xref: /titanic_52/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision c0dbe950192897247de6003f2dea1ca3c466fe01)
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 			int zaperr = zap_contains(mos, dsobj,
427 			    DS_FIELD_LARGE_BLOCKS);
428 			if (zaperr != ENOENT) {
429 				VERIFY0(zaperr);
430 				ds->ds_large_blocks = B_TRUE;
431 			}
432 		}
433 
434 		if (err == 0) {
435 			err = dsl_dir_hold_obj(dp,
436 			    dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds,
437 			    &ds->ds_dir);
438 		}
439 		if (err != 0) {
440 			mutex_destroy(&ds->ds_lock);
441 			mutex_destroy(&ds->ds_opening_lock);
442 			mutex_destroy(&ds->ds_sendstream_lock);
443 			refcount_destroy(&ds->ds_longholds);
444 			bplist_destroy(&ds->ds_pending_deadlist);
445 			dsl_deadlist_close(&ds->ds_deadlist);
446 			kmem_free(ds, sizeof (dsl_dataset_t));
447 			dmu_buf_rele(dbuf, tag);
448 			return (err);
449 		}
450 
451 		if (!ds->ds_is_snapshot) {
452 			ds->ds_snapname[0] = '\0';
453 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
454 				err = dsl_dataset_hold_obj(dp,
455 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
456 				    ds, &ds->ds_prev);
457 			}
458 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
459 				int zaperr = zap_lookup(mos, ds->ds_object,
460 				    DS_FIELD_BOOKMARK_NAMES,
461 				    sizeof (ds->ds_bookmarks), 1,
462 				    &ds->ds_bookmarks);
463 				if (zaperr != ENOENT)
464 					VERIFY0(zaperr);
465 			}
466 		} else {
467 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
468 				err = dsl_dataset_get_snapname(ds);
469 			if (err == 0 &&
470 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
471 				err = zap_count(
472 				    ds->ds_dir->dd_pool->dp_meta_objset,
473 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
474 				    &ds->ds_userrefs);
475 			}
476 		}
477 
478 		if (err == 0 && !ds->ds_is_snapshot) {
479 			err = dsl_prop_get_int_ds(ds,
480 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
481 			    &ds->ds_reserved);
482 			if (err == 0) {
483 				err = dsl_prop_get_int_ds(ds,
484 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
485 				    &ds->ds_quota);
486 			}
487 		} else {
488 			ds->ds_reserved = ds->ds_quota = 0;
489 		}
490 
491 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
492 		if (err == 0)
493 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
494 
495 		if (err != 0 || winner != NULL) {
496 			bplist_destroy(&ds->ds_pending_deadlist);
497 			dsl_deadlist_close(&ds->ds_deadlist);
498 			if (ds->ds_prev)
499 				dsl_dataset_rele(ds->ds_prev, ds);
500 			dsl_dir_rele(ds->ds_dir, ds);
501 			mutex_destroy(&ds->ds_lock);
502 			mutex_destroy(&ds->ds_opening_lock);
503 			mutex_destroy(&ds->ds_sendstream_lock);
504 			refcount_destroy(&ds->ds_longholds);
505 			kmem_free(ds, sizeof (dsl_dataset_t));
506 			if (err != 0) {
507 				dmu_buf_rele(dbuf, tag);
508 				return (err);
509 			}
510 			ds = winner;
511 		} else {
512 			ds->ds_fsid_guid =
513 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
514 		}
515 	}
516 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
517 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
518 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
519 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
520 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
521 	*dsp = ds;
522 	return (0);
523 }
524 
525 int
526 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
527     void *tag, dsl_dataset_t **dsp)
528 {
529 	dsl_dir_t *dd;
530 	const char *snapname;
531 	uint64_t obj;
532 	int err = 0;
533 
534 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
535 	if (err != 0)
536 		return (err);
537 
538 	ASSERT(dsl_pool_config_held(dp));
539 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
540 	if (obj != 0)
541 		err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
542 	else
543 		err = SET_ERROR(ENOENT);
544 
545 	/* we may be looking for a snapshot */
546 	if (err == 0 && snapname != NULL) {
547 		dsl_dataset_t *ds;
548 
549 		if (*snapname++ != '@') {
550 			dsl_dataset_rele(*dsp, tag);
551 			dsl_dir_rele(dd, FTAG);
552 			return (SET_ERROR(ENOENT));
553 		}
554 
555 		dprintf("looking for snapshot '%s'\n", snapname);
556 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
557 		if (err == 0)
558 			err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
559 		dsl_dataset_rele(*dsp, tag);
560 
561 		if (err == 0) {
562 			mutex_enter(&ds->ds_lock);
563 			if (ds->ds_snapname[0] == 0)
564 				(void) strlcpy(ds->ds_snapname, snapname,
565 				    sizeof (ds->ds_snapname));
566 			mutex_exit(&ds->ds_lock);
567 			*dsp = ds;
568 		}
569 	}
570 
571 	dsl_dir_rele(dd, FTAG);
572 	return (err);
573 }
574 
575 int
576 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
577     void *tag, dsl_dataset_t **dsp)
578 {
579 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
580 	if (err != 0)
581 		return (err);
582 	if (!dsl_dataset_tryown(*dsp, tag)) {
583 		dsl_dataset_rele(*dsp, tag);
584 		*dsp = NULL;
585 		return (SET_ERROR(EBUSY));
586 	}
587 	return (0);
588 }
589 
590 int
591 dsl_dataset_own(dsl_pool_t *dp, const char *name,
592     void *tag, dsl_dataset_t **dsp)
593 {
594 	int err = dsl_dataset_hold(dp, name, tag, dsp);
595 	if (err != 0)
596 		return (err);
597 	if (!dsl_dataset_tryown(*dsp, tag)) {
598 		dsl_dataset_rele(*dsp, tag);
599 		return (SET_ERROR(EBUSY));
600 	}
601 	return (0);
602 }
603 
604 /*
605  * See the comment above dsl_pool_hold() for details.  In summary, a long
606  * hold is used to prevent destruction of a dataset while the pool hold
607  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
608  *
609  * The dataset and pool must be held when this function is called.  After it
610  * is called, the pool hold may be released while the dataset is still held
611  * and accessed.
612  */
613 void
614 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
615 {
616 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
617 	(void) refcount_add(&ds->ds_longholds, tag);
618 }
619 
620 void
621 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
622 {
623 	(void) refcount_remove(&ds->ds_longholds, tag);
624 }
625 
626 /* Return B_TRUE if there are any long holds on this dataset. */
627 boolean_t
628 dsl_dataset_long_held(dsl_dataset_t *ds)
629 {
630 	return (!refcount_is_zero(&ds->ds_longholds));
631 }
632 
633 void
634 dsl_dataset_name(dsl_dataset_t *ds, char *name)
635 {
636 	if (ds == NULL) {
637 		(void) strcpy(name, "mos");
638 	} else {
639 		dsl_dir_name(ds->ds_dir, name);
640 		VERIFY0(dsl_dataset_get_snapname(ds));
641 		if (ds->ds_snapname[0]) {
642 			(void) strcat(name, "@");
643 			/*
644 			 * We use a "recursive" mutex so that we
645 			 * can call dprintf_ds() with ds_lock held.
646 			 */
647 			if (!MUTEX_HELD(&ds->ds_lock)) {
648 				mutex_enter(&ds->ds_lock);
649 				(void) strcat(name, ds->ds_snapname);
650 				mutex_exit(&ds->ds_lock);
651 			} else {
652 				(void) strcat(name, ds->ds_snapname);
653 			}
654 		}
655 	}
656 }
657 
658 void
659 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
660 {
661 	dmu_buf_rele(ds->ds_dbuf, tag);
662 }
663 
664 void
665 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
666 {
667 	ASSERT3P(ds->ds_owner, ==, tag);
668 	ASSERT(ds->ds_dbuf != NULL);
669 
670 	mutex_enter(&ds->ds_lock);
671 	ds->ds_owner = NULL;
672 	mutex_exit(&ds->ds_lock);
673 	dsl_dataset_long_rele(ds, tag);
674 	dsl_dataset_rele(ds, tag);
675 }
676 
677 boolean_t
678 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
679 {
680 	boolean_t gotit = FALSE;
681 
682 	mutex_enter(&ds->ds_lock);
683 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
684 		ds->ds_owner = tag;
685 		dsl_dataset_long_hold(ds, tag);
686 		gotit = TRUE;
687 	}
688 	mutex_exit(&ds->ds_lock);
689 	return (gotit);
690 }
691 
692 uint64_t
693 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
694     uint64_t flags, dmu_tx_t *tx)
695 {
696 	dsl_pool_t *dp = dd->dd_pool;
697 	dmu_buf_t *dbuf;
698 	dsl_dataset_phys_t *dsphys;
699 	uint64_t dsobj;
700 	objset_t *mos = dp->dp_meta_objset;
701 
702 	if (origin == NULL)
703 		origin = dp->dp_origin_snap;
704 
705 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
706 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
707 	ASSERT(dmu_tx_is_syncing(tx));
708 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
709 
710 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
711 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
712 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
713 	dmu_buf_will_dirty(dbuf, tx);
714 	dsphys = dbuf->db_data;
715 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
716 	dsphys->ds_dir_obj = dd->dd_object;
717 	dsphys->ds_flags = flags;
718 	dsphys->ds_fsid_guid = unique_create();
719 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
720 	    sizeof (dsphys->ds_guid));
721 	dsphys->ds_snapnames_zapobj =
722 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
723 	    DMU_OT_NONE, 0, tx);
724 	dsphys->ds_creation_time = gethrestime_sec();
725 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
726 
727 	if (origin == NULL) {
728 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
729 	} else {
730 		dsl_dataset_t *ohds; /* head of the origin snapshot */
731 
732 		dsphys->ds_prev_snap_obj = origin->ds_object;
733 		dsphys->ds_prev_snap_txg =
734 		    dsl_dataset_phys(origin)->ds_creation_txg;
735 		dsphys->ds_referenced_bytes =
736 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
737 		dsphys->ds_compressed_bytes =
738 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
739 		dsphys->ds_uncompressed_bytes =
740 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
741 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
742 
743 		/*
744 		 * Inherit flags that describe the dataset's contents
745 		 * (INCONSISTENT) or properties (Case Insensitive).
