xref: /titanic_44/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision fd435bccec40cb841f712413108d8577422fd9ab)
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, 2015 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 #include <sys/dmu_send.h>
54 #include <sys/zio_checksum.h>
55 #include <sys/zio_compress.h>
56 #include <zfs_fletcher.h>
57 
58 /*
59  * The SPA supports block sizes up to 16MB.  However, very large blocks
60  * can have an impact on i/o latency (e.g. tying up a spinning disk for
61  * ~300ms), and also potentially on the memory allocator.  Therefore,
62  * we do not allow the recordsize to be set larger than zfs_max_recordsize
63  * (default 1MB).  Larger blocks can be created by changing this tunable,
64  * and pools with larger blocks can always be imported and used, regardless
65  * of this setting.
66  */
67 int zfs_max_recordsize = 1 * 1024 * 1024;
68 
69 #define	SWITCH64(x, y) \
70 	{ \
71 		uint64_t __tmp = (x); \
72 		(x) = (y); \
73 		(y) = __tmp; \
74 	}
75 
76 #define	DS_REF_MAX	(1ULL << 62)
77 
78 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
79 
80 /*
81  * Figure out how much of this delta should be propogated to the dsl_dir
82  * layer.  If there's a refreservation, that space has already been
83  * partially accounted for in our ancestors.
84  */
85 static int64_t
86 parent_delta(dsl_dataset_t *ds, int64_t delta)
87 {
88 	dsl_dataset_phys_t *ds_phys;
89 	uint64_t old_bytes, new_bytes;
90 
91 	if (ds->ds_reserved == 0)
92 		return (delta);
93 
94 	ds_phys = dsl_dataset_phys(ds);
95 	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
96 	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
97 
98 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
99 	return (new_bytes - old_bytes);
100 }
101 
102 void
103 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
104 {
105 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
106 	int compressed = BP_GET_PSIZE(bp);
107 	int uncompressed = BP_GET_UCSIZE(bp);
108 	int64_t delta;
109 
110 	dprintf_bp(bp, "ds=%p", ds);
111 
112 	ASSERT(dmu_tx_is_syncing(tx));
113 	/* It could have been compressed away to nothing */
114 	if (BP_IS_HOLE(bp))
115 		return;
116 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
117 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
118 	if (ds == NULL) {
119 		dsl_pool_mos_diduse_space(tx->tx_pool,
120 		    used, compressed, uncompressed);
121 		return;
122 	}
123 
124 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
125 	mutex_enter(&ds->ds_lock);
126 	delta = parent_delta(ds, used);
127 	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
128 	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
129 	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
130 	dsl_dataset_phys(ds)->ds_unique_bytes += used;
131 	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
132 		ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
133 		    B_TRUE;
134 	}
135 	mutex_exit(&ds->ds_lock);
136 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
137 	    compressed, uncompressed, tx);
138 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
139 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
140 }
141 
142 int
143 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
144     boolean_t async)
145 {
146 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
147 	int compressed = BP_GET_PSIZE(bp);
148 	int uncompressed = BP_GET_UCSIZE(bp);
149 
150 	if (BP_IS_HOLE(bp))
151 		return (0);
152 
153 	ASSERT(dmu_tx_is_syncing(tx));
154 	ASSERT(bp->blk_birth <= tx->tx_txg);
155 
156 	if (ds == NULL) {
157 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
158 		dsl_pool_mos_diduse_space(tx->tx_pool,
159 		    -used, -compressed, -uncompressed);
160 		return (used);
161 	}
162 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
163 
164 	ASSERT(!ds->ds_is_snapshot);
165 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
166 
167 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
168 		int64_t delta;
169 
170 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
171 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
172 
173 		mutex_enter(&ds->ds_lock);
174 		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
175 		    !DS_UNIQUE_IS_ACCURATE(ds));
176 		delta = parent_delta(ds, -used);
177 		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
178 		mutex_exit(&ds->ds_lock);
179 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
180 		    delta, -compressed, -uncompressed, tx);
181 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
182 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
183 	} else {
184 		dprintf_bp(bp, "putting on dead list: %s", "");
185 		if (async) {
186 			/*
187 			 * We are here as part of zio's write done callback,
188 			 * which means we're a zio interrupt thread.  We can't
189 			 * call dsl_deadlist_insert() now because it may block
190 			 * waiting for I/O.  Instead, put bp on the deferred
191 			 * queue and let dsl_pool_sync() finish the job.
192 			 */
193 			bplist_append(&ds->ds_pending_deadlist, bp);
194 		} else {
195 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
196 		}
197 		ASSERT3U(ds->ds_prev->ds_object, ==,
198 		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
199 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
200 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
201 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
202 		    ds->ds_object && bp->blk_birth >
203 		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
204 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
205 			mutex_enter(&ds->ds_prev->ds_lock);
206 			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
207 			mutex_exit(&ds->ds_prev->ds_lock);
208 		}
209 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
210 			dsl_dir_transfer_space(ds->ds_dir, used,
211 			    DD_USED_HEAD, DD_USED_SNAP, tx);
212 		}
213 	}
214 	mutex_enter(&ds->ds_lock);
215 	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
216 	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
217 	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
218 	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
219 	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
220 	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
221 	mutex_exit(&ds->ds_lock);
222 
223 	return (used);
224 }
225 
226 uint64_t
227 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
228 {
229 	uint64_t trysnap = 0;
230 
231 	if (ds == NULL)
232 		return (0);
233 	/*
234 	 * The snapshot creation could fail, but that would cause an
235 	 * incorrect FALSE return, which would only result in an
236 	 * overestimation of the amount of space that an operation would
237 	 * consume, which is OK.
238 	 *
239 	 * There's also a small window where we could miss a pending
240 	 * snapshot, because we could set the sync task in the quiescing
241 	 * phase.  So this should only be used as a guess.
242 	 */
243 	if (ds->ds_trysnap_txg >
244 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
245 		trysnap = ds->ds_trysnap_txg;
246 	return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
247 }
248 
249 boolean_t
250 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
251     uint64_t blk_birth)
252 {
253 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
254 	    (bp != NULL && BP_IS_HOLE(bp)))
255 		return (B_FALSE);
256 
257 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
258 
259 	return (B_TRUE);
260 }
261 
262 static void
263 dsl_dataset_evict(void *dbu)
264 {
265 	dsl_dataset_t *ds = dbu;
266 
267 	ASSERT(ds->ds_owner == NULL);
268 
269 	ds->ds_dbuf = NULL;
270 
271 	unique_remove(ds->ds_fsid_guid);
272 
273 	if (ds->ds_objset != NULL)
274 		dmu_objset_evict(ds->ds_objset);
275 
276 	if (ds->ds_prev) {
277 		dsl_dataset_rele(ds->ds_prev, ds);
278 		ds->ds_prev = NULL;
279 	}
280 
281 	bplist_destroy(&ds->ds_pending_deadlist);
282 	if (ds->ds_deadlist.dl_os != NULL)
283 		dsl_deadlist_close(&ds->ds_deadlist);
284 	if (ds->ds_dir)
285 		dsl_dir_async_rele(ds->ds_dir, ds);
286 
287 	ASSERT(!list_link_active(&ds->ds_synced_link));
288 
289 	mutex_destroy(&ds->ds_lock);
290 	mutex_destroy(&ds->ds_opening_lock);
291 	mutex_destroy(&ds->ds_sendstream_lock);
292 	refcount_destroy(&ds->ds_longholds);
293 
294 	kmem_free(ds, sizeof (dsl_dataset_t));
295 }
296 
297 int
298 dsl_dataset_get_snapname(dsl_dataset_t *ds)
299 {
300 	dsl_dataset_phys_t *headphys;
301 	int err;
302 	dmu_buf_t *headdbuf;
303 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
304 	objset_t *mos = dp->dp_meta_objset;
305 
306 	if (ds->ds_snapname[0])
307 		return (0);
308 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
309 		return (0);
310 
311 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
312 	    FTAG, &headdbuf);
313 	if (err != 0)
314 		return (err);
315 	headphys = headdbuf->db_data;
316 	err = zap_value_search(dp->dp_meta_objset,
317 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
318 	dmu_buf_rele(headdbuf, FTAG);
319 	return (err);
320 }
321 
322 int
323 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
324 {
325 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
326 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
327 	matchtype_t mt;
328 	int err;
329 
330 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
331 		mt = MT_FIRST;
332 	else
333 		mt = MT_EXACT;
334 
335 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
336 	    value, mt, NULL, 0, NULL);
337 	if (err == ENOTSUP && mt == MT_FIRST)
338 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
339 	return (err);
340 }
341 
342 int
343 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
344     boolean_t adj_cnt)
345 {
346 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
347 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
348 	matchtype_t mt;
349 	int err;
350 
351 	dsl_dir_snap_cmtime_update(ds->ds_dir);
352 
353 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
354 		mt = MT_FIRST;
355 	else
356 		mt = MT_EXACT;
357 
358 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
359 	if (err == ENOTSUP && mt == MT_FIRST)
360 		err = zap_remove(mos, snapobj, name, tx);
361 
362 	if (err == 0 && adj_cnt)
363 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
364 		    DD_FIELD_SNAPSHOT_COUNT, tx);
365 
366 	return (err);
367 }
368 
369 boolean_t
370 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
371 {
372 	dmu_buf_t *dbuf = ds->ds_dbuf;
373 	boolean_t result = B_FALSE;
374 
375 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
376 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
377 
378 		if (ds == dmu_buf_get_user(dbuf))
379 			result = B_TRUE;
380 		else
381 			dmu_buf_rele(dbuf, tag);
382 	}
383 
384 	return (result);
385 }
386 
387 int
388 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
389     dsl_dataset_t **dsp)
390 {
391 	objset_t *mos = dp->dp_meta_objset;
392 	dmu_buf_t *dbuf;
393 	dsl_dataset_t *ds;
394 	int err;
395 	dmu_object_info_t doi;
396 
397 	ASSERT(dsl_pool_config_held(dp));
398 
399 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
400 	if (err != 0)
401 		return (err);
402 
403 	/* Make sure dsobj has the correct object type. */
404 	dmu_object_info_from_db(dbuf, &doi);
405 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
406 		dmu_buf_rele(dbuf, tag);
407 		return (SET_ERROR(EINVAL));
408 	}
409 
410 	ds = dmu_buf_get_user(dbuf);
411 	if (ds == NULL) {
412 		dsl_dataset_t *winner = NULL;
413 
414 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
415 		ds->ds_dbuf = dbuf;
416 		ds->ds_object = dsobj;
417 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
418 
419 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
420 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
421 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
422 		refcount_create(&ds->ds_longholds);
423 
424 		bplist_create(&ds->ds_pending_deadlist);
425 		dsl_deadlist_open(&ds->ds_deadlist,
426 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
427 
428 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
429 		    offsetof(dmu_sendarg_t, dsa_link));
430 
431 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
432 			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
433 				if (!(spa_feature_table[f].fi_flags &
434 				    ZFEATURE_FLAG_PER_DATASET))
435 					continue;
436 				err = zap_contains(mos, dsobj,
437 				    spa_feature_table[f].fi_guid);
438 				if (err == 0) {
439 					ds->ds_feature_inuse[f] = B_TRUE;
440 				} else {
441 					ASSERT3U(err, ==, ENOENT);
442 					err = 0;
443 				}
444 			}
445 		}
446 
447 		err = dsl_dir_hold_obj(dp,
448 		    dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
449 		if (err != 0) {
450 			mutex_destroy(&ds->ds_lock);
451 			mutex_destroy(&ds->ds_opening_lock);
452 			mutex_destroy(&ds->ds_sendstream_lock);
453 			refcount_destroy(&ds->ds_longholds);
454 			bplist_destroy(&ds->ds_pending_deadlist);
455 			dsl_deadlist_close(&ds->ds_deadlist);
456 			kmem_free(ds, sizeof (dsl_dataset_t));
457 			dmu_buf_rele(dbuf, tag);
458 			return (err);
459 		}
460 
461 		if (!ds->ds_is_snapshot) {
462 			ds->ds_snapname[0] = '\0';
463 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
464 				err = dsl_dataset_hold_obj(dp,
465 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
466 				    ds, &ds->ds_prev);
467 			}
468 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
469 				int zaperr = zap_lookup(mos, ds->ds_object,
470 				    DS_FIELD_BOOKMARK_NAMES,
471 				    sizeof (ds->ds_bookmarks), 1,
472 				    &ds->ds_bookmarks);
473 				if (zaperr != ENOENT)
474 					VERIFY0(zaperr);
