl_dir_t *dd;
2486 	struct promotenode *snap;
2487 
2488 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2489 	    &ddpa->ddpa_clone);
2490 	if (error != 0)
2491 		return (error);
2492 	dd = ddpa->ddpa_clone->ds_dir;
2493 
2494 	if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2495 	    !dsl_dir_is_clone(dd)) {
2496 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2497 		return (SET_ERROR(EINVAL));
2498 	}
2499 
2500 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2501 	    &ddpa->shared_snaps, tag);
2502 	if (error != 0)
2503 		goto out;
2504 
2505 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2506 	    &ddpa->clone_snaps, tag);
2507 	if (error != 0)
2508 		goto out;
2509 
2510 	snap = list_head(&ddpa->shared_snaps);
2511 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2512 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2513 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2514 	    &ddpa->origin_snaps, tag);
2515 	if (error != 0)
2516 		goto out;
2517 
2518 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2519 		error = dsl_dataset_hold_obj(dp,
2520 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2521 		    tag, &ddpa->origin_origin);
2522 		if (error != 0)
2523 			goto out;
2524 	}
2525 out:
2526 	if (error != 0)
2527 		promote_rele(ddpa, tag);
2528 	return (error);
2529 }
2530 
2531 static void
2532 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2533 {
2534 	snaplist_destroy(&ddpa->shared_snaps, tag);
2535 	snaplist_destroy(&ddpa->clone_snaps, tag);
2536 	snaplist_destroy(&ddpa->origin_snaps, tag);
2537 	if (ddpa->origin_origin != NULL)
2538 		dsl_dataset_rele(ddpa->origin_origin, tag);
2539 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2540 }
2541 
2542 /*
2543  * Promote a clone.
2544  *
2545  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2546  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2547  */
2548 int
2549 dsl_dataset_promote(const char *name, char *conflsnap)
2550 {
2551 	dsl_dataset_promote_arg_t ddpa = { 0 };
2552 	uint64_t numsnaps;
2553 	int error;
2554 	objset_t *os;
2555 
2556 	/*
2557 	 * We will modify space proportional to the number of
2558 	 * snapshots.  Compute numsnaps.
2559 	 */
2560 	error = dmu_objset_hold(name, FTAG, &os);
2561 	if (error != 0)
2562 		return (error);
2563 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2564 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2565 	    &numsnaps);
2566 	dmu_objset_rele(os, FTAG);
2567 	if (error != 0)
2568 		return (error);
2569 
2570 	ddpa.ddpa_clonename = name;
2571 	ddpa.err_ds = conflsnap;
2572 	ddpa.cr = CRED();
2573 
2574 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2575 	    dsl_dataset_promote_sync, &ddpa,
2576 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2577 }
2578 
2579 int
2580 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2581     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2582 {
2583 	int64_t unused_refres_delta;
2584 
2585 	/* they should both be heads */
2586 	if (dsl_dataset_is_snapshot(clone) ||
2587 	    dsl_dataset_is_snapshot(origin_head))
2588 		return (SET_ERROR(EINVAL));
2589 
2590 	/* if we are not forcing, the branch point should be just before them */
2591 	if (!force && clone->ds_prev != origin_head->ds_prev)
2592 		return (SET_ERROR(EINVAL));
2593 
2594 	/* clone should be the clone (unless they are unrelated) */
2595 	if (clone->ds_prev != NULL &&
2596 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2597 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2598 		return (SET_ERROR(EINVAL));
2599 
2600 	/* the clone should be a child of the origin */
2601 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2602 		return (SET_ERROR(EINVAL));
2603 
2604 	/* origin_head shouldn't be modified unless 'force' */
2605 	if (!force &&
2606 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2607 		return (SET_ERROR(ETXTBSY));
2608 
2609 	/* origin_head should have no long holds (e.g. is not mounted) */
2610 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2611 		return (SET_ERROR(EBUSY));
2612 
2613 	/* check amount of any unconsumed refreservation */
2614 	unused_refres_delta =
2615 	    (int64_t)MIN(origin_head->ds_reserved,
2616 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2617 	    (int64_t)MIN(origin_head->ds_reserved,
2618 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2619 
2620 	if (unused_refres_delta > 0 &&
2621 	    unused_refres_delta >
2622 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2623 		return (SET_ERROR(ENOSPC));
2624 
2625 	/* clone can't be over the head's refquota */