746 		 */
747 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
748 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
749 
750 		if (origin->ds_large_blocks)
751 			dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
752 
753 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
754 		dsl_dataset_phys(origin)->ds_num_children++;
755 
756 		VERIFY0(dsl_dataset_hold_obj(dp,
757 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
758 		    FTAG, &ohds));
759 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
760 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
761 		dsl_dataset_rele(ohds, FTAG);
762 
763 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
764 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
765 				dsl_dataset_phys(origin)->ds_next_clones_obj =
766 				    zap_create(mos,
767 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
768 			}
769 			VERIFY0(zap_add_int(mos,
770 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
771 			    dsobj, tx));
772 		}
773 
774 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
775 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
776 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
777 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
778 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
779 				dsl_dir_phys(origin->ds_dir)->dd_clones =
780 				    zap_create(mos,
781 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
782 			}
783 			VERIFY0(zap_add_int(mos,
784 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
785 			    dsobj, tx));
786 		}
787 	}
788 
789 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
790 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
791 
792 	dmu_buf_rele(dbuf, FTAG);
793 
794 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
795 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
796 
797 	return (dsobj);
798 }
799 
800 static void
801 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
802 {
803 	objset_t *os;
804 
805 	VERIFY0(dmu_objset_from_ds(ds, &os));
806 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
807 	dsl_dataset_dirty(ds, tx);
808 }
809 
810 uint64_t
811 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
812     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
813 {
814 	dsl_pool_t *dp = pdd->dd_pool;
815 	uint64_t dsobj, ddobj;
816 	dsl_dir_t *dd;
817 
818 	ASSERT(dmu_tx_is_syncing(tx));
819 	ASSERT(lastname[0] != '@');
820 
821 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
822 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
823 
824 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
825 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
826 
827 	dsl_deleg_set_create_perms(dd, tx, cr);
828 
829 	/*
830 	 * Since we're creating a new node we know it's a leaf, so we can
831 	 * initialize the counts if the limit feature is active.
832 	 */
833 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
834 		uint64_t cnt = 0;
835 		objset_t *os = dd->dd_pool->dp_meta_objset;
836 
837 		dsl_dir_zapify(dd, tx);
838 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
839 		    sizeof (cnt), 1, &cnt, tx));
840 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
841 		    sizeof (cnt), 1, &cnt, tx));
842 	}
843 
844 	dsl_dir_rele(dd, FTAG);
845 
846 	/*
847 	 * If we are creating a clone, make sure we zero out any stale
848 	 * data from the origin snapshots zil header.
849 	 */
850 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
851 		dsl_dataset_t *ds;
852 
853 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
854 		dsl_dataset_zero_zil(ds, tx);
855 		dsl_dataset_rele(ds, FTAG);
856 	}
857 
858 	return (dsobj);
859 }
860 
861 /*
862  * The unique space in the head dataset can be calculated by subtracting
863  * the space used in the most recent snapshot, that is still being used
864  * in this file system, from the space currently in use.  To figure out
865  * the space in the most recent snapshot still in use, we need to take
866  * the total space used in the snapshot and subtract out the space that
867  * has been freed up since the snapshot was taken.
868  */
869 void
870 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
871 {
872 	uint64_t mrs_used;
873 	uint64_t dlused, dlcomp, dluncomp;
874 
875 	ASSERT(!ds->ds_is_snapshot);
876 
877 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
878 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
879 	else
880 		mrs_used = 0;
881 
882 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
883 
884 	ASSERT3U(dlused, <=, mrs_used);
885 	dsl_dataset_phys(ds)->ds_unique_bytes =
886 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
887 
888 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
889 	    SPA_VERSION_UNIQUE_ACCURATE)
890 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
891 }
892 
893 void
894 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
895     dmu_tx_t *tx)
896 {
897 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
898 	uint64_t count;
899 	int err;
900 
901 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
902 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
903 	    obj, tx);
904 	/*
905 	 * The err should not be ENOENT, but a bug in a previous version
906 	 * of the code could cause upgrade_clones_cb() to not set
907 	 * ds_next_snap_obj when it should, leading to a missing entry.
908 	 * If we knew that the pool was created after
909 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
910 	 * ENOENT.  However, at least we can check that we don't have
911 	 * too many entries in the next_clones_obj even after failing to
912 	 * remove this one.
913 	 */
914 	if (err != ENOENT)
915 		VERIFY0(err);
916 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
917 	    &count));
918 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
919 }
920 
921 
922 blkptr_t *
923 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
924 {
925 	return (&dsl_dataset_phys(ds)->ds_bp);
926 }
927 
928 void
929 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
930 {
931 	ASSERT(dmu_tx_is_syncing(tx));
932 	/* If it's the meta-objset, set dp_meta_rootbp */
933 	if (ds == NULL) {
934 		tx->tx_pool->dp_meta_rootbp = *bp;
935 	} else {
936 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
937 		dsl_dataset_phys(ds)->ds_bp = *bp;
938 	}
939 }
940 
941 spa_t *
942 dsl_dataset_get_spa(dsl_dataset_t *ds)
943 {
944 	return (ds->ds_dir->dd_pool->dp_spa);
945 }
946 
947 void
948 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
949 {
950 	dsl_pool_t *dp;
951 
952 	if (ds == NULL) /* this is the meta-objset */
953 		return;
954 
955 	ASSERT(ds->ds_objset != NULL);
956 
957 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
958 		panic("dirtying snapshot!");
959 
960 	dp = ds->ds_dir->dd_pool;
961 
962 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
963 		/* up the hold count until we can be written out */
964 		dmu_buf_add_ref(ds->ds_dbuf, ds);
965 	}
966 }
967 
968 boolean_t
969 dsl_dataset_is_dirty(dsl_dataset_t *ds)
970 {
971 	for (int t = 0; t < TXG_SIZE; t++) {
972 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
973 		    ds, t))
974 			return (B_TRUE);
975 	}
976 	return (B_FALSE);
977 }
978 
979 static int
980 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
981 {
982 	uint64_t asize;
983 
984 	if (!dmu_tx_is_syncing(tx))
985 		return (0);
986 
987 	/*
988 	 * If there's an fs-only reservation, any blocks that might become
989 	 * owned by the snapshot dataset must be accommodated by space
990 	 * outside of the reservation.
991 	 */
992 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
993 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
994 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
995 		return (SET_ERROR(ENOSPC));
996 
997 	/*
998 	 * Propagate any reserved space for this snapshot to other
999 	 * snapshot checks in this sync group.
1000 	 */
1001 	if (asize > 0)
1002 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1003 
1004 	return (0);
1005 }
1006 
1007 typedef struct dsl_dataset_snapshot_arg {
1008 	nvlist_t *ddsa_snaps;
1009 	nvlist_t *ddsa_props;
1010 	nvlist_t *ddsa_errors;
1011 	cred_t *ddsa_cr;
1012 } dsl_dataset_snapshot_arg_t;
1013 
1014 int
1015 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1016     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1017 {
1018 	int error;
1019 	uint64_t value;
1020 
1021 	ds->ds_trysnap_txg = tx->tx_txg;
1022 
1023 	if (!dmu_tx_is_syncing(tx))
1024 		return (0);
1025 
1026 	/*
1027 	 * We don't allow multiple snapshots of the same txg.  If there
1028 	 * is already one, try again.
1029 	 */
1030 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1031 		return (SET_ERROR(EAGAIN));
1032 
1033 	/*
1034 	 * Check for conflicting snapshot name.
1035 	 */
1036 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1037 	if (error == 0)
1038 		return (SET_ERROR(EEXIST));
1039 	if (error != ENOENT)
1040 		return (error);
1041 
1042 	/*
1043 	 * We don't allow taking snapshots of inconsistent datasets, such as
1044 	 * those into which we are currently receiving.  However, if we are
1045 	 * creating this snapshot as part of a receive, this check will be
1046 	 * executed atomically with respect to the completion of the receive
1047 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1048 	 * case we ignore this, knowing it will be fixed up for us shortly in
1049 	 * dmu_recv_end_sync().
1050 	 */
1051 	if (!recv && DS_IS_INCONSISTENT(ds))
1052 		return (SET_ERROR(EBUSY));
1053 
1054 	/*
1055 	 * Skip the check for temporary snapshots or if we have already checked
1056 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1057 	 * check the count here when we're receiving a stream.
1058 	 */
1059 	if (cnt != 0 && cr != NULL) {
1060 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1061 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1062 		if (error != 0)
1063 			return (error);
1064 	}
1065 
1066 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1067 	if (error != 0)
1068 		return (error);
1069 
1070 	return (0);
1071 }
1072 
1073 static int
1074 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1075 {
1076 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1077 	dsl_pool_t *dp = dmu_tx_pool(tx);
1078 	nvpair_t *pair;
1079 	int rv = 0;
1080 
1081 	/*
1082 	 * Pre-compute how many total new snapshots will be created for each
1083 	 * level in the tree and below. This is needed for validating the
1084 	 * snapshot limit when either taking a recursive snapshot or when
1085 	 * taking multiple snapshots.
1086 	 *
1087 	 * The problem is that the counts are not actually adjusted when
1088 	 * we are checking, only when we finally sync. For a single snapshot,
1089 	 * this is easy, the count will increase by 1 at each node up the tree,
1090 	 * but its more complicated for the recursive/multiple snapshot case.
1091 	 *
1092 	 * The dsl_fs_ss_limit_check function does recursively check the count
1093 	 * at each level up the tree but since it is validating each snapshot
1094 	 * independently we need to be sure that we are validating the complete
1095 	 * count for the entire set of snapshots. We do this by rolling up the
1096 	 * counts for each component of the name into an nvlist and then
1097 	 * checking each of those cases with the aggregated count.
1098 	 *
1099 	 * This approach properly handles not only the recursive snapshot
1100 	 * case (where we get all of those on the ddsa_snaps list) but also
1101 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1102 	 * validate the limit on 'a' using a count of 2).
1103 	 *
1104 	 * We validate the snapshot names in the third loop and only report
1105 	 * name errors once.