475 			}
476 		} else {
477 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
478 				err = dsl_dataset_get_snapname(ds);
479 			if (err == 0 &&
480 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
481 				err = zap_count(
482 				    ds->ds_dir->dd_pool->dp_meta_objset,
483 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
484 				    &ds->ds_userrefs);
485 			}
486 		}
487 
488 		if (err == 0 && !ds->ds_is_snapshot) {
489 			err = dsl_prop_get_int_ds(ds,
490 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
491 			    &ds->ds_reserved);
492 			if (err == 0) {
493 				err = dsl_prop_get_int_ds(ds,
494 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
495 				    &ds->ds_quota);
496 			}
497 		} else {
498 			ds->ds_reserved = ds->ds_quota = 0;
499 		}
500 
501 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
502 		if (err == 0)
503 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
504 
505 		if (err != 0 || winner != NULL) {
506 			bplist_destroy(&ds->ds_pending_deadlist);
507 			dsl_deadlist_close(&ds->ds_deadlist);
508 			if (ds->ds_prev)
509 				dsl_dataset_rele(ds->ds_prev, ds);
510 			dsl_dir_rele(ds->ds_dir, ds);
511 			mutex_destroy(&ds->ds_lock);
512 			mutex_destroy(&ds->ds_opening_lock);
513 			mutex_destroy(&ds->ds_sendstream_lock);
514 			refcount_destroy(&ds->ds_longholds);
515 			kmem_free(ds, sizeof (dsl_dataset_t));
516 			if (err != 0) {
517 				dmu_buf_rele(dbuf, tag);
518 				return (err);
519 			}
520 			ds = winner;
521 		} else {
522 			ds->ds_fsid_guid =
523 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
524 		}
525 	}
526 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
527 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
528 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
529 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
530 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
531 	*dsp = ds;
532 	return (0);
533 }
534 
535 int
536 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
537     void *tag, dsl_dataset_t **dsp)
538 {
539 	dsl_dir_t *dd;
540 	const char *snapname;
541 	uint64_t obj;
542 	int err = 0;
543 	dsl_dataset_t *ds;
544 
545 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
546 	if (err != 0)
547 		return (err);
548 
549 	ASSERT(dsl_pool_config_held(dp));
550 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
551 	if (obj != 0)
552 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
553 	else
554 		err = SET_ERROR(ENOENT);
555 
556 	/* we may be looking for a snapshot */
557 	if (err == 0 && snapname != NULL) {
558 		dsl_dataset_t *snap_ds;
559 
560 		if (*snapname++ != '@') {
561 			dsl_dataset_rele(ds, tag);
562 			dsl_dir_rele(dd, FTAG);
563 			return (SET_ERROR(ENOENT));
564 		}
565 
566 		dprintf("looking for snapshot '%s'\n", snapname);
567 		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
568 		if (err == 0)
569 			err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
570 		dsl_dataset_rele(ds, tag);
571 
572 		if (err == 0) {
573 			mutex_enter(&snap_ds->ds_lock);
574 			if (snap_ds->ds_snapname[0] == 0)
575 				(void) strlcpy(snap_ds->ds_snapname, snapname,
576 				    sizeof (snap_ds->ds_snapname));
577 			mutex_exit(&snap_ds->ds_lock);
578 			ds = snap_ds;
579 		}
580 	}
581 	if (err == 0)
582 		*dsp = ds;
583 	dsl_dir_rele(dd, FTAG);
584 	return (err);
585 }
586 
587 int
588 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
589     void *tag, dsl_dataset_t **dsp)
590 {
591 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
592 	if (err != 0)
593 		return (err);
594 	if (!dsl_dataset_tryown(*dsp, tag)) {
595 		dsl_dataset_rele(*dsp, tag);
596 		*dsp = NULL;
597 		return (SET_ERROR(EBUSY));
598 	}
599 	return (0);
600 }
601 
602 int
603 dsl_dataset_own(dsl_pool_t *dp, const char *name,
604     void *tag, dsl_dataset_t **dsp)
605 {
606 	int err = dsl_dataset_hold(dp, name, tag, dsp);
607 	if (err != 0)
608 		return (err);
609 	if (!dsl_dataset_tryown(*dsp, tag)) {
610 		dsl_dataset_rele(*dsp, tag);
611 		return (SET_ERROR(EBUSY));
612 	}
613 	return (0);
614 }
615 
616 /*
617  * See the comment above dsl_pool_hold() for details.  In summary, a long
618  * hold is used to prevent destruction of a dataset while the pool hold
619  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
620  *
621  * The dataset and pool must be held when this function is called.  After it
622  * is called, the pool hold may be released while the dataset is still held
623  * and accessed.
624  */
625 void
626 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
627 {
628 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
629 	(void) refcount_add(&ds->ds_longholds, tag);
630 }
631 
632 void
633 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
634 {
635 	(void) refcount_remove(&ds->ds_longholds, tag);
636 }
637 
638 /* Return B_TRUE if there are any long holds on this dataset. */
639 boolean_t
640 dsl_dataset_long_held(dsl_dataset_t *ds)
641 {
642 	return (!refcount_is_zero(&ds->ds_longholds));
643 }
644 
645 void
646 dsl_dataset_name(dsl_dataset_t *ds, char *name)
647 {
648 	if (ds == NULL) {
649 		(void) strcpy(name, "mos");
650 	} else {
651 		dsl_dir_name(ds->ds_dir, name);
652 		VERIFY0(dsl_dataset_get_snapname(ds));
653 		if (ds->ds_snapname[0]) {
654 			VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
655 			    <, ZFS_MAX_DATASET_NAME_LEN);
656 			/*
657 			 * We use a "recursive" mutex so that we
658 			 * can call dprintf_ds() with ds_lock held.
659 			 */
660 			if (!MUTEX_HELD(&ds->ds_lock)) {
661 				mutex_enter(&ds->ds_lock);
662 				VERIFY3U(strlcat(name, ds->ds_snapname,
663 				    ZFS_MAX_DATASET_NAME_LEN), <,
664 				    ZFS_MAX_DATASET_NAME_LEN);
665 				mutex_exit(&ds->ds_lock);
666 			} else {
667 				VERIFY3U(strlcat(name, ds->ds_snapname,
668 				    ZFS_MAX_DATASET_NAME_LEN), <,
669 				    ZFS_MAX_DATASET_NAME_LEN);
670 			}
671 		}
672 	}
673 }
674 
675 int
676 dsl_dataset_namelen(dsl_dataset_t *ds)
677 {
678 	VERIFY0(dsl_dataset_get_snapname(ds));
679 	mutex_enter(&ds->ds_lock);
680 	int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
681 	mutex_exit(&ds->ds_lock);
682 	return (len);
683 }
684 
685 void
686 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
687 {
688 	dmu_buf_rele(ds->ds_dbuf, tag);
689 }
690 
691 void
692 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
693 {
694 	ASSERT3P(ds->ds_owner, ==, tag);
695 	ASSERT(ds->ds_dbuf != NULL);
696 
697 	mutex_enter(&ds->ds_lock);
698 	ds->ds_owner = NULL;
699 	mutex_exit(&ds->ds_lock);
700 	dsl_dataset_long_rele(ds, tag);
701 	dsl_dataset_rele(ds, tag);
702 }
703 
704 boolean_t
705 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
706 {
707 	boolean_t gotit = FALSE;
708 
709 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
710 	mutex_enter(&ds->ds_lock);
711 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
712 		ds->ds_owner = tag;
713 		dsl_dataset_long_hold(ds, tag);
714 		gotit = TRUE;
715 	}
716 	mutex_exit(&ds->ds_lock);
717 	return (gotit);
718 }
719 
720 boolean_t
721 dsl_dataset_has_owner(dsl_dataset_t *ds)
722 {
723 	boolean_t rv;
724 	mutex_enter(&ds->ds_lock);
725 	rv = (ds->ds_owner != NULL);
726 	mutex_exit(&ds->ds_lock);
727 	return (rv);
728 }
729 
730 static void
731 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
732 {
733 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
734 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
735 	uint64_t zero = 0;
736 
737 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
738 
739 	spa_feature_incr(spa, f, tx);
740 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
741 
742 	VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
743 	    sizeof (zero), 1, &zero, tx));
744 }
745 
746 void
747 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
748 {
749 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
750 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
751 
752 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
753 
754 	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
755 	spa_feature_decr(spa, f, tx);
756 }
757 
758 uint64_t
759 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
760     uint64_t flags, dmu_tx_t *tx)
761 {
762 	dsl_pool_t *dp = dd->dd_pool;
763 	dmu_buf_t *dbuf;
764 	dsl_dataset_phys_t *dsphys;
765 	uint64_t dsobj;
766 	objset_t *mos = dp->dp_meta_objset;
767 
768 	if (origin == NULL)
769 		origin = dp->dp_origin_snap;
770 
771 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
772 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
773 	ASSERT(dmu_tx_is_syncing(tx));
774 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
775 
776 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
777 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
778 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
779 	dmu_buf_will_dirty(dbuf, tx);
780 	dsphys = dbuf->db_data;
781 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
782 	dsphys->ds_dir_obj = dd->dd_object;
783 	dsphys->ds_flags = flags;
784 	dsphys->ds_fsid_guid = unique_create();
785 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
786 	    sizeof (dsphys->ds_guid));
787 	dsphys->ds_snapnames_zapobj =
788 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
789 	    DMU_OT_NONE, 0, tx);
790 	dsphys->ds_creation_time = gethrestime_sec();
791 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
792 
793 	if (origin == NULL) {
794 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
795 	} else {
796 		dsl_dataset_t *ohds; /* head of the origin snapshot */
797 
798 		dsphys->ds_prev_snap_obj = origin->ds_object;
799 		dsphys->ds_prev_snap_txg =
800 		    dsl_dataset_phys(origin)->ds_creation_txg;
801 		dsphys->ds_referenced_bytes =
802 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
803 		dsphys->ds_compressed_bytes =
804 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
805 		dsphys->ds_uncompressed_bytes =
806 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
807 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
808 
809 		/*
810 		 * Inherit flags that describe the dataset's contents
811 		 * (INCONSISTENT) or properties (Case Insensitive).
812 		 */
813 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
814 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
815 
816 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
817 			if (origin->ds_feature_inuse[f])
818 				dsl_dataset_activate_feature(dsobj, f, tx);
819 		}
820 
821 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
822 		dsl_dataset_phys(origin)->ds_num_children++;
823 
824 		VERIFY0(dsl_dataset_hold_obj(dp,
825 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
826 		    FTAG, &ohds));
827 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
828 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
829 		dsl_dataset_rele(ohds, FTAG);
830 
831 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
832 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
833 				dsl_dataset_phys(origin)->ds_next_clones_obj =
834 				    zap_create(mos,
835 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
836 			}
837 			VERIFY0(zap_add_int(mos,
838 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
839 			    dsobj, tx));
840 		}
841 
842 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
843 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
844 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
845 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
846 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
847 				dsl_dir_phys(origin->ds_dir)->dd_clones =
848 				    zap_create(mos,
849 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
850 			}
851 			VERIFY0(zap_add_int(mos,
852 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
853 			    dsobj, tx));
854 		}
855 	}
856 
857 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
858 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
859 
860 	dmu_buf_rele(dbuf, FTAG);
861 
862 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
863 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
864 
865 	return (dsobj);
866 }
867 
868 static void
869 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
870 {
871 	objset_t *os;
872 
873 	VERIFY0(dmu_objset_from_ds(ds, &os));
874 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
875 	dsl_dataset_dirty(ds, tx);
876 }
877 
878 uint64_t
879 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
880     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
881 {
882 	dsl_pool_t *dp = pdd->dd_pool;
883 	uint64_t dsobj, ddobj;
884 	dsl_dir_t *dd;
885 
886 	ASSERT(dmu_tx_is_syncing(tx));
887 	ASSERT(lastname[0] != '@');
888 
889 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
890 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
891 
892 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
893 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
894 
895 	dsl_deleg_set_create_perms(dd, tx, cr);
896 
897 	/*
898 	 * Since we're creating a new node we know it's a leaf, so we can
899 	 * initialize the counts if the limit feature is active.
900 	 */
901 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
902 		uint64_t cnt = 0;
903 		objset_t *os = dd->dd_pool->dp_meta_objset;
904 
905 		dsl_dir_zapify(dd, tx);
906 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
907 		    sizeof (cnt), 1, &cnt, tx));
908 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
909 		    sizeof (cnt), 1, &cnt, tx));
910 	}
911 
912 	dsl_dir_rele(dd, FTAG);
913 
914 	/*
915 	 * If we are creating a clone, make sure we zero out any stale
916 	 * data from the origin snapshots zil header.
