xref: /illumos-gate/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 2bbdd445a21f9d61f4a0ca0faf05d5ceb2bd91f3)
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  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27  */
28 
29 #include <sys/dmu_objset.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dmu_traverse.h>
35 #include <sys/dmu_impl.h>
36 #include <sys/dmu_tx.h>
37 #include <sys/arc.h>
38 #include <sys/zio.h>
39 #include <sys/zap.h>
40 #include <sys/zfeature.h>
41 #include <sys/unique.h>
42 #include <sys/zfs_context.h>
43 #include <sys/zfs_ioctl.h>
44 #include <sys/spa.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/zfs_onexit.h>
47 #include <sys/zvol.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/dsl_deadlist.h>
50 #include <sys/dsl_destroy.h>
51 #include <sys/dsl_userhold.h>
52 #include <sys/dsl_bookmark.h>
53 
54 /*
55  * The SPA supports block sizes up to 16MB.  However, very large blocks
56  * can have an impact on i/o latency (e.g. tying up a spinning disk for
57  * ~300ms), and also potentially on the memory allocator.  Therefore,
58  * we do not allow the recordsize to be set larger than zfs_max_recordsize
59  * (default 1MB).  Larger blocks can be created by changing this tunable,
60  * and pools with larger blocks can always be imported and used, regardless
61  * of this setting.
62  */
63 int zfs_max_recordsize = 1 * 1024 * 1024;
64 
65 #define	SWITCH64(x, y) \
66 	{ \
67 		uint64_t __tmp = (x); \
68 		(x) = (y); \
69 		(y) = __tmp; \
70 	}
71 
72 #define	DS_REF_MAX	(1ULL << 62)
73 
74 extern inline dsl_dataset_phys_t *dsl_dataset_phys(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(!ds->ds_is_snapshot);
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 static void
257 dsl_dataset_evict(void *dbu)
258 {
259 	dsl_dataset_t *ds = dbu;
260 
261 	ASSERT(ds->ds_owner == NULL);
262 
263 	ds->ds_dbuf = NULL;
264 
265 	unique_remove(ds->ds_fsid_guid);
266 
267 	if (ds->ds_objset != NULL)
268 		dmu_objset_evict(ds->ds_objset);
269 
270 	if (ds->ds_prev) {
271 		dsl_dataset_rele(ds->ds_prev, ds);
272 		ds->ds_prev = NULL;
273 	}
274 
275 	bplist_destroy(&ds->ds_pending_deadlist);
276 	if (ds->ds_deadlist.dl_os != NULL)
277 		dsl_deadlist_close(&ds->ds_deadlist);
278 	if (ds->ds_dir)
279 		dsl_dir_async_rele(ds->ds_dir, ds);
280 
281 	ASSERT(!list_link_active(&ds->ds_synced_link));
282 
283 	mutex_destroy(&ds->ds_lock);
284 	mutex_destroy(&ds->ds_opening_lock);
285 	mutex_destroy(&ds->ds_sendstream_lock);
286 	refcount_destroy(&ds->ds_longholds);
287 
288 	kmem_free(ds, sizeof (dsl_dataset_t));
289 }
290 
291 int
292 dsl_dataset_get_snapname(dsl_dataset_t *ds)
293 {
294 	dsl_dataset_phys_t *headphys;
295 	int err;
296 	dmu_buf_t *headdbuf;
297 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
298 	objset_t *mos = dp->dp_meta_objset;
299 
300 	if (ds->ds_snapname[0])
301 		return (0);
302 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
303 		return (0);
304 
305 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
306 	    FTAG, &headdbuf);
307 	if (err != 0)
308 		return (err);
309 	headphys = headdbuf->db_data;
310 	err = zap_value_search(dp->dp_meta_objset,
311 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
312 	dmu_buf_rele(headdbuf, FTAG);
313 	return (err);
314 }
315 
316 int
317 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
318 {
319 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
320 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
321 	matchtype_t mt;
322 	int err;
323 
324 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
325 		mt = MT_FIRST;
326 	else
327 		mt = MT_EXACT;
328 
329 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
330 	    value, mt, NULL, 0, NULL);
331 	if (err == ENOTSUP && mt == MT_FIRST)
332 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
333 	return (err);
334 }
335 
336 int
337 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
338     boolean_t adj_cnt)
339 {
340 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
341 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
342 	matchtype_t mt;
343 	int err;
344 
345 	dsl_dir_snap_cmtime_update(ds->ds_dir);
346 
347 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
348 		mt = MT_FIRST;
349 	else
350 		mt = MT_EXACT;
351 
352 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
353 	if (err == ENOTSUP && mt == MT_FIRST)
354 		err = zap_remove(mos, snapobj, name, tx);
355 
356 	if (err == 0 && adj_cnt)
357 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
358 		    DD_FIELD_SNAPSHOT_COUNT, tx);
359 
360 	return (err);
361 }
362 
363 boolean_t
364 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
365 {
366 	dmu_buf_t *dbuf = ds->ds_dbuf;
367 	boolean_t result = B_FALSE;
368 
369 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
370 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
371 
372 		if (ds == dmu_buf_get_user(dbuf))
373 			result = B_TRUE;
374 		else
375 			dmu_buf_rele(dbuf, tag);
376 	}
377 
378 	return (result);
379 }
380 
381 int
382 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
383     dsl_dataset_t **dsp)
384 {
385 	objset_t *mos = dp->dp_meta_objset;
386 	dmu_buf_t *dbuf;
387 	dsl_dataset_t *ds;
388 	int err;
389 	dmu_object_info_t doi;
390 
391 	ASSERT(dsl_pool_config_held(dp));
392 
393 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
394 	if (err != 0)
395 		return (err);
396 
397 	/* Make sure dsobj has the correct object type. */
398 	dmu_object_info_from_db(dbuf, &doi);
399 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
400 		dmu_buf_rele(dbuf, tag);
401 		return (SET_ERROR(EINVAL));
402 	}
403 
404 	ds = dmu_buf_get_user(dbuf);
405 	if (ds == NULL) {
406 		dsl_dataset_t *winner = NULL;
407 
408 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
409 		ds->ds_dbuf = dbuf;
410 		ds->ds_object = dsobj;
411 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
412 
413 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
414 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
415 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
416 		refcount_create(&ds->ds_longholds);
417 
418 		bplist_create(&ds->ds_pending_deadlist);
419 		dsl_deadlist_open(&ds->ds_deadlist,
420 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
421 
422 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
423 		    offsetof(dmu_sendarg_t, dsa_link));
424 
425 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
426 			err = zap_contains(mos, dsobj, DS_FIELD_LARGE_BLOCKS);
427 			if (err == 0)
428 				ds->ds_large_blocks = B_TRUE;
429 			else
430 				ASSERT3U(err, ==, ENOENT);
431 		}
432 
433 		if (err == 0) {
434 			err = dsl_dir_hold_obj(dp,
435 			    dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds,
436 			    &ds->ds_dir);
437 		}
438 		if (err != 0) {
439 			mutex_destroy(&ds->ds_lock);
440 			mutex_destroy(&ds->ds_opening_lock);
441 			mutex_destroy(&ds->ds_sendstream_lock);
442 			refcount_destroy(&ds->ds_longholds);
443 			bplist_destroy(&ds->ds_pending_deadlist);
444 			dsl_deadlist_close(&ds->ds_deadlist);
445 			kmem_free(ds, sizeof (dsl_dataset_t));
446 			dmu_buf_rele(dbuf, tag);
447 			return (err);
448 		}
449 
450 		if (!ds->ds_is_snapshot) {
451 			ds->ds_snapname[0] = '\0';
452 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
453 				err = dsl_dataset_hold_obj(dp,
454 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
455 				    ds, &ds->ds_prev);
456 			}
457 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
458 				int zaperr = zap_lookup(mos, ds->ds_object,
459 				    DS_FIELD_BOOKMARK_NAMES,
460 				    sizeof (ds->ds_bookmarks), 1,
461 				    &ds->ds_bookmarks);
462 				if (zaperr != ENOENT)
463 					VERIFY0(zaperr);
464 			}
465 		} else {
466 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
467 				err = dsl_dataset_get_snapname(ds);
468 			if (err == 0 &&
469 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
470 				err = zap_count(
471 				    ds->ds_dir->dd_pool->dp_meta_objset,
472 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
473 				    &ds->ds_userrefs);
474 			}
475 		}
476 
477 		if (err == 0 && !ds->ds_is_snapshot) {
478 			err = dsl_prop_get_int_ds(ds,
479 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
480 			    &ds->ds_reserved);
481 			if (err == 0) {
482 				err = dsl_prop_get_int_ds(ds,
483 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
484 				    &ds->ds_quota);
485 			}
486 		} else {
487 			ds->ds_reserved = ds->ds_quota = 0;
488 		}
489 
490 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
491 		if (err == 0)
492 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
493 
494 		if (err != 0 || winner != NULL) {
495 			bplist_destroy(&ds->ds_pending_deadlist);
496 			dsl_deadlist_close(&ds->ds_deadlist);
497 			if (ds->ds_prev)
498 				dsl_dataset_rele(ds->ds_prev, ds);
499 			dsl_dir_rele(ds->ds_dir, ds);
500 			mutex_destroy(&ds->ds_lock);
501 			mutex_destroy(&ds->ds_opening_lock);
502 			mutex_destroy(&ds->ds_sendstream_lock);
503 			refcount_destroy(&ds->ds_longholds);
504 			kmem_free(ds, sizeof (dsl_dataset_t));
505 			if (err != 0) {
506 				dmu_buf_rele(dbuf, tag);
507 				return (err);
508 			}
509 			ds = winner;
510 		} else {
511 			ds->ds_fsid_guid =
512 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
513 		}
514 	}
515 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
516 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
517 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
518 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
519 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
520 	*dsp = ds;
521 	return (0);
522 }
523 
524 int
525 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
526     void *tag, dsl_dataset_t **dsp)
527 {
528 	dsl_dir_t *dd;
529 	const char *snapname;
530 	uint64_t obj;
531 	int err = 0;
532 
533 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
534 	if (err != 0)
535 		return (err);
536 
537 	ASSERT(dsl_pool_config_held(dp));
538 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
539 	if (obj != 0)
540 		err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
541 	else
542 		err = SET_ERROR(ENOENT);
543 
544 	/* we may be looking for a snapshot */
545 	if (err == 0 && snapname != NULL) {
546 		dsl_dataset_t *ds;
547 
548 		if (*snapname++ != '@') {
549 			dsl_dataset_rele(*dsp, tag);
550 			dsl_dir_rele(dd, FTAG);
551 			return (SET_ERROR(ENOENT));
552 		}
553 
554 		dprintf("looking for snapshot '%s'\n", snapname);
555 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
556 		if (err == 0)
557 			err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
558 		dsl_dataset_rele(*dsp, tag);
559 
560 		if (err == 0) {
561 			mutex_enter(&ds->ds_lock);
562 			if (ds->ds_snapname[0] == 0)
563 				(void) strlcpy(ds->ds_snapname, snapname,
564 				    sizeof (ds->ds_snapname));
565 			mutex_exit(&ds->ds_lock);
566 			*dsp = ds;
567 		}
568 	}
569 
570 	dsl_dir_rele(dd, FTAG);
571 	return (err);
572 }
573 
574 int
575 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
576     void *tag, dsl_dataset_t **dsp)
577 {
578 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
579 	if (err != 0)
580 		return (err);
581 	if (!dsl_dataset_tryown(*dsp, tag)) {
582 		dsl_dataset_rele(*dsp, tag);
583 		*dsp = NULL;
584 		return (SET_ERROR(EBUSY));
585 	}
586 	return (0);
587 }
588 
589 int
590 dsl_dataset_own(dsl_pool_t *dp, const char *name,
591     void *tag, dsl_dataset_t **dsp)
592 {
593 	int err = dsl_dataset_hold(dp, name, tag, dsp);
594 	if (err != 0)
595 		return (err);
596 	if (!dsl_dataset_tryown(*dsp, tag)) {
597 		dsl_dataset_rele(*dsp, tag);
598 		return (SET_ERROR(EBUSY));
599 	}
600 	return (0);
601 }
602 
603 /*
604  * See the comment above dsl_pool_hold() for details.  In summary, a long
605  * hold is used to prevent destruction of a dataset while the pool hold
606  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
607  *
608  * The dataset and pool must be held when this function is called.  After it
609  * is called, the pool hold may be released while the dataset is still held
610  * and accessed.