1106 	 */
1107 	if (dmu_tx_is_syncing(tx)) {
1108 		nvlist_t *cnt_track = NULL;
1109 		cnt_track = fnvlist_alloc();
1110 
1111 		/* Rollup aggregated counts into the cnt_track list */
1112 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1113 		    pair != NULL;
1114 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1115 			char *pdelim;
1116 			uint64_t val;
1117 			char nm[MAXPATHLEN];
1118 
1119 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1120 			pdelim = strchr(nm, '@');
1121 			if (pdelim == NULL)
1122 				continue;
1123 			*pdelim = '\0';
1124 
1125 			do {
1126 				if (nvlist_lookup_uint64(cnt_track, nm,
1127 				    &val) == 0) {
1128 					/* update existing entry */
1129 					fnvlist_add_uint64(cnt_track, nm,
1130 					    val + 1);
1131 				} else {
1132 					/* add to list */
1133 					fnvlist_add_uint64(cnt_track, nm, 1);
1134 				}
1135 
1136 				pdelim = strrchr(nm, '/');
1137 				if (pdelim != NULL)
1138 					*pdelim = '\0';
1139 			} while (pdelim != NULL);
1140 		}
1141 
1142 		/* Check aggregated counts at each level */
1143 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1144 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1145 			int error = 0;
1146 			char *name;
1147 			uint64_t cnt = 0;
1148 			dsl_dataset_t *ds;
1149 
1150 			name = nvpair_name(pair);
1151 			cnt = fnvpair_value_uint64(pair);
1152 			ASSERT(cnt > 0);
1153 
1154 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1155 			if (error == 0) {
1156 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1157 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1158 				    ddsa->ddsa_cr);
1159 				dsl_dataset_rele(ds, FTAG);
1160 			}
1161 
1162 			if (error != 0) {
1163 				if (ddsa->ddsa_errors != NULL)
1164 					fnvlist_add_int32(ddsa->ddsa_errors,
1165 					    name, error);
1166 				rv = error;
1167 				/* only report one error for this check */
1168 				break;
1169 			}
1170 		}
1171 		nvlist_free(cnt_track);
1172 	}
1173 
1174 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1175 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1176 		int error = 0;
1177 		dsl_dataset_t *ds;
1178 		char *name, *atp;
1179 		char dsname[MAXNAMELEN];
1180 
1181 		name = nvpair_name(pair);
1182 		if (strlen(name) >= MAXNAMELEN)
1183 			error = SET_ERROR(ENAMETOOLONG);
1184 		if (error == 0) {
1185 			atp = strchr(name, '@');
1186 			if (atp == NULL)
1187 				error = SET_ERROR(EINVAL);
1188 			if (error == 0)
1189 				(void) strlcpy(dsname, name, atp - name + 1);
1190 		}
1191 		if (error == 0)
1192 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1193 		if (error == 0) {
1194 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1195 			error = dsl_dataset_snapshot_check_impl(ds,
1196 			    atp + 1, tx, B_FALSE, 0, NULL);
1197 			dsl_dataset_rele(ds, FTAG);
1198 		}
1199 
1200 		if (error != 0) {
1201 			if (ddsa->ddsa_errors != NULL) {
1202 				fnvlist_add_int32(ddsa->ddsa_errors,
1203 				    name, error);
1204 			}
1205 			rv = error;
1206 		}
1207 	}
1208 
1209 	return (rv);
1210 }
1211 
1212 void
1213 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1214     dmu_tx_t *tx)
1215 {
1216 	static zil_header_t zero_zil;
1217 
1218 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1219 	dmu_buf_t *dbuf;
1220 	dsl_dataset_phys_t *dsphys;
1221 	uint64_t dsobj, crtxg;
1222 	objset_t *mos = dp->dp_meta_objset;
1223 	objset_t *os;
1224 
1225 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1226 
1227 	/*
1228 	 * If we are on an old pool, the zil must not be active, in which
1229 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1230 	 */
1231 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1232 	    dmu_objset_from_ds(ds, &os) != 0 ||
1233 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1234 	    sizeof (zero_zil)) == 0);
1235 
1236 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1237 
1238 	/*
1239 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1240 	 */
1241 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1242 		crtxg = 1;
1243 	else
1244 		crtxg = tx->tx_txg;
1245 
1246 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1247 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1248 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1249 	dmu_buf_will_dirty(dbuf, tx);
1250 	dsphys = dbuf->db_data;
1251 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1252 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1253 	dsphys->ds_fsid_guid = unique_create();
1254 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1255 	    sizeof (dsphys->ds_guid));
1256 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1257 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1258 	dsphys->ds_next_snap_obj = ds->ds_object;
1259 	dsphys->ds_num_children = 1;
1260 	dsphys->ds_creation_time = gethrestime_sec();
1261 	dsphys->ds_creation_txg = crtxg;
1262 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1263 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1264 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1265 	dsphys->ds_uncompressed_bytes =
1266 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1267 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1268 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1269 	dmu_buf_rele(dbuf, FTAG);
1270 
1271 	if (ds->ds_large_blocks)
1272 		dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1273 
1274 	ASSERT3U(ds->ds_prev != 0, ==,
1275 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1276 	if (ds->ds_prev) {
1277 		uint64_t next_clones_obj =
1278 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1279 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1280 		    ds->ds_object ||
1281 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1282 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1283 		    ds->ds_object) {
1284 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1285 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1286 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1287 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1288 		} else if (next_clones_obj != 0) {
1289 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1290 			    dsphys->ds_next_snap_obj, tx);
1291 			VERIFY0(zap_add_int(mos,
1292 			    next_clones_obj, dsobj, tx));
1293 		}
1294 	}
1295 
1296 	/*
1297 	 * If we have a reference-reservation on this dataset, we will
1298 	 * need to increase the amount of refreservation being charged
1299 	 * since our unique space is going to zero.
1300 	 */
1301 	if (ds->ds_reserved) {
1302 		int64_t delta;
1303 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1304 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1305 		    ds->ds_reserved);
1306 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1307 		    delta, 0, 0, tx);
1308 	}
1309 
1310 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1311 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1312 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1313 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1314 	dsl_deadlist_close(&ds->ds_deadlist);
1315 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1316 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1317 	dsl_deadlist_add_key(&ds->ds_deadlist,
1318 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1319 
1320 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1321 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1322 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1323 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1324 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1325 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1326 
1327 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1328 	    snapname, 8, 1, &dsobj, tx));
1329 
1330 	if (ds->ds_prev)
1331 		dsl_dataset_rele(ds->ds_prev, ds);
1332 	VERIFY0(dsl_dataset_hold_obj(dp,
1333 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1334 
1335 	dsl_scan_ds_snapshotted(ds, tx);
1336 
1337 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1338 
1339 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1340 }
1341 
1342 static void
1343 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1344 {
1345 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1346 	dsl_pool_t *dp = dmu_tx_pool(tx);
1347 	nvpair_t *pair;
1348 
1349 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1350 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1351 		dsl_dataset_t *ds;
1352 		char *name, *atp;
1353 		char dsname[MAXNAMELEN];
1354 
1355 		name = nvpair_name(pair);
1356 		atp = strchr(name, '@');
1357 		(void) strlcpy(dsname, name, atp - name + 1);
1358 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1359 
1360 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1361 		if (ddsa->ddsa_props != NULL) {
1362 			dsl_props_set_sync_impl(ds->ds_prev,
1363 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1364 		}
1365 		dsl_dataset_rele(ds, FTAG);
1366 	}
1367 }
1368 
1369 /*
1370  * The snapshots must all be in the same pool.
1371  * All-or-nothing: if there are any failures, nothing will be modified.
1372  */
1373 int
1374 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1375 {
1376 	dsl_dataset_snapshot_arg_t ddsa;
1377 	nvpair_t *pair;
1378 	boolean_t needsuspend;
1379 	int error;
1380 	spa_t *spa;
1381 	char *firstname;
1382 	nvlist_t *suspended = NULL;
1383 
1384 	pair = nvlist_next_nvpair(snaps, NULL);
1385 	if (pair == NULL)
1386 		return (0);
1387 	firstname = nvpair_name(pair);
1388 
1389 	error = spa_open(firstname, &spa, FTAG);
1390 	if (error != 0)
1391 		return (error);
1392 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1393 	spa_close(spa, FTAG);
1394 
1395 	if (needsuspend) {
1396 		suspended = fnvlist_alloc();
1397 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1398 		    pair = nvlist_next_nvpair(snaps, pair)) {
1399 			char fsname[MAXNAMELEN];
1400 			char *snapname = nvpair_name(pair);
1401 			char *atp;
1402 			void *cookie;
1403 
1404 			atp = strchr(snapname, '@');
1405 			if (atp == NULL) {
1406 				error = SET_ERROR(EINVAL);
1407 				break;
1408 			}
1409 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1410 
1411 			error = zil_suspend(fsname, &cookie);
1412 			if (error != 0)
1413 				break;
1414 			fnvlist_add_uint64(suspended, fsname,
1415 			    (uintptr_t)cookie);
1416 		}
1417 	}
1418 
1419 	ddsa.ddsa_snaps = snaps;
1420 	ddsa.ddsa_props = props;
1421 	ddsa.ddsa_errors = errors;
1422 	ddsa.ddsa_cr = CRED();
1423 
1424 	if (error == 0) {
1425 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1426 		    dsl_dataset_snapshot_sync, &ddsa,
1427 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1428 	}
1429 
1430 	if (suspended != NULL) {
1431 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1432 		    pair = nvlist_next_nvpair(suspended, pair)) {
1433 			zil_resume((void *)(uintptr_t)
1434 			    fnvpair_value_uint64(pair));
1435 		}
1436 		fnvlist_free(suspended);
1437 	}
1438 
1439 	return (error);
1440 }
1441 
1442 typedef struct dsl_dataset_snapshot_tmp_arg {
1443 	const char *ddsta_fsname;
1444 	const char *ddsta_snapname;
1445 	minor_t ddsta_cleanup_minor;
1446 	const char *ddsta_htag;
1447 } dsl_dataset_snapshot_tmp_arg_t;
1448 
1449 static int
1450 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1451 {
1452 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1453 	dsl_pool_t *dp = dmu_tx_pool(tx);
1454 	dsl_dataset_t *ds;
1455 	int error;
1456 
1457 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1458 	if (error != 0)
1459 		return (error);
1460 
1461 	/* NULL cred means no limit check for tmp snapshot */
1462 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1463 	    tx, B_FALSE, 0, NULL);
1464 	if (error != 0) {
1465 		dsl_dataset_rele(ds, FTAG);
1466 		return (error);
1467 	}
1468 
1469 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1470 		dsl_dataset_rele(ds, FTAG);
1471 		return (SET_ERROR(ENOTSUP));
1472 	}
1473 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1474 	    B_TRUE, tx);
1475 	if (error != 0) {
1476 		dsl_dataset_rele(ds, FTAG);
1477 		return (error);
1478 	}
1479 
1480 	dsl_dataset_rele(ds, FTAG);
1481 	return (0);
1482 }
1483 
1484 static void
1485 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1486 {
1487 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1488 	dsl_pool_t *dp = dmu_tx_pool(tx);
1489 	dsl_dataset_t *ds;
1490 
1491 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1492 
1493 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1494 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1495 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1496 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1497 
1498 	dsl_dataset_rele(ds, FTAG);
1499 }
1500 
1501 int
1502 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1503     minor_t cleanup_minor, const char *htag)
1504 {
1505 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1506 	int error;
1507 	spa_t *spa;
1508 	boolean_t needsuspend;
1509 	void *cookie;
1510 
1511 	ddsta.ddsta_fsname = fsname;
1512 	ddsta.ddsta_snapname = snapname;
1513 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1514 	ddsta.ddsta_htag = htag;
1515 
1516 	error = spa_open(fsname, &spa, FTAG);
1517 	if (error != 0)
1518 		return (error);
1519 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1520 	spa_close(spa, FTAG);
1521 
1522 	if (needsuspend) {
1523 		error = zil_suspend(fsname, &cookie);
1524 		if (error != 0)
1525 			return (error);
1526 	}
1527 
1528 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1529 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1530 
1531 	if (needsuspend)
1532 		zil_resume(cookie);
1533 	return (error);
1534 }
1535 
1536 
1537 void
1538 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1539 {
1540 	ASSERT(dmu_tx_is_syncing(tx));
1541 	ASSERT(ds->ds_objset != NULL);
1542 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1543 
1544 	/*
1545 	 * in case we had to change ds_fsid_guid when we opened it,
1546 	 * sync it out now.