917 	 */
918 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
919 		dsl_dataset_t *ds;
920 
921 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
922 		dsl_dataset_zero_zil(ds, tx);
923 		dsl_dataset_rele(ds, FTAG);
924 	}
925 
926 	return (dsobj);
927 }
928 
929 /*
930  * The unique space in the head dataset can be calculated by subtracting
931  * the space used in the most recent snapshot, that is still being used
932  * in this file system, from the space currently in use.  To figure out
933  * the space in the most recent snapshot still in use, we need to take
934  * the total space used in the snapshot and subtract out the space that
935  * has been freed up since the snapshot was taken.
936  */
937 void
938 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
939 {
940 	uint64_t mrs_used;
941 	uint64_t dlused, dlcomp, dluncomp;
942 
943 	ASSERT(!ds->ds_is_snapshot);
944 
945 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
946 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
947 	else
948 		mrs_used = 0;
949 
950 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
951 
952 	ASSERT3U(dlused, <=, mrs_used);
953 	dsl_dataset_phys(ds)->ds_unique_bytes =
954 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
955 
956 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
957 	    SPA_VERSION_UNIQUE_ACCURATE)
958 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
959 }
960 
961 void
962 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
963     dmu_tx_t *tx)
964 {
965 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
966 	uint64_t count;
967 	int err;
968 
969 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
970 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
971 	    obj, tx);
972 	/*
973 	 * The err should not be ENOENT, but a bug in a previous version
974 	 * of the code could cause upgrade_clones_cb() to not set
975 	 * ds_next_snap_obj when it should, leading to a missing entry.
976 	 * If we knew that the pool was created after
977 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
978 	 * ENOENT.  However, at least we can check that we don't have
979 	 * too many entries in the next_clones_obj even after failing to
980 	 * remove this one.
981 	 */
982 	if (err != ENOENT)
983 		VERIFY0(err);
984 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
985 	    &count));
986 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
987 }
988 
989 
990 blkptr_t *
991 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
992 {
993 	return (&dsl_dataset_phys(ds)->ds_bp);
994 }
995 
996 void
997 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
998 {
999 	ASSERT(dmu_tx_is_syncing(tx));
1000 	/* If it's the meta-objset, set dp_meta_rootbp */
1001 	if (ds == NULL) {
1002 		tx->tx_pool->dp_meta_rootbp = *bp;
1003 	} else {
1004 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1005 		dsl_dataset_phys(ds)->ds_bp = *bp;
1006 	}
1007 }
1008 
1009 spa_t *
1010 dsl_dataset_get_spa(dsl_dataset_t *ds)
1011 {
1012 	return (ds->ds_dir->dd_pool->dp_spa);
1013 }
1014 
1015 void
1016 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1017 {
1018 	dsl_pool_t *dp;
1019 
1020 	if (ds == NULL) /* this is the meta-objset */
1021 		return;
1022 
1023 	ASSERT(ds->ds_objset != NULL);
1024 
1025 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1026 		panic("dirtying snapshot!");
1027 
1028 	dp = ds->ds_dir->dd_pool;
1029 
1030 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1031 		/* up the hold count until we can be written out */
1032 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1033 	}
1034 }
1035 
1036 boolean_t
1037 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1038 {
1039 	for (int t = 0; t < TXG_SIZE; t++) {
1040 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1041 		    ds, t))
1042 			return (B_TRUE);
1043 	}
1044 	return (B_FALSE);
1045 }
1046 
1047 static int
1048 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1049 {
1050 	uint64_t asize;
1051 
1052 	if (!dmu_tx_is_syncing(tx))
1053 		return (0);
1054 
1055 	/*
1056 	 * If there's an fs-only reservation, any blocks that might become
1057 	 * owned by the snapshot dataset must be accommodated by space
1058 	 * outside of the reservation.
1059 	 */
1060 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1061 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1062 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1063 		return (SET_ERROR(ENOSPC));
1064 
1065 	/*
1066 	 * Propagate any reserved space for this snapshot to other
1067 	 * snapshot checks in this sync group.
1068 	 */
1069 	if (asize > 0)
1070 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1071 
1072 	return (0);
1073 }
1074 
1075 typedef struct dsl_dataset_snapshot_arg {
1076 	nvlist_t *ddsa_snaps;
1077 	nvlist_t *ddsa_props;
1078 	nvlist_t *ddsa_errors;
1079 	cred_t *ddsa_cr;
1080 } dsl_dataset_snapshot_arg_t;
1081 
1082 int
1083 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1084     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1085 {
1086 	int error;
1087 	uint64_t value;
1088 
1089 	ds->ds_trysnap_txg = tx->tx_txg;
1090 
1091 	if (!dmu_tx_is_syncing(tx))
1092 		return (0);
1093 
1094 	/*
1095 	 * We don't allow multiple snapshots of the same txg.  If there
1096 	 * is already one, try again.
1097 	 */
1098 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1099 		return (SET_ERROR(EAGAIN));
1100 
1101 	/*
1102 	 * Check for conflicting snapshot name.
1103 	 */
1104 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1105 	if (error == 0)
1106 		return (SET_ERROR(EEXIST));
1107 	if (error != ENOENT)
1108 		return (error);
1109 
1110 	/*
1111 	 * We don't allow taking snapshots of inconsistent datasets, such as
1112 	 * those into which we are currently receiving.  However, if we are
1113 	 * creating this snapshot as part of a receive, this check will be
1114 	 * executed atomically with respect to the completion of the receive
1115 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1116 	 * case we ignore this, knowing it will be fixed up for us shortly in
1117 	 * dmu_recv_end_sync().
1118 	 */
1119 	if (!recv && DS_IS_INCONSISTENT(ds))
1120 		return (SET_ERROR(EBUSY));
1121 
1122 	/*
1123 	 * Skip the check for temporary snapshots or if we have already checked
1124 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1125 	 * check the count here when we're receiving a stream.
1126 	 */
1127 	if (cnt != 0 && cr != NULL) {
1128 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1129 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1130 		if (error != 0)
1131 			return (error);
1132 	}
1133 
1134 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1135 	if (error != 0)
1136 		return (error);
1137 
1138 	return (0);
1139 }
1140 
1141 static int
1142 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1143 {
1144 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1145 	dsl_pool_t *dp = dmu_tx_pool(tx);
1146 	nvpair_t *pair;
1147 	int rv = 0;
1148 
1149 	/*
1150 	 * Pre-compute how many total new snapshots will be created for each
1151 	 * level in the tree and below. This is needed for validating the
1152 	 * snapshot limit when either taking a recursive snapshot or when
1153 	 * taking multiple snapshots.
1154 	 *
1155 	 * The problem is that the counts are not actually adjusted when
1156 	 * we are checking, only when we finally sync. For a single snapshot,
1157 	 * this is easy, the count will increase by 1 at each node up the tree,
1158 	 * but its more complicated for the recursive/multiple snapshot case.
1159 	 *
1160 	 * The dsl_fs_ss_limit_check function does recursively check the count
1161 	 * at each level up the tree but since it is validating each snapshot
1162 	 * independently we need to be sure that we are validating the complete
1163 	 * count for the entire set of snapshots. We do this by rolling up the
1164 	 * counts for each component of the name into an nvlist and then
1165 	 * checking each of those cases with the aggregated count.
1166 	 *
1167 	 * This approach properly handles not only the recursive snapshot
1168 	 * case (where we get all of those on the ddsa_snaps list) but also
1169 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1170 	 * validate the limit on 'a' using a count of 2).
1171 	 *
1172 	 * We validate the snapshot names in the third loop and only report
1173 	 * name errors once.
1174 	 */
1175 	if (dmu_tx_is_syncing(tx)) {
1176 		nvlist_t *cnt_track = NULL;
1177 		cnt_track = fnvlist_alloc();
1178 
1179 		/* Rollup aggregated counts into the cnt_track list */
1180 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1181 		    pair != NULL;
1182 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1183 			char *pdelim;
1184 			uint64_t val;
1185 			char nm[MAXPATHLEN];
1186 
1187 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1188 			pdelim = strchr(nm, '@');
1189 			if (pdelim == NULL)
1190 				continue;
1191 			*pdelim = '\0';
1192 
1193 			do {
1194 				if (nvlist_lookup_uint64(cnt_track, nm,
1195 				    &val) == 0) {
1196 					/* update existing entry */
1197 					fnvlist_add_uint64(cnt_track, nm,
1198 					    val + 1);
1199 				} else {
1200 					/* add to list */
1201 					fnvlist_add_uint64(cnt_track, nm, 1);
1202 				}
1203 
1204 				pdelim = strrchr(nm, '/');
1205 				if (pdelim != NULL)
1206 					*pdelim = '\0';
1207 			} while (pdelim != NULL);
1208 		}
1209 
1210 		/* Check aggregated counts at each level */
1211 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1212 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1213 			int error = 0;
1214 			char *name;
1215 			uint64_t cnt = 0;
1216 			dsl_dataset_t *ds;
1217 
1218 			name = nvpair_name(pair);
1219 			cnt = fnvpair_value_uint64(pair);
1220 			ASSERT(cnt > 0);
1221 
1222 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1223 			if (error == 0) {
1224 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1225 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1226 				    ddsa->ddsa_cr);
1227 				dsl_dataset_rele(ds, FTAG);
1228 			}
1229 
1230 			if (error != 0) {
1231 				if (ddsa->ddsa_errors != NULL)
1232 					fnvlist_add_int32(ddsa->ddsa_errors,
1233 					    name, error);
1234 				rv = error;
1235 				/* only report one error for this check */
1236 				break;
1237 			}
1238 		}
1239 		nvlist_free(cnt_track);
1240 	}
1241 
1242 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1243 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1244 		int error = 0;
1245 		dsl_dataset_t *ds;
1246 		char *name, *atp;
1247 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1248 
1249 		name = nvpair_name(pair);
1250 		if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1251 			error = SET_ERROR(ENAMETOOLONG);
1252 		if (error == 0) {
1253 			atp = strchr(name, '@');
1254 			if (atp == NULL)
1255 				error = SET_ERROR(EINVAL);
1256 			if (error == 0)
1257 				(void) strlcpy(dsname, name, atp - name + 1);
1258 		}
1259 		if (error == 0)
1260 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1261 		if (error == 0) {
1262 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1263 			error = dsl_dataset_snapshot_check_impl(ds,
1264 			    atp + 1, tx, B_FALSE, 0, NULL);
1265 			dsl_dataset_rele(ds, FTAG);
1266 		}
1267 
1268 		if (error != 0) {
1269 			if (ddsa->ddsa_errors != NULL) {
1270 				fnvlist_add_int32(ddsa->ddsa_errors,
1271 				    name, error);
1272 			}
1273 			rv = error;
1274 		}
1275 	}
1276 
1277 	return (rv);
1278 }
1279 
1280 void
1281 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1282     dmu_tx_t *tx)
1283 {
1284 	static zil_header_t zero_zil;
1285 
1286 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1287 	dmu_buf_t *dbuf;
1288 	dsl_dataset_phys_t *dsphys;
1289 	uint64_t dsobj, crtxg;
1290 	objset_t *mos = dp->dp_meta_objset;
1291 	objset_t *os;
1292 
1293 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1294 
1295 	/*
1296 	 * If we are on an old pool, the zil must not be active, in which
1297 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1298 	 */
1299 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1300 	    dmu_objset_from_ds(ds, &os) != 0 ||
1301 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1302 	    sizeof (zero_zil)) == 0);
1303 
1304 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1305 
1306 	/*
1307 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1308 	 */
1309 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1310 		crtxg = 1;
1311 	else
1312 		crtxg = tx->tx_txg;
1313 
1314 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1315 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1316 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1317 	dmu_buf_will_dirty(dbuf, tx);
1318 	dsphys = dbuf->db_data;
1319 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1320 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1321 	dsphys->ds_fsid_guid = unique_create();
1322 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1323 	    sizeof (dsphys->ds_guid));
1324 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1325 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1326 	dsphys->ds_next_snap_obj = ds->ds_object;
1327 	dsphys->ds_num_children = 1;
1328 	dsphys->ds_creation_time = gethrestime_sec();
1329 	dsphys->ds_creation_txg = crtxg;
1330 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1331 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1332 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1333 	dsphys->ds_uncompressed_bytes =
1334 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1335 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1336 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1337 	dmu_buf_rele(dbuf, FTAG);
1338 
1339 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1340 		if (ds->ds_feature_inuse[f])
1341 			dsl_dataset_activate_feature(dsobj, f, tx);
1342 	}
1343 
1344 	ASSERT3U(ds->ds_prev != 0, ==,
1345 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1346 	if (ds->ds_prev) {
1347 		uint64_t next_clones_obj =
1348 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1349 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1350 		    ds->ds_object ||
1351 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1352 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1353 		    ds->ds_object) {
1354 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1355 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1356 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1357 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1358 		} else if (next_clones_obj != 0) {
1359 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1360 			    dsphys->ds_next_snap_obj, tx);
1361 			VERIFY0(zap_add_int(mos,
1362 			    next_clones_obj, dsobj, tx));
1363 		}
1364 	}
1365 
1366 	/*
1367 	 * If we have a reference-reservation on this dataset, we will
1368 	 * need to increase the amount of refreservation being charged
1369 	 * since our unique space is going to zero.