2626 	if (origin_head->ds_quota != 0 &&
2627 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
2628 	    origin_head->ds_quota)
2629 		return (SET_ERROR(EDQUOT));
2630 
2631 	return (0);
2632 }
2633 
2634 void
2635 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2636     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2637 {
2638 	dsl_pool_t *dp = dmu_tx_pool(tx);
2639 	int64_t unused_refres_delta;
2640 
2641 	ASSERT(clone->ds_reserved == 0);
2642 	ASSERT(origin_head->ds_quota == 0 ||
2643 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2644 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2645 
2646 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2647 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2648 
2649 	if (clone->ds_objset != NULL) {
2650 		dmu_objset_evict(clone->ds_objset);
2651 		clone->ds_objset = NULL;
2652 	}
2653 
2654 	if (origin_head->ds_objset != NULL) {
2655 		dmu_objset_evict(origin_head->ds_objset);
2656 		origin_head->ds_objset = NULL;
2657 	}
2658 
2659 	unused_refres_delta =
2660 	    (int64_t)MIN(origin_head->ds_reserved,
2661 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2662 	    (int64_t)MIN(origin_head->ds_reserved,
2663 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2664 
2665 	/*
2666 	 * Reset origin's unique bytes, if it exists.
2667 	 */
2668 	if (clone->ds_prev) {
2669 		dsl_dataset_t *origin = clone->ds_prev;
2670 		uint64_t comp, uncomp;
2671 
2672 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2673 		dsl_deadlist_space_range(&clone->ds_deadlist,
2674 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2675 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2676 	}
2677 
2678 	/* swap blkptrs */
2679 	{
2680 		blkptr_t tmp;
2681 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
2682 		dsl_dataset_phys(origin_head)->ds_bp =
2683 		    dsl_dataset_phys(clone)->ds_bp;
2684 		dsl_dataset_phys(clone)->ds_bp = tmp;
2685 	}
2686 
2687 	/* set dd_*_bytes */
2688 	{
2689 		int64_t dused, dcomp, duncomp;
2690 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2691 		uint64_t odl_used, odl_comp, odl_uncomp;
2692 
2693 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2694 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2695 
2696 		dsl_deadlist_space(&clone->ds_deadlist,
2697 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2698 		dsl_deadlist_space(&origin_head->ds_deadlist,
2699 		    &odl_used, &odl_comp, &odl_uncomp);
2700 
2701 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2702 		    cdl_used -
2703 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2704 		    odl_used);
2705 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2706 		    cdl_comp -
2707 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2708 		    odl_comp);
2709 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2710 		    cdl_uncomp -
2711 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2712 		    odl_uncomp);
2713 
2714 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2715 		    dused, dcomp, duncomp, tx);
2716 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2717 		    -dused, -dcomp, -duncomp, tx);
2718 
2719 		/*
2720 		 * The difference in the space used by snapshots is the
2721 		 * difference in snapshot space due to the head's
2722 		 * deadlist (since that's the only thing that's
2723 		 * changing that affects the snapused).
2724 		 */
2725 		dsl_deadlist_space_range(&clone->ds_deadlist,
2726 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2727 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2728 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2729 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2730 		    &odl_used, &odl_comp, &odl_uncomp);
2731 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2732 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2733 	}
2734 
2735 	/* swap ds_*_bytes */
2736 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2737 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
2738 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2739 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
2740 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2741 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2742 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2743 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2744 
2745 	/* apply any parent delta for change in unconsumed refreservation */
2746 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2747 	    unused_refres_delta, 0, 0, tx);
2748 
2749 	/*
2750 	 * Swap deadlists.