1547 	 */
1548 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1549 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1550 
1551 	dmu_objset_sync(ds->ds_objset, zio, tx);
1552 
1553 	if (ds->ds_need_large_blocks && !ds->ds_large_blocks) {
1554 		dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
1555 		ds->ds_large_blocks = B_TRUE;
1556 	}
1557 }
1558 
1559 static void
1560 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1561 {
1562 	uint64_t count = 0;
1563 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1564 	zap_cursor_t zc;
1565 	zap_attribute_t za;
1566 	nvlist_t *propval = fnvlist_alloc();
1567 	nvlist_t *val = fnvlist_alloc();
1568 
1569 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1570 
1571 	/*
1572 	 * There may be missing entries in ds_next_clones_obj
1573 	 * due to a bug in a previous version of the code.
1574 	 * Only trust it if it has the right number of entries.
1575 	 */
1576 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1577 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1578 		    &count));
1579 	}
1580 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1581 		goto fail;
1582 	for (zap_cursor_init(&zc, mos,
1583 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1584 	    zap_cursor_retrieve(&zc, &za) == 0;
1585 	    zap_cursor_advance(&zc)) {
1586 		dsl_dataset_t *clone;
1587 		char buf[ZFS_MAXNAMELEN];
1588 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1589 		    za.za_first_integer, FTAG, &clone));
1590 		dsl_dir_name(clone->ds_dir, buf);
1591 		fnvlist_add_boolean(val, buf);
1592 		dsl_dataset_rele(clone, FTAG);
1593 	}
1594 	zap_cursor_fini(&zc);
1595 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1596 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1597 fail:
1598 	nvlist_free(val);
1599 	nvlist_free(propval);
1600 }
1601 
1602 void
1603 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1604 {
1605 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1606 	uint64_t refd, avail, uobjs, aobjs, ratio;
1607 
1608 	ASSERT(dsl_pool_config_held(dp));
1609 
1610 	ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1611 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1612 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1613 
1614 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1615 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1616 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1617 
1618 	if (ds->ds_is_snapshot) {
1619 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1620 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1621 		    dsl_dataset_phys(ds)->ds_unique_bytes);
1622 		get_clones_stat(ds, nv);
1623 	} else {
1624 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1625 			char buf[MAXNAMELEN];
1626 			dsl_dataset_name(ds->ds_prev, buf);
1627 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1628 		}
1629 
1630 		dsl_dir_stats(ds->ds_dir, nv);
1631 	}
1632 
1633 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1634 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1635 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1636 
1637 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1638 	    dsl_dataset_phys(ds)->ds_creation_time);
1639 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1640 	    dsl_dataset_phys(ds)->ds_creation_txg);
1641 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1642 	    ds->ds_quota);
1643 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1644 	    ds->ds_reserved);
1645 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1646 	    dsl_dataset_phys(ds)->ds_guid);
1647 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1648 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1649 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1650 	    ds->ds_object);
1651 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1652 	    ds->ds_userrefs);
1653 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1654 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1655 
1656 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1657 		uint64_t written, comp, uncomp;
1658 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1659 		dsl_dataset_t *prev;
1660 
1661 		int err = dsl_dataset_hold_obj(dp,
1662 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1663 		if (err == 0) {
1664 			err = dsl_dataset_space_written(prev, ds, &written,
1665 			    &comp, &uncomp);
1666 			dsl_dataset_rele(prev, FTAG);
1667 			if (err == 0) {
1668 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1669 				    written);
1670 			}
1671 		}
1672 	}
1673 }
1674 
1675 void
1676 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1677 {
1678 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1679 	ASSERT(dsl_pool_config_held(dp));
1680 
1681 	stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1682 	stat->dds_inconsistent =
1683 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1684 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1685 	stat->dds_origin[0] = '\0';
1686 	if (ds->ds_is_snapshot) {
1687 		stat->dds_is_snapshot = B_TRUE;
1688 		stat->dds_num_clones =
1689 		    dsl_dataset_phys(ds)->ds_num_children - 1;
1690 	} else {
1691 		stat->dds_is_snapshot = B_FALSE;
1692 		stat->dds_num_clones = 0;
1693 
1694 		if (dsl_dir_is_clone(ds->ds_dir)) {
1695 			dsl_dataset_t *ods;
1696 
1697 			VERIFY0(dsl_dataset_hold_obj(dp,
1698 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1699 			    FTAG, &ods));
1700 			dsl_dataset_name(ods, stat->dds_origin);
1701 			dsl_dataset_rele(ods, FTAG);
1702 		}
1703 	}
1704 }
1705 
1706 uint64_t
1707 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1708 {
1709 	return (ds->ds_fsid_guid);
1710 }
1711 
1712 void
1713 dsl_dataset_space(dsl_dataset_t *ds,
1714     uint64_t *refdbytesp, uint64_t *availbytesp,
1715     uint64_t *usedobjsp, uint64_t *availobjsp)
1716 {
1717 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1718 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1719 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1720 		*availbytesp +=
1721 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1722 	if (ds->ds_quota != 0) {
1723 		/*
1724 		 * Adjust available bytes according to refquota
1725 		 */
1726 		if (*refdbytesp < ds->ds_quota)
1727 			*availbytesp = MIN(*availbytesp,
1728 			    ds->ds_quota - *refdbytesp);
1729 		else
1730 			*availbytesp = 0;
1731 	}
1732 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1733 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1734 }
1735 
1736 boolean_t
1737 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1738 {
1739 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1740 
1741 	ASSERT(dsl_pool_config_held(dp));
1742 	if (snap == NULL)
1743 		return (B_FALSE);
1744 	if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1745 	    dsl_dataset_phys(snap)->ds_creation_txg) {
1746 		objset_t *os, *os_snap;
1747 		/*
1748 		 * It may be that only the ZIL differs, because it was
1749 		 * reset in the head.  Don't count that as being
1750 		 * modified.
1751 		 */
1752 		if (dmu_objset_from_ds(ds, &os) != 0)
1753 			return (B_TRUE);
1754 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1755 			return (B_TRUE);
1756 		return (bcmp(&os->os_phys->os_meta_dnode,
1757 		    &os_snap->os_phys->os_meta_dnode,
1758 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1759 	}
1760 	return (B_FALSE);
1761 }
1762 
1763 typedef struct dsl_dataset_rename_snapshot_arg {
1764 	const char *ddrsa_fsname;
1765 	const char *ddrsa_oldsnapname;
1766 	const char *ddrsa_newsnapname;
1767 	boolean_t ddrsa_recursive;
1768 	dmu_tx_t *ddrsa_tx;
1769 } dsl_dataset_rename_snapshot_arg_t;
1770 
1771 /* ARGSUSED */
1772 static int
1773 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1774     dsl_dataset_t *hds, void *arg)
1775 {
1776 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1777 	int error;
1778 	uint64_t val;
1779 
1780 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1781 	if (error != 0) {
1782 		/* ignore nonexistent snapshots */
1783 		return (error == ENOENT ? 0 : error);
1784 	}
1785 
1786 	/* new name should not exist */
1787 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1788 	if (error == 0)
1789 		error = SET_ERROR(EEXIST);
1790 	else if (error == ENOENT)
1791 		error = 0;
1792 
1793 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1794 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1795 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1796 		error = SET_ERROR(ENAMETOOLONG);
1797 
1798 	return (error);
1799 }
1800 
1801 static int
1802 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1803 {
1804 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1805 	dsl_pool_t *dp = dmu_tx_pool(tx);
1806 	dsl_dataset_t *hds;
1807 	int error;
1808 
1809 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1810 	if (error != 0)
1811 		return (error);
1812 
1813 	if (ddrsa->ddrsa_recursive) {
1814 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1815 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1816 		    DS_FIND_CHILDREN);
1817 	} else {
1818 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1819 	}
1820 	dsl_dataset_rele(hds, FTAG);
1821 	return (error);
1822 }
1823 
1824 static int
1825 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1826     dsl_dataset_t *hds, void *arg)
1827 {
1828 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1829 	dsl_dataset_t *ds;
1830 	uint64_t val;
1831 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1832 	int error;
1833 
1834 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1835 	ASSERT(error == 0 || error == ENOENT);
1836 	if (error == ENOENT) {
1837 		/* ignore nonexistent snapshots */
1838 		return (0);
1839 	}
1840 
1841 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1842 
1843 	/* log before we change the name */
1844 	spa_history_log_internal_ds(ds, "rename", tx,
1845 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1846 
1847 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1848 	    B_FALSE));
1849 	mutex_enter(&ds->ds_lock);
1850 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1851 	mutex_exit(&ds->ds_lock);
1852 	VERIFY0(zap_add(dp->dp_meta_objset,
1853 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
1854 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1855 
1856 	dsl_dataset_rele(ds, FTAG);
1857 	return (0);
1858 }
1859 
1860 static void
1861 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1862 {
1863 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1864 	dsl_pool_t *dp = dmu_tx_pool(tx);
1865 	dsl_dataset_t *hds;
1866 
1867 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1868 	ddrsa->ddrsa_tx = tx;
1869 	if (ddrsa->ddrsa_recursive) {
1870 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1871 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1872 		    DS_FIND_CHILDREN));
1873 	} else {
1874 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1875 	}
1876 	dsl_dataset_rele(hds, FTAG);
1877 }
1878 
1879 int
1880 dsl_dataset_rename_snapshot(const char *fsname,
1881     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1882 {
1883 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1884 
1885 	ddrsa.ddrsa_fsname = fsname;
1886 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1887 	ddrsa.ddrsa_newsnapname = newsnapname;
1888 	ddrsa.ddrsa_recursive = recursive;
1889 
1890 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1891 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
1892 	    1, ZFS_SPACE_CHECK_RESERVED));
1893 }
1894 
1895 /*
1896  * If we're doing an ownership handoff, we need to make sure that there is
1897  * only one long hold on the dataset.  We're not allowed to change anything here
1898  * so we don't permanently release the long hold or regular hold here.  We want
1899  * to do this only when syncing to avoid the dataset unexpectedly going away
1900  * when we release the long hold.