1370 	 */
1371 	if (ds->ds_reserved) {
1372 		int64_t delta;
1373 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1374 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1375 		    ds->ds_reserved);
1376 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1377 		    delta, 0, 0, tx);
1378 	}
1379 
1380 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1381 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1382 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1383 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1384 	dsl_deadlist_close(&ds->ds_deadlist);
1385 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1386 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1387 	dsl_deadlist_add_key(&ds->ds_deadlist,
1388 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1389 
1390 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1391 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1392 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1393 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1394 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1395 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1396 
1397 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1398 	    snapname, 8, 1, &dsobj, tx));
1399 
1400 	if (ds->ds_prev)
1401 		dsl_dataset_rele(ds->ds_prev, ds);
1402 	VERIFY0(dsl_dataset_hold_obj(dp,
1403 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1404 
1405 	dsl_scan_ds_snapshotted(ds, tx);
1406 
1407 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1408 
1409 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1410 }
1411 
1412 static void
1413 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1414 {
1415 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1416 	dsl_pool_t *dp = dmu_tx_pool(tx);
1417 	nvpair_t *pair;
1418 
1419 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1420 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1421 		dsl_dataset_t *ds;
1422 		char *name, *atp;
1423 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1424 
1425 		name = nvpair_name(pair);
1426 		atp = strchr(name, '@');
1427 		(void) strlcpy(dsname, name, atp - name + 1);
1428 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1429 
1430 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1431 		if (ddsa->ddsa_props != NULL) {
1432 			dsl_props_set_sync_impl(ds->ds_prev,
1433 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1434 		}
1435 		dsl_dataset_rele(ds, FTAG);
1436 	}
1437 }
1438 
1439 /*
1440  * The snapshots must all be in the same pool.
1441  * All-or-nothing: if there are any failures, nothing will be modified.
1442  */
1443 int
1444 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1445 {
1446 	dsl_dataset_snapshot_arg_t ddsa;
1447 	nvpair_t *pair;
1448 	boolean_t needsuspend;
1449 	int error;
1450 	spa_t *spa;
1451 	char *firstname;
1452 	nvlist_t *suspended = NULL;
1453 
1454 	pair = nvlist_next_nvpair(snaps, NULL);
1455 	if (pair == NULL)
1456 		return (0);
1457 	firstname = nvpair_name(pair);
1458 
1459 	error = spa_open(firstname, &spa, FTAG);
1460 	if (error != 0)
1461 		return (error);
1462 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1463 	spa_close(spa, FTAG);
1464 
1465 	if (needsuspend) {
1466 		suspended = fnvlist_alloc();
1467 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1468 		    pair = nvlist_next_nvpair(snaps, pair)) {
1469 			char fsname[ZFS_MAX_DATASET_NAME_LEN];
1470 			char *snapname = nvpair_name(pair);
1471 			char *atp;
1472 			void *cookie;
1473 
1474 			atp = strchr(snapname, '@');
1475 			if (atp == NULL) {
1476 				error = SET_ERROR(EINVAL);
1477 				break;
1478 			}
1479 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1480 
1481 			error = zil_suspend(fsname, &cookie);
1482 			if (error != 0)
1483 				break;
1484 			fnvlist_add_uint64(suspended, fsname,
1485 			    (uintptr_t)cookie);
1486 		}
1487 	}
1488 
1489 	ddsa.ddsa_snaps = snaps;
1490 	ddsa.ddsa_props = props;
1491 	ddsa.ddsa_errors = errors;
1492 	ddsa.ddsa_cr = CRED();
1493 
1494 	if (error == 0) {
1495 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1496 		    dsl_dataset_snapshot_sync, &ddsa,
1497 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1498 	}
1499 
1500 	if (suspended != NULL) {
1501 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1502 		    pair = nvlist_next_nvpair(suspended, pair)) {
1503 			zil_resume((void *)(uintptr_t)
1504 			    fnvpair_value_uint64(pair));
1505 		}
1506 		fnvlist_free(suspended);
1507 	}
1508 
1509 	return (error);
1510 }
1511 
1512 typedef struct dsl_dataset_snapshot_tmp_arg {
1513 	const char *ddsta_fsname;
1514 	const char *ddsta_snapname;
1515 	minor_t ddsta_cleanup_minor;
1516 	const char *ddsta_htag;
1517 } dsl_dataset_snapshot_tmp_arg_t;
1518 
1519 static int
1520 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1521 {
1522 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1523 	dsl_pool_t *dp = dmu_tx_pool(tx);
1524 	dsl_dataset_t *ds;
1525 	int error;
1526 
1527 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1528 	if (error != 0)
1529 		return (error);
1530 
1531 	/* NULL cred means no limit check for tmp snapshot */
1532 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1533 	    tx, B_FALSE, 0, NULL);
1534 	if (error != 0) {
1535 		dsl_dataset_rele(ds, FTAG);
1536 		return (error);
1537 	}
1538 
1539 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1540 		dsl_dataset_rele(ds, FTAG);
1541 		return (SET_ERROR(ENOTSUP));
1542 	}
1543 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1544 	    B_TRUE, tx);
1545 	if (error != 0) {
1546 		dsl_dataset_rele(ds, FTAG);
1547 		return (error);
1548 	}
1549 
1550 	dsl_dataset_rele(ds, FTAG);
1551 	return (0);
1552 }
1553 
1554 static void
1555 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1556 {
1557 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1558 	dsl_pool_t *dp = dmu_tx_pool(tx);
1559 	dsl_dataset_t *ds;
1560 
1561 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1562 
1563 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1564 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1565 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1566 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1567 
1568 	dsl_dataset_rele(ds, FTAG);
1569 }
1570 
1571 int
1572 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1573     minor_t cleanup_minor, const char *htag)
1574 {
1575 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1576 	int error;
1577 	spa_t *spa;
1578 	boolean_t needsuspend;
1579 	void *cookie;
1580 
1581 	ddsta.ddsta_fsname = fsname;
1582 	ddsta.ddsta_snapname = snapname;
1583 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1584 	ddsta.ddsta_htag = htag;
1585 
1586 	error = spa_open(fsname, &spa, FTAG);
1587 	if (error != 0)
1588 		return (error);
1589 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1590 	spa_close(spa, FTAG);
1591 
1592 	if (needsuspend) {
1593 		error = zil_suspend(fsname, &cookie);
1594 		if (error != 0)
1595 			return (error);
1596 	}
1597 
1598 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1599 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1600 
1601 	if (needsuspend)
1602 		zil_resume(cookie);
1603 	return (error);
1604 }
1605 
1606 
1607 void
1608 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1609 {
1610 	ASSERT(dmu_tx_is_syncing(tx));
1611 	ASSERT(ds->ds_objset != NULL);
1612 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1613 
1614 	/*
1615 	 * in case we had to change ds_fsid_guid when we opened it,
1616 	 * sync it out now.
1617 	 */
1618 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1619 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1620 
1621 	if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1622 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1623 		    ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1624 		    &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1625 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1626 		    ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1627 		    &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1628 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1629 		    ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1630 		    &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1631 		ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1632 		ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1633 		ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1634 	}
1635 
1636 	dmu_objset_sync(ds->ds_objset, zio, tx);
1637 
1638 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1639 		if (ds->ds_feature_activation_needed[f]) {
1640 			if (ds->ds_feature_inuse[f])
1641 				continue;
1642 			dsl_dataset_activate_feature(ds->ds_object, f, tx);
1643 			ds->ds_feature_inuse[f] = B_TRUE;
1644 		}
1645 	}
1646 }
1647 
1648 static void
1649 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1650 {
1651 	uint64_t count = 0;
1652 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1653 	zap_cursor_t zc;
1654 	zap_attribute_t za;
1655 	nvlist_t *propval = fnvlist_alloc();
1656 	nvlist_t *val = fnvlist_alloc();
1657 
1658 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1659 
1660 	/*
1661 	 * There may be missing entries in ds_next_clones_obj
1662 	 * due to a bug in a previous version of the code.
1663 	 * Only trust it if it has the right number of entries.
1664 	 */
1665 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1666 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1667 		    &count));
1668 	}
1669 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1670 		goto fail;
1671 	for (zap_cursor_init(&zc, mos,
1672 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1673 	    zap_cursor_retrieve(&zc, &za) == 0;
1674 	    zap_cursor_advance(&zc)) {
1675 		dsl_dataset_t *clone;
1676 		char buf[ZFS_MAX_DATASET_NAME_LEN];
1677 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1678 		    za.za_first_integer, FTAG, &clone));
1679 		dsl_dir_name(clone->ds_dir, buf);
1680 		fnvlist_add_boolean(val, buf);
1681 		dsl_dataset_rele(clone, FTAG);
1682 	}
1683 	zap_cursor_fini(&zc);
1684 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1685 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1686 fail:
1687 	nvlist_free(val);
1688 	nvlist_free(propval);
1689 }
1690 
1691 static void
1692 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1693 {
1694 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1695 
1696 	if (dsl_dataset_has_resume_receive_state(ds)) {
1697 		char *str;
1698 		void *packed;
1699 		uint8_t *compressed;
1700 		uint64_t val;
1701 		nvlist_t *token_nv = fnvlist_alloc();
1702 		size_t packed_size, compressed_size;
1703 
1704 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1705 		    DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1706 			fnvlist_add_uint64(token_nv, "fromguid", val);
1707 		}
1708 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1709 		    DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1710 			fnvlist_add_uint64(token_nv, "object", val);
1711 		}
1712 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1713 		    DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1714 			fnvlist_add_uint64(token_nv, "offset", val);
1715 		}
1716 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1717 		    DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1718 			fnvlist_add_uint64(token_nv, "bytes", val);
1719 		}
1720 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1721 		    DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1722 			fnvlist_add_uint64(token_nv, "toguid", val);
1723 		}
1724 		char buf[256];
1725 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1726 		    DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1727 			fnvlist_add_string(token_nv, "toname", buf);
1728 		}
1729 		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1730 		    DS_FIELD_RESUME_EMBEDOK) == 0) {
1731 			fnvlist_add_boolean(token_nv, "embedok");
1732 		}
1733 		packed = fnvlist_pack(token_nv, &packed_size);
1734 		fnvlist_free(token_nv);
1735 		compressed = kmem_alloc(packed_size, KM_SLEEP);
1736 
1737 		compressed_size = gzip_compress(packed, compressed,
1738 		    packed_size, packed_size, 6);
1739 
1740 		zio_cksum_t cksum;
1741 		fletcher_4_native(compressed, compressed_size, &cksum);
1742 
1743 		str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1744 		for (int i = 0; i < compressed_size; i++) {
1745 			(void) sprintf(str + i * 2, "%02x", compressed[i]);
1746 		}
1747 		str[compressed_size * 2] = '\0';
1748 		char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1749 		    ZFS_SEND_RESUME_TOKEN_VERSION,
1750 		    (longlong_t)cksum.zc_word[0],
1751 		    (longlong_t)packed_size, str);
1752 		dsl_prop_nvlist_add_string(nv,
1753 		    ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1754 		kmem_free(packed, packed_size);
1755 		kmem_free(str, compressed_size * 2 + 1);
1756 		kmem_free(compressed, packed_size);
1757 		strfree(propval);
1758 	}
1759 }
1760 
1761 void
1762 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1763 {
1764 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1765 	uint64_t refd, avail, uobjs, aobjs, ratio;
1766 
1767 	ASSERT(dsl_pool_config_held(dp));
1768 
1769 	ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1770 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1771 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1772 
1773 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1774 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1775 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1776 
1777 	if (ds->ds_is_snapshot) {
1778 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1779 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1780 		    dsl_dataset_phys(ds)->ds_unique_bytes);
1781 		get_clones_stat(ds, nv);
1782 	} else {
1783 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1784 			char buf[ZFS_MAX_DATASET_NAME_LEN];
1785 			dsl_dataset_name(ds->ds_prev, buf);
1786 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1787 		}
1788 
1789 		dsl_dir_stats(ds->ds_dir, nv);
1790 	}
1791 
1792 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1793 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1794 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1795 
1796 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1797 	    dsl_dataset_phys(ds)->ds_creation_time);
1798 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1799 	    dsl_dataset_phys(ds)->ds_creation_txg);
1800 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1801 	    ds->ds_quota);
1802 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1803 	    ds->ds_reserved);
1804 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1805 	    dsl_dataset_phys(ds)->ds_guid);
1806 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1807 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1808 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1809 	    ds->ds_object);
1810 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1811 	    ds->ds_userrefs);
1812 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1813 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1814 
1815 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1816 		uint64_t written, comp, uncomp;
1817 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1818 		dsl_dataset_t *prev;
1819 
1820 		int err = dsl_dataset_hold_obj(dp,
1821 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1822 		if (err == 0) {
1823 			err = dsl_dataset_space_written(prev, ds, &written,
1824 			    &comp, &uncomp);
1825 			dsl_dataset_rele(prev, FTAG);
1826 			if (err == 0) {
1827 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1828 				    written);
1829 			}
1830 		}
1831 	}
1832 
1833 	if (!dsl_dataset_is_snapshot(ds)) {
1834 		/*
1835 		 * A failed "newfs" (e.g. full) resumable receive leaves
1836 		 * the stats set on this dataset.  Check here for the prop.