2751 	 */
2752 	dsl_deadlist_close(&clone->ds_deadlist);
2753 	dsl_deadlist_close(&origin_head->ds_deadlist);
2754 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2755 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2756 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2757 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2758 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2759 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
2760 
2761 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2762 
2763 	spa_history_log_internal_ds(clone, "clone swap", tx,
2764 	    "parent=%s", origin_head->ds_dir->dd_myname);
2765 }
2766 
2767 /*
2768  * Given a pool name and a dataset object number in that pool,
2769  * return the name of that dataset.
2770  */
2771 int
2772 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2773 {
2774 	dsl_pool_t *dp;
2775 	dsl_dataset_t *ds;
2776 	int error;
2777 
2778 	error = dsl_pool_hold(pname, FTAG, &dp);
2779 	if (error != 0)
2780 		return (error);
2781 
2782 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2783 	if (error == 0) {
2784 		dsl_dataset_name(ds, buf);
2785 		dsl_dataset_rele(ds, FTAG);
2786 	}
2787 	dsl_pool_rele(dp, FTAG);
2788 
2789 	return (error);
2790 }
2791 
2792 int
2793 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2794     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2795 {
2796 	int error = 0;
2797 
2798 	ASSERT3S(asize, >, 0);
2799 
2800 	/*
2801 	 * *ref_rsrv is the portion of asize that will come from any
2802 	 * unconsumed refreservation space.
2803 	 */
2804 	*ref_rsrv = 0;
2805 
2806 	mutex_enter(&ds->ds_lock);
2807 	/*
2808 	 * Make a space adjustment for reserved bytes.
2809 	 */
2810 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2811 		ASSERT3U(*used, >=,
2812 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2813 		*used -=
2814 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2815 		*ref_rsrv =
2816 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2817 	}
2818 
2819 	if (!check_quota || ds->ds_quota == 0) {
2820 		mutex_exit(&ds->ds_lock);
2821 		return (0);
2822 	}
2823 	/*
2824 	 * If they are requesting more space, and our current estimate
2825 	 * is over quota, they get to try again unless the actual
2826 	 * on-disk is over quota and there are no pending changes (which
2827 	 * may free up space for us).
2828 	 */
2829 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
2830 	    ds->ds_quota) {
2831 		if (inflight > 0 ||
2832 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
2833 			error = SET_ERROR(ERESTART);
2834 		else
2835 			error = SET_ERROR(EDQUOT);
2836 	}
2837 	mutex_exit(&ds->ds_lock);
2838 
2839 	return (error);
2840 }
2841 
2842 typedef struct dsl_dataset_set_qr_arg {
2843 	const char *ddsqra_name;
2844 	zprop_source_t ddsqra_source;
2845 	uint64_t ddsqra_value;
2846 } dsl_dataset_set_qr_arg_t;
2847 
2848 
2849 /* ARGSUSED */
2850 static int
2851 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2852 {
2853 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2854 	dsl_pool_t *dp = dmu_tx_pool(tx);
2855 	dsl_dataset_t *ds;
2856 	int error;
2857 	uint64_t newval;
2858 
2859 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2860 		return (SET_ERROR(ENOTSUP));
2861 
2862 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2863 	if (error != 0)
2864 		return (error);
2865 
2866 	if (dsl_dataset_is_snapshot(ds)) {
2867 		dsl_dataset_rele(ds, FTAG);
2868 		return (SET_ERROR(EINVAL));
2869 	}
2870 
2871 	error = dsl_prop_predict(ds->ds_dir,
2872 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2873 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2874 	if (error != 0) {
2875 		dsl_dataset_rele(ds, FTAG);
2876 		return (error);
2877 	}
2878 
2879 	if (newval == 0) {
2880 		dsl_dataset_rele(ds, FTAG);
2881 		return (0);
2882 	}
2883 
2884 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
2885 	    newval < ds->ds_reserved) {
2886 		dsl_dataset_rele(ds, FTAG);
2887 		return (SET_ERROR(ENOSPC));
2888 	}
2889 
2890 	dsl_dataset_rele(ds, FTAG);
2891 	return (0);
2892 }
2893 
2894 static void
2895 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2896 {
2897 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2898 	dsl_pool_t *dp = dmu_tx_pool(tx);
2899 	dsl_dataset_t *ds;
2900 	uint64_t newval;
2901 
2902 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2903 
2904 	dsl_prop_set_sync_impl(ds,
2905 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2906 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2907 	    &ddsqra->ddsqra_value, tx);