1901  */
1902 static int
1903 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1904 {
1905 	boolean_t held;
1906 
1907 	if (!dmu_tx_is_syncing(tx))
1908 		return (0);
1909 
1910 	if (owner != NULL) {
1911 		VERIFY3P(ds->ds_owner, ==, owner);
1912 		dsl_dataset_long_rele(ds, owner);
1913 	}
1914 
1915 	held = dsl_dataset_long_held(ds);
1916 
1917 	if (owner != NULL)
1918 		dsl_dataset_long_hold(ds, owner);
1919 
1920 	if (held)
1921 		return (SET_ERROR(EBUSY));
1922 
1923 	return (0);
1924 }
1925 
1926 typedef struct dsl_dataset_rollback_arg {
1927 	const char *ddra_fsname;
1928 	void *ddra_owner;
1929 	nvlist_t *ddra_result;
1930 } dsl_dataset_rollback_arg_t;
1931 
1932 static int
1933 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1934 {
1935 	dsl_dataset_rollback_arg_t *ddra = arg;
1936 	dsl_pool_t *dp = dmu_tx_pool(tx);
1937 	dsl_dataset_t *ds;
1938 	int64_t unused_refres_delta;
1939 	int error;
1940 
1941 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1942 	if (error != 0)
1943 		return (error);
1944 
1945 	/* must not be a snapshot */
1946 	if (ds->ds_is_snapshot) {
1947 		dsl_dataset_rele(ds, FTAG);
1948 		return (SET_ERROR(EINVAL));
1949 	}
1950 
1951 	/* must have a most recent snapshot */
1952 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
1953 		dsl_dataset_rele(ds, FTAG);
1954 		return (SET_ERROR(EINVAL));
1955 	}
1956 
1957 	/* must not have any bookmarks after the most recent snapshot */
1958 	nvlist_t *proprequest = fnvlist_alloc();
1959 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
1960 	nvlist_t *bookmarks = fnvlist_alloc();
1961 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
1962 	fnvlist_free(proprequest);
1963 	if (error != 0)
1964 		return (error);
1965 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
1966 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
1967 		nvlist_t *valuenv =
1968 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
1969 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
1970 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
1971 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
1972 			fnvlist_free(bookmarks);
1973 			dsl_dataset_rele(ds, FTAG);
1974 			return (SET_ERROR(EEXIST));
1975 		}
1976 	}
1977 	fnvlist_free(bookmarks);
1978 
1979 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
1980 	if (error != 0) {
1981 		dsl_dataset_rele(ds, FTAG);
1982 		return (error);
1983 	}
1984 
1985 	/*
1986 	 * Check if the snap we are rolling back to uses more than
1987 	 * the refquota.
1988 	 */
1989 	if (ds->ds_quota != 0 &&
1990 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
1991 		dsl_dataset_rele(ds, FTAG);
1992 		return (SET_ERROR(EDQUOT));
1993 	}
1994 
1995 	/*
1996 	 * When we do the clone swap, we will temporarily use more space
1997 	 * due to the refreservation (the head will no longer have any
1998 	 * unique space, so the entire amount of the refreservation will need
1999 	 * to be free).  We will immediately destroy the clone, freeing
2000 	 * this space, but the freeing happens over many txg's.
2001 	 */
2002 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2003 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2004 
2005 	if (unused_refres_delta > 0 &&
2006 	    unused_refres_delta >
2007 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2008 		dsl_dataset_rele(ds, FTAG);
2009 		return (SET_ERROR(ENOSPC));
2010 	}
2011 
2012 	dsl_dataset_rele(ds, FTAG);
2013 	return (0);
2014 }
2015 
2016 static void
2017 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2018 {
2019 	dsl_dataset_rollback_arg_t *ddra = arg;
2020 	dsl_pool_t *dp = dmu_tx_pool(tx);
2021 	dsl_dataset_t *ds, *clone;
2022 	uint64_t cloneobj;
2023 	char namebuf[ZFS_MAXNAMELEN];
2024 
2025 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2026 
2027 	dsl_dataset_name(ds->ds_prev, namebuf);
2028 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2029 
2030 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2031 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2032 
2033 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2034 
2035 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2036 	dsl_dataset_zero_zil(ds, tx);
2037 
2038 	dsl_destroy_head_sync_impl(clone, tx);
2039 
2040 	dsl_dataset_rele(clone, FTAG);
2041 	dsl_dataset_rele(ds, FTAG);
2042 }
2043 
2044 /*
2045  * Rolls back the given filesystem or volume to the most recent snapshot.
2046  * The name of the most recent snapshot will be returned under key "target"
2047  * in the result nvlist.
2048  *
2049  * If owner != NULL:
2050  * - The existing dataset MUST be owned by the specified owner at entry
2051  * - Upon return, dataset will still be held by the same owner, whether we
2052  *   succeed or not.
2053  *
2054  * This mode is required any time the existing filesystem is mounted.  See
2055  * notes above zfs_suspend_fs() for further details.
2056  */
2057 int
2058 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2059 {
2060 	dsl_dataset_rollback_arg_t ddra;
2061 
2062 	ddra.ddra_fsname = fsname;
2063 	ddra.ddra_owner = owner;
2064 	ddra.ddra_result = result;
2065 
2066 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2067 	    dsl_dataset_rollback_sync, &ddra,
2068 	    1, ZFS_SPACE_CHECK_RESERVED));
2069 }
2070 
2071 struct promotenode {
2072 	list_node_t link;
2073 	dsl_dataset_t *ds;
2074 };
2075 
2076 typedef struct dsl_dataset_promote_arg {
2077 	const char *ddpa_clonename;
2078 	dsl_dataset_t *ddpa_clone;
2079 	list_t shared_snaps, origin_snaps, clone_snaps;
2080 	dsl_dataset_t *origin_origin; /* origin of the origin */
2081 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2082 	char *err_ds;
2083 	cred_t *cr;
2084 } dsl_dataset_promote_arg_t;
2085 
2086 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2087 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2088     void *tag);
2089 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2090 
2091 static int
2092 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2093 {
2094 	dsl_dataset_promote_arg_t *ddpa = arg;
2095 	dsl_pool_t *dp = dmu_tx_pool(tx);
2096 	dsl_dataset_t *hds;
2097 	struct promotenode *snap;
2098 	dsl_dataset_t *origin_ds;
2099 	int err;
2100 	uint64_t unused;
2101 	uint64_t ss_mv_cnt;
2102 	size_t max_snap_len;
2103 
2104 	err = promote_hold(ddpa, dp, FTAG);
2105 	if (err != 0)
2106 		return (err);
2107 
2108 	hds = ddpa->ddpa_clone;
2109 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2110 
2111 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2112 		promote_rele(ddpa, FTAG);
2113 		return (SET_ERROR(EXDEV));
2114 	}
2115 
2116 	/*
2117 	 * Compute and check the amount of space to transfer.  Since this is
2118 	 * so expensive, don't do the preliminary check.
2119 	 */
2120 	if (!dmu_tx_is_syncing(tx)) {
2121 		promote_rele(ddpa, FTAG);
2122 		return (0);
2123 	}
2124 
2125 	snap = list_head(&ddpa->shared_snaps);
2126 	origin_ds = snap->ds;
2127 
2128 	/* compute origin's new unique space */
2129 	snap = list_tail(&ddpa->clone_snaps);
2130 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2131 	    origin_ds->ds_object);
2132 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2133 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2134 	    &ddpa->unique, &unused, &unused);
2135 
2136 	/*
2137 	 * Walk the snapshots that we are moving
2138 	 *
2139 	 * Compute space to transfer.  Consider the incremental changes
2140 	 * to used by each snapshot:
2141 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2142 	 * So each snapshot gave birth to:
2143 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2144 	 * So a sequence would look like:
2145 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2146 	 * Which simplifies to:
2147 	 * uN + kN + kN-1 + ... + k1 + k0
2148 	 * Note however, if we stop before we reach the ORIGIN we get:
2149 	 * uN + kN + kN-1 + ... + kM - uM-1
2150 	 */
2151 	ss_mv_cnt = 0;
2152 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2153 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2154 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2155 	for (snap = list_head(&ddpa->shared_snaps); snap;
2156 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2157 		uint64_t val, dlused, dlcomp, dluncomp;
2158 		dsl_dataset_t *ds = snap->ds;
2159 
2160 		ss_mv_cnt++;
2161 
2162 		/*
2163 		 * If there are long holds, we won't be able to evict
2164 		 * the objset.
2165 		 */
2166 		if (dsl_dataset_long_held(ds)) {
2167 			err = SET_ERROR(EBUSY);
2168 			goto out;
2169 		}
2170 
2171 		/* Check that the snapshot name does not conflict */
2172 		VERIFY0(dsl_dataset_get_snapname(ds));
2173 		if (strlen(ds->ds_snapname) >= max_snap_len) {
2174 			err = SET_ERROR(ENAMETOOLONG);
2175 			goto out;
2176 		}
2177 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2178 		if (err == 0) {
2179 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2180 			err = SET_ERROR(EEXIST);
2181 			goto out;
2182 		}
2183 		if (err != ENOENT)
2184 			goto out;
2185 
2186 		/* The very first snapshot does not have a deadlist */
2187 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2188 			continue;
2189 
2190 		dsl_deadlist_space(&ds->ds_deadlist,
2191 		    &dlused, &dlcomp, &dluncomp);
2192 		ddpa->used += dlused;
2193 		ddpa->comp += dlcomp;
2194 		ddpa->uncomp += dluncomp;
2195 	}
2196 
2197 	/*
2198 	 * If we are a clone of a clone then we never reached ORIGIN,
2199 	 * so we need to subtract out the clone origin's used space.
2200 	 */
2201 	if (ddpa->origin_origin) {
2202 		ddpa->used -=
2203 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2204 		ddpa->comp -=
2205 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2206 		ddpa->uncomp -=
2207 		    dsl_dataset_phys(ddpa->origin_origin)->
2208 		    ds_uncompressed_bytes;
2209 	}
2210 
2211 	/* Check that there is enough space and limit headroom here */
2212 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2213 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2214 	if (err != 0)
2215 		goto out;
2216 
2217 	/*
2218 	 * Compute the amounts of space that will be used by snapshots
2219 	 * after the promotion (for both origin and clone).  For each,
2220 	 * it is the amount of space that will be on all of their
2221 	 * deadlists (that was not born before their new origin).
2222 	 */
2223 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2224 		uint64_t space;
2225 
2226 		/*
2227 		 * Note, typically this will not be a clone of a clone,
2228 		 * so dd_origin_txg will be < TXG_INITIAL, so
2229 		 * these snaplist_space() -> dsl_deadlist_space_range()
2230 		 * calls will be fast because they do not have to
2231 		 * iterate over all bps.