1837 		 */
1838 		get_receive_resume_stats(ds, nv);
1839 
1840 		/*
1841 		 * A failed incremental resumable receive leaves the
1842 		 * stats set on our child named "%recv".  Check the child
1843 		 * for the prop.
1844 		 */
1845 		/* 6 extra bytes for /%recv */
1846 		char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1847 		dsl_dataset_t *recv_ds;
1848 		dsl_dataset_name(ds, recvname);
1849 		if (strlcat(recvname, "/", sizeof (recvname)) <
1850 		    sizeof (recvname) &&
1851 		    strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1852 		    sizeof (recvname) &&
1853 		    dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1854 			get_receive_resume_stats(recv_ds, nv);
1855 			dsl_dataset_rele(recv_ds, FTAG);
1856 		}
1857 	}
1858 }
1859 
1860 void
1861 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1862 {
1863 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1864 	ASSERT(dsl_pool_config_held(dp));
1865 
1866 	stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1867 	stat->dds_inconsistent =
1868 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1869 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1870 	stat->dds_origin[0] = '\0';
1871 	if (ds->ds_is_snapshot) {
1872 		stat->dds_is_snapshot = B_TRUE;
1873 		stat->dds_num_clones =
1874 		    dsl_dataset_phys(ds)->ds_num_children - 1;
1875 	} else {
1876 		stat->dds_is_snapshot = B_FALSE;
1877 		stat->dds_num_clones = 0;
1878 
1879 		if (dsl_dir_is_clone(ds->ds_dir)) {
1880 			dsl_dataset_t *ods;
1881 
1882 			VERIFY0(dsl_dataset_hold_obj(dp,
1883 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1884 			    FTAG, &ods));
1885 			dsl_dataset_name(ods, stat->dds_origin);
1886 			dsl_dataset_rele(ods, FTAG);
1887 		}
1888 	}
1889 }
1890 
1891 uint64_t
1892 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1893 {
1894 	return (ds->ds_fsid_guid);
1895 }
1896 
1897 void
1898 dsl_dataset_space(dsl_dataset_t *ds,
1899     uint64_t *refdbytesp, uint64_t *availbytesp,
1900     uint64_t *usedobjsp, uint64_t *availobjsp)
1901 {
1902 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1903 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1904 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1905 		*availbytesp +=
1906 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1907 	if (ds->ds_quota != 0) {
1908 		/*
1909 		 * Adjust available bytes according to refquota
1910 		 */
1911 		if (*refdbytesp < ds->ds_quota)
1912 			*availbytesp = MIN(*availbytesp,
1913 			    ds->ds_quota - *refdbytesp);
1914 		else
1915 			*availbytesp = 0;
1916 	}
1917 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1918 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1919 }
1920 
1921 boolean_t
1922 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1923 {
1924 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1925 
1926 	ASSERT(dsl_pool_config_held(dp));
1927 	if (snap == NULL)
1928 		return (B_FALSE);
1929 	if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1930 	    dsl_dataset_phys(snap)->ds_creation_txg) {
1931 		objset_t *os, *os_snap;
1932 		/*
1933 		 * It may be that only the ZIL differs, because it was
1934 		 * reset in the head.  Don't count that as being
1935 		 * modified.
1936 		 */
1937 		if (dmu_objset_from_ds(ds, &os) != 0)
1938 			return (B_TRUE);
1939 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1940 			return (B_TRUE);
1941 		return (bcmp(&os->os_phys->os_meta_dnode,
1942 		    &os_snap->os_phys->os_meta_dnode,
1943 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1944 	}
1945 	return (B_FALSE);
1946 }
1947 
1948 typedef struct dsl_dataset_rename_snapshot_arg {
1949 	const char *ddrsa_fsname;
1950 	const char *ddrsa_oldsnapname;
1951 	const char *ddrsa_newsnapname;
1952 	boolean_t ddrsa_recursive;
1953 	dmu_tx_t *ddrsa_tx;
1954 } dsl_dataset_rename_snapshot_arg_t;
1955 
1956 /* ARGSUSED */
1957 static int
1958 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1959     dsl_dataset_t *hds, void *arg)
1960 {
1961 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1962 	int error;
1963 	uint64_t val;
1964 
1965 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1966 	if (error != 0) {
1967 		/* ignore nonexistent snapshots */
1968 		return (error == ENOENT ? 0 : error);
1969 	}
1970 
1971 	/* new name should not exist */
1972 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1973 	if (error == 0)
1974 		error = SET_ERROR(EEXIST);
1975 	else if (error == ENOENT)
1976 		error = 0;
1977 
1978 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1979 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1980 	    strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
1981 		error = SET_ERROR(ENAMETOOLONG);
1982 
1983 	return (error);
1984 }
1985 
1986 static int
1987 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1988 {
1989 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1990 	dsl_pool_t *dp = dmu_tx_pool(tx);
1991 	dsl_dataset_t *hds;
1992 	int error;
1993 
1994 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1995 	if (error != 0)
1996 		return (error);
1997 
1998 	if (ddrsa->ddrsa_recursive) {
1999 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2000 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
2001 		    DS_FIND_CHILDREN);
2002 	} else {
2003 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2004 	}
2005 	dsl_dataset_rele(hds, FTAG);
2006 	return (error);
2007 }
2008 
2009 static int
2010 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2011     dsl_dataset_t *hds, void *arg)
2012 {
2013 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2014 	dsl_dataset_t *ds;
2015 	uint64_t val;
2016 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
2017 	int error;
2018 
2019 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2020 	ASSERT(error == 0 || error == ENOENT);
2021 	if (error == ENOENT) {
2022 		/* ignore nonexistent snapshots */
2023 		return (0);
2024 	}
2025 
2026 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2027 
2028 	/* log before we change the name */
2029 	spa_history_log_internal_ds(ds, "rename", tx,
2030 	    "-> @%s", ddrsa->ddrsa_newsnapname);
2031 
2032 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2033 	    B_FALSE));
2034 	mutex_enter(&ds->ds_lock);
2035 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2036 	mutex_exit(&ds->ds_lock);
2037 	VERIFY0(zap_add(dp->dp_meta_objset,
2038 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2039 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2040 
2041 	dsl_dataset_rele(ds, FTAG);
2042 	return (0);
2043 }
2044 
2045 static void
2046 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2047 {
2048 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2049 	dsl_pool_t *dp = dmu_tx_pool(tx);
2050 	dsl_dataset_t *hds;
2051 
2052 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2053 	ddrsa->ddrsa_tx = tx;
2054 	if (ddrsa->ddrsa_recursive) {
2055 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2056 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2057 		    DS_FIND_CHILDREN));
2058 	} else {
2059 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2060 	}
2061 	dsl_dataset_rele(hds, FTAG);
2062 }
2063 
2064 int
2065 dsl_dataset_rename_snapshot(const char *fsname,
2066     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2067 {
2068 	dsl_dataset_rename_snapshot_arg_t ddrsa;
2069 
2070 	ddrsa.ddrsa_fsname = fsname;
2071 	ddrsa.ddrsa_oldsnapname = oldsnapname;
2072 	ddrsa.ddrsa_newsnapname = newsnapname;
2073 	ddrsa.ddrsa_recursive = recursive;
2074 
2075 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2076 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
2077 	    1, ZFS_SPACE_CHECK_RESERVED));
2078 }
2079 
2080 /*
2081  * If we're doing an ownership handoff, we need to make sure that there is
2082  * only one long hold on the dataset.  We're not allowed to change anything here
2083  * so we don't permanently release the long hold or regular hold here.  We want
2084  * to do this only when syncing to avoid the dataset unexpectedly going away
2085  * when we release the long hold.
2086  */
2087 static int
2088 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2089 {
2090 	boolean_t held;
2091 
2092 	if (!dmu_tx_is_syncing(tx))
2093 		return (0);
2094 
2095 	if (owner != NULL) {
2096 		VERIFY3P(ds->ds_owner, ==, owner);
2097 		dsl_dataset_long_rele(ds, owner);
2098 	}
2099 
2100 	held = dsl_dataset_long_held(ds);
2101 
2102 	if (owner != NULL)
2103 		dsl_dataset_long_hold(ds, owner);
2104 
2105 	if (held)
2106 		return (SET_ERROR(EBUSY));
2107 
2108 	return (0);
2109 }
2110 
2111 typedef struct dsl_dataset_rollback_arg {
2112 	const char *ddra_fsname;
2113 	void *ddra_owner;
2114 	nvlist_t *ddra_result;
2115 } dsl_dataset_rollback_arg_t;
2116 
2117 static int
2118 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2119 {
2120 	dsl_dataset_rollback_arg_t *ddra = arg;
2121 	dsl_pool_t *dp = dmu_tx_pool(tx);
2122 	dsl_dataset_t *ds;
2123 	int64_t unused_refres_delta;
2124 	int error;
2125 
2126 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2127 	if (error != 0)
2128 		return (error);
2129 
2130 	/* must not be a snapshot */
2131 	if (ds->ds_is_snapshot) {
2132 		dsl_dataset_rele(ds, FTAG);
2133 		return (SET_ERROR(EINVAL));
2134 	}
2135 
2136 	/* must have a most recent snapshot */
2137 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2138 		dsl_dataset_rele(ds, FTAG);
2139 		return (SET_ERROR(EINVAL));
2140 	}
2141 
2142 	/* must not have any bookmarks after the most recent snapshot */
2143 	nvlist_t *proprequest = fnvlist_alloc();
2144 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2145 	nvlist_t *bookmarks = fnvlist_alloc();
2146 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2147 	fnvlist_free(proprequest);
2148 	if (error != 0)
2149 		return (error);
2150 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2151 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2152 		nvlist_t *valuenv =
2153 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2154 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
2155 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2156 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2157 			fnvlist_free(bookmarks);
2158 			dsl_dataset_rele(ds, FTAG);
2159 			return (SET_ERROR(EEXIST));
2160 		}
2161 	}
2162 	fnvlist_free(bookmarks);
2163 
2164 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2165 	if (error != 0) {
2166 		dsl_dataset_rele(ds, FTAG);
2167 		return (error);
2168 	}
2169 
2170 	/*
2171 	 * Check if the snap we are rolling back to uses more than
2172 	 * the refquota.
2173 	 */
2174 	if (ds->ds_quota != 0 &&
2175 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2176 		dsl_dataset_rele(ds, FTAG);
2177 		return (SET_ERROR(EDQUOT));
2178 	}
2179 
2180 	/*
2181 	 * When we do the clone swap, we will temporarily use more space
2182 	 * due to the refreservation (the head will no longer have any
2183 	 * unique space, so the entire amount of the refreservation will need
2184 	 * to be free).  We will immediately destroy the clone, freeing
2185 	 * this space, but the freeing happens over many txg's.
2186 	 */
2187 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2188 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2189 
2190 	if (unused_refres_delta > 0 &&
2191 	    unused_refres_delta >
2192 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2193 		dsl_dataset_rele(ds, FTAG);
2194 		return (SET_ERROR(ENOSPC));
2195 	}
2196 
2197 	dsl_dataset_rele(ds, FTAG);
2198 	return (0);
2199 }
2200 
2201 static void
2202 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2203 {
2204 	dsl_dataset_rollback_arg_t *ddra = arg;
2205 	dsl_pool_t *dp = dmu_tx_pool(tx);
2206 	dsl_dataset_t *ds, *clone;
2207 	uint64_t cloneobj;
2208 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2209 
2210 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2211 
2212 	dsl_dataset_name(ds->ds_prev, namebuf);
2213 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2214 
2215 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2216 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2217 
2218 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2219 
2220 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2221 	dsl_dataset_zero_zil(ds, tx);
2222 
2223 	dsl_destroy_head_sync_impl(clone, tx);
2224 
2225 	dsl_dataset_rele(clone, FTAG);
2226 	dsl_dataset_rele(ds, FTAG);
2227 }
2228 
2229 /*
2230  * Rolls back the given filesystem or volume to the most recent snapshot.