611  */
612 void
613 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
614 {
615 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
616 	(void) refcount_add(&ds->ds_longholds, tag);
617 }
618 
619 void
620 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
621 {
622 	(void) refcount_remove(&ds->ds_longholds, tag);
623 }
624 
625 /* Return B_TRUE if there are any long holds on this dataset. */
626 boolean_t
627 dsl_dataset_long_held(dsl_dataset_t *ds)
628 {
629 	return (!refcount_is_zero(&ds->ds_longholds));
630 }
631 
632 void
633 dsl_dataset_name(dsl_dataset_t *ds, char *name)
634 {
635 	if (ds == NULL) {
636 		(void) strcpy(name, "mos");
637 	} else {
638 		dsl_dir_name(ds->ds_dir, name);
639 		VERIFY0(dsl_dataset_get_snapname(ds));
640 		if (ds->ds_snapname[0]) {
641 			(void) strcat(name, "@");
642 			/*
643 			 * We use a "recursive" mutex so that we
644 			 * can call dprintf_ds() with ds_lock held.
645 			 */
646 			if (!MUTEX_HELD(&ds->ds_lock)) {
647 				mutex_enter(&ds->ds_lock);
648 				(void) strcat(name, ds->ds_snapname);
649 				mutex_exit(&ds->ds_lock);
650 			} else {
651 				(void) strcat(name, ds->ds_snapname);
652 			}
653 		}
654 	}
655 }
656 
657 void
658 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
659 {
660 	dmu_buf_rele(ds->ds_dbuf, tag);
661 }
662 
663 void
664 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
665 {
666 	ASSERT3P(ds->ds_owner, ==, tag);
667 	ASSERT(ds->ds_dbuf != NULL);
668 
669 	mutex_enter(&ds->ds_lock);
670 	ds->ds_owner = NULL;
671 	mutex_exit(&ds->ds_lock);
672 	dsl_dataset_long_rele(ds, tag);
673 	dsl_dataset_rele(ds, tag);
674 }
675 
676 boolean_t
677 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
678 {
679 	boolean_t gotit = FALSE;
680 
681 	mutex_enter(&ds->ds_lock);
682 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
683 		ds->ds_owner = tag;
684 		dsl_dataset_long_hold(ds, tag);
685 		gotit = TRUE;
686 	}
687 	mutex_exit(&ds->ds_lock);
688 	return (gotit);
689 }
690 
691 uint64_t
692 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
693     uint64_t flags, dmu_tx_t *tx)
694 {
695 	dsl_pool_t *dp = dd->dd_pool;
696 	dmu_buf_t *dbuf;
697 	dsl_dataset_phys_t *dsphys;
698 	uint64_t dsobj;
699 	objset_t *mos = dp->dp_meta_objset;
700 
701 	if (origin == NULL)
702 		origin = dp->dp_origin_snap;
703 
704 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
705 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
706 	ASSERT(dmu_tx_is_syncing(tx));
707 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
708 
709 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
710 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
711 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
712 	dmu_buf_will_dirty(dbuf, tx);
713 	dsphys = dbuf->db_data;
714 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
715 	dsphys->ds_dir_obj = dd->dd_object;
716 	dsphys->ds_flags = flags;
717 	dsphys->ds_fsid_guid = unique_create();
718 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
719 	    sizeof (dsphys->ds_guid));
720 	dsphys->ds_snapnames_zapobj =
721 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
722 	    DMU_OT_NONE, 0, tx);
723 	dsphys->ds_creation_time = gethrestime_sec();
724 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
725 
726 	if (origin == NULL) {
727 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
728 	} else {
729 		dsl_dataset_t *ohds; /* head of the origin snapshot */
730 
731 		dsphys->ds_prev_snap_obj = origin->ds_object;
732 		dsphys->ds_prev_snap_txg =
733 		    dsl_dataset_phys(origin)->ds_creation_txg;
734 		dsphys->ds_referenced_bytes =
735 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
736 		dsphys->ds_compressed_bytes =
737 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
738 		dsphys->ds_uncompressed_bytes =
739 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
740 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
741 
742 		/*
743 		 * Inherit flags that describe the dataset's contents
744 		 * (INCONSISTENT) or properties (Case Insensitive).
745 		 */
746 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
747 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
748 
749 		if (origin->ds_large_blocks)
750 			dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
751 
752 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
753 		dsl_dataset_phys(origin)->ds_num_children++;
754 
755 		VERIFY0(dsl_dataset_hold_obj(dp,
756 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
757 		    FTAG, &ohds));
758 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
759 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
760 		dsl_dataset_rele(ohds, FTAG);
761 
762 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
763 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
764 				dsl_dataset_phys(origin)->ds_next_clones_obj =
765 				    zap_create(mos,
766 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
767 			}
768 			VERIFY0(zap_add_int(mos,
769 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
770 			    dsobj, tx));
771 		}
772 
773 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
774 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
775 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
776 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
777 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
778 				dsl_dir_phys(origin->ds_dir)->dd_clones =
779 				    zap_create(mos,
780 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
781 			}
782 			VERIFY0(zap_add_int(mos,
783 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
784 			    dsobj, tx));
785 		}
786 	}
787 
788 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
789 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
790 
791 	dmu_buf_rele(dbuf, FTAG);
792 
793 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
794 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
795 
796 	return (dsobj);
797 }
798 
799 static void
800 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
801 {
802 	objset_t *os;
803 
804 	VERIFY0(dmu_objset_from_ds(ds, &os));
805 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
806 	dsl_dataset_dirty(ds, tx);
807 }
808 
809 uint64_t
810 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
811     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
812 {
813 	dsl_pool_t *dp = pdd->dd_pool;
814 	uint64_t dsobj, ddobj;
815 	dsl_dir_t *dd;
816 
817 	ASSERT(dmu_tx_is_syncing(tx));
818 	ASSERT(lastname[0] != '@');
819 
820 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
821 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
822 
823 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
824 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
825 
826 	dsl_deleg_set_create_perms(dd, tx, cr);
827 
828 	/*
829 	 * Since we're creating a new node we know it's a leaf, so we can
830 	 * initialize the counts if the limit feature is active.
831 	 */
832 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
833 		uint64_t cnt = 0;
834 		objset_t *os = dd->dd_pool->dp_meta_objset;
835 
836 		dsl_dir_zapify(dd, tx);
837 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
838 		    sizeof (cnt), 1, &cnt, tx));
839 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
840 		    sizeof (cnt), 1, &cnt, tx));
841 	}
842 
843 	dsl_dir_rele(dd, FTAG);
844 
845 	/*
846 	 * If we are creating a clone, make sure we zero out any stale
847 	 * data from the origin snapshots zil header.
848 	 */
849 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
850 		dsl_dataset_t *ds;
851 
852 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
853 		dsl_dataset_zero_zil(ds, tx);
854 		dsl_dataset_rele(ds, FTAG);
855 	}
856 
857 	return (dsobj);
858 }
859 
860 /*
861  * The unique space in the head dataset can be calculated by subtracting
862  * the space used in the most recent snapshot, that is still being used
863  * in this file system, from the space currently in use.  To figure out
864  * the space in the most recent snapshot still in use, we need to take
865  * the total space used in the snapshot and subtract out the space that
866  * has been freed up since the snapshot was taken.
867  */
868 void
869 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
870 {
871 	uint64_t mrs_used;
872 	uint64_t dlused, dlcomp, dluncomp;
873 
874 	ASSERT(!ds->ds_is_snapshot);
875 
876 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
877 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
878 	else
879 		mrs_used = 0;
880 
881 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
882 
883 	ASSERT3U(dlused, <=, mrs_used);
884 	dsl_dataset_phys(ds)->ds_unique_bytes =
885 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
886 
887 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
888 	    SPA_VERSION_UNIQUE_ACCURATE)
889 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
890 }
891 
892 void
893 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
894     dmu_tx_t *tx)
895 {
896 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
897 	uint64_t count;
898 	int err;
899 
900 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
901 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
902 	    obj, tx);
903 	/*
904 	 * The err should not be ENOENT, but a bug in a previous version
905 	 * of the code could cause upgrade_clones_cb() to not set
906 	 * ds_next_snap_obj when it should, leading to a missing entry.
907 	 * If we knew that the pool was created after
908 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
909 	 * ENOENT.  However, at least we can check that we don't have
910 	 * too many entries in the next_clones_obj even after failing to
911 	 * remove this one.
912 	 */
913 	if (err != ENOENT)
914 		VERIFY0(err);
915 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
916 	    &count));
917 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
918 }
919 
920 
921 blkptr_t *
922 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
923 {
924 	return (&dsl_dataset_phys(ds)->ds_bp);
925 }
926 
927 void
928 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
929 {
930 	ASSERT(dmu_tx_is_syncing(tx));
931 	/* If it's the meta-objset, set dp_meta_rootbp */
932 	if (ds == NULL) {
933 		tx->tx_pool->dp_meta_rootbp = *bp;
934 	} else {
935 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
936 		dsl_dataset_phys(ds)->ds_bp = *bp;
937 	}
938 }
939 
940 spa_t *
941 dsl_dataset_get_spa(dsl_dataset_t *ds)
942 {
943 	return (ds->ds_dir->dd_pool->dp_spa);
944 }
945 
946 void
947 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
948 {
949 	dsl_pool_t *dp;
950 
951 	if (ds == NULL) /* this is the meta-objset */
952 		return;
953 
954 	ASSERT(ds->ds_objset != NULL);
955 
956 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
957 		panic("dirtying snapshot!");
958 
959 	dp = ds->ds_dir->dd_pool;
960 
961 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
962 		/* up the hold count until we can be written out */
963 		dmu_buf_add_ref(ds->ds_dbuf, ds);
964 	}
965 }
966 
967 boolean_t
968 dsl_dataset_is_dirty(dsl_dataset_t *ds)
969 {
970 	for (int t = 0; t < TXG_SIZE; t++) {
971 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
972 		    ds, t))
973 			return (B_TRUE);
974 	}
975 	return (B_FALSE);
976 }
977 
978 static int
979 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
980 {
981 	uint64_t asize;
982 
983 	if (!dmu_tx_is_syncing(tx))
984 		return (0);
985 
986 	/*
987 	 * If there's an fs-only reservation, any blocks that might become
988 	 * owned by the snapshot dataset must be accommodated by space
989 	 * outside of the reservation.
990 	 */
991 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
992 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
993 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
994 		return (SET_ERROR(ENOSPC));
995 
996 	/*
997 	 * Propagate any reserved space for this snapshot to other
998 	 * snapshot checks in this sync group.