2232 		 */
2233 		snap = list_head(&ddpa->origin_snaps);
2234 		err = snaplist_space(&ddpa->shared_snaps,
2235 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2236 		if (err != 0)
2237 			goto out;
2238 
2239 		err = snaplist_space(&ddpa->clone_snaps,
2240 		    snap->ds->ds_dir->dd_origin_txg, &space);
2241 		if (err != 0)
2242 			goto out;
2243 		ddpa->cloneusedsnap += space;
2244 	}
2245 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2246 	    DD_FLAG_USED_BREAKDOWN) {
2247 		err = snaplist_space(&ddpa->origin_snaps,
2248 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2249 		    &ddpa->originusedsnap);
2250 		if (err != 0)
2251 			goto out;
2252 	}
2253 
2254 out:
2255 	promote_rele(ddpa, FTAG);
2256 	return (err);
2257 }
2258 
2259 static void
2260 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2261 {
2262 	dsl_dataset_promote_arg_t *ddpa = arg;
2263 	dsl_pool_t *dp = dmu_tx_pool(tx);
2264 	dsl_dataset_t *hds;
2265 	struct promotenode *snap;
2266 	dsl_dataset_t *origin_ds;
2267 	dsl_dataset_t *origin_head;
2268 	dsl_dir_t *dd;
2269 	dsl_dir_t *odd = NULL;
2270 	uint64_t oldnext_obj;
2271 	int64_t delta;
2272 
2273 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2274 	hds = ddpa->ddpa_clone;
2275 
2276 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2277 
2278 	snap = list_head(&ddpa->shared_snaps);
2279 	origin_ds = snap->ds;
2280 	dd = hds->ds_dir;
2281 
2282 	snap = list_head(&ddpa->origin_snaps);
2283 	origin_head = snap->ds;
2284 
2285 	/*
2286 	 * We need to explicitly open odd, since origin_ds's dd will be
2287 	 * changing.
2288 	 */
2289 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2290 	    NULL, FTAG, &odd));
2291 
2292 	/* change origin's next snap */
2293 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2294 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2295 	snap = list_tail(&ddpa->clone_snaps);
2296 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2297 	    origin_ds->ds_object);
2298 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2299 
2300 	/* change the origin's next clone */
2301 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2302 		dsl_dataset_remove_from_next_clones(origin_ds,
2303 		    snap->ds->ds_object, tx);
2304 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2305 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2306 		    oldnext_obj, tx));
2307 	}
2308 
2309 	/* change origin */
2310 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2311 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2312 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2313 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2314 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2315 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2316 	origin_head->ds_dir->dd_origin_txg =
2317 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2318 
2319 	/* change dd_clone entries */
2320 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2321 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2322 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2323 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2324 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2325 		    hds->ds_object, tx));
2326 
2327 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2328 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2329 		    origin_head->ds_object, tx));
2330 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2331 			dsl_dir_phys(dd)->dd_clones =
2332 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2333 			    DMU_OT_NONE, 0, tx);
2334 		}
2335 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2336 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2337 	}
2338 
2339 	/* move snapshots to this dir */
2340 	for (snap = list_head(&ddpa->shared_snaps); snap;
2341 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2342 		dsl_dataset_t *ds = snap->ds;
2343 
2344 		/*
2345 		 * Property callbacks are registered to a particular
2346 		 * dsl_dir.  Since ours is changing, evict the objset
2347 		 * so that they will be unregistered from the old dsl_dir.
2348 		 */
2349 		if (ds->ds_objset) {
2350 			dmu_objset_evict(ds->ds_objset);
2351 			ds->ds_objset = NULL;
2352 		}
2353 
2354 		/* move snap name entry */
2355 		VERIFY0(dsl_dataset_get_snapname(ds));
2356 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2357 		    ds->ds_snapname, tx, B_TRUE));
2358 		VERIFY0(zap_add(dp->dp_meta_objset,
2359 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2360 		    8, 1, &ds->ds_object, tx));
2361 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2362 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2363 
2364 		/* change containing dsl_dir */
2365 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2366 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2367 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2368 		ASSERT3P(ds->ds_dir, ==, odd);
2369 		dsl_dir_rele(ds->ds_dir, ds);
2370 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2371 		    NULL, ds, &ds->ds_dir));
2372 
2373 		/* move any clone references */
2374 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2375 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2376 			zap_cursor_t zc;
2377 			zap_attribute_t za;
2378 
2379 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2380 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
2381 			    zap_cursor_retrieve(&zc, &za) == 0;
2382 			    zap_cursor_advance(&zc)) {
2383 				dsl_dataset_t *cnds;
2384 				uint64_t o;
2385 
2386 				if (za.za_first_integer == oldnext_obj) {
2387 					/*
2388 					 * We've already moved the
2389 					 * origin's reference.
2390 					 */
2391 					continue;
2392 				}
2393 
2394 				VERIFY0(dsl_dataset_hold_obj(dp,
2395 				    za.za_first_integer, FTAG, &cnds));
2396 				o = dsl_dir_phys(cnds->ds_dir)->
2397 				    dd_head_dataset_obj;
2398 
2399 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2400 				    dsl_dir_phys(odd)->dd_clones, o, tx));
2401 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2402 				    dsl_dir_phys(dd)->dd_clones, o, tx));
2403 				dsl_dataset_rele(cnds, FTAG);
2404 			}
2405 			zap_cursor_fini(&zc);
2406 		}
2407 
2408 		ASSERT(!dsl_prop_hascb(ds));
2409 	}
2410 
2411 	/*
2412 	 * Change space accounting.
2413 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2414 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2415 	 * is true for each of {clone,origin} independently.
2416 	 */
2417 
2418 	delta = ddpa->cloneusedsnap -
2419 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2420 	ASSERT3S(delta, >=, 0);
2421 	ASSERT3U(ddpa->used, >=, delta);
2422 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2423 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2424 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2425 
2426 	delta = ddpa->originusedsnap -
2427 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2428 	ASSERT3S(delta, <=, 0);
2429 	ASSERT3U(ddpa->used, >=, -delta);
2430 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2431 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2432 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2433 
2434 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2435 
2436 	/* log history record */
2437 	spa_history_log_internal_ds(hds, "promote", tx, "");
2438 
2439 	dsl_dir_rele(odd, FTAG);
2440 	promote_rele(ddpa, FTAG);
2441 }
2442 
2443 /*
2444  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2445  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2446  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2447  * snapshots back to this dataset's origin.
2448  */
2449 static int
2450 snaplist_make(dsl_pool_t *dp,
2451     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2452 {
2453 	uint64_t obj = last_obj;
2454 
2455 	list_create(l, sizeof (struct promotenode),
2456 	    offsetof(struct promotenode, link));
2457 
2458 	while (obj != first_obj) {
2459 		dsl_dataset_t *ds;
2460 		struct promotenode *snap;
2461 		int err;
2462 
2463 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2464 		ASSERT(err != ENOENT);
2465 		if (err != 0)
2466 			return (err);
2467 
2468 		if (first_obj == 0)
2469 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2470 
2471 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2472 		snap->ds = ds;
2473 		list_insert_tail(l, snap);
2474 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2475 	}
2476 
2477 	return (0);
2478 }
2479 
2480 static int
2481 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2482 {
2483 	struct promotenode *snap;
2484 
2485 	*spacep = 0;
2486 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2487 		uint64_t used, comp, uncomp;
2488 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2489 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2490 		*spacep += used;
2491 	}
2492 	return (0);
2493 }
2494 
2495 static void
2496 snaplist_destroy(list_t *l, void *tag)
2497 {
2498 	struct promotenode *snap;
2499 
2500 	if (l == NULL || !list_link_active(&l->list_head))
2501 		return;
2502 
2503 	while ((snap = list_tail(l)) != NULL) {
2504 		list_remove(l, snap);
2505 		dsl_dataset_rele(snap->ds, tag);
2506 		kmem_free(snap, sizeof (*snap));
2507 	}
2508 	list_destroy(l);
2509 }
2510 
2511 static int
2512 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2513 {
2514 	int error;
2515 	dsl_dir_t *dd;
2516 	struct promotenode *snap;
2517 
2518 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2519 	    &ddpa->ddpa_clone);
2520 	if (error != 0)
2521 		return (error);
2522 	dd = ddpa->ddpa_clone->ds_dir;
2523 
2524 	if (ddpa->ddpa_clone->ds_is_snapshot ||
2525 	    !dsl_dir_is_clone(dd)) {
2526 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2527 		return (SET_ERROR(EINVAL));
2528 	}
2529 
2530 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2531 	    &ddpa->shared_snaps, tag);
2532 	if (error != 0)
2533 		goto out;
2534 
2535 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2536 	    &ddpa->clone_snaps, tag);
2537 	if (error != 0)
2538 		goto out;
2539 
2540 	snap = list_head(&ddpa->shared_snaps);
2541 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2542 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2543 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2544 	    &ddpa->origin_snaps, tag);
2545 	if (error != 0)
2546 		goto out;
2547 
2548 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2549 		error = dsl_dataset_hold_obj(dp,
2550 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2551 		    tag, &ddpa->origin_origin);
2552 		if (error != 0)
2553 			goto out;
2554 	}
2555 out:
2556 	if (error != 0)
2557 		promote_rele(ddpa, tag);
2558 	return (error);
2559 }
2560 
2561 static void
2562 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2563 {
2564 	snaplist_destroy(&ddpa->shared_snaps, tag);
2565 	snaplist_destroy(&ddpa->clone_snaps, tag);
2566 	snaplist_destroy(&ddpa->origin_snaps, tag);
2567 	if (ddpa->origin_origin != NULL)
2568 		dsl_dataset_rele(ddpa->origin_origin, tag);
2569 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2570 }
2571 
2572 /*
2573  * Promote a clone.
2574  *
2575  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2576  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2577  */
2578 int
2579 dsl_dataset_promote(const char *name, char *conflsnap)
2580 {
2581 	dsl_dataset_promote_arg_t ddpa = { 0 };
2582 	uint64_t numsnaps;
2583 	int error;
2584 	objset_t *os;
2585 
2586 	/*
2587 	 * We will modify space proportional to the number of
2588 	 * snapshots.  Compute numsnaps.