2231  * The name of the most recent snapshot will be returned under key "target"
2232  * in the result nvlist.
2233  *
2234  * If owner != NULL:
2235  * - The existing dataset MUST be owned by the specified owner at entry
2236  * - Upon return, dataset will still be held by the same owner, whether we
2237  *   succeed or not.
2238  *
2239  * This mode is required any time the existing filesystem is mounted.  See
2240  * notes above zfs_suspend_fs() for further details.
2241  */
2242 int
2243 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2244 {
2245 	dsl_dataset_rollback_arg_t ddra;
2246 
2247 	ddra.ddra_fsname = fsname;
2248 	ddra.ddra_owner = owner;
2249 	ddra.ddra_result = result;
2250 
2251 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2252 	    dsl_dataset_rollback_sync, &ddra,
2253 	    1, ZFS_SPACE_CHECK_RESERVED));
2254 }
2255 
2256 struct promotenode {
2257 	list_node_t link;
2258 	dsl_dataset_t *ds;
2259 };
2260 
2261 typedef struct dsl_dataset_promote_arg {
2262 	const char *ddpa_clonename;
2263 	dsl_dataset_t *ddpa_clone;
2264 	list_t shared_snaps, origin_snaps, clone_snaps;
2265 	dsl_dataset_t *origin_origin; /* origin of the origin */
2266 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2267 	char *err_ds;
2268 	cred_t *cr;
2269 } dsl_dataset_promote_arg_t;
2270 
2271 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2272 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2273     void *tag);
2274 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2275 
2276 static int
2277 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2278 {
2279 	dsl_dataset_promote_arg_t *ddpa = arg;
2280 	dsl_pool_t *dp = dmu_tx_pool(tx);
2281 	dsl_dataset_t *hds;
2282 	struct promotenode *snap;
2283 	dsl_dataset_t *origin_ds;
2284 	int err;
2285 	uint64_t unused;
2286 	uint64_t ss_mv_cnt;
2287 	size_t max_snap_len;
2288 
2289 	err = promote_hold(ddpa, dp, FTAG);
2290 	if (err != 0)
2291 		return (err);
2292 
2293 	hds = ddpa->ddpa_clone;
2294 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2295 
2296 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2297 		promote_rele(ddpa, FTAG);
2298 		return (SET_ERROR(EXDEV));
2299 	}
2300 
2301 	/*
2302 	 * Compute and check the amount of space to transfer.  Since this is
2303 	 * so expensive, don't do the preliminary check.
2304 	 */
2305 	if (!dmu_tx_is_syncing(tx)) {
2306 		promote_rele(ddpa, FTAG);
2307 		return (0);
2308 	}
2309 
2310 	snap = list_head(&ddpa->shared_snaps);
2311 	origin_ds = snap->ds;
2312 
2313 	/* compute origin's new unique space */
2314 	snap = list_tail(&ddpa->clone_snaps);
2315 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2316 	    origin_ds->ds_object);
2317 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2318 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2319 	    &ddpa->unique, &unused, &unused);
2320 
2321 	/*
2322 	 * Walk the snapshots that we are moving
2323 	 *
2324 	 * Compute space to transfer.  Consider the incremental changes
2325 	 * to used by each snapshot:
2326 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2327 	 * So each snapshot gave birth to:
2328 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2329 	 * So a sequence would look like:
2330 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2331 	 * Which simplifies to:
2332 	 * uN + kN + kN-1 + ... + k1 + k0
2333 	 * Note however, if we stop before we reach the ORIGIN we get:
2334 	 * uN + kN + kN-1 + ... + kM - uM-1
2335 	 */
2336 	ss_mv_cnt = 0;
2337 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2338 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2339 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2340 	for (snap = list_head(&ddpa->shared_snaps); snap;
2341 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2342 		uint64_t val, dlused, dlcomp, dluncomp;
2343 		dsl_dataset_t *ds = snap->ds;
2344 
2345 		ss_mv_cnt++;
2346 
2347 		/*
2348 		 * If there are long holds, we won't be able to evict
2349 		 * the objset.
2350 		 */
2351 		if (dsl_dataset_long_held(ds)) {
2352 			err = SET_ERROR(EBUSY);
2353 			goto out;
2354 		}
2355 
2356 		/* Check that the snapshot name does not conflict */
2357 		VERIFY0(dsl_dataset_get_snapname(ds));
2358 		if (strlen(ds->ds_snapname) >= max_snap_len) {
2359 			err = SET_ERROR(ENAMETOOLONG);
2360 			goto out;
2361 		}
2362 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2363 		if (err == 0) {
2364 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2365 			err = SET_ERROR(EEXIST);
2366 			goto out;
2367 		}
2368 		if (err != ENOENT)
2369 			goto out;
2370 
2371 		/* The very first snapshot does not have a deadlist */
2372 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2373 			continue;
2374 
2375 		dsl_deadlist_space(&ds->ds_deadlist,
2376 		    &dlused, &dlcomp, &dluncomp);
2377 		ddpa->used += dlused;
2378 		ddpa->comp += dlcomp;
2379 		ddpa->uncomp += dluncomp;
2380 	}
2381 
2382 	/*
2383 	 * If we are a clone of a clone then we never reached ORIGIN,
2384 	 * so we need to subtract out the clone origin's used space.
2385 	 */
2386 	if (ddpa->origin_origin) {
2387 		ddpa->used -=
2388 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2389 		ddpa->comp -=
2390 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2391 		ddpa->uncomp -=
2392 		    dsl_dataset_phys(ddpa->origin_origin)->
2393 		    ds_uncompressed_bytes;
2394 	}
2395 
2396 	/* Check that there is enough space and limit headroom here */
2397 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2398 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2399 	if (err != 0)
2400 		goto out;
2401 
2402 	/*
2403 	 * Compute the amounts of space that will be used by snapshots
2404 	 * after the promotion (for both origin and clone).  For each,
2405 	 * it is the amount of space that will be on all of their
2406 	 * deadlists (that was not born before their new origin).
2407 	 */
2408 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2409 		uint64_t space;
2410 
2411 		/*
2412 		 * Note, typically this will not be a clone of a clone,
2413 		 * so dd_origin_txg will be < TXG_INITIAL, so
2414 		 * these snaplist_space() -> dsl_deadlist_space_range()
2415 		 * calls will be fast because they do not have to
2416 		 * iterate over all bps.
2417 		 */
2418 		snap = list_head(&ddpa->origin_snaps);
2419 		err = snaplist_space(&ddpa->shared_snaps,
2420 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2421 		if (err != 0)
2422 			goto out;
2423 
2424 		err = snaplist_space(&ddpa->clone_snaps,
2425 		    snap->ds->ds_dir->dd_origin_txg, &space);
2426 		if (err != 0)
2427 			goto out;
2428 		ddpa->cloneusedsnap += space;
2429 	}
2430 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2431 	    DD_FLAG_USED_BREAKDOWN) {
2432 		err = snaplist_space(&ddpa->origin_snaps,
2433 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2434 		    &ddpa->originusedsnap);
2435 		if (err != 0)
2436 			goto out;
2437 	}
2438 
2439 out:
2440 	promote_rele(ddpa, FTAG);
2441 	return (err);
2442 }
2443 
2444 static void
2445 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2446 {
2447 	dsl_dataset_promote_arg_t *ddpa = arg;
2448 	dsl_pool_t *dp = dmu_tx_pool(tx);
2449 	dsl_dataset_t *hds;
2450 	struct promotenode *snap;
2451 	dsl_dataset_t *origin_ds;
2452 	dsl_dataset_t *origin_head;
2453 	dsl_dir_t *dd;
2454 	dsl_dir_t *odd = NULL;
2455 	uint64_t oldnext_obj;
2456 	int64_t delta;
2457 
2458 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2459 	hds = ddpa->ddpa_clone;
2460 
2461 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2462 
2463 	snap = list_head(&ddpa->shared_snaps);
2464 	origin_ds = snap->ds;
2465 	dd = hds->ds_dir;
2466 
2467 	snap = list_head(&ddpa->origin_snaps);
2468 	origin_head = snap->ds;
2469 
2470 	/*
2471 	 * We need to explicitly open odd, since origin_ds's dd will be
2472 	 * changing.
2473 	 */
2474 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2475 	    NULL, FTAG, &odd));
2476 
2477 	/* change origin's next snap */
2478 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2479 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2480 	snap = list_tail(&ddpa->clone_snaps);
2481 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2482 	    origin_ds->ds_object);
2483 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2484 
2485 	/* change the origin's next clone */
2486 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2487 		dsl_dataset_remove_from_next_clones(origin_ds,
2488 		    snap->ds->ds_object, tx);
2489 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2490 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2491 		    oldnext_obj, tx));
2492 	}
2493 
2494 	/* change origin */
2495 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2496 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2497 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2498 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2499 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2500 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2501 	origin_head->ds_dir->dd_origin_txg =
2502 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2503 
2504 	/* change dd_clone entries */
2505 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2506 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2507 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2508 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2509 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2510 		    hds->ds_object, tx));
2511 
2512 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2513 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2514 		    origin_head->ds_object, tx));
2515 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2516 			dsl_dir_phys(dd)->dd_clones =
2517 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2518 			    DMU_OT_NONE, 0, tx);
2519 		}
2520 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2521 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2522 	}
2523 
2524 	/* move snapshots to this dir */
2525 	for (snap = list_head(&ddpa->shared_snaps); snap;
2526 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2527 		dsl_dataset_t *ds = snap->ds;
2528 
2529 		/*
2530 		 * Property callbacks are registered to a particular
2531 		 * dsl_dir.  Since ours is changing, evict the objset
2532 		 * so that they will be unregistered from the old dsl_dir.
2533 		 */
2534 		if (ds->ds_objset) {
2535 			dmu_objset_evict(ds->ds_objset);
2536 			ds->ds_objset = NULL;
2537 		}
2538 
2539 		/* move snap name entry */
2540 		VERIFY0(dsl_dataset_get_snapname(ds));
2541 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2542 		    ds->ds_snapname, tx, B_TRUE));
2543 		VERIFY0(zap_add(dp->dp_meta_objset,
2544 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2545 		    8, 1, &ds->ds_object, tx));
2546 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2547 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2548 
2549 		/* change containing dsl_dir */
2550 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2551 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2552 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2553 		ASSERT3P(ds->ds_dir, ==, odd);
2554 		dsl_dir_rele(ds->ds_dir, ds);
2555 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2556 		    NULL, ds, &ds->ds_dir));
2557 
2558 		/* move any clone references */
2559 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2560 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2561 			zap_cursor_t zc;
2562 			zap_attribute_t za;
2563 
2564 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2565 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
2566 			    zap_cursor_retrieve(&zc, &za) == 0;
2567 			    zap_cursor_advance(&zc)) {
2568 				dsl_dataset_t *cnds;
2569 				uint64_t o;
2570 
2571 				if (za.za_first_integer == oldnext_obj) {
2572 					/*
2573 					 * We've already moved the
2574 					 * origin's reference.
2575 					 */
2576 					continue;
2577 				}
2578 
2579 				VERIFY0(dsl_dataset_hold_obj(dp,
2580 				    za.za_first_integer, FTAG, &cnds));
2581 				o = dsl_dir_phys(cnds->ds_dir)->
2582 				    dd_head_dataset_obj;
2583 
2584 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2585 				    dsl_dir_phys(odd)->dd_clones, o, tx));
2586 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2587 				    dsl_dir_phys(dd)->dd_clones, o, tx));
2588 				dsl_dataset_rele(cnds, FTAG);
2589 			}
2590 			zap_cursor_fini(&zc);
2591 		}
2592 
2593 		ASSERT(!dsl_prop_hascb(ds));
2594 	}
2595 
2596 	/*
2597 	 * Change space accounting.
2598 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2599 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2600 	 * is true for each of {clone,origin} independently.
2601 	 */
2602 
2603 	delta = ddpa->cloneusedsnap -
2604 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2605 	ASSERT3S(delta, >=, 0);
2606 	ASSERT3U(ddpa->used, >=, delta);
2607 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2608 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2609 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2610 
2611 	delta = ddpa->originusedsnap -
2612 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2613 	ASSERT3S(delta, <=, 0);
2614 	ASSERT3U(ddpa->used, >=, -delta);
2615 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2616 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2617 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2618 
2619 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2620 
2621 	/* log history record */
2622 	spa_history_log_internal_ds(hds, "promote", tx, "");
2623 
2624 	dsl_dir_rele(odd, FTAG);
2625 	promote_rele(ddpa, FTAG);
2626 }
2627 
2628 /*
2629  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2630  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2631  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2632  * snapshots back to this dataset's origin.