999 	 */
1000 	if (asize > 0)
1001 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1002 
1003 	return (0);
1004 }
1005 
1006 typedef struct dsl_dataset_snapshot_arg {
1007 	nvlist_t *ddsa_snaps;
1008 	nvlist_t *ddsa_props;
1009 	nvlist_t *ddsa_errors;
1010 	cred_t *ddsa_cr;
1011 } dsl_dataset_snapshot_arg_t;
1012 
1013 int
1014 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1015     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1016 {
1017 	int error;
1018 	uint64_t value;
1019 
1020 	ds->ds_trysnap_txg = tx->tx_txg;
1021 
1022 	if (!dmu_tx_is_syncing(tx))
1023 		return (0);
1024 
1025 	/*
1026 	 * We don't allow multiple snapshots of the same txg.  If there
1027 	 * is already one, try again.
1028 	 */
1029 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1030 		return (SET_ERROR(EAGAIN));
1031 
1032 	/*
1033 	 * Check for conflicting snapshot name.
1034 	 */
1035 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1036 	if (error == 0)
1037 		return (SET_ERROR(EEXIST));
1038 	if (error != ENOENT)
1039 		return (error);
1040 
1041 	/*
1042 	 * We don't allow taking snapshots of inconsistent datasets, such as
1043 	 * those into which we are currently receiving.  However, if we are
1044 	 * creating this snapshot as part of a receive, this check will be
1045 	 * executed atomically with respect to the completion of the receive
1046 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1047 	 * case we ignore this, knowing it will be fixed up for us shortly in
1048 	 * dmu_recv_end_sync().
1049 	 */
1050 	if (!recv && DS_IS_INCONSISTENT(ds))
1051 		return (SET_ERROR(EBUSY));
1052 
1053 	/*
1054 	 * Skip the check for temporary snapshots or if we have already checked
1055 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1056 	 * check the count here when we're receiving a stream.
1057 	 */
1058 	if (cnt != 0 && cr != NULL) {
1059 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1060 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1061 		if (error != 0)
1062 			return (error);
1063 	}
1064 
1065 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1066 	if (error != 0)
1067 		return (error);
1068 
1069 	return (0);
1070 }
1071 
1072 static int
1073 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1074 {
1075 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1076 	dsl_pool_t *dp = dmu_tx_pool(tx);
1077 	nvpair_t *pair;
1078 	int rv = 0;
1079 
1080 	/*
1081 	 * Pre-compute how many total new snapshots will be created for each
1082 	 * level in the tree and below. This is needed for validating the
1083 	 * snapshot limit when either taking a recursive snapshot or when
1084 	 * taking multiple snapshots.
1085 	 *
1086 	 * The problem is that the counts are not actually adjusted when
1087 	 * we are checking, only when we finally sync. For a single snapshot,
1088 	 * this is easy, the count will increase by 1 at each node up the tree,
1089 	 * but its more complicated for the recursive/multiple snapshot case.
1090 	 *
1091 	 * The dsl_fs_ss_limit_check function does recursively check the count
1092 	 * at each level up the tree but since it is validating each snapshot
1093 	 * independently we need to be sure that we are validating the complete
1094 	 * count for the entire set of snapshots. We do this by rolling up the
1095 	 * counts for each component of the name into an nvlist and then
1096 	 * checking each of those cases with the aggregated count.
1097 	 *
1098 	 * This approach properly handles not only the recursive snapshot
1099 	 * case (where we get all of those on the ddsa_snaps list) but also
1100 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1101 	 * validate the limit on 'a' using a count of 2).
1102 	 *
1103 	 * We validate the snapshot names in the third loop and only report
1104 	 * name errors once.
1105 	 */
1106 	if (dmu_tx_is_syncing(tx)) {
1107 		nvlist_t *cnt_track = NULL;
1108 		cnt_track = fnvlist_alloc();
1109 
1110 		/* Rollup aggregated counts into the cnt_track list */
1111 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1112 		    pair != NULL;
1113 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1114 			char *pdelim;
1115 			uint64_t val;
1116 			char nm[MAXPATHLEN];
1117 
1118 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1119 			pdelim = strchr(nm, '@');
1120 			if (pdelim == NULL)
1121 				continue;
1122 			*pdelim = '\0';
1123 
1124 			do {
1125 				if (nvlist_lookup_uint64(cnt_track, nm,
1126 				    &val) == 0) {
1127 					/* update existing entry */
1128 					fnvlist_add_uint64(cnt_track, nm,
1129 					    val + 1);
1130 				} else {
1131 					/* add to list */
1132 					fnvlist_add_uint64(cnt_track, nm, 1);
1133 				}
1134 
1135 				pdelim = strrchr(nm, '/');
1136 				if (pdelim != NULL)
1137 					*pdelim = '\0';
1138 			} while (pdelim != NULL);
1139 		}
1140 
1141 		/* Check aggregated counts at each level */
1142 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1143 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1144 			int error = 0;
1145 			char *name;
1146 			uint64_t cnt = 0;
1147 			dsl_dataset_t *ds;
1148 
1149 			name = nvpair_name(pair);
1150 			cnt = fnvpair_value_uint64(pair);
1151 			ASSERT(cnt > 0);
1152 
1153 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1154 			if (error == 0) {
1155 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1156 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1157 				    ddsa->ddsa_cr);
1158 				dsl_dataset_rele(ds, FTAG);
1159 			}
1160 
1161 			if (error != 0) {
1162 				if (ddsa->ddsa_errors != NULL)
1163 					fnvlist_add_int32(ddsa->ddsa_errors,
1164 					    name, error);
1165 				rv = error;
1166 				/* only report one error for this check */
1167 				break;
1168 			}
1169 		}
1170 		nvlist_free(cnt_track);
1171 	}
1172 
1173 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1174 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1175 		int error = 0;
1176 		dsl_dataset_t *ds;
1177 		char *name, *atp;
1178 		char dsname[MAXNAMELEN];
1179 
1180 		name = nvpair_name(pair);
1181 		if (strlen(name) >= MAXNAMELEN)
1182 			error = SET_ERROR(ENAMETOOLONG);
1183 		if (error == 0) {
1184 			atp = strchr(name, '@');
1185 			if (atp == NULL)
1186 				error = SET_ERROR(EINVAL);
1187 			if (error == 0)
1188 				(void) strlcpy(dsname, name, atp - name + 1);
1189 		}
1190 		if (error == 0)
1191 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1192 		if (error == 0) {
1193 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1194 			error = dsl_dataset_snapshot_check_impl(ds,
1195 			    atp + 1, tx, B_FALSE, 0, NULL);
1196 			dsl_dataset_rele(ds, FTAG);
1197 		}
1198 
1199 		if (error != 0) {
1200 			if (ddsa->ddsa_errors != NULL) {
1201 				fnvlist_add_int32(ddsa->ddsa_errors,
1202 				    name, error);
1203 			}
1204 			rv = error;
1205 		}
1206 	}
1207 
1208 	return (rv);
1209 }
1210 
1211 void
1212 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1213     dmu_tx_t *tx)
1214 {
1215 	static zil_header_t zero_zil;
1216 
1217 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1218 	dmu_buf_t *dbuf;
1219 	dsl_dataset_phys_t *dsphys;
1220 	uint64_t dsobj, crtxg;
1221 	objset_t *mos = dp->dp_meta_objset;
1222 	objset_t *os;
1223 
1224 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1225 
1226 	/*
1227 	 * If we are on an old pool, the zil must not be active, in which
1228 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1229 	 */
1230 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1231 	    dmu_objset_from_ds(ds, &os) != 0 ||
1232 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1233 	    sizeof (zero_zil)) == 0);
1234 
1235 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1236 
1237 	/*
1238 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1239 	 */
1240 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1241 		crtxg = 1;
1242 	else
1243 		crtxg = tx->tx_txg;
1244 
1245 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1246 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1247 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1248 	dmu_buf_will_dirty(dbuf, tx);
1249 	dsphys = dbuf->db_data;
1250 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1251 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1252 	dsphys->ds_fsid_guid = unique_create();
1253 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1254 	    sizeof (dsphys->ds_guid));
1255 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1256 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1257 	dsphys->ds_next_snap_obj = ds->ds_object;
1258 	dsphys->ds_num_children = 1;
1259 	dsphys->ds_creation_time = gethrestime_sec();
1260 	dsphys->ds_creation_txg = crtxg;
1261 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1262 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1263 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1264 	dsphys->ds_uncompressed_bytes =
1265 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1266 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1267 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1268 	dmu_buf_rele(dbuf, FTAG);
1269 
1270 	if (ds->ds_large_blocks)
1271 		dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1272 
1273 	ASSERT3U(ds->ds_prev != 0, ==,
1274 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1275 	if (ds->ds_prev) {
1276 		uint64_t next_clones_obj =
1277 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1278 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1279 		    ds->ds_object ||
1280 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1281 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1282 		    ds->ds_object) {
1283 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1284 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1285 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1286 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1287 		} else if (next_clones_obj != 0) {
1288 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1289 			    dsphys->ds_next_snap_obj, tx);
1290 			VERIFY0(zap_add_int(mos,
1291 			    next_clones_obj, dsobj, tx));
1292 		}
1293 	}
1294 
1295 	/*
1296 	 * If we have a reference-reservation on this dataset, we will
1297 	 * need to increase the amount of refreservation being charged
1298 	 * since our unique space is going to zero.
1299 	 */
1300 	if (ds->ds_reserved) {
1301 		int64_t delta;
1302 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1303 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1304 		    ds->ds_reserved);
1305 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1306 		    delta, 0, 0, tx);
1307 	}
1308 
1309 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1310 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1311 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1312 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1313 	dsl_deadlist_close(&ds->ds_deadlist);
1314 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1315 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1316 	dsl_deadlist_add_key(&ds->ds_deadlist,
1317 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1318 
1319 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1320 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1321 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1322 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1323 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1324 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1325 
1326 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1327 	    snapname, 8, 1, &dsobj, tx));
1328 
1329 	if (ds->ds_prev)
1330 		dsl_dataset_rele(ds->ds_prev, ds);
1331 	VERIFY0(dsl_dataset_hold_obj(dp,
1332 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1333 
1334 	dsl_scan_ds_snapshotted(ds, tx);
1335 
1336 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1337 
1338 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1339 }
1340 
1341 static void
1342 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1343 {
1344 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1345 	dsl_pool_t *dp = dmu_tx_pool(tx);
1346 	nvpair_t *pair;
1347 
1348 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1349 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1350 		dsl_dataset_t *ds;
1351 		char *name, *atp;
1352 		char dsname[MAXNAMELEN];
1353 
1354 		name = nvpair_name(pair);
1355 		atp = strchr(name, '@');
1356 		(void) strlcpy(dsname, name, atp - name + 1);
1357 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1358 
1359 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1360 		if (ddsa->ddsa_props != NULL) {
1361 			dsl_props_set_sync_impl(ds->ds_prev,
1362 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1363 		}
1364 		dsl_dataset_rele(ds, FTAG);
1365 	}
1366 }
1367 
1368 /*
1369  * The snapshots must all be in the same pool.
1370  * All-or-nothing: if there are any failures, nothing will be modified.
