xref: /titanic_52/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 4b7f25f92fce04a513df62afed73561f9216a4fd)
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  * Copyright (c) 2014 Integros [integros.com]
28  * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
29  */
30 
31 #include <sys/dmu_objset.h>
32 #include <sys/dsl_dataset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_prop.h>
35 #include <sys/dsl_synctask.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dmu_impl.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/arc.h>
40 #include <sys/zio.h>
41 #include <sys/zap.h>
42 #include <sys/zfeature.h>
43 #include <sys/unique.h>
44 #include <sys/zfs_context.h>
45 #include <sys/zfs_ioctl.h>
46 #include <sys/spa.h>
47 #include <sys/zfs_znode.h>
48 #include <sys/zfs_onexit.h>
49 #include <sys/zvol.h>
50 #include <sys/dsl_scan.h>
51 #include <sys/dsl_deadlist.h>
52 #include <sys/dsl_destroy.h>
53 #include <sys/dsl_userhold.h>
54 #include <sys/dsl_bookmark.h>
55 #include <sys/dmu_send.h>
56 #include <sys/zio_checksum.h>
57 #include <sys/zio_compress.h>
58 #include <zfs_fletcher.h>
59 
60 /*
61  * The SPA supports block sizes up to 16MB.  However, very large blocks
62  * can have an impact on i/o latency (e.g. tying up a spinning disk for
63  * ~300ms), and also potentially on the memory allocator.  Therefore,
64  * we do not allow the recordsize to be set larger than zfs_max_recordsize
65  * (default 1MB).  Larger blocks can be created by changing this tunable,
66  * and pools with larger blocks can always be imported and used, regardless
67  * of this setting.
68  */
69 int zfs_max_recordsize = 1 * 1024 * 1024;
70 
71 #define	SWITCH64(x, y) \
72 	{ \
73 		uint64_t __tmp = (x); \
74 		(x) = (y); \
75 		(y) = __tmp; \
76 	}
77 
78 #define	DS_REF_MAX	(1ULL << 62)
79 
80 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
81 
82 extern int spa_asize_inflation;
83 
84 /*
85  * Figure out how much of this delta should be propogated to the dsl_dir
86  * layer.  If there's a refreservation, that space has already been
87  * partially accounted for in our ancestors.
88  */
89 static int64_t
90 parent_delta(dsl_dataset_t *ds, int64_t delta)
91 {
92 	dsl_dataset_phys_t *ds_phys;
93 	uint64_t old_bytes, new_bytes;
94 
95 	if (ds->ds_reserved == 0)
96 		return (delta);
97 
98 	ds_phys = dsl_dataset_phys(ds);
99 	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
100 	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
101 
102 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
103 	return (new_bytes - old_bytes);
104 }
105 
106 void
107 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
108 {
109 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
110 	int compressed = BP_GET_PSIZE(bp);
111 	int uncompressed = BP_GET_UCSIZE(bp);
112 	int64_t delta;
113 
114 	dprintf_bp(bp, "ds=%p", ds);
115 
116 	ASSERT(dmu_tx_is_syncing(tx));
117 	/* It could have been compressed away to nothing */
118 	if (BP_IS_HOLE(bp))
119 		return;
120 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
121 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
122 	if (ds == NULL) {
123 		dsl_pool_mos_diduse_space(tx->tx_pool,
124 		    used, compressed, uncompressed);
125 		return;
126 	}
127 
128 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
129 	mutex_enter(&ds->ds_lock);
130 	delta = parent_delta(ds, used);
131 	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
132 	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
133 	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
134 	dsl_dataset_phys(ds)->ds_unique_bytes += used;
135 
136 	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
137 		ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
138 		    B_TRUE;
139 	}
140 
141 	spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
142 	if (f != SPA_FEATURE_NONE)
143 		ds->ds_feature_activation_needed[f] = B_TRUE;
144 
145 	mutex_exit(&ds->ds_lock);