2633  */
2634 static int
2635 snaplist_make(dsl_pool_t *dp,
2636     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2637 {
2638 	uint64_t obj = last_obj;
2639 
2640 	list_create(l, sizeof (struct promotenode),
2641 	    offsetof(struct promotenode, link));
2642 
2643 	while (obj != first_obj) {
2644 		dsl_dataset_t *ds;
2645 		struct promotenode *snap;
2646 		int err;
2647 
2648 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2649 		ASSERT(err != ENOENT);
2650 		if (err != 0)
2651 			return (err);
2652 
2653 		if (first_obj == 0)
2654 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2655 
2656 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2657 		snap->ds = ds;
2658 		list_insert_tail(l, snap);
2659 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2660 	}
2661 
2662 	return (0);
2663 }
2664 
2665 static int
2666 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2667 {
2668 	struct promotenode *snap;
2669 
2670 	*spacep = 0;
2671 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2672 		uint64_t used, comp, uncomp;
2673 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2674 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2675 		*spacep += used;
2676 	}
2677 	return (0);
2678 }
2679 
2680 static void
2681 snaplist_destroy(list_t *l, void *tag)
2682 {
2683 	struct promotenode *snap;
2684 
2685 	if (l == NULL || !list_link_active(&l->list_head))
2686 		return;
2687 
2688 	while ((snap = list_tail(l)) != NULL) {
2689 		list_remove(l, snap);
2690 		dsl_dataset_rele(snap->ds, tag);
2691 		kmem_free(snap, sizeof (*snap));
2692 	}
2693 	list_destroy(l);
2694 }
2695 
2696 static int
2697 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2698 {
2699 	int error;
2700 	dsl_dir_t *dd;
2701 	struct promotenode *snap;
2702 
2703 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2704 	    &ddpa->ddpa_clone);
2705 	if (error != 0)
2706 		return (error);
2707 	dd = ddpa->ddpa_clone->ds_dir;
2708 
2709 	if (ddpa->ddpa_clone->ds_is_snapshot ||
2710 	    !dsl_dir_is_clone(dd)) {
2711 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2712 		return (SET_ERROR(EINVAL));
2713 	}
2714 
2715 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2716 	    &ddpa->shared_snaps, tag);
2717 	if (error != 0)
2718 		goto out;
2719 
2720 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2721 	    &ddpa->clone_snaps, tag);
2722 	if (error != 0)
2723 		goto out;
2724 
2725 	snap = list_head(&ddpa->shared_snaps);
2726 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2727 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2728 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2729 	    &ddpa->origin_snaps, tag);
2730 	if (error != 0)
2731 		goto out;
2732 
2733 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2734 		error = dsl_dataset_hold_obj(dp,
2735 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2736 		    tag, &ddpa->origin_origin);
2737 		if (error != 0)
2738 			goto out;
2739 	}
2740 out:
2741 	if (error != 0)
2742 		promote_rele(ddpa, tag);
2743 	return (error);
2744 }
2745 
2746 static void
2747 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2748 {
2749 	snaplist_destroy(&ddpa->shared_snaps, tag);
2750 	snaplist_destroy(&ddpa->clone_snaps, tag);
2751 	snaplist_destroy(&ddpa->origin_snaps, tag);
2752 	if (ddpa->origin_origin != NULL)
2753 		dsl_dataset_rele(ddpa->origin_origin, tag);
2754 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2755 }
2756 
2757 /*
2758  * Promote a clone.
2759  *
2760  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2761  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2762  */
2763 int
2764 dsl_dataset_promote(const char *name, char *conflsnap)
2765 {
2766 	dsl_dataset_promote_arg_t ddpa = { 0 };
2767 	uint64_t numsnaps;
2768 	int error;
2769 	objset_t *os;
2770 
2771 	/*
2772 	 * We will modify space proportional to the number of
2773 	 * snapshots.  Compute numsnaps.
2774 	 */
2775 	error = dmu_objset_hold(name, FTAG, &os);
2776 	if (error != 0)
2777 		return (error);
2778 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2779 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2780 	    &numsnaps);
2781 	dmu_objset_rele(os, FTAG);
2782 	if (error != 0)
2783 		return (error);
2784 
2785 	ddpa.ddpa_clonename = name;
2786 	ddpa.err_ds = conflsnap;
2787 	ddpa.cr = CRED();
2788 
2789 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2790 	    dsl_dataset_promote_sync, &ddpa,
2791 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2792 }
2793 
2794 int
2795 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2796     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2797 {
2798 	int64_t unused_refres_delta;
2799 
2800 	/* they should both be heads */
2801 	if (clone->ds_is_snapshot ||
2802 	    origin_head->ds_is_snapshot)
2803 		return (SET_ERROR(EINVAL));
2804 
2805 	/* if we are not forcing, the branch point should be just before them */
2806 	if (!force && clone->ds_prev != origin_head->ds_prev)
2807 		return (SET_ERROR(EINVAL));
2808 
2809 	/* clone should be the clone (unless they are unrelated) */
2810 	if (clone->ds_prev != NULL &&
2811 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2812 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2813 		return (SET_ERROR(EINVAL));
2814 
2815 	/* the clone should be a child of the origin */
2816 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2817 		return (SET_ERROR(EINVAL));
2818 
2819 	/* origin_head shouldn't be modified unless 'force' */
2820 	if (!force &&
2821 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2822 		return (SET_ERROR(ETXTBSY));
2823 
2824 	/* origin_head should have no long holds (e.g. is not mounted) */
2825 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2826 		return (SET_ERROR(EBUSY));
2827 
2828 	/* check amount of any unconsumed refreservation */
2829 	unused_refres_delta =
2830 	    (int64_t)MIN(origin_head->ds_reserved,
2831 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2832 	    (int64_t)MIN(origin_head->ds_reserved,
2833 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2834 
2835 	if (unused_refres_delta > 0 &&
2836 	    unused_refres_delta >
2837 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2838 		return (SET_ERROR(ENOSPC));
2839 
2840 	/* clone can't be over the head's refquota */
2841 	if (origin_head->ds_quota != 0 &&
2842 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
2843 	    origin_head->ds_quota)
2844 		return (SET_ERROR(EDQUOT));
2845 
2846 	return (0);
2847 }
2848 
2849 void
2850 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2851     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2852 {
2853 	dsl_pool_t *dp = dmu_tx_pool(tx);
2854 	int64_t unused_refres_delta;
2855 
2856 	ASSERT(clone->ds_reserved == 0);
2857 	ASSERT(origin_head->ds_quota == 0 ||
2858 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2859 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2860 
2861 	/*
2862 	 * Swap per-dataset feature flags.
2863 	 */
2864 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2865 		if (!(spa_feature_table[f].fi_flags &
2866 		    ZFEATURE_FLAG_PER_DATASET)) {
2867 			ASSERT(!clone->ds_feature_inuse[f]);
2868 			ASSERT(!origin_head->ds_feature_inuse[f]);
2869 			continue;
2870 		}
2871 
2872 		boolean_t clone_inuse = clone->ds_feature_inuse[f];
2873 		boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2874 
2875 		if (clone_inuse) {
2876 			dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2877 			clone->ds_feature_inuse[f] = B_FALSE;
2878 		}
2879 		if (origin_head_inuse) {
2880 			dsl_dataset_deactivate_feature(origin_head->ds_object,
2881 			    f, tx);
2882 			origin_head->ds_feature_inuse[f] = B_FALSE;
2883 		}
2884 		if (clone_inuse) {
2885 			dsl_dataset_activate_feature(origin_head->ds_object,
2886 			    f, tx);
2887 			origin_head->ds_feature_inuse[f] = B_TRUE;
2888 		}
2889 		if (origin_head_inuse) {
2890 			dsl_dataset_activate_feature(clone->ds_object, f, tx);
2891 			clone->ds_feature_inuse[f] = B_TRUE;
2892 		}
2893 	}
2894 
2895 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2896 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2897 
2898 	if (clone->ds_objset != NULL) {
2899 		dmu_objset_evict(clone->ds_objset);
2900 		clone->ds_objset = NULL;
2901 	}
2902 
2903 	if (origin_head->ds_objset != NULL) {
2904 		dmu_objset_evict(origin_head->ds_objset);
2905 		origin_head->ds_objset = NULL;
2906 	}
2907 
2908 	unused_refres_delta =
2909 	    (int64_t)MIN(origin_head->ds_reserved,
2910 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2911 	    (int64_t)MIN(origin_head->ds_reserved,
2912 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2913 
2914 	/*
2915 	 * Reset origin's unique bytes, if it exists.
2916 	 */
2917 	if (clone->ds_prev) {
2918 		dsl_dataset_t *origin = clone->ds_prev;
2919 		uint64_t comp, uncomp;
2920 
2921 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2922 		dsl_deadlist_space_range(&clone->ds_deadlist,
2923 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2924 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2925 	}
2926 
2927 	/* swap blkptrs */
2928 	{
2929 		blkptr_t tmp;
2930 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
2931 		dsl_dataset_phys(origin_head)->ds_bp =
2932 		    dsl_dataset_phys(clone)->ds_bp;
2933 		dsl_dataset_phys(clone)->ds_bp = tmp;
2934 	}
2935 
2936 	/* set dd_*_bytes */
2937 	{
2938 		int64_t dused, dcomp, duncomp;
2939 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2940 		uint64_t odl_used, odl_comp, odl_uncomp;
2941 
2942 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2943 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2944 
2945 		dsl_deadlist_space(&clone->ds_deadlist,
2946 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2947 		dsl_deadlist_space(&origin_head->ds_deadlist,
2948 		    &odl_used, &odl_comp, &odl_uncomp);
2949 
2950 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2951 		    cdl_used -
2952 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2953 		    odl_used);
2954 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2955 		    cdl_comp -
2956 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2957 		    odl_comp);
2958 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2959 		    cdl_uncomp -
2960 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2961 		    odl_uncomp);
2962 
2963 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2964 		    dused, dcomp, duncomp, tx);
2965 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2966 		    -dused, -dcomp, -duncomp, tx);
2967 
2968 		/*
2969 		 * The difference in the space used by snapshots is the
2970 		 * difference in snapshot space due to the head's
2971 		 * deadlist (since that's the only thing that's
2972 		 * changing that affects the snapused).
2973 		 */
2974 		dsl_deadlist_space_range(&clone->ds_deadlist,
2975 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2976 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2977 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2978 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2979 		    &odl_used, &odl_comp, &odl_uncomp);
2980 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2981 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2982 	}
2983 
2984 	/* swap ds_*_bytes */
2985 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2986 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
2987 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2988 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
2989 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2990 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2991 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2992 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2993 
2994 	/* apply any parent delta for change in unconsumed refreservation */
2995 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2996 	    unused_refres_delta, 0, 0, tx);
2997 
2998 	/*
2999 	 * Swap deadlists.
3000 	 */
3001 	dsl_deadlist_close(&clone->ds_deadlist);
3002 	dsl_deadlist_close(&origin_head->ds_deadlist);
3003 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3004 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3005 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3006 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3007 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3008 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3009 
3010 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3011 
3012 	spa_history_log_internal_ds(clone, "clone swap", tx,
3013 	    "parent=%s", origin_head->ds_dir->dd_myname);
3014 }
3015 
3016 /*
3017  * Given a pool name and a dataset object number in that pool,
3018  * return the name of that dataset.
3019  */
3020 int
3021 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3022 {
3023 	dsl_pool_t *dp;
3024 	dsl_dataset_t *ds;
3025 	int error;
3026 
3027 	error = dsl_pool_hold(pname, FTAG, &dp);
3028 	if (error != 0)
3029 		return (error);
3030 
3031 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3032 	if (error == 0) {
3033 		dsl_dataset_name(ds, buf);
3034 		dsl_dataset_rele(ds, FTAG);
3035 	}
3036 	dsl_pool_rele(dp, FTAG);
3037 
3038 	return (error);
3039 }
3040 
3041 int
3042 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3043     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3044 {
3045 	int error = 0;
3046 
3047 	ASSERT3S(asize, >, 0);
3048 
3049 	/*
3050 	 * *ref_rsrv is the portion of asize that will come from any
3051 	 * unconsumed refreservation space.
3052 	 */
3053 	*ref_rsrv = 0;
3054 
3055 	mutex_enter(&ds->ds_lock);
3056 	/*
3057 	 * Make a space adjustment for reserved bytes.