1371  */
1372 int
1373 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1374 {
1375 	dsl_dataset_snapshot_arg_t ddsa;
1376 	nvpair_t *pair;
1377 	boolean_t needsuspend;
1378 	int error;
1379 	spa_t *spa;
1380 	char *firstname;
1381 	nvlist_t *suspended = NULL;
1382 
1383 	pair = nvlist_next_nvpair(snaps, NULL);
1384 	if (pair == NULL)
1385 		return (0);
1386 	firstname = nvpair_name(pair);
1387 
1388 	error = spa_open(firstname, &spa, FTAG);
1389 	if (error != 0)
1390 		return (error);
1391 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1392 	spa_close(spa, FTAG);
1393 
1394 	if (needsuspend) {
1395 		suspended = fnvlist_alloc();
1396 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1397 		    pair = nvlist_next_nvpair(snaps, pair)) {
1398 			char fsname[MAXNAMELEN];
1399 			char *snapname = nvpair_name(pair);
1400 			char *atp;
1401 			void *cookie;
1402 
1403 			atp = strchr(snapname, '@');
1404 			if (atp == NULL) {
1405 				error = SET_ERROR(EINVAL);
1406 				break;
1407 			}
1408 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1409 
1410 			error = zil_suspend(fsname, &cookie);
1411 			if (error != 0)
1412 				break;
1413 			fnvlist_add_uint64(suspended, fsname,
1414 			    (uintptr_t)cookie);
1415 		}
1416 	}
1417 
1418 	ddsa.ddsa_snaps = snaps;
1419 	ddsa.ddsa_props = props;
1420 	ddsa.ddsa_errors = errors;
1421 	ddsa.ddsa_cr = CRED();
1422 
1423 	if (error == 0) {
1424 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1425 		    dsl_dataset_snapshot_sync, &ddsa,
1426 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1427 	}
1428 
1429 	if (suspended != NULL) {
1430 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1431 		    pair = nvlist_next_nvpair(suspended, pair)) {
1432 			zil_resume((void *)(uintptr_t)
1433 			    fnvpair_value_uint64(pair));
1434 		}
1435 		fnvlist_free(suspended);
1436 	}
1437 
1438 	return (error);
1439 }
1440 
1441 typedef struct dsl_dataset_snapshot_tmp_arg {
1442 	const char *ddsta_fsname;
1443 	const char *ddsta_snapname;
1444 	minor_t ddsta_cleanup_minor;
1445 	const char *ddsta_htag;
1446 } dsl_dataset_snapshot_tmp_arg_t;
1447 
1448 static int
1449 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1450 {
1451 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1452 	dsl_pool_t *dp = dmu_tx_pool(tx);
1453 	dsl_dataset_t *ds;
1454 	int error;
1455 
1456 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1457 	if (error != 0)
1458 		return (error);
1459 
1460 	/* NULL cred means no limit check for tmp snapshot */
1461 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1462 	    tx, B_FALSE, 0, NULL);
1463 	if (error != 0) {
1464 		dsl_dataset_rele(ds, FTAG);
1465 		return (error);
1466 	}
1467 
1468 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1469 		dsl_dataset_rele(ds, FTAG);
1470 		return (SET_ERROR(ENOTSUP));
1471 	}
1472 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1473 	    B_TRUE, tx);
1474 	if (error != 0) {
1475 		dsl_dataset_rele(ds, FTAG);
1476 		return (error);
1477 	}
1478 
1479 	dsl_dataset_rele(ds, FTAG);
1480 	return (0);
1481 }
1482 
1483 static void
1484 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1485 {
1486 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1487 	dsl_pool_t *dp = dmu_tx_pool(tx);
1488 	dsl_dataset_t *ds;
1489 
1490 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1491 
1492 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1493 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1494 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1495 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1496 
1497 	dsl_dataset_rele(ds, FTAG);
1498 }
1499 
1500 int
1501 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1502     minor_t cleanup_minor, const char *htag)
1503 {
1504 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1505 	int error;
1506 	spa_t *spa;
1507 	boolean_t needsuspend;
1508 	void *cookie;
1509 
1510 	ddsta.ddsta_fsname = fsname;
1511 	ddsta.ddsta_snapname = snapname;
1512 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1513 	ddsta.ddsta_htag = htag;
1514 
1515 	error = spa_open(fsname, &spa, FTAG);
1516 	if (error != 0)
1517 		return (error);
1518 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1519 	spa_close(spa, FTAG);
1520 
1521 	if (needsuspend) {
1522 		error = zil_suspend(fsname, &cookie);
1523 		if (error != 0)
1524 			return (error);
1525 	}
1526 
1527 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1528 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1529 
1530 	if (needsuspend)
1531 		zil_resume(cookie);
1532 	return (error);
1533 }
1534 
1535 
1536 void
1537 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1538 {
1539 	ASSERT(dmu_tx_is_syncing(tx));
1540 	ASSERT(ds->ds_objset != NULL);
1541 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1542 
1543 	/*
1544 	 * in case we had to change ds_fsid_guid when we opened it,
1545 	 * sync it out now.
1546 	 */
1547 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1548 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1549 
1550 	dmu_objset_sync(ds->ds_objset, zio, tx);
1551 
1552 	if (ds->ds_need_large_blocks && !ds->ds_large_blocks) {
1553 		dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
1554 		ds->ds_large_blocks = B_TRUE;
1555 	}
1556 }
1557 
1558 static void
1559 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1560 {
1561 	uint64_t count = 0;
1562 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1563 	zap_cursor_t zc;
1564 	zap_attribute_t za;
1565 	nvlist_t *propval = fnvlist_alloc();
1566 	nvlist_t *val = fnvlist_alloc();
1567 
1568 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1569 
1570 	/*
1571 	 * There may be missing entries in ds_next_clones_obj
1572 	 * due to a bug in a previous version of the code.
1573 	 * Only trust it if it has the right number of entries.
1574 	 */
1575 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1576 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1577 		    &count));
1578 	}
1579 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1580 		goto fail;
1581 	for (zap_cursor_init(&zc, mos,
1582 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1583 	    zap_cursor_retrieve(&zc, &za) == 0;
1584 	    zap_cursor_advance(&zc)) {
1585 		dsl_dataset_t *clone;
1586 		char buf[ZFS_MAXNAMELEN];
1587 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1588 		    za.za_first_integer, FTAG, &clone));
1589 		dsl_dir_name(clone->ds_dir, buf);
1590 		fnvlist_add_boolean(val, buf);
1591 		dsl_dataset_rele(clone, FTAG);
1592 	}
1593 	zap_cursor_fini(&zc);
1594 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1595 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1596 fail:
1597 	nvlist_free(val);
1598 	nvlist_free(propval);
1599 }
1600 
1601 void
1602 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1603 {
1604 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1605 	uint64_t refd, avail, uobjs, aobjs, ratio;
1606 
1607 	ASSERT(dsl_pool_config_held(dp));
1608 
1609 	ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1610 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1611 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1612 
1613 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1614 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1615 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1616 
1617 	if (ds->ds_is_snapshot) {
1618 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1619 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1620 		    dsl_dataset_phys(ds)->ds_unique_bytes);
1621 		get_clones_stat(ds, nv);
1622 	} else {
1623 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1624 			char buf[MAXNAMELEN];
1625 			dsl_dataset_name(ds->ds_prev, buf);
1626 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1627 		}
1628 
1629 		dsl_dir_stats(ds->ds_dir, nv);
1630 	}
1631 
1632 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1633 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1634 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1635 
1636 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1637 	    dsl_dataset_phys(ds)->ds_creation_time);
1638 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1639 	    dsl_dataset_phys(ds)->ds_creation_txg);
1640 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1641 	    ds->ds_quota);
1642 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1643 	    ds->ds_reserved);
1644 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1645 	    dsl_dataset_phys(ds)->ds_guid);
1646 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1647 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1648 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1649 	    ds->ds_object);
1650 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1651 	    ds->ds_userrefs);
1652 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1653 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1654 
1655 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1656 		uint64_t written, comp, uncomp;
1657 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1658 		dsl_dataset_t *prev;
1659 
1660 		int err = dsl_dataset_hold_obj(dp,
1661 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1662 		if (err == 0) {
1663 			err = dsl_dataset_space_written(prev, ds, &written,
1664 			    &comp, &uncomp);
1665 			dsl_dataset_rele(prev, FTAG);
1666 			if (err == 0) {
1667 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1668 				    written);
1669 			}
1670 		}
1671 	}
1672 }
1673 
1674 void
1675 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1676 {
1677 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1678 	ASSERT(dsl_pool_config_held(dp));
1679 
1680 	stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1681 	stat->dds_inconsistent =
1682 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1683 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1684 	stat->dds_origin[0] = '\0';
1685 	if (ds->ds_is_snapshot) {
1686 		stat->dds_is_snapshot = B_TRUE;
1687 		stat->dds_num_clones =
1688 		    dsl_dataset_phys(ds)->ds_num_children - 1;
1689 	} else {
1690 		stat->dds_is_snapshot = B_FALSE;
1691 		stat->dds_num_clones = 0;
1692 
1693 		if (dsl_dir_is_clone(ds->ds_dir)) {
1694 			dsl_dataset_t *ods;
1695 
1696 			VERIFY0(dsl_dataset_hold_obj(dp,
1697 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1698 			    FTAG, &ods));
1699 			dsl_dataset_name(ods, stat->dds_origin);
1700 			dsl_dataset_rele(ods, FTAG);
1701 		}
1702 	}
1703 }
1704 
1705 uint64_t
1706 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1707 {
1708 	return (ds->ds_fsid_guid);
1709 }
1710 
1711 void
1712 dsl_dataset_space(dsl_dataset_t *ds,
1713     uint64_t *refdbytesp, uint64_t *availbytesp,
1714     uint64_t *usedobjsp, uint64_t *availobjsp)
1715 {
1716 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1717 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1718 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1719 		*availbytesp +=
1720 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1721 	if (ds->ds_quota != 0) {
1722 		/*
1723 		 * Adjust available bytes according to refquota
1724 		 */
1725 		if (*refdbytesp < ds->ds_quota)
1726 			*availbytesp = MIN(*availbytesp,
1727 			    ds->ds_quota - *refdbytesp);
1728 		else
1729 			*availbytesp = 0;
1730 	}
1731 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1732 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1733 }
1734 
1735 boolean_t
1736 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1737 {
1738 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1739 
1740 	ASSERT(dsl_pool_config_held(dp));
1741 	if (snap == NULL)
1742 		return (B_FALSE);
1743 	if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1744 	    dsl_dataset_phys(snap)->ds_creation_txg) {
1745 		objset_t *os, *os_snap;
1746 		/*
1747 		 * It may be that only the ZIL differs, because it was
1748 		 * reset in the head.  Don't count that as being
1749 		 * modified.