146 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
147 	    compressed, uncompressed, tx);
148 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
149 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
150 }
151 
152 int
153 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
154     boolean_t async)
155 {
156 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
157 	int compressed = BP_GET_PSIZE(bp);
158 	int uncompressed = BP_GET_UCSIZE(bp);
159 
160 	if (BP_IS_HOLE(bp))
161 		return (0);
162 
163 	ASSERT(dmu_tx_is_syncing(tx));
164 	ASSERT(bp->blk_birth <= tx->tx_txg);
165 
166 	if (ds == NULL) {
167 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
168 		dsl_pool_mos_diduse_space(tx->tx_pool,
169 		    -used, -compressed, -uncompressed);
170 		return (used);
171 	}
172 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
173 
174 	ASSERT(!ds->ds_is_snapshot);
175 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
176 
177 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
178 		int64_t delta;
179 
180 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
181 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
182 
183 		mutex_enter(&ds->ds_lock);
184 		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
185 		    !DS_UNIQUE_IS_ACCURATE(ds));
186 		delta = parent_delta(ds, -used);
187 		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
188 		mutex_exit(&ds->ds_lock);
189 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
190 		    delta, -compressed, -uncompressed, tx);
191 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
192 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
193 	} else {
194 		dprintf_bp(bp, "putting on dead list: %s", "");
195 		if (async) {
196 			/*
197 			 * We are here as part of zio's write done callback,
198 			 * which means we're a zio interrupt thread.  We can't
199 			 * call dsl_deadlist_insert() now because it may block
200 			 * waiting for I/O.  Instead, put bp on the deferred
201 			 * queue and let dsl_pool_sync() finish the job.
202 			 */
203 			bplist_append(&ds->ds_pending_deadlist, bp);
204 		} else {
205 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
206 		}
207 		ASSERT3U(ds->ds_prev->ds_object, ==,
208 		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
209 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
210 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
211 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
212 		    ds->ds_object && bp->blk_birth >
213 		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
214 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
215 			mutex_enter(&ds->ds_prev->ds_lock);
216 			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
217 			mutex_exit(&ds->ds_prev->ds_lock);
218 		}
219 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
220 			dsl_dir_transfer_space(ds->ds_dir, used,
221 			    DD_USED_HEAD, DD_USED_SNAP, tx);
222 		}
223 	}
224 	mutex_enter(&ds->ds_lock);
225 	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
226 	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
227 	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
228 	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
229 	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
230 	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
231 	mutex_exit(&ds->ds_lock);
232 
233 	return (used);
234 }
235 
236 uint64_t
237 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
238 {
239 	uint64_t trysnap = 0;
240 
241 	if (ds == NULL)
242 		return (0);
243 	/*
244 	 * The snapshot creation could fail, but that would cause an
245 	 * incorrect FALSE return, which would only result in an
246 	 * overestimation of the amount of space that an operation would
247 	 * consume, which is OK.
248 	 *
249 	 * There's also a small window where we could miss a pending
250 	 * snapshot, because we could set the sync task in the quiescing
251 	 * phase.  So this should only be used as a guess.