2908 
2909 	VERIFY0(dsl_prop_get_int_ds(ds,
2910 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2911 
2912 	if (ds->ds_quota != newval) {
2913 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2914 		ds->ds_quota = newval;
2915 	}
2916 	dsl_dataset_rele(ds, FTAG);
2917 }
2918 
2919 int
2920 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2921     uint64_t refquota)
2922 {
2923 	dsl_dataset_set_qr_arg_t ddsqra;
2924 
2925 	ddsqra.ddsqra_name = dsname;
2926 	ddsqra.ddsqra_source = source;
2927 	ddsqra.ddsqra_value = refquota;
2928 
2929 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2930 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
2931 }
2932 
2933 static int
2934 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2935 {
2936 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2937 	dsl_pool_t *dp = dmu_tx_pool(tx);
2938 	dsl_dataset_t *ds;
2939 	int error;
2940 	uint64_t newval, unique;
2941 
2942 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2943 		return (SET_ERROR(ENOTSUP));
2944 
2945 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2946 	if (error != 0)
2947 		return (error);
2948 
2949 	if (dsl_dataset_is_snapshot(ds)) {
2950 		dsl_dataset_rele(ds, FTAG);
2951 		return (SET_ERROR(EINVAL));
2952 	}
2953 
2954 	error = dsl_prop_predict(ds->ds_dir,
2955 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2956 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2957 	if (error != 0) {
2958 		dsl_dataset_rele(ds, FTAG);
2959 		return (error);
2960 	}
2961 
2962 	/*
2963 	 * If we are doing the preliminary check in open context, the
2964 	 * space estimates may be inaccurate.
2965 	 */
2966 	if (!dmu_tx_is_syncing(tx)) {
2967 		dsl_dataset_rele(ds, FTAG);
2968 		return (0);
2969 	}
2970 
2971 	mutex_enter(&ds->ds_lock);
2972 	if (!DS_UNIQUE_IS_ACCURATE(ds))
2973 		dsl_dataset_recalc_head_uniq(ds);
2974 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
2975 	mutex_exit(&ds->ds_lock);
2976 
2977 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2978 		uint64_t delta = MAX(unique, newval) -
2979 		    MAX(unique, ds->ds_reserved);
2980 
2981 		if (delta >
2982 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2983 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2984 			dsl_dataset_rele(ds, FTAG);
2985 			return (SET_ERROR(ENOSPC));
2986 		}
2987 	}
2988 
2989 	dsl_dataset_rele(ds, FTAG);
2990 	return (0);
2991 }
2992 
2993 void
2994 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2995     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2996 {
2997 	uint64_t newval;
2998 	uint64_t unique;
2999 	int64_t delta;
3000 
3001 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3002 	    source, sizeof (value), 1, &value, tx);
3003 
3004 	VERIFY0(dsl_prop_get_int_ds(ds,
3005 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3006 
3007 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3008 	mutex_enter(&ds->ds_dir->dd_lock);
3009 	mutex_enter(&ds->ds_lock);
3010 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3011 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3012 	delta = MAX(0, (int64_t)(newval - unique)) -
3013 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3014 	ds->ds_reserved = newval;
3015 	mutex_exit(&ds->ds_lock);
3016 
3017 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3018 	mutex_exit(&ds->ds_dir->dd_lock);
3019 }
3020 
3021 static void
3022 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3023 {
3024 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3025 	dsl_pool_t *dp = dmu_tx_pool(tx);
3026 	dsl_dataset_t *ds;
3027 
3028 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3029 	dsl_dataset_set_refreservation_sync_impl(ds,
3030 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3031 	dsl_dataset_rele(ds, FTAG);
3032 }
3033 
3034 int
3035 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3036     uint64_t refreservation)
3037 {
3038 	dsl_dataset_set_qr_arg_t ddsqra;
3039 
3040 	ddsqra.ddsqra_name = dsname;
3041 	ddsqra.ddsqra_source = source;
3042 	ddsqra.ddsqra_value = refreservation;
3043 
3044 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3045 	    dsl_dataset_set_refreservation_sync, &ddsqra,
3046 	    0, ZFS_SPACE_CHECK_NONE));
3047 }
3048 
3049 /*
3050  * Return (in *usedp) the amount of space written in new that is not
3051  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3052  * a snapshot before new, in new's filesystem (or its origin).  If not then
3053  * fail and return EINVAL.