1750 		 */
1751 		if (dmu_objset_from_ds(ds, &os) != 0)
1752 			return (B_TRUE);
1753 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1754 			return (B_TRUE);
1755 		return (bcmp(&os->os_phys->os_meta_dnode,
1756 		    &os_snap->os_phys->os_meta_dnode,
1757 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1758 	}
1759 	return (B_FALSE);
1760 }
1761 
1762 typedef struct dsl_dataset_rename_snapshot_arg {
1763 	const char *ddrsa_fsname;
1764 	const char *ddrsa_oldsnapname;
1765 	const char *ddrsa_newsnapname;
1766 	boolean_t ddrsa_recursive;
1767 	dmu_tx_t *ddrsa_tx;
1768 } dsl_dataset_rename_snapshot_arg_t;
1769 
1770 /* ARGSUSED */
1771 static int
1772 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1773     dsl_dataset_t *hds, void *arg)
1774 {
1775 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1776 	int error;
1777 	uint64_t val;
1778 
1779 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1780 	if (error != 0) {
1781 		/* ignore nonexistent snapshots */
1782 		return (error == ENOENT ? 0 : error);
1783 	}
1784 
1785 	/* new name should not exist */
1786 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1787 	if (error == 0)
1788 		error = SET_ERROR(EEXIST);
1789 	else if (error == ENOENT)
1790 		error = 0;
1791 
1792 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1793 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1794 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1795 		error = SET_ERROR(ENAMETOOLONG);
1796 
1797 	return (error);
1798 }
1799 
1800 static int
1801 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1802 {
1803 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1804 	dsl_pool_t *dp = dmu_tx_pool(tx);
1805 	dsl_dataset_t *hds;
1806 	int error;
1807 
1808 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1809 	if (error != 0)
1810 		return (error);
1811 
1812 	if (ddrsa->ddrsa_recursive) {
1813 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1814 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1815 		    DS_FIND_CHILDREN);
1816 	} else {
1817 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1818 	}
1819 	dsl_dataset_rele(hds, FTAG);
1820 	return (error);
1821 }
1822 
1823 static int
1824 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1825     dsl_dataset_t *hds, void *arg)
1826 {
1827 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1828 	dsl_dataset_t *ds;
1829 	uint64_t val;
1830 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1831 	int error;
1832 
1833 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1834 	ASSERT(error == 0 || error == ENOENT);
1835 	if (error == ENOENT) {
1836 		/* ignore nonexistent snapshots */
1837 		return (0);
1838 	}
1839 
1840 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1841 
1842 	/* log before we change the name */
1843 	spa_history_log_internal_ds(ds, "rename", tx,
1844 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1845 
1846 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1847 	    B_FALSE));
1848 	mutex_enter(&ds->ds_lock);
1849 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1850 	mutex_exit(&ds->ds_lock);
1851 	VERIFY0(zap_add(dp->dp_meta_objset,
1852 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
1853 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1854 
1855 	dsl_dataset_rele(ds, FTAG);
1856 	return (0);
1857 }
1858 
1859 static void
1860 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1861 {
1862 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1863 	dsl_pool_t *dp = dmu_tx_pool(tx);
1864 	dsl_dataset_t *hds;
1865 
1866 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1867 	ddrsa->ddrsa_tx = tx;
1868 	if (ddrsa->ddrsa_recursive) {
1869 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1870 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1871 		    DS_FIND_CHILDREN));
1872 	} else {
1873 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1874 	}
1875 	dsl_dataset_rele(hds, FTAG);
1876 }
1877 
1878 int
1879 dsl_dataset_rename_snapshot(const char *fsname,
1880     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1881 {
1882 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1883 
1884 	ddrsa.ddrsa_fsname = fsname;
1885 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1886 	ddrsa.ddrsa_newsnapname = newsnapname;
1887 	ddrsa.ddrsa_recursive = recursive;
1888 
1889 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1890 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
1891 	    1, ZFS_SPACE_CHECK_RESERVED));
1892 }
1893 
1894 /*
1895  * If we're doing an ownership handoff, we need to make sure that there is
1896  * only one long hold on the dataset.  We're not allowed to change anything here
1897  * so we don't permanently release the long hold or regular hold here.  We want
1898  * to do this only when syncing to avoid the dataset unexpectedly going away
1899  * when we release the long hold.
1900  */
1901 static int
1902 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1903 {
1904 	boolean_t held;
1905 
1906 	if (!dmu_tx_is_syncing(tx))
1907 		return (0);
1908 
1909 	if (owner != NULL) {
1910 		VERIFY3P(ds->ds_owner, ==, owner);
1911 		dsl_dataset_long_rele(ds, owner);
1912 	}
1913 
1914 	held = dsl_dataset_long_held(ds);
1915 
1916 	if (owner != NULL)
1917 		dsl_dataset_long_hold(ds, owner);
1918 
1919 	if (held)
1920 		return (SET_ERROR(EBUSY));
1921 
1922 	return (0);
1923 }
1924 
1925 typedef struct dsl_dataset_rollback_arg {
1926 	const char *ddra_fsname;
1927 	void *ddra_owner;
1928 	nvlist_t *ddra_result;
1929 } dsl_dataset_rollback_arg_t;
1930 
1931 static int
1932 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1933 {
1934 	dsl_dataset_rollback_arg_t *ddra = arg;
1935 	dsl_pool_t *dp = dmu_tx_pool(tx);
1936 	dsl_dataset_t *ds;
1937 	int64_t unused_refres_delta;
1938 	int error;
1939 
1940 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1941 	if (error != 0)
1942 		return (error);
1943 
1944 	/* must not be a snapshot */
1945 	if (ds->ds_is_snapshot) {
1946 		dsl_dataset_rele(ds, FTAG);
1947 		return (SET_ERROR(EINVAL));
1948 	}
1949 
1950 	/* must have a most recent snapshot */
1951 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
1952 		dsl_dataset_rele(ds, FTAG);
1953 		return (SET_ERROR(EINVAL));
1954 	}
1955 
1956 	/* must not have any bookmarks after the most recent snapshot */
1957 	nvlist_t *proprequest = fnvlist_alloc();
1958 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
1959 	nvlist_t *bookmarks = fnvlist_alloc();
1960 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
1961 	fnvlist_free(proprequest);
1962 	if (error != 0)
1963 		return (error);
1964 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
1965 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
1966 		nvlist_t *valuenv =
1967 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
1968 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
1969 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
1970 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
1971 			fnvlist_free(bookmarks);
1972 			dsl_dataset_rele(ds, FTAG);
1973 			return (SET_ERROR(EEXIST));
1974 		}
1975 	}
1976 	fnvlist_free(bookmarks);
1977 
1978 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
1979 	if (error != 0) {
1980 		dsl_dataset_rele(ds, FTAG);
1981 		return (error);
1982 	}
1983 
1984 	/*
1985 	 * Check if the snap we are rolling back to uses more than
1986 	 * the refquota.
1987 	 */
1988 	if (ds->ds_quota != 0 &&
1989 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
1990 		dsl_dataset_rele(ds, FTAG);
1991 		return (SET_ERROR(EDQUOT));
1992 	}
1993 
1994 	/*
1995 	 * When we do the clone swap, we will temporarily use more space
1996 	 * due to the refreservation (the head will no longer have any
1997 	 * unique space, so the entire amount of the refreservation will need
1998 	 * to be free).  We will immediately destroy the clone, freeing
1999 	 * this space, but the freeing happens over many txg's.
2000 	 */
2001 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2002 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2003 
2004 	if (unused_refres_delta > 0 &&
2005 	    unused_refres_delta >
2006 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2007 		dsl_dataset_rele(ds, FTAG);
2008 		return (SET_ERROR(ENOSPC));
2009 	}
2010 
2011 	dsl_dataset_rele(ds, FTAG);
2012 	return (0);
2013 }
2014 
2015 static void
2016 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2017 {
2018 	dsl_dataset_rollback_arg_t *ddra = arg;
2019 	dsl_pool_t *dp = dmu_tx_pool(tx);
2020 	dsl_dataset_t *ds, *clone;
2021 	uint64_t cloneobj;
2022 	char namebuf[ZFS_MAXNAMELEN];
2023 
2024 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2025 
2026 	dsl_dataset_name(ds->ds_prev, namebuf);
2027 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2028 
2029 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2030 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2031 
2032 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2033 
2034 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2035 	dsl_dataset_zero_zil(ds, tx);
2036 
2037 	dsl_destroy_head_sync_impl(clone, tx);
2038 
2039 	dsl_dataset_rele(clone, FTAG);
2040 	dsl_dataset_rele(ds, FTAG);
2041 }
2042 
2043 /*
2044  * Rolls back the given filesystem or volume to the most recent snapshot.
2045  * The name of the most recent snapshot will be returned under key "target"
2046  * in the result nvlist.
2047  *
2048  * If owner != NULL:
2049  * - The existing dataset MUST be owned by the specified owner at entry
2050  * - Upon return, dataset will still be held by the same owner, whether we
2051  *   succeed or not.
2052  *
2053  * This mode is required any time the existing filesystem is mounted.  See
2054  * notes above zfs_suspend_fs() for further details.
2055  */
2056 int
2057 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2058 {
2059 	dsl_dataset_rollback_arg_t ddra;
2060 
2061 	ddra.ddra_fsname = fsname;
2062 	ddra.ddra_owner = owner;
2063 	ddra.ddra_result = result;
2064 
2065 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2066 	    dsl_dataset_rollback_sync, &ddra,
2067 	    1, ZFS_SPACE_CHECK_RESERVED));
2068 }
2069 
2070 struct promotenode {
2071 	list_node_t link;
2072 	dsl_dataset_t *ds;
2073 };
2074 
2075 typedef struct dsl_dataset_promote_arg {
2076 	const char *ddpa_clonename;
2077 	dsl_dataset_t *ddpa_clone;
2078 	list_t shared_snaps, origin_snaps, clone_snaps;
2079 	dsl_dataset_t *origin_origin; /* origin of the origin */
2080 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2081 	char *err_ds;
2082 	cred_t *cr;
2083 } dsl_dataset_promote_arg_t;
2084 
2085 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2086 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2087     void *tag);
2088 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2089 
2090 static int
2091 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2092 {
2093 	dsl_dataset_promote_arg_t *ddpa = arg;
2094 	dsl_pool_t *dp = dmu_tx_pool(tx);
2095 	dsl_dataset_t *hds;
2096 	struct promotenode *snap;
2097 	dsl_dataset_t *origin_ds;
2098 	int err;
2099 	uint64_t unused;
2100 	uint64_t ss_mv_cnt;
2101 
2102 	err = promote_hold(ddpa, dp, FTAG);
2103 	if (err != 0)
2104 		return (err);
2105 
2106 	hds = ddpa->ddpa_clone;
2107 
2108 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2109 		promote_rele(ddpa, FTAG);
2110 		return (SET_ERROR(EXDEV));
2111 	}
2112 
2113 	/*
2114 	 * Compute and check the amount of space to transfer.  Since this is
2115 	 * so expensive, don't do the preliminary check.
2116 	 */
2117 	if (!dmu_tx_is_syncing(tx)) {
2118 		promote_rele(ddpa, FTAG);
2119 		return (0);
2120 	}
2121 
2122 	snap = list_head(&ddpa->shared_snaps);
2123 	origin_ds = snap->ds;
2124 
2125 	/* compute origin's new unique space */
2126 	snap = list_tail(&ddpa->clone_snaps);
2127 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2128 	    origin_ds->ds_object);
2129 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2130 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2131 	    &ddpa->unique, &unused, &unused);
2132 
2133 	/*
2134 	 * Walk the snapshots that we are moving
2135 	 *
2136 	 * Compute space to transfer.  Consider the incremental changes
2137 	 * to used by each snapshot:
2138 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2139 	 * So each snapshot gave birth to:
2140 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2141 	 * So a sequence would look like:
2142 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2143 	 * Which simplifies to:
2144 	 * uN + kN + kN-1 + ... + k1 + k0
2145 	 * Note however, if we stop before we reach the ORIGIN we get:
2146 	 * uN + kN + kN-1 + ... + kM - uM-1
2147 	 */
2148 	ss_mv_cnt = 0;
2149 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2150 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2151 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2152 	for (snap = list_head(&ddpa->shared_snaps); snap;
2153 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2154 		uint64_t val, dlused, dlcomp, dluncomp;
2155 		dsl_dataset_t *ds = snap->ds;
2156 
2157 		ss_mv_cnt++;
2158 
2159 		/*
2160 		 * If there are long holds, we won't be able to evict
2161 		 * the objset.