252 	 */
253 	if (ds->ds_trysnap_txg >
254 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
255 		trysnap = ds->ds_trysnap_txg;
256 	return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
257 }
258 
259 boolean_t
260 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
261     uint64_t blk_birth)
262 {
263 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
264 	    (bp != NULL && BP_IS_HOLE(bp)))
265 		return (B_FALSE);
266 
267 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
268 
269 	return (B_TRUE);
270 }
271 
272 static void
273 dsl_dataset_evict(void *dbu)
274 {
275 	dsl_dataset_t *ds = dbu;
276 
277 	ASSERT(ds->ds_owner == NULL);
278 
279 	ds->ds_dbuf = NULL;
280 
281 	unique_remove(ds->ds_fsid_guid);
282 
283 	if (ds->ds_objset != NULL)
284 		dmu_objset_evict(ds->ds_objset);
285 
286 	if (ds->ds_prev) {
287 		dsl_dataset_rele(ds->ds_prev, ds);
288 		ds->ds_prev = NULL;
289 	}
290 
291 	bplist_destroy(&ds->ds_pending_deadlist);
292 	if (ds->ds_deadlist.dl_os != NULL)
293 		dsl_deadlist_close(&ds->ds_deadlist);
294 	if (ds->ds_dir)
295 		dsl_dir_async_rele(ds->ds_dir, ds);
296 
297 	ASSERT(!list_link_active(&ds->ds_synced_link));
298 
299 	list_destroy(&ds->ds_prop_cbs);
300 	mutex_destroy(&ds->ds_lock);
301 	mutex_destroy(&ds->ds_opening_lock);
302 	mutex_destroy(&ds->ds_sendstream_lock);
303 	refcount_destroy(&ds->ds_longholds);
304 
305 	kmem_free(ds, sizeof (dsl_dataset_t));
306 }
307 
308 int
309 dsl_dataset_get_snapname(dsl_dataset_t *ds)
310 {
311 	dsl_dataset_phys_t *headphys;
312 	int err;
313 	dmu_buf_t *headdbuf;
314 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
315 	objset_t *mos = dp->dp_meta_objset;
316 
317 	if (ds->ds_snapname[0])
318 		return (0);
319 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
320 		return (0);
321 
322 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
323 	    FTAG, &headdbuf);
324 	if (err != 0)
325 		return (err);
326 	headphys = headdbuf->db_data;
327 	err = zap_value_search(dp->dp_meta_objset,
328 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
329 	dmu_buf_rele(headdbuf, FTAG);
330 	return (err);
331 }
332 
333 int
334 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
335 {
336 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
337 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
338 	matchtype_t mt;
339 	int err;
340 
341 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
342 		mt = MT_FIRST;
343 	else
344 		mt = MT_EXACT;
345 
346 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
347 	    value, mt, NULL, 0, NULL);
348 	if (err == ENOTSUP && mt == MT_FIRST)
349 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
350 	return (err);
351 }
352 
353 int
354 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
355     boolean_t adj_cnt)
356 {
357 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
358 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
359 	matchtype_t mt;
360 	int err;
361 
362 	dsl_dir_snap_cmtime_update(ds->ds_dir);
363 
364 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
365 		mt = MT_FIRST;
366 	else
367 		mt = MT_EXACT;
368 
369 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
370 	if (err == ENOTSUP && mt == MT_FIRST)
371 		err = zap_remove(mos, snapobj, name, tx);
372 
373 	if (err == 0 && adj_cnt)
374 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
375 		    DD_FIELD_SNAPSHOT_COUNT, tx);
376 
377 	return (err);
378 }
379 
380 boolean_t
381 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
382 {
383 	dmu_buf_t *dbuf = ds->ds_dbuf;
384 	boolean_t result = B_FALSE;
385 
386 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
387 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
388 
389 		if (ds == dmu_buf_get_user(dbuf))
390 			result = B_TRUE;
391 		else
392 			dmu_buf_rele(dbuf, tag);
393 	}
394 
395 	return (result);
396 }
397 
398 int
399 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
400     dsl_dataset_t **dsp)
401 {
402 	objset_t *mos = dp->dp_meta_objset;
403 	dmu_buf_t *dbuf;
404 	dsl_dataset_t *ds;