2589 	 */
2590 	error = dmu_objset_hold(name, FTAG, &os);
2591 	if (error != 0)
2592 		return (error);
2593 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2594 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2595 	    &numsnaps);
2596 	dmu_objset_rele(os, FTAG);
2597 	if (error != 0)
2598 		return (error);
2599 
2600 	ddpa.ddpa_clonename = name;
2601 	ddpa.err_ds = conflsnap;
2602 	ddpa.cr = CRED();
2603 
2604 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2605 	    dsl_dataset_promote_sync, &ddpa,
2606 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2607 }
2608 
2609 int
2610 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2611     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2612 {
2613 	int64_t unused_refres_delta;
2614 
2615 	/* they should both be heads */
2616 	if (clone->ds_is_snapshot ||
2617 	    origin_head->ds_is_snapshot)
2618 		return (SET_ERROR(EINVAL));
2619 
2620 	/* if we are not forcing, the branch point should be just before them */
2621 	if (!force && clone->ds_prev != origin_head->ds_prev)
2622 		return (SET_ERROR(EINVAL));
2623 
2624 	/* clone should be the clone (unless they are unrelated) */
2625 	if (clone->ds_prev != NULL &&
2626 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2627 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2628 		return (SET_ERROR(EINVAL));
2629 
2630 	/* the clone should be a child of the origin */
2631 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2632 		return (SET_ERROR(EINVAL));
2633 
2634 	/* origin_head shouldn't be modified unless 'force' */
2635 	if (!force &&
2636 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2637 		return (SET_ERROR(ETXTBSY));
2638 
2639 	/* origin_head should have no long holds (e.g. is not mounted) */
2640 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2641 		return (SET_ERROR(EBUSY));
2642 
2643 	/* check amount of any unconsumed refreservation */
2644 	unused_refres_delta =
2645 	    (int64_t)MIN(origin_head->ds_reserved,
2646 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2647 	    (int64_t)MIN(origin_head->ds_reserved,
2648 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2649 
2650 	if (unused_refres_delta > 0 &&
2651 	    unused_refres_delta >
2652 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2653 		return (SET_ERROR(ENOSPC));
2654 
2655 	/* clone can't be over the head's refquota */
2656 	if (origin_head->ds_quota != 0 &&
2657 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
2658 	    origin_head->ds_quota)
2659 		return (SET_ERROR(EDQUOT));
2660 
2661 	return (0);
2662 }
2663 
2664 void
2665 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2666     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2667 {
2668 	dsl_pool_t *dp = dmu_tx_pool(tx);
2669 	int64_t unused_refres_delta;
2670 
2671 	ASSERT(clone->ds_reserved == 0);
2672 	ASSERT(origin_head->ds_quota == 0 ||
2673 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2674 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2675 
2676 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2677 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2678 
2679 	if (clone->ds_objset != NULL) {
2680 		dmu_objset_evict(clone->ds_objset);
2681 		clone->ds_objset = NULL;
2682 	}
2683 
2684 	if (origin_head->ds_objset != NULL) {
2685 		dmu_objset_evict(origin_head->ds_objset);
2686 		origin_head->ds_objset = NULL;
2687 	}
2688 
2689 	unused_refres_delta =
2690 	    (int64_t)MIN(origin_head->ds_reserved,
2691 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2692 	    (int64_t)MIN(origin_head->ds_reserved,
2693 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2694 
2695 	/*
2696 	 * Reset origin's unique bytes, if it exists.
2697 	 */
2698 	if (clone->ds_prev) {
2699 		dsl_dataset_t *origin = clone->ds_prev;
2700 		uint64_t comp, uncomp;
2701 
2702 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2703 		dsl_deadlist_space_range(&clone->ds_deadlist,
2704 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2705 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2706 	}
2707 
2708 	/* swap blkptrs */
2709 	{
2710 		blkptr_t tmp;
2711 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
2712 		dsl_dataset_phys(origin_head)->ds_bp =
2713 		    dsl_dataset_phys(clone)->ds_bp;
2714 		dsl_dataset_phys(clone)->ds_bp = tmp;
2715 	}
2716 
2717 	/* set dd_*_bytes */
2718 	{
2719 		int64_t dused, dcomp, duncomp;
2720 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2721 		uint64_t odl_used, odl_comp, odl_uncomp;
2722 
2723 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2724 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2725 
2726 		dsl_deadlist_space(&clone->ds_deadlist,
2727 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2728 		dsl_deadlist_space(&origin_head->ds_deadlist,
2729 		    &odl_used, &odl_comp, &odl_uncomp);
2730 
2731 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2732 		    cdl_used -
2733 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2734 		    odl_used);
2735 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2736 		    cdl_comp -
2737 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2738 		    odl_comp);
2739 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2740 		    cdl_uncomp -
2741 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2742 		    odl_uncomp);
2743 
2744 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2745 		    dused, dcomp, duncomp, tx);
2746 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2747 		    -dused, -dcomp, -duncomp, tx);
2748 
2749 		/*
2750 		 * The difference in the space used by snapshots is the
2751 		 * difference in snapshot space due to the head's
2752 		 * deadlist (since that's the only thing that's
2753 		 * changing that affects the snapused).
2754 		 */
2755 		dsl_deadlist_space_range(&clone->ds_deadlist,
2756 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2757 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2758 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2759 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2760 		    &odl_used, &odl_comp, &odl_uncomp);
2761 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2762 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2763 	}
2764 
2765 	/* swap ds_*_bytes */
2766 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2767 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
2768 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2769 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
2770 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2771 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2772 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2773 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2774 
2775 	/* apply any parent delta for change in unconsumed refreservation */
2776 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2777 	    unused_refres_delta, 0, 0, tx);
2778 
2779 	/*
2780 	 * Swap deadlists.
2781 	 */
2782 	dsl_deadlist_close(&clone->ds_deadlist);
2783 	dsl_deadlist_close(&origin_head->ds_deadlist);
2784 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2785 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2786 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2787 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2788 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2789 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
2790 
2791 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2792 
2793 	spa_history_log_internal_ds(clone, "clone swap", tx,
2794 	    "parent=%s", origin_head->ds_dir->dd_myname);
2795 }
2796 
2797 /*
2798  * Given a pool name and a dataset object number in that pool,
2799  * return the name of that dataset.
2800  */
2801 int
2802 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2803 {
2804 	dsl_pool_t *dp;
2805 	dsl_dataset_t *ds;
2806 	int error;
2807 
2808 	error = dsl_pool_hold(pname, FTAG, &dp);
2809 	if (error != 0)
2810 		return (error);
2811 
2812 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2813 	if (error == 0) {
2814 		dsl_dataset_name(ds, buf);
2815 		dsl_dataset_rele(ds, FTAG);
2816 	}
2817 	dsl_pool_rele(dp, FTAG);
2818 
2819 	return (error);
2820 }
2821 
2822 int
2823 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2824     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2825 {
2826 	int error = 0;
2827 
2828 	ASSERT3S(asize, >, 0);
2829 
2830 	/*
2831 	 * *ref_rsrv is the portion of asize that will come from any
2832 	 * unconsumed refreservation space.
2833 	 */
2834 	*ref_rsrv = 0;
2835 
2836 	mutex_enter(&ds->ds_lock);
2837 	/*
2838 	 * Make a space adjustment for reserved bytes.
2839 	 */
2840 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2841 		ASSERT3U(*used, >=,
2842 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2843 		*used -=
2844 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2845 		*ref_rsrv =
2846 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2847 	}
2848 
2849 	if (!check_quota || ds->ds_quota == 0) {
2850 		mutex_exit(&ds->ds_lock);
2851 		return (0);
2852 	}
2853 	/*
2854 	 * If they are requesting more space, and our current estimate
2855 	 * is over quota, they get to try again unless the actual
2856 	 * on-disk is over quota and there are no pending changes (which
2857 	 * may free up space for us).
2858 	 */
2859 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
2860 	    ds->ds_quota) {
2861 		if (inflight > 0 ||
2862 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
2863 			error = SET_ERROR(ERESTART);
2864 		else
2865 			error = SET_ERROR(EDQUOT);
2866 	}
2867 	mutex_exit(&ds->ds_lock);
2868 
2869 	return (error);
2870 }
2871 
2872 typedef struct dsl_dataset_set_qr_arg {
2873 	const char *ddsqra_name;
2874 	zprop_source_t ddsqra_source;
2875 	uint64_t ddsqra_value;
2876 } dsl_dataset_set_qr_arg_t;
2877 
2878 
2879 /* ARGSUSED */
2880 static int
2881 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2882 {
2883 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2884 	dsl_pool_t *dp = dmu_tx_pool(tx);
2885 	dsl_dataset_t *ds;
2886 	int error;
2887 	uint64_t newval;
2888 
2889 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2890 		return (SET_ERROR(ENOTSUP));
2891 
2892 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2893 	if (error != 0)
2894 		return (error);
2895 
2896 	if (ds->ds_is_snapshot) {
2897 		dsl_dataset_rele(ds, FTAG);
2898 		return (SET_ERROR(EINVAL));
2899 	}
2900 
2901 	error = dsl_prop_predict(ds->ds_dir,
2902 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2903 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2904 	if (error != 0) {
2905 		dsl_dataset_rele(ds, FTAG);
2906 		return (error);
2907 	}
2908 
2909 	if (newval == 0) {
2910 		dsl_dataset_rele(ds, FTAG);
2911 		return (0);
2912 	}
2913 
2914 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
2915 	    newval < ds->ds_reserved) {
2916 		dsl_dataset_rele(ds, FTAG);
2917 		return (SET_ERROR(ENOSPC));
2918 	}
2919 
2920 	dsl_dataset_rele(ds, FTAG);
2921 	return (0);
2922 }
2923 
2924 static void
2925 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2926 {
2927 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2928 	dsl_pool_t *dp = dmu_tx_pool(tx);
2929 	dsl_dataset_t *ds;
2930 	uint64_t newval;
2931 
2932 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2933 
2934 	dsl_prop_set_sync_impl(ds,
2935 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2936 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2937 	    &ddsqra->ddsqra_value, tx);
2938 
2939 	VERIFY0(dsl_prop_get_int_ds(ds,
2940 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2941 
2942 	if (ds->ds_quota != newval) {
2943 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2944 		ds->ds_quota = newval;
2945 	}
2946 	dsl_dataset_rele(ds, FTAG);
2947 }
2948 
2949 int
2950 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2951     uint64_t refquota)
2952 {
2953 	dsl_dataset_set_qr_arg_t ddsqra;
2954 
2955 	ddsqra.ddsqra_name = dsname;
2956 	ddsqra.ddsqra_source = source;
2957 	ddsqra.ddsqra_value = refquota;
2958 
2959 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2960 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
2961 }
2962 
2963 static int
2964 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2965 {
2966 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2967 	dsl_pool_t *dp = dmu_tx_pool(tx);
2968 	dsl_dataset_t *ds;
2969 	int error;
2970 	uint64_t newval, unique;
2971 
2972 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2973 		return (SET_ERROR(ENOTSUP));
2974 
2975 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2976 	if (error != 0)
2977 		return (error);
2978 
2979 	if (ds->ds_is_snapshot) {
2980 		dsl_dataset_rele(ds, FTAG);
2981 		return (SET_ERROR(EINVAL));
2982 	}
2983 
2984 	error = dsl_prop_predict(ds->ds_dir,
2985 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2986 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2987 	if (error != 0) {
2988 		dsl_dataset_rele(ds, FTAG);
2989 		return (error);
2990 	}
2991 
2992 	/*
2993 	 * If we are doing the preliminary check in open context, the
2994 	 * space estimates may be inaccurate.