2162 		 */
2163 		if (dsl_dataset_long_held(ds)) {
2164 			err = SET_ERROR(EBUSY);
2165 			goto out;
2166 		}
2167 
2168 		/* Check that the snapshot name does not conflict */
2169 		VERIFY0(dsl_dataset_get_snapname(ds));
2170 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2171 		if (err == 0) {
2172 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2173 			err = SET_ERROR(EEXIST);
2174 			goto out;
2175 		}
2176 		if (err != ENOENT)
2177 			goto out;
2178 
2179 		/* The very first snapshot does not have a deadlist */
2180 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2181 			continue;
2182 
2183 		dsl_deadlist_space(&ds->ds_deadlist,
2184 		    &dlused, &dlcomp, &dluncomp);
2185 		ddpa->used += dlused;
2186 		ddpa->comp += dlcomp;
2187 		ddpa->uncomp += dluncomp;
2188 	}
2189 
2190 	/*
2191 	 * If we are a clone of a clone then we never reached ORIGIN,
2192 	 * so we need to subtract out the clone origin's used space.
2193 	 */
2194 	if (ddpa->origin_origin) {
2195 		ddpa->used -=
2196 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2197 		ddpa->comp -=
2198 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2199 		ddpa->uncomp -=
2200 		    dsl_dataset_phys(ddpa->origin_origin)->
2201 		    ds_uncompressed_bytes;
2202 	}
2203 
2204 	/* Check that there is enough space and limit headroom here */
2205 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2206 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2207 	if (err != 0)
2208 		goto out;
2209 
2210 	/*
2211 	 * Compute the amounts of space that will be used by snapshots
2212 	 * after the promotion (for both origin and clone).  For each,
2213 	 * it is the amount of space that will be on all of their
2214 	 * deadlists (that was not born before their new origin).
2215 	 */
2216 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2217 		uint64_t space;
2218 
2219 		/*
2220 		 * Note, typically this will not be a clone of a clone,
2221 		 * so dd_origin_txg will be < TXG_INITIAL, so
2222 		 * these snaplist_space() -> dsl_deadlist_space_range()
2223 		 * calls will be fast because they do not have to
2224 		 * iterate over all bps.
2225 		 */
2226 		snap = list_head(&ddpa->origin_snaps);
2227 		err = snaplist_space(&ddpa->shared_snaps,
2228 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2229 		if (err != 0)
2230 			goto out;
2231 
2232 		err = snaplist_space(&ddpa->clone_snaps,
2233 		    snap->ds->ds_dir->dd_origin_txg, &space);
2234 		if (err != 0)
2235 			goto out;
2236 		ddpa->cloneusedsnap += space;
2237 	}
2238 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2239 	    DD_FLAG_USED_BREAKDOWN) {
2240 		err = snaplist_space(&ddpa->origin_snaps,
2241 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2242 		    &ddpa->originusedsnap);
2243 		if (err != 0)
2244 			goto out;
2245 	}
2246 
2247 out:
2248 	promote_rele(ddpa, FTAG);
2249 	return (err);
2250 }
2251 
2252 static void
2253 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2254 {
2255 	dsl_dataset_promote_arg_t *ddpa = arg;
2256 	dsl_pool_t *dp = dmu_tx_pool(tx);
2257 	dsl_dataset_t *hds;
2258 	struct promotenode *snap;
2259 	dsl_dataset_t *origin_ds;
2260 	dsl_dataset_t *origin_head;
2261 	dsl_dir_t *dd;
2262 	dsl_dir_t *odd = NULL;
2263 	uint64_t oldnext_obj;
2264 	int64_t delta;
2265 
2266 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2267 	hds = ddpa->ddpa_clone;
2268 
2269 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2270 
2271 	snap = list_head(&ddpa->shared_snaps);
2272 	origin_ds = snap->ds;
2273 	dd = hds->ds_dir;
2274 
2275 	snap = list_head(&ddpa->origin_snaps);
2276 	origin_head = snap->ds;
2277 
2278 	/*
2279 	 * We need to explicitly open odd, since origin_ds's dd will be
2280 	 * changing.
2281 	 */
2282 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2283 	    NULL, FTAG, &odd));
2284 
2285 	/* change origin's next snap */
2286 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2287 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2288 	snap = list_tail(&ddpa->clone_snaps);
2289 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2290 	    origin_ds->ds_object);
2291 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2292 
2293 	/* change the origin's next clone */
2294 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2295 		dsl_dataset_remove_from_next_clones(origin_ds,
2296 		    snap->ds->ds_object, tx);
2297 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2298 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2299 		    oldnext_obj, tx));
2300 	}
2301 
2302 	/* change origin */
2303 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2304 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2305 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2306 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2307 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2308 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2309 	origin_head->ds_dir->dd_origin_txg =
2310 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2311 
2312 	/* change dd_clone entries */
2313 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2314 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2315 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2316 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2317 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2318 		    hds->ds_object, tx));
2319 
2320 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2321 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2322 		    origin_head->ds_object, tx));
2323 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2324 			dsl_dir_phys(dd)->dd_clones =
2325 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2326 			    DMU_OT_NONE, 0, tx);
2327 		}
2328 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2329 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2330 	}
2331 
2332 	/* move snapshots to this dir */
2333 	for (snap = list_head(&ddpa->shared_snaps); snap;
2334 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2335 		dsl_dataset_t *ds = snap->ds;
2336 
2337 		/*
2338 		 * Property callbacks are registered to a particular
2339 		 * dsl_dir.  Since ours is changing, evict the objset
2340 		 * so that they will be unregistered from the old dsl_dir.
2341 		 */
2342 		if (ds->ds_objset) {
2343 			dmu_objset_evict(ds->ds_objset);
2344 			ds->ds_objset = NULL;
2345 		}
2346 
2347 		/* move snap name entry */
2348 		VERIFY0(dsl_dataset_get_snapname(ds));
2349 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2350 		    ds->ds_snapname, tx, B_TRUE));
2351 		VERIFY0(zap_add(dp->dp_meta_objset,
2352 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2353 		    8, 1, &ds->ds_object, tx));
2354 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2355 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2356 
2357 		/* change containing dsl_dir */
2358 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2359 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2360 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2361 		ASSERT3P(ds->ds_dir, ==, odd);
2362 		dsl_dir_rele(ds->ds_dir, ds);
2363 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2364 		    NULL, ds, &ds->ds_dir));
2365 
2366 		/* move any clone references */
2367 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2368 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2369 			zap_cursor_t zc;
2370 			zap_attribute_t za;
2371 
2372 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2373 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
2374 			    zap_cursor_retrieve(&zc, &za) == 0;
2375 			    zap_cursor_advance(&zc)) {
2376 				dsl_dataset_t *cnds;
2377 				uint64_t o;
2378 
2379 				if (za.za_first_integer == oldnext_obj) {
2380 					/*
2381 					 * We've already moved the
2382 					 * origin's reference.
2383 					 */
2384 					continue;
2385 				}
2386 
2387 				VERIFY0(dsl_dataset_hold_obj(dp,
2388 				    za.za_first_integer, FTAG, &cnds));
2389 				o = dsl_dir_phys(cnds->ds_dir)->
2390 				    dd_head_dataset_obj;
2391 
2392 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2393 				    dsl_dir_phys(odd)->dd_clones, o, tx));
2394 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2395 				    dsl_dir_phys(dd)->dd_clones, o, tx));
2396 				dsl_dataset_rele(cnds, FTAG);
2397 			}
2398 			zap_cursor_fini(&zc);
2399 		}
2400 
2401 		ASSERT(!dsl_prop_hascb(ds));
2402 	}
2403 
2404 	/*
2405 	 * Change space accounting.
2406 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2407 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2408 	 * is true for each of {clone,origin} independently.
2409 	 */
2410 
2411 	delta = ddpa->cloneusedsnap -
2412 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2413 	ASSERT3S(delta, >=, 0);
2414 	ASSERT3U(ddpa->used, >=, delta);
2415 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2416 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2417 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2418 
2419 	delta = ddpa->originusedsnap -
2420 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2421 	ASSERT3S(delta, <=, 0);
2422 	ASSERT3U(ddpa->used, >=, -delta);
2423 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2424 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2425 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2426 
2427 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2428 
2429 	/* log history record */
2430 	spa_history_log_internal_ds(hds, "promote", tx, "");
2431 
2432 	dsl_dir_rele(odd, FTAG);
2433 	promote_rele(ddpa, FTAG);
2434 }
2435 
2436 /*
2437  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2438  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2439  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2440  * snapshots back to this dataset's origin.
2441  */
2442 static int
2443 snaplist_make(dsl_pool_t *dp,
2444     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2445 {
2446 	uint64_t obj = last_obj;
2447 
2448 	list_create(l, sizeof (struct promotenode),
2449 	    offsetof(struct promotenode, link));
2450 
2451 	while (obj != first_obj) {
2452 		dsl_dataset_t *ds;
2453 		struct promotenode *snap;
2454 		int err;
2455 
2456 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2457 		ASSERT(err != ENOENT);
2458 		if (err != 0)
2459 			return (err);
2460 
2461 		if (first_obj == 0)
2462 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2463 
2464 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2465 		snap->ds = ds;
2466 		list_insert_tail(l, snap);
2467 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2468 	}
2469 
2470 	return (0);
2471 }
2472 
2473 static int
2474 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2475 {
2476 	struct promotenode *snap;
2477 
2478 	*spacep = 0;
2479 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2480 		uint64_t used, comp, uncomp;
2481 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2482 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2483 		*spacep += used;
2484 	}
2485 	return (0);
2486 }
2487 
2488 static void
2489 snaplist_destroy(list_t *l, void *tag)
2490 {
2491 	struct promotenode *snap;
2492 
2493 	if (l == NULL || !list_link_active(&l->list_head))
2494 		return;
2495 
2496 	while ((snap = list_tail(l)) != NULL) {
2497 		list_remove(l, snap);
2498 		dsl_dataset_rele(snap->ds, tag);
2499 		kmem_free(snap, sizeof (*snap));
2500 	}
2501 	list_destroy(l);
2502 }
2503 
2504 static int
2505 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2506 {
2507 	int error;
2508 	dsl_dir_t *dd;
2509 	struct promotenode *snap;
2510 
2511 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2512 	    &ddpa->ddpa_clone);
2513 	if (error != 0)
2514 		return (error);
2515 	dd = ddpa->ddpa_clone->ds_dir;
2516 
2517 	if (ddpa->ddpa_clone->ds_is_snapshot ||
2518 	    !dsl_dir_is_clone(dd)) {
2519 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2520 		return (SET_ERROR(EINVAL));
2521 	}
2522 
2523 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2524 	    &ddpa->shared_snaps, tag);
2525 	if (error != 0)
2526 		goto out;
2527 
2528 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2529 	    &ddpa->clone_snaps, tag);
2530 	if (error != 0)
2531 		goto out;
2532 
2533 	snap = list_head(&ddpa->shared_snaps);
2534 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2535 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2536 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2537 	    &ddpa->origin_snaps, tag);
2538 	if (error != 0)
2539 		goto out;
2540 
2541 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2542 		error = dsl_dataset_hold_obj(dp,
2543 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2544 		    tag, &ddpa->origin_origin);
2545 		if (error != 0)
2546 			goto out;
2547 	}
2548 out:
2549 	if (error != 0)
2550 		promote_rele(ddpa, tag);
2551 	return (error);
2552 }
2553 
2554 static void
2555 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2556 {
2557 	snaplist_destroy(&ddpa->shared_snaps, tag);
2558 	snaplist_destroy(&ddpa->clone_snaps, tag);
2559 	snaplist_destroy(&ddpa->origin_snaps, tag);
2560 	if (ddpa->origin_origin != NULL)
2561 		dsl_dataset_rele(ddpa->origin_origin, tag);
2562 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2563 }
2564 
2565 /*
2566  * Promote a clone.