405 	int err;
406 	dmu_object_info_t doi;
407 
408 	ASSERT(dsl_pool_config_held(dp));
409 
410 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
411 	if (err != 0)
412 		return (err);
413 
414 	/* Make sure dsobj has the correct object type. */
415 	dmu_object_info_from_db(dbuf, &doi);
416 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
417 		dmu_buf_rele(dbuf, tag);
418 		return (SET_ERROR(EINVAL));
419 	}
420 
421 	ds = dmu_buf_get_user(dbuf);
422 	if (ds == NULL) {
423 		dsl_dataset_t *winner = NULL;
424 
425 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
426 		ds->ds_dbuf = dbuf;
427 		ds->ds_object = dsobj;
428 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
429 
430 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
431 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
432 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
433 		refcount_create(&ds->ds_longholds);
434 
435 		bplist_create(&ds->ds_pending_deadlist);
436 		dsl_deadlist_open(&ds->ds_deadlist,
437 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
438 
439 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
440 		    offsetof(dmu_sendarg_t, dsa_link));
441 
442 		list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
443 		    offsetof(dsl_prop_cb_record_t, cbr_ds_node));
444 
445 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
446 			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
447 				if (!(spa_feature_table[f].fi_flags &
448 				    ZFEATURE_FLAG_PER_DATASET))
449 					continue;
450 				err = zap_contains(mos, dsobj,
451 				    spa_feature_table[f].fi_guid);
452 				if (err == 0) {
453 					ds->ds_feature_inuse[f] = B_TRUE;
454 				} else {
455 					ASSERT3U(err, ==, ENOENT);
456 					err = 0;
457 				}
458 			}
459 		}
460 
461 		err = dsl_dir_hold_obj(dp,
462 		    dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
463 		if (err != 0) {
464 			mutex_destroy(&ds->ds_lock);
465 			mutex_destroy(&ds->ds_opening_lock);
466 			mutex_destroy(&ds->ds_sendstream_lock);
467 			refcount_destroy(&ds->ds_longholds);
468 			bplist_destroy(&ds->ds_pending_deadlist);
469 			dsl_deadlist_close(&ds->ds_deadlist);
470 			kmem_free(ds, sizeof (dsl_dataset_t));
471 			dmu_buf_rele(dbuf, tag);
472 			return (err);
473 		}
474 
475 		if (!ds->ds_is_snapshot) {
476 			ds->ds_snapname[0] = '\0';
477 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
478 				err = dsl_dataset_hold_obj(dp,
479 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
480 				    ds, &ds->ds_prev);
481 			}
482 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
483 				int zaperr = zap_lookup(mos, ds->ds_object,
484 				    DS_FIELD_BOOKMARK_NAMES,
485 				    sizeof (ds->ds_bookmarks), 1,
486 				    &ds->ds_bookmarks);
487 				if (zaperr != ENOENT)
488 					VERIFY0(zaperr);
489 			}
490 		} else {
491 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
492 				err = dsl_dataset_get_snapname(ds);
493 			if (err == 0 &&
494 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
495 				err = zap_count(
496 				    ds->ds_dir->dd_pool->dp_meta_objset,
497 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
498 				    &ds->ds_userrefs);
499 			}
500 		}
501 
502 		if (err == 0 && !ds->ds_is_snapshot) {
503 			err = dsl_prop_get_int_ds(ds,
504 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
505 			    &ds->ds_reserved);
506 			if (err == 0) {
507 				err = dsl_prop_get_int_ds(ds,
508 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
509 				    &ds->ds_quota);
510 			}
511 		} else {
512 			ds->ds_reserved = ds->ds_quota = 0;
513 		}
514 
515 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
516 		if (err == 0)
517 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
518 
519 		if (err != 0 || winner != NULL) {
520 			bplist_destroy(&ds->ds_pending_deadlist);
521 			dsl_deadlist_close(&ds->ds_deadlist);
522 			if (ds->ds_prev)
523 				dsl_dataset_rele(ds->ds_prev, ds);
524 			dsl_dir_rele(ds->ds_dir, ds);
525 			mutex_destroy(&ds->ds_lock);
526 			mutex_destroy(&ds->ds_opening_lock);