3054  *
3055  * The written space is calculated by considering two components:  First, we
3056  * ignore any freed space, and calculate the written as new's used space
3057  * minus old's used space.  Next, we add in the amount of space that was freed
3058  * between the two snapshots, thus reducing new's used space relative to old's.
3059  * Specifically, this is the space that was born before old->ds_creation_txg,
3060  * and freed before new (ie. on new's deadlist or a previous deadlist).
3061  *
3062  * space freed                         [---------------------]
3063  * snapshots                       ---O-------O--------O-------O------
3064  *                                         oldsnap            new
3065  */
3066 int
3067 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3068     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3069 {
3070 	int err = 0;
3071 	uint64_t snapobj;
3072 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3073 
3074 	ASSERT(dsl_pool_config_held(dp));
3075 
3076 	*usedp = 0;
3077 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3078 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3079 
3080 	*compp = 0;
3081 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3082 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3083 
3084 	*uncompp = 0;
3085 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3086 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3087 
3088 	snapobj = new->ds_object;
3089 	while (snapobj != oldsnap->ds_object) {
3090 		dsl_dataset_t *snap;
3091 		uint64_t used, comp, uncomp;
3092 
3093 		if (snapobj == new->ds_object) {
3094 			snap = new;
3095 		} else {
3096 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3097 			if (err != 0)
3098 				break;
3099 		}
3100 
3101 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3102 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3103 			/*
3104 			 * The blocks in the deadlist can not be born after
3105 			 * ds_prev_snap_txg, so get the whole deadlist space,
3106 			 * which is more efficient (especially for old-format
3107 			 * deadlists).  Unfortunately the deadlist code
3108 			 * doesn't have enough information to make this
3109 			 * optimization itself.
3110 			 */
3111 			dsl_deadlist_space(&snap->ds_deadlist,
3112 			    &used, &comp, &uncomp);
3113 		} else {
3114 			dsl_deadlist_space_range(&snap->ds_deadlist,
3115 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3116 			    &used, &comp, &uncomp);
3117 		}
3118 		*usedp += used;
3119 		*compp += comp;
3120 		*uncompp += uncomp;
3121 
3122 		/*
3123 		 * If we get to the beginning of the chain of snapshots
3124 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3125 		 * was not a snapshot of/before new.
3126 		 */
3127 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3128 		if (snap != new)
3129 			dsl_dataset_rele(snap, FTAG);
3130 		if (snapobj == 0) {
3131 			err = SET_ERROR(EINVAL);
3132 			break;
3133 		}
3134 
3135 	}
3136 	return (err);
3137 }
3138 
3139 /*
3140  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3141  * lastsnap, and all snapshots in between are deleted.
3142  *
3143  * blocks that would be freed            [---------------------------]
3144  * snapshots                       ---O-------O--------O-------O--------O
3145  *                                        firstsnap        lastsnap
3146  *
3147  * This is the set of blocks that were born after the snap before firstsnap,
3148  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3149  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3150  * We calculate this by iterating over the relevant deadlists (from the snap
3151  * after lastsnap, backward to the snap after firstsnap), summing up the
3152  * space on the deadlist that was born after the snap before firstsnap.
3153  */
3154 int
3155 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3156     dsl_dataset_t *lastsnap,
3157     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3158 {
3159 	int err = 0;
3160 	uint64_t snapobj;
3161 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3162 
3163 	ASSERT(dsl_dataset_is_snapshot(firstsnap));
3164 	ASSERT(dsl_dataset_is_snapshot(lastsnap));
3165 
3166 	/*
3167 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3168 	 * is before lastsnap.