2567  *
2568  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2569  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2570  */
2571 int
2572 dsl_dataset_promote(const char *name, char *conflsnap)
2573 {
2574 	dsl_dataset_promote_arg_t ddpa = { 0 };
2575 	uint64_t numsnaps;
2576 	int error;
2577 	objset_t *os;
2578 
2579 	/*
2580 	 * We will modify space proportional to the number of
2581 	 * snapshots.  Compute numsnaps.
2582 	 */
2583 	error = dmu_objset_hold(name, FTAG, &os);
2584 	if (error != 0)
2585 		return (error);
2586 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2587 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2588 	    &numsnaps);
2589 	dmu_objset_rele(os, FTAG);
2590 	if (error != 0)
2591 		return (error);
2592 
2593 	ddpa.ddpa_clonename = name;
2594 	ddpa.err_ds = conflsnap;
2595 	ddpa.cr = CRED();
2596 
2597 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2598 	    dsl_dataset_promote_sync, &ddpa,
2599 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2600 }
2601 
2602 int
2603 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2604     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2605 {
2606 	int64_t unused_refres_delta;
2607 
2608 	/* they should both be heads */
2609 	if (clone->ds_is_snapshot ||
2610 	    origin_head->ds_is_snapshot)
2611 		return (SET_ERROR(EINVAL));
2612 
2613 	/* if we are not forcing, the branch point should be just before them */
2614 	if (!force && clone->ds_prev != origin_head->ds_prev)
2615 		return (SET_ERROR(EINVAL));
2616 
2617 	/* clone should be the clone (unless they are unrelated) */
2618 	if (clone->ds_prev != NULL &&
2619 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2620 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2621 		return (SET_ERROR(EINVAL));
2622 
2623 	/* the clone should be a child of the origin */
2624 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2625 		return (SET_ERROR(EINVAL));
2626 
2627 	/* origin_head shouldn't be modified unless 'force' */
2628 	if (!force &&
2629 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2630 		return (SET_ERROR(ETXTBSY));
2631 
2632 	/* origin_head should have no long holds (e.g. is not mounted) */
2633 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2634 		return (SET_ERROR(EBUSY));
2635 
2636 	/* check amount of any unconsumed refreservation */
2637 	unused_refres_delta =
2638 	    (int64_t)MIN(origin_head->ds_reserved,
2639 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2640 	    (int64_t)MIN(origin_head->ds_reserved,
2641 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2642 
2643 	if (unused_refres_delta > 0 &&
2644 	    unused_refres_delta >
2645 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2646 		return (SET_ERROR(ENOSPC));
2647 
2648 	/* clone can't be over the head's refquota */
2649 	if (origin_head->ds_quota != 0 &&
2650 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
2651 	    origin_head->ds_quota)
2652 		return (SET_ERROR(EDQUOT));
2653 
2654 	return (0);
2655 }
2656 
2657 void
2658 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2659     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2660 {
2661 	dsl_pool_t *dp = dmu_tx_pool(tx);
2662 	int64_t unused_refres_delta;
2663 
2664 	ASSERT(clone->ds_reserved == 0);
2665 	ASSERT(origin_head->ds_quota == 0 ||
2666 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2667 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2668 
2669 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2670 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2671 
2672 	if (clone->ds_objset != NULL) {
2673 		dmu_objset_evict(clone->ds_objset);
2674 		clone->ds_objset = NULL;
2675 	}
2676 
2677 	if (origin_head->ds_objset != NULL) {
2678 		dmu_objset_evict(origin_head->ds_objset);
2679 		origin_head->ds_objset = NULL;
2680 	}
2681 
2682 	unused_refres_delta =
2683 	    (int64_t)MIN(origin_head->ds_reserved,
2684 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2685 	    (int64_t)MIN(origin_head->ds_reserved,
2686 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2687 
2688 	/*
2689 	 * Reset origin's unique bytes, if it exists.
2690 	 */
2691 	if (clone->ds_prev) {
2692 		dsl_dataset_t *origin = clone->ds_prev;
2693 		uint64_t comp, uncomp;
2694 
2695 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2696 		dsl_deadlist_space_range(&clone->ds_deadlist,
2697 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2698 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2699 	}
2700 
2701 	/* swap blkptrs */
2702 	{
2703 		blkptr_t tmp;
2704 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
2705 		dsl_dataset_phys(origin_head)->ds_bp =
2706 		    dsl_dataset_phys(clone)->ds_bp;
2707 		dsl_dataset_phys(clone)->ds_bp = tmp;
2708 	}
2709 
2710 	/* set dd_*_bytes */
2711 	{
2712 		int64_t dused, dcomp, duncomp;
2713 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2714 		uint64_t odl_used, odl_comp, odl_uncomp;
2715 
2716 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2717 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2718 
2719 		dsl_deadlist_space(&clone->ds_deadlist,
2720 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2721 		dsl_deadlist_space(&origin_head->ds_deadlist,
2722 		    &odl_used, &odl_comp, &odl_uncomp);
2723 
2724 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2725 		    cdl_used -
2726 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2727 		    odl_used);
2728 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2729 		    cdl_comp -
2730 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2731 		    odl_comp);
2732 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2733 		    cdl_uncomp -
2734 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2735 		    odl_uncomp);
2736 
2737 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2738 		    dused, dcomp, duncomp, tx);
2739 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2740 		    -dused, -dcomp, -duncomp, tx);
2741 
2742 		/*
2743 		 * The difference in the space used by snapshots is the
2744 		 * difference in snapshot space due to the head's
2745 		 * deadlist (since that's the only thing that's
2746 		 * changing that affects the snapused).
2747 		 */
2748 		dsl_deadlist_space_range(&clone->ds_deadlist,
2749 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2750 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2751 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2752 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2753 		    &odl_used, &odl_comp, &odl_uncomp);
2754 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2755 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2756 	}
2757 
2758 	/* swap ds_*_bytes */
2759 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2760 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
2761 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2762 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
2763 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2764 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2765 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2766 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2767 
2768 	/* apply any parent delta for change in unconsumed refreservation */
2769 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2770 	    unused_refres_delta, 0, 0, tx);
2771 
2772 	/*
2773 	 * Swap deadlists.
2774 	 */
2775 	dsl_deadlist_close(&clone->ds_deadlist);
2776 	dsl_deadlist_close(&origin_head->ds_deadlist);
2777 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2778 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2779 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2780 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2781 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2782 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
2783 
2784 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2785 
2786 	spa_history_log_internal_ds(clone, "clone swap", tx,
2787 	    "parent=%s", origin_head->ds_dir->dd_myname);
2788 }
2789 
2790 /*
2791  * Given a pool name and a dataset object number in that pool,
2792  * return the name of that dataset.
2793  */
2794 int
2795 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2796 {
2797 	dsl_pool_t *dp;
2798 	dsl_dataset_t *ds;
2799 	int error;
2800 
2801 	error = dsl_pool_hold(pname, FTAG, &dp);
2802 	if (error != 0)
2803 		return (error);
2804 
2805 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2806 	if (error == 0) {
2807 		dsl_dataset_name(ds, buf);
2808 		dsl_dataset_rele(ds, FTAG);
2809 	}
2810 	dsl_pool_rele(dp, FTAG);
2811 
2812 	return (error);
2813 }
2814 
2815 int
2816 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2817     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2818 {
2819 	int error = 0;
2820 
2821 	ASSERT3S(asize, >, 0);
2822 
2823 	/*
2824 	 * *ref_rsrv is the portion of asize that will come from any
2825 	 * unconsumed refreservation space.
2826 	 */
2827 	*ref_rsrv = 0;
2828 
2829 	mutex_enter(&ds->ds_lock);
2830 	/*
2831 	 * Make a space adjustment for reserved bytes.
2832 	 */
2833 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2834 		ASSERT3U(*used, >=,
2835 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2836 		*used -=
2837 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2838 		*ref_rsrv =
2839 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2840 	}
2841 
2842 	if (!check_quota || ds->ds_quota == 0) {
2843 		mutex_exit(&ds->ds_lock);
2844 		return (0);
2845 	}
2846 	/*
2847 	 * If they are requesting more space, and our current estimate
2848 	 * is over quota, they get to try again unless the actual
2849 	 * on-disk is over quota and there are no pending changes (which
2850 	 * may free up space for us).
2851 	 */
2852 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
2853 	    ds->ds_quota) {
2854 		if (inflight > 0 ||
2855 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
2856 			error = SET_ERROR(ERESTART);
2857 		else
2858 			error = SET_ERROR(EDQUOT);
2859 	}
2860 	mutex_exit(&ds->ds_lock);
2861 
2862 	return (error);
2863 }
2864 
2865 typedef struct dsl_dataset_set_qr_arg {
2866 	const char *ddsqra_name;
2867 	zprop_source_t ddsqra_source;
2868 	uint64_t ddsqra_value;
2869 } dsl_dataset_set_qr_arg_t;
2870 
2871 
2872 /* ARGSUSED */
2873 static int
2874 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2875 {
2876 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2877 	dsl_pool_t *dp = dmu_tx_pool(tx);
2878 	dsl_dataset_t *ds;
2879 	int error;
2880 	uint64_t newval;
2881 
2882 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2883 		return (SET_ERROR(ENOTSUP));
2884 
2885 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2886 	if (error != 0)
2887 		return (error);
2888 
2889 	if (ds->ds_is_snapshot) {
2890 		dsl_dataset_rele(ds, FTAG);
2891 		return (SET_ERROR(EINVAL));
2892 	}
2893 
2894 	error = dsl_prop_predict(ds->ds_dir,
2895 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2896 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2897 	if (error != 0) {
2898 		dsl_dataset_rele(ds, FTAG);
2899 		return (error);
2900 	}
2901 
2902 	if (newval == 0) {
2903 		dsl_dataset_rele(ds, FTAG);
2904 		return (0);
2905 	}
2906 
2907 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
2908 	    newval < ds->ds_reserved) {
2909 		dsl_dataset_rele(ds, FTAG);
2910 		return (SET_ERROR(ENOSPC));
2911 	}
2912 
2913 	dsl_dataset_rele(ds, FTAG);
2914 	return (0);
2915 }
2916 
2917 static void
2918 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2919 {
2920 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2921 	dsl_pool_t *dp = dmu_tx_pool(tx);
2922 	dsl_dataset_t *ds;
2923 	uint64_t newval;
2924 
2925 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2926 
2927 	dsl_prop_set_sync_impl(ds,
2928 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2929 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2930 	    &ddsqra->ddsqra_value, tx);
2931 
2932 	VERIFY0(dsl_prop_get_int_ds(ds,
2933 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2934 
2935 	if (ds->ds_quota != newval) {
2936 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2937 		ds->ds_quota = newval;
2938 	}
2939 	dsl_dataset_rele(ds, FTAG);
2940 }
2941 
2942 int
2943 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2944     uint64_t refquota)
2945 {
2946 	dsl_dataset_set_qr_arg_t ddsqra;
2947 
2948 	ddsqra.ddsqra_name = dsname;
2949 	ddsqra.ddsqra_source = source;
2950 	ddsqra.ddsqra_value = refquota;
2951 
2952 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2953 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
2954 }
2955 
2956 static int
2957 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2958 {
2959 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2960 	dsl_pool_t *dp = dmu_tx_pool(tx);
2961 	dsl_dataset_t *ds;
2962 	int error;
2963 	uint64_t newval, unique;
2964 
2965 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2966 		return (SET_ERROR(ENOTSUP));
2967 
2968 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2969 	if (error != 0)
2970 		return (error);
2971 
2972 	if (ds->ds_is_snapshot) {
2973 		dsl_dataset_rele(ds, FTAG);
2974 		return (SET_ERROR(EINVAL));
2975 	}
2976 
2977 	error = dsl_prop_predict(ds->ds_dir,
2978 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2979 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2980 	if (error != 0) {
2981 		dsl_dataset_rele(ds, FTAG);
2982 		return (error);
2983 	}
2984 
2985 	/*
2986 	 * If we are doing the preliminary check in open context, the
2987 	 * space estimates may be inaccurate.