2995 	 */
2996 	if (!dmu_tx_is_syncing(tx)) {
2997 		dsl_dataset_rele(ds, FTAG);
2998 		return (0);
2999 	}
3000 
3001 	mutex_enter(&ds->ds_lock);
3002 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3003 		dsl_dataset_recalc_head_uniq(ds);
3004 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3005 	mutex_exit(&ds->ds_lock);
3006 
3007 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3008 		uint64_t delta = MAX(unique, newval) -
3009 		    MAX(unique, ds->ds_reserved);
3010 
3011 		if (delta >
3012 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3013 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3014 			dsl_dataset_rele(ds, FTAG);
3015 			return (SET_ERROR(ENOSPC));
3016 		}
3017 	}
3018 
3019 	dsl_dataset_rele(ds, FTAG);
3020 	return (0);
3021 }
3022 
3023 void
3024 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3025     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3026 {
3027 	uint64_t newval;
3028 	uint64_t unique;
3029 	int64_t delta;
3030 
3031 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3032 	    source, sizeof (value), 1, &value, tx);
3033 
3034 	VERIFY0(dsl_prop_get_int_ds(ds,
3035 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3036 
3037 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3038 	mutex_enter(&ds->ds_dir->dd_lock);
3039 	mutex_enter(&ds->ds_lock);
3040 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3041 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3042 	delta = MAX(0, (int64_t)(newval - unique)) -
3043 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3044 	ds->ds_reserved = newval;
3045 	mutex_exit(&ds->ds_lock);
3046 
3047 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3048 	mutex_exit(&ds->ds_dir->dd_lock);
3049 }
3050 
3051 static void
3052 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3053 {
3054 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3055 	dsl_pool_t *dp = dmu_tx_pool(tx);
3056 	dsl_dataset_t *ds;
3057 
3058 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3059 	dsl_dataset_set_refreservation_sync_impl(ds,
3060 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3061 	dsl_dataset_rele(ds, FTAG);
3062 }
3063 
3064 int
3065 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3066     uint64_t refreservation)
3067 {
3068 	dsl_dataset_set_qr_arg_t ddsqra;
3069 
3070 	ddsqra.ddsqra_name = dsname;
3071 	ddsqra.ddsqra_source = source;
3072 	ddsqra.ddsqra_value = refreservation;
3073 
3074 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3075 	    dsl_dataset_set_refreservation_sync, &ddsqra,
3076 	    0, ZFS_SPACE_CHECK_NONE));
3077 }
3078 
3079 /*
3080  * Return (in *usedp) the amount of space written in new that is not
3081  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3082  * a snapshot before new, in new's filesystem (or its origin).  If not then
3083  * fail and return EINVAL.
3084  *
3085  * The written space is calculated by considering two components:  First, we
3086  * ignore any freed space, and calculate the written as new's used space
3087  * minus old's used space.  Next, we add in the amount of space that was freed
3088  * between the two snapshots, thus reducing new's used space relative to old's.
3089  * Specifically, this is the space that was born before old->ds_creation_txg,
3090  * and freed before new (ie. on new's deadlist or a previous deadlist).
3091  *
3092  * space freed                         [---------------------]
3093  * snapshots                       ---O-------O--------O-------O------
3094  *                                         oldsnap            new
3095  */
3096 int
3097 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3098     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3099 {
3100 	int err = 0;
3101 	uint64_t snapobj;
3102 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3103 
3104 	ASSERT(dsl_pool_config_held(dp));
3105 
3106 	*usedp = 0;
3107 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3108 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3109 
3110 	*compp = 0;
3111 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3112 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3113 
3114 	*uncompp = 0;
3115 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3116 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3117 
3118 	snapobj = new->ds_object;
3119 	while (snapobj != oldsnap->ds_object) {
3120 		dsl_dataset_t *snap;
3121 		uint64_t used, comp, uncomp;
3122 
3123 		if (snapobj == new->ds_object) {
3124 			snap = new;
3125 		} else {
3126 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3127 			if (err != 0)
3128 				break;
3129 		}
3130 
3131 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3132 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3133 			/*
3134 			 * The blocks in the deadlist can not be born after
3135 			 * ds_prev_snap_txg, so get the whole deadlist space,
3136 			 * which is more efficient (especially for old-format
3137 			 * deadlists).  Unfortunately the deadlist code
3138 			 * doesn't have enough information to make this
3139 			 * optimization itself.
3140 			 */
3141 			dsl_deadlist_space(&snap->ds_deadlist,
3142 			    &used, &comp, &uncomp);
3143 		} else {
3144 			dsl_deadlist_space_range(&snap->ds_deadlist,
3145 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3146 			    &used, &comp, &uncomp);
3147 		}
3148 		*usedp += used;
3149 		*compp += comp;
3150 		*uncompp += uncomp;
3151 
3152 		/*
3153 		 * If we get to the beginning of the chain of snapshots
3154 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3155 		 * was not a snapshot of/before new.
3156 		 */
3157 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3158 		if (snap != new)
3159 			dsl_dataset_rele(snap, FTAG);
3160 		if (snapobj == 0) {
3161 			err = SET_ERROR(EINVAL);
3162 			break;
3163 		}
3164 
3165 	}
3166 	return (err);
3167 }
3168 
3169 /*
3170  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3171  * lastsnap, and all snapshots in between are deleted.
3172  *
3173  * blocks that would be freed            [---------------------------]
3174  * snapshots                       ---O-------O--------O-------O--------O
3175  *                                        firstsnap        lastsnap
3176  *
3177  * This is the set of blocks that were born after the snap before firstsnap,
3178  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3179  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3180  * We calculate this by iterating over the relevant deadlists (from the snap
3181  * after lastsnap, backward to the snap after firstsnap), summing up the
3182  * space on the deadlist that was born after the snap before firstsnap.
3183  */
3184 int
3185 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3186     dsl_dataset_t *lastsnap,
3187     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3188 {
3189 	int err = 0;
3190 	uint64_t snapobj;
3191 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3192 
3193 	ASSERT(firstsnap->ds_is_snapshot);
3194 	ASSERT(lastsnap->ds_is_snapshot);
3195 
3196 	/*
3197 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3198 	 * is before lastsnap.
3199 	 */
3200 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3201 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3202 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3203 		return (SET_ERROR(EINVAL));
3204 
3205 	*usedp = *compp = *uncompp = 0;
3206 
3207 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3208 	while (snapobj != firstsnap->ds_object) {
3209 		dsl_dataset_t *ds;
3210 		uint64_t used, comp, uncomp;
3211 
3212 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3213 		if (err != 0)
3214 			break;
3215 
3216 		dsl_deadlist_space_range(&ds->ds_deadlist,
3217 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3218 		    &used, &comp, &uncomp);
3219 		*usedp += used;
3220 		*compp += comp;
3221 		*uncompp += uncomp;
3222 
3223 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3224 		ASSERT3U(snapobj, !=, 0);
3225 		dsl_dataset_rele(ds, FTAG);
3226 	}
3227 	return (err);
3228 }
3229 
3230 static int
3231 dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx)
3232 {
3233 	const char *dsname = arg;
3234 	dsl_dataset_t *ds;
3235 	dsl_pool_t *dp = dmu_tx_pool(tx);
3236 	int error = 0;
3237 
3238 	if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
3239 		return (SET_ERROR(ENOTSUP));
3240 
3241 	ASSERT(spa_feature_is_enabled(dp->dp_spa,
3242 	    SPA_FEATURE_EXTENSIBLE_DATASET));
3243 
3244 	error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
3245 	if (error != 0)
3246 		return (error);
3247 
3248 	if (ds->ds_large_blocks)
3249 		error = EALREADY;
3250 	dsl_dataset_rele(ds, FTAG);
3251 
3252 	return (error);
3253 }
3254 
3255 void
3256 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx)
3257 {
3258 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
3259 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
3260 	uint64_t zero = 0;
3261 
3262 	spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx);
3263 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
3264 
3265 	VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS,
3266 	    sizeof (zero), 1, &zero, tx));
3267 }
3268 
3269 static void
3270 dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx)
3271 {
3272 	const char *dsname = arg;
3273 	dsl_dataset_t *ds;
3274 
3275 	VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds));
3276 
3277 	dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
3278 	ASSERT(!ds->ds_large_blocks);
3279 	ds->ds_large_blocks = B_TRUE;
3280 	dsl_dataset_rele(ds, FTAG);
3281 }
3282 
3283 int
3284 dsl_dataset_activate_large_blocks(const char *dsname)
3285 {
3286 	int error;
3287 
3288 	error = dsl_sync_task(dsname,
3289 	    dsl_dataset_activate_large_blocks_check,
3290 	    dsl_dataset_activate_large_blocks_sync, (void *)dsname,
3291 	    1, ZFS_SPACE_CHECK_RESERVED);
3292 
3293 	/*
3294 	 * EALREADY indicates that this dataset already supports large blocks.
3295 	 */
3296 	if (error == EALREADY)
3297 		error = 0;
3298 	return (error);
3299 }
3300 
3301 /*
3302  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3303  * For example, they could both be snapshots of the same filesystem, and
3304  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3305  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3306  * filesystem.  Or 'earlier' could be the origin's origin.
3307  *
3308  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3309  */
3310 boolean_t
3311 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3312 	uint64_t earlier_txg)
3313 {
3314 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3315 	int error;
3316 	boolean_t ret;
3317 
3318 	ASSERT(dsl_pool_config_held(dp));
3319 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3320 
3321 	if (earlier_txg == 0)
3322 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3323 
3324 	if (later->ds_is_snapshot &&
3325 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3326 		return (B_FALSE);
3327 
3328 	if (later->ds_dir == earlier->ds_dir)
3329 		return (B_TRUE);
3330 	if (!dsl_dir_is_clone(later->ds_dir))
3331 		return (B_FALSE);
3332 
3333 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3334 		return (B_TRUE);
3335 	dsl_dataset_t *origin;
3336 	error = dsl_dataset_hold_obj(dp,
3337 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3338 	if (error != 0)
3339 		return (B_FALSE);
3340 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3341 	dsl_dataset_rele(origin, FTAG);
3342 	return (ret);
3343 }
3344 
3345 
3346 void
3347 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3348 {
3349 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3350 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3351 }
3352