3169 	 */
3170 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3171 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3172 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3173 		return (SET_ERROR(EINVAL));
3174 
3175 	*usedp = *compp = *uncompp = 0;
3176 
3177 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3178 	while (snapobj != firstsnap->ds_object) {
3179 		dsl_dataset_t *ds;
3180 		uint64_t used, comp, uncomp;
3181 
3182 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3183 		if (err != 0)
3184 			break;
3185 
3186 		dsl_deadlist_space_range(&ds->ds_deadlist,
3187 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3188 		    &used, &comp, &uncomp);
3189 		*usedp += used;
3190 		*compp += comp;
3191 		*uncompp += uncomp;
3192 
3193 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3194 		ASSERT3U(snapobj, !=, 0);
3195 		dsl_dataset_rele(ds, FTAG);
3196 	}
3197 	return (err);
3198 }
3199 
3200 static int
3201 dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx)
3202 {
3203 	const char *dsname = arg;
3204 	dsl_dataset_t *ds;
3205 	dsl_pool_t *dp = dmu_tx_pool(tx);
3206 	int error = 0;
3207 
3208 	if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
3209 		return (SET_ERROR(ENOTSUP));
3210 
3211 	ASSERT(spa_feature_is_enabled(dp->dp_spa,
3212 	    SPA_FEATURE_EXTENSIBLE_DATASET));
3213 
3214 	error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
3215 	if (error != 0)
3216 		return (error);
3217 
3218 	if (ds->ds_large_blocks)
3219 		error = EALREADY;
3220 	dsl_dataset_rele(ds, FTAG);
3221 
3222 	return (error);
3223 }
3224 
3225 void
3226 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx)
3227 {
3228 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
3229 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
3230 	uint64_t zero = 0;
3231 
3232 	spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx);
3233 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
3234 
3235 	VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS,
3236 	    sizeof (zero), 1, &zero, tx));
3237 }
3238 
3239 static void
3240 dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx)
3241 {
3242 	const char *dsname = arg;
3243 	dsl_dataset_t *ds;
3244 
3245 	VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds));
3246 
3247 	dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
3248 	ASSERT(!ds->ds_large_blocks);
3249 	ds->ds_large_blocks = B_TRUE;
3250 	dsl_dataset_rele(ds, FTAG);
3251 }
3252 
3253 int
3254 dsl_dataset_activate_large_blocks(const char *dsname)
3255 {
3256 	int error;
3257 
3258 	error = dsl_sync_task(dsname,
3259 	    dsl_dataset_activate_large_blocks_check,
3260 	    dsl_dataset_activate_large_blocks_sync, (void *)dsname,
3261 	    1, ZFS_SPACE_CHECK_RESERVED);
3262 
3263 	/*
3264 	 * EALREADY indicates that this dataset already supports large blocks.
3265 	 */
3266 	if (error == EALREADY)
3267 		error = 0;
3268 	return (error);
3269 }
3270 
3271 /*
3272  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3273  * For example, they could both be snapshots of the same filesystem, and
3274  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3275  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3276  * filesystem.  Or 'earlier' could be the origin's origin.
3277  *
3278  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3279  */
3280 boolean_t
3281 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3282 	uint64_t earlier_txg)
3283 {
3284 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3285 	int error;
3286 	boolean_t ret;
3287 
3288 	ASSERT(dsl_pool_config_held(dp));
3289 	ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
3290 
3291 	if (earlier_txg == 0)
3292 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3293 
3294 	if (dsl_dataset_is_snapshot(later) &&
3295 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3296 		return (B_FALSE);
3297 
3298 	if (later->ds_dir == earlier->ds_dir)
3299 		return (B_TRUE);
3300 	if (!dsl_dir_is_clone(later->ds_dir))
3301 		return (B_FALSE);
3302 
3303 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3304 		return (B_TRUE);
3305 	dsl_dataset_t *origin;
3306 	error = dsl_dataset_hold_obj(dp,
3307 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3308 	if (error != 0)
3309 		return (B_FALSE);
3310 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3311 	dsl_dataset_rele(origin, FTAG);
3312 	return (ret);
3313 }
3314 
3315 
3316 void
3317 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3318 {
3319 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3320 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3321 }
3322