2988 	 */
2989 	if (!dmu_tx_is_syncing(tx)) {
2990 		dsl_dataset_rele(ds, FTAG);
2991 		return (0);
2992 	}
2993 
2994 	mutex_enter(&ds->ds_lock);
2995 	if (!DS_UNIQUE_IS_ACCURATE(ds))
2996 		dsl_dataset_recalc_head_uniq(ds);
2997 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
2998 	mutex_exit(&ds->ds_lock);
2999 
3000 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3001 		uint64_t delta = MAX(unique, newval) -
3002 		    MAX(unique, ds->ds_reserved);
3003 
3004 		if (delta >
3005 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3006 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3007 			dsl_dataset_rele(ds, FTAG);
3008 			return (SET_ERROR(ENOSPC));
3009 		}
3010 	}
3011 
3012 	dsl_dataset_rele(ds, FTAG);
3013 	return (0);
3014 }
3015 
3016 void
3017 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3018     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3019 {
3020 	uint64_t newval;
3021 	uint64_t unique;
3022 	int64_t delta;
3023 
3024 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3025 	    source, sizeof (value), 1, &value, tx);
3026 
3027 	VERIFY0(dsl_prop_get_int_ds(ds,
3028 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3029 
3030 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3031 	mutex_enter(&ds->ds_dir->dd_lock);
3032 	mutex_enter(&ds->ds_lock);
3033 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3034 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3035 	delta = MAX(0, (int64_t)(newval - unique)) -
3036 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3037 	ds->ds_reserved = newval;
3038 	mutex_exit(&ds->ds_lock);
3039 
3040 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3041 	mutex_exit(&ds->ds_dir->dd_lock);
3042 }
3043 
3044 static void
3045 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3046 {
3047 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3048 	dsl_pool_t *dp = dmu_tx_pool(tx);
3049 	dsl_dataset_t *ds;
3050 
3051 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3052 	dsl_dataset_set_refreservation_sync_impl(ds,
3053 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3054 	dsl_dataset_rele(ds, FTAG);
3055 }
3056 
3057 int
3058 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3059     uint64_t refreservation)
3060 {
3061 	dsl_dataset_set_qr_arg_t ddsqra;
3062 
3063 	ddsqra.ddsqra_name = dsname;
3064 	ddsqra.ddsqra_source = source;
3065 	ddsqra.ddsqra_value = refreservation;
3066 
3067 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3068 	    dsl_dataset_set_refreservation_sync, &ddsqra,
3069 	    0, ZFS_SPACE_CHECK_NONE));
3070 }
3071 
3072 /*
3073  * Return (in *usedp) the amount of space written in new that is not
3074  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3075  * a snapshot before new, in new's filesystem (or its origin).  If not then
3076  * fail and return EINVAL.
3077  *
3078  * The written space is calculated by considering two components:  First, we
3079  * ignore any freed space, and calculate the written as new's used space
3080  * minus old's used space.  Next, we add in the amount of space that was freed
3081  * between the two snapshots, thus reducing new's used space relative to old's.
3082  * Specifically, this is the space that was born before old->ds_creation_txg,
3083  * and freed before new (ie. on new's deadlist or a previous deadlist).
3084  *
3085  * space freed                         [---------------------]
3086  * snapshots                       ---O-------O--------O-------O------
3087  *                                         oldsnap            new
3088  */
3089 int
3090 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3091     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3092 {
3093 	int err = 0;
3094 	uint64_t snapobj;
3095 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3096 
3097 	ASSERT(dsl_pool_config_held(dp));
3098 
3099 	*usedp = 0;
3100 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3101 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3102 
3103 	*compp = 0;
3104 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3105 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3106 
3107 	*uncompp = 0;
3108 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3109 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3110 
3111 	snapobj = new->ds_object;
3112 	while (snapobj != oldsnap->ds_object) {
3113 		dsl_dataset_t *snap;
3114 		uint64_t used, comp, uncomp;
3115 
3116 		if (snapobj == new->ds_object) {
3117 			snap = new;
3118 		} else {
3119 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3120 			if (err != 0)
3121 				break;
3122 		}
3123 
3124 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3125 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3126 			/*
3127 			 * The blocks in the deadlist can not be born after
3128 			 * ds_prev_snap_txg, so get the whole deadlist space,
3129 			 * which is more efficient (especially for old-format
3130 			 * deadlists).  Unfortunately the deadlist code
3131 			 * doesn't have enough information to make this
3132 			 * optimization itself.
3133 			 */
3134 			dsl_deadlist_space(&snap->ds_deadlist,
3135 			    &used, &comp, &uncomp);
3136 		} else {
3137 			dsl_deadlist_space_range(&snap->ds_deadlist,
3138 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3139 			    &used, &comp, &uncomp);
3140 		}
3141 		*usedp += used;
3142 		*compp += comp;
3143 		*uncompp += uncomp;
3144 
3145 		/*
3146 		 * If we get to the beginning of the chain of snapshots
3147 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3148 		 * was not a snapshot of/before new.
3149 		 */
3150 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3151 		if (snap != new)
3152 			dsl_dataset_rele(snap, FTAG);
3153 		if (snapobj == 0) {
3154 			err = SET_ERROR(EINVAL);
3155 			break;
3156 		}
3157 
3158 	}
3159 	return (err);
3160 }
3161 
3162 /*
3163  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3164  * lastsnap, and all snapshots in between are deleted.
3165  *
3166  * blocks that would be freed            [---------------------------]
3167  * snapshots                       ---O-------O--------O-------O--------O
3168  *                                        firstsnap        lastsnap
3169  *
3170  * This is the set of blocks that were born after the snap before firstsnap,
3171  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3172  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3173  * We calculate this by iterating over the relevant deadlists (from the snap
3174  * after lastsnap, backward to the snap after firstsnap), summing up the
3175  * space on the deadlist that was born after the snap before firstsnap.
3176  */
3177 int
3178 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3179     dsl_dataset_t *lastsnap,
3180     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3181 {
3182 	int err = 0;
3183 	uint64_t snapobj;
3184 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3185 
3186 	ASSERT(firstsnap->ds_is_snapshot);
3187 	ASSERT(lastsnap->ds_is_snapshot);
3188 
3189 	/*
3190 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3191 	 * is before lastsnap.
3192 	 */
3193 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3194 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3195 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3196 		return (SET_ERROR(EINVAL));
3197 
3198 	*usedp = *compp = *uncompp = 0;
3199 
3200 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3201 	while (snapobj != firstsnap->ds_object) {
3202 		dsl_dataset_t *ds;
3203 		uint64_t used, comp, uncomp;
3204 
3205 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3206 		if (err != 0)
3207 			break;
3208 
3209 		dsl_deadlist_space_range(&ds->ds_deadlist,
3210 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3211 		    &used, &comp, &uncomp);
3212 		*usedp += used;
3213 		*compp += comp;
3214 		*uncompp += uncomp;
3215 
3216 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3217 		ASSERT3U(snapobj, !=, 0);
3218 		dsl_dataset_rele(ds, FTAG);
3219 	}
3220 	return (err);
3221 }
3222 
3223 static int
3224 dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx)
3225 {
3226 	const char *dsname = arg;
3227 	dsl_dataset_t *ds;
3228 	dsl_pool_t *dp = dmu_tx_pool(tx);
3229 	int error = 0;
3230 
3231 	if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
3232 		return (SET_ERROR(ENOTSUP));
3233 
3234 	ASSERT(spa_feature_is_enabled(dp->dp_spa,
3235 	    SPA_FEATURE_EXTENSIBLE_DATASET));
3236 
3237 	error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
3238 	if (error != 0)
3239 		return (error);
3240 
3241 	if (ds->ds_large_blocks)
3242 		error = EALREADY;
3243 	dsl_dataset_rele(ds, FTAG);
3244 
3245 	return (error);
3246 }
3247 
3248 void
3249 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx)
3250 {
3251 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
3252 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
3253 	uint64_t zero = 0;
3254 
3255 	spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx);
3256 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
3257 
3258 	VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS,
3259 	    sizeof (zero), 1, &zero, tx));
3260 }
3261 
3262 static void
3263 dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx)
3264 {
3265 	const char *dsname = arg;
3266 	dsl_dataset_t *ds;
3267 
3268 	VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds));
3269 
3270 	dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
3271 	ASSERT(!ds->ds_large_blocks);
3272 	ds->ds_large_blocks = B_TRUE;
3273 	dsl_dataset_rele(ds, FTAG);
3274 }
3275 
3276 int
3277 dsl_dataset_activate_large_blocks(const char *dsname)
3278 {
3279 	int error;
3280 
3281 	error = dsl_sync_task(dsname,
3282 	    dsl_dataset_activate_large_blocks_check,
3283 	    dsl_dataset_activate_large_blocks_sync, (void *)dsname,
3284 	    1, ZFS_SPACE_CHECK_RESERVED);
3285 
3286 	/*
3287 	 * EALREADY indicates that this dataset already supports large blocks.
3288 	 */
3289 	if (error == EALREADY)
3290 		error = 0;
3291 	return (error);
3292 }
3293 
3294 /*
3295  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3296  * For example, they could both be snapshots of the same filesystem, and
3297  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3298  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3299  * filesystem.  Or 'earlier' could be the origin's origin.
3300  *
3301  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3302  */
3303 boolean_t
3304 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3305 	uint64_t earlier_txg)
3306 {
3307 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3308 	int error;
3309 	boolean_t ret;
3310 
3311 	ASSERT(dsl_pool_config_held(dp));
3312 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3313 
3314 	if (earlier_txg == 0)
3315 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3316 
3317 	if (later->ds_is_snapshot &&
3318 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3319 		return (B_FALSE);
3320 
3321 	if (later->ds_dir == earlier->ds_dir)
3322 		return (B_TRUE);
3323 	if (!dsl_dir_is_clone(later->ds_dir))
3324 		return (B_FALSE);
3325 
3326 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3327 		return (B_TRUE);
3328 	dsl_dataset_t *origin;
3329 	error = dsl_dataset_hold_obj(dp,
3330 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3331 	if (error != 0)
3332 		return (B_FALSE);
3333 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3334 	dsl_dataset_rele(origin, FTAG);
3335 	return (ret);
3336 }
3337 
3338 
3339 void
3340 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3341 {
3342 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3343 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3344 }
3345