3058 	 */
3059 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3060 		ASSERT3U(*used, >=,
3061 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3062 		*used -=
3063 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3064 		*ref_rsrv =
3065 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3066 	}
3067 
3068 	if (!check_quota || ds->ds_quota == 0) {
3069 		mutex_exit(&ds->ds_lock);
3070 		return (0);
3071 	}
3072 	/*
3073 	 * If they are requesting more space, and our current estimate
3074 	 * is over quota, they get to try again unless the actual
3075 	 * on-disk is over quota and there are no pending changes (which
3076 	 * may free up space for us).
3077 	 */
3078 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3079 	    ds->ds_quota) {
3080 		if (inflight > 0 ||
3081 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3082 			error = SET_ERROR(ERESTART);
3083 		else
3084 			error = SET_ERROR(EDQUOT);
3085 	}
3086 	mutex_exit(&ds->ds_lock);
3087 
3088 	return (error);
3089 }
3090 
3091 typedef struct dsl_dataset_set_qr_arg {
3092 	const char *ddsqra_name;
3093 	zprop_source_t ddsqra_source;
3094 	uint64_t ddsqra_value;
3095 } dsl_dataset_set_qr_arg_t;
3096 
3097 
3098 /* ARGSUSED */
3099 static int
3100 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3101 {
3102 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3103 	dsl_pool_t *dp = dmu_tx_pool(tx);
3104 	dsl_dataset_t *ds;
3105 	int error;
3106 	uint64_t newval;
3107 
3108 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3109 		return (SET_ERROR(ENOTSUP));
3110 
3111 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3112 	if (error != 0)
3113 		return (error);
3114 
3115 	if (ds->ds_is_snapshot) {
3116 		dsl_dataset_rele(ds, FTAG);
3117 		return (SET_ERROR(EINVAL));
3118 	}
3119 
3120 	error = dsl_prop_predict(ds->ds_dir,
3121 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3122 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3123 	if (error != 0) {
3124 		dsl_dataset_rele(ds, FTAG);
3125 		return (error);
3126 	}
3127 
3128 	if (newval == 0) {
3129 		dsl_dataset_rele(ds, FTAG);
3130 		return (0);
3131 	}
3132 
3133 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3134 	    newval < ds->ds_reserved) {
3135 		dsl_dataset_rele(ds, FTAG);
3136 		return (SET_ERROR(ENOSPC));
3137 	}
3138 
3139 	dsl_dataset_rele(ds, FTAG);
3140 	return (0);
3141 }
3142 
3143 static void
3144 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3145 {
3146 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3147 	dsl_pool_t *dp = dmu_tx_pool(tx);
3148 	dsl_dataset_t *ds;
3149 	uint64_t newval;
3150 
3151 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3152 
3153 	dsl_prop_set_sync_impl(ds,
3154 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3155 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3156 	    &ddsqra->ddsqra_value, tx);
3157 
3158 	VERIFY0(dsl_prop_get_int_ds(ds,
3159 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3160 
3161 	if (ds->ds_quota != newval) {
3162 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3163 		ds->ds_quota = newval;
3164 	}
3165 	dsl_dataset_rele(ds, FTAG);
3166 }
3167 
3168 int
3169 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3170     uint64_t refquota)
3171 {
3172 	dsl_dataset_set_qr_arg_t ddsqra;
3173 
3174 	ddsqra.ddsqra_name = dsname;
3175 	ddsqra.ddsqra_source = source;
3176 	ddsqra.ddsqra_value = refquota;
3177 
3178 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3179 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3180 }
3181 
3182 static int
3183 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3184 {
3185 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3186 	dsl_pool_t *dp = dmu_tx_pool(tx);
3187 	dsl_dataset_t *ds;
3188 	int error;
3189 	uint64_t newval, unique;
3190 
3191 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3192 		return (SET_ERROR(ENOTSUP));
3193 
3194 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3195 	if (error != 0)
3196 		return (error);
3197 
3198 	if (ds->ds_is_snapshot) {
3199 		dsl_dataset_rele(ds, FTAG);
3200 		return (SET_ERROR(EINVAL));
3201 	}
3202 
3203 	error = dsl_prop_predict(ds->ds_dir,
3204 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3205 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3206 	if (error != 0) {
3207 		dsl_dataset_rele(ds, FTAG);
3208 		return (error);
3209 	}
3210 
3211 	/*
3212 	 * If we are doing the preliminary check in open context, the
3213 	 * space estimates may be inaccurate.
3214 	 */
3215 	if (!dmu_tx_is_syncing(tx)) {
3216 		dsl_dataset_rele(ds, FTAG);
3217 		return (0);
3218 	}
3219 
3220 	mutex_enter(&ds->ds_lock);
3221 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3222 		dsl_dataset_recalc_head_uniq(ds);
3223 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3224 	mutex_exit(&ds->ds_lock);
3225 
3226 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3227 		uint64_t delta = MAX(unique, newval) -
3228 		    MAX(unique, ds->ds_reserved);
3229 
3230 		if (delta >
3231 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3232 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3233 			dsl_dataset_rele(ds, FTAG);
3234 			return (SET_ERROR(ENOSPC));
3235 		}
3236 	}
3237 
3238 	dsl_dataset_rele(ds, FTAG);
3239 	return (0);
3240 }
3241 
3242 void
3243 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3244     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3245 {
3246 	uint64_t newval;
3247 	uint64_t unique;
3248 	int64_t delta;
3249 
3250 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3251 	    source, sizeof (value), 1, &value, tx);
3252 
3253 	VERIFY0(dsl_prop_get_int_ds(ds,
3254 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3255 
3256 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3257 	mutex_enter(&ds->ds_dir->dd_lock);
3258 	mutex_enter(&ds->ds_lock);
3259 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3260 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3261 	delta = MAX(0, (int64_t)(newval - unique)) -
3262 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3263 	ds->ds_reserved = newval;
3264 	mutex_exit(&ds->ds_lock);
3265 
3266 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3267 	mutex_exit(&ds->ds_dir->dd_lock);
3268 }
3269 
3270 static void
3271 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3272 {
3273 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3274 	dsl_pool_t *dp = dmu_tx_pool(tx);
3275 	dsl_dataset_t *ds;
3276 
3277 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3278 	dsl_dataset_set_refreservation_sync_impl(ds,
3279 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3280 	dsl_dataset_rele(ds, FTAG);
3281 }
3282 
3283 int
3284 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3285     uint64_t refreservation)
3286 {
3287 	dsl_dataset_set_qr_arg_t ddsqra;
3288 
3289 	ddsqra.ddsqra_name = dsname;
3290 	ddsqra.ddsqra_source = source;
3291 	ddsqra.ddsqra_value = refreservation;
3292 
3293 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3294 	    dsl_dataset_set_refreservation_sync, &ddsqra,
3295 	    0, ZFS_SPACE_CHECK_NONE));
3296 }
3297 
3298 /*
3299  * Return (in *usedp) the amount of space written in new that is not
3300  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3301  * a snapshot before new, in new's filesystem (or its origin).  If not then
3302  * fail and return EINVAL.
3303  *
3304  * The written space is calculated by considering two components:  First, we
3305  * ignore any freed space, and calculate the written as new's used space
3306  * minus old's used space.  Next, we add in the amount of space that was freed
3307  * between the two snapshots, thus reducing new's used space relative to old's.
3308  * Specifically, this is the space that was born before old->ds_creation_txg,
3309  * and freed before new (ie. on new's deadlist or a previous deadlist).
3310  *
3311  * space freed                         [---------------------]
3312  * snapshots                       ---O-------O--------O-------O------
3313  *                                         oldsnap            new
3314  */
3315 int
3316 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3317     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3318 {
3319 	int err = 0;
3320 	uint64_t snapobj;
3321 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3322 
3323 	ASSERT(dsl_pool_config_held(dp));
3324 
3325 	*usedp = 0;
3326 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3327 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3328 
3329 	*compp = 0;
3330 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3331 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3332 
3333 	*uncompp = 0;
3334 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3335 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3336 
3337 	snapobj = new->ds_object;
3338 	while (snapobj != oldsnap->ds_object) {
3339 		dsl_dataset_t *snap;
3340 		uint64_t used, comp, uncomp;
3341 
3342 		if (snapobj == new->ds_object) {
3343 			snap = new;
3344 		} else {
3345 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3346 			if (err != 0)
3347 				break;
3348 		}
3349 
3350 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3351 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3352 			/*
3353 			 * The blocks in the deadlist can not be born after
3354 			 * ds_prev_snap_txg, so get the whole deadlist space,
3355 			 * which is more efficient (especially for old-format
3356 			 * deadlists).  Unfortunately the deadlist code
3357 			 * doesn't have enough information to make this
3358 			 * optimization itself.
3359 			 */
3360 			dsl_deadlist_space(&snap->ds_deadlist,
3361 			    &used, &comp, &uncomp);
3362 		} else {
3363 			dsl_deadlist_space_range(&snap->ds_deadlist,
3364 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3365 			    &used, &comp, &uncomp);
3366 		}
3367 		*usedp += used;
3368 		*compp += comp;
3369 		*uncompp += uncomp;
3370 
3371 		/*
3372 		 * If we get to the beginning of the chain of snapshots
3373 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3374 		 * was not a snapshot of/before new.
3375 		 */
3376 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3377 		if (snap != new)
3378 			dsl_dataset_rele(snap, FTAG);
3379 		if (snapobj == 0) {
3380 			err = SET_ERROR(EINVAL);
3381 			break;
3382 		}
3383 
3384 	}
3385 	return (err);
3386 }
3387 
3388 /*
3389  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3390  * lastsnap, and all snapshots in between are deleted.
3391  *
3392  * blocks that would be freed            [---------------------------]
3393  * snapshots                       ---O-------O--------O-------O--------O
3394  *                                        firstsnap        lastsnap
3395  *
3396  * This is the set of blocks that were born after the snap before firstsnap,
3397  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3398  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3399  * We calculate this by iterating over the relevant deadlists (from the snap
3400  * after lastsnap, backward to the snap after firstsnap), summing up the
3401  * space on the deadlist that was born after the snap before firstsnap.
3402  */
3403 int
3404 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3405     dsl_dataset_t *lastsnap,
3406     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3407 {
3408 	int err = 0;
3409 	uint64_t snapobj;
3410 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3411 
3412 	ASSERT(firstsnap->ds_is_snapshot);
3413 	ASSERT(lastsnap->ds_is_snapshot);
3414 
3415 	/*
3416 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3417 	 * is before lastsnap.
3418 	 */
3419 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3420 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3421 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3422 		return (SET_ERROR(EINVAL));
3423 
3424 	*usedp = *compp = *uncompp = 0;
3425 
3426 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3427 	while (snapobj != firstsnap->ds_object) {
3428 		dsl_dataset_t *ds;
3429 		uint64_t used, comp, uncomp;
3430 
3431 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3432 		if (err != 0)
3433 			break;
3434 
3435 		dsl_deadlist_space_range(&ds->ds_deadlist,
3436 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3437 		    &used, &comp, &uncomp);
3438 		*usedp += used;
3439 		*compp += comp;
3440 		*uncompp += uncomp;
3441 
3442 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3443 		ASSERT3U(snapobj, !=, 0);
3444 		dsl_dataset_rele(ds, FTAG);
3445 	}
3446 	return (err);
3447 }
3448 
3449 /*
3450  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3451  * For example, they could both be snapshots of the same filesystem, and
3452  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3453  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3454  * filesystem.  Or 'earlier' could be the origin's origin.
3455  *
3456  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3457  */
3458 boolean_t
3459 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3460 	uint64_t earlier_txg)
3461 {
3462 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3463 	int error;
3464 	boolean_t ret;
3465 
3466 	ASSERT(dsl_pool_config_held(dp));
3467 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3468 
3469 	if (earlier_txg == 0)
3470 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3471 
3472 	if (later->ds_is_snapshot &&
3473 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3474 		return (B_FALSE);
3475 
3476 	if (later->ds_dir == earlier->ds_dir)
3477 		return (B_TRUE);
3478 	if (!dsl_dir_is_clone(later->ds_dir))
3479 		return (B_FALSE);
3480 
3481 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3482 		return (B_TRUE);
3483 	dsl_dataset_t *origin;
3484 	error = dsl_dataset_hold_obj(dp,
3485 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3486 	if (error != 0)
3487 		return (B_FALSE);
3488 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3489 	dsl_dataset_rele(origin, FTAG);
3490 	return (ret);
3491 }
3492 
3493 void
3494 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3495 {
3496 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3497 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3498 }
3499 
3500 boolean_t
3501 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3502 {
3503 	dmu_object_info_t doi;
3504 
3505 	dmu_object_info_from_db(ds->ds_dbuf, &doi);
3506 	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3507 }
3508 
3509 boolean_t
3510 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3511 {
3512 	return (dsl_dataset_is_zapified(ds) &&
3513 	    zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3514 	    ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3515 }
3516