527 			mutex_destroy(&ds->ds_sendstream_lock);
528 			refcount_destroy(&ds->ds_longholds);
529 			kmem_free(ds, sizeof (dsl_dataset_t));
530 			if (err != 0) {
531 				dmu_buf_rele(dbuf, tag);
532 				return (err);
533 			}
534 			ds = winner;
535 		} else {
536 			ds->ds_fsid_guid =
537 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
538 		}
539 	}
540 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
541 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
542 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
543 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
544 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
545 	*dsp = ds;
546 	return (0);
547 }
548 
549 int
550 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
551     void *tag, dsl_dataset_t **dsp)
552 {
553 	dsl_dir_t *dd;
554 	const char *snapname;
555 	uint64_t obj;
556 	int err = 0;
557 	dsl_dataset_t *ds;
558 
559 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
560 	if (err != 0)
561 		return (err);
562 
563 	ASSERT(dsl_pool_config_held(dp));
564 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
565 	if (obj != 0)
566 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
567 	else
568 		err = SET_ERROR(ENOENT);
569 
570 	/* we may be looking for a snapshot */
571 	if (err == 0 && snapname != NULL) {
572 		dsl_dataset_t *snap_ds;
573 
574 		if (*snapname++ != '@') {
575 			dsl_dataset_rele(ds, tag);
576 			dsl_dir_rele(dd, FTAG);
577 			return (SET_ERROR(ENOENT));
578 		}
579 
580 		dprintf("looking for snapshot '%s'\n", snapname);
581 		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
582 		if (err == 0)
583 			err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
584 		dsl_dataset_rele(ds, tag);
585 
586 		if (err == 0) {
587 			mutex_enter(&snap_ds->ds_lock);
588 			if (snap_ds->ds_snapname[0] == 0)
589 				(void) strlcpy(snap_ds->ds_snapname, snapname,
590 				    sizeof (snap_ds->ds_snapname));
591 			mutex_exit(&snap_ds->ds_lock);
592 			ds = snap_ds;
593 		}
594 	}
595 	if (err == 0)
596 		*dsp = ds;
597 	dsl_dir_rele(dd, FTAG);
598 	return (err);
599 }
600 
601 int
602 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
603     void *tag, dsl_dataset_t **dsp)
604 {
605 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
606 	if (err != 0)
607 		return (err);
608 	if (!dsl_dataset_tryown(*dsp, tag)) {
609 		dsl_dataset_rele(*dsp, tag);
610 		*dsp = NULL;
611 		return (SET_ERROR(EBUSY));
612 	}
613 	return (0);
614 }
615 
616 int
617 dsl_dataset_own(dsl_pool_t *dp, const char *name,
618     void *tag, dsl_dataset_t **dsp)
619 {
620 	int err = dsl_dataset_hold(dp, name, tag, dsp);
621 	if (err != 0)
622 		return (err);
623 	if (!dsl_dataset_tryown(*dsp, tag)) {
624 		dsl_dataset_rele(*dsp, tag);
625 		return (SET_ERROR(EBUSY));
626 	}
627 	return (0);
628 }
629 
630 /*
631  * See the comment above dsl_pool_hold() for details.  In summary, a long
632  * hold is used to prevent destruction of a dataset while the pool hold
633  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
634  *
635  * The dataset and pool must be held when this function is called.  After it
636  * is called, the pool hold may be released while the dataset is still held
637  * and accessed.
638  */
639 void
640 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
641 {
642 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
643 	(void) refcount_add(&ds->ds_longholds, tag);
644 }
645 
646 void
647 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
648 {
649 	(void) refcount_remove(&ds->ds_longholds, tag);
650 }
651 
652 /* Return B_TRUE if there are any long holds on this dataset. */
653 boolean_t
654 dsl_dataset_long_held(dsl_dataset_t *ds)
655 {
656 	return (!refcount_is_zero(&ds->ds_longholds));
657 }
658 
659 void
660 dsl_dataset_name(dsl_dataset_t *ds, char *name)
661 {
662 	if (ds == NULL) {
663 		(void) strcpy(name, "mos");
664 	} else {
665 		dsl_dir_name(ds->ds_dir, name);
666 		VERIFY0(dsl_dataset_get_snapname(ds));
667 		if (ds->ds_snapname[0]) {
668 			(void) strcat(name, "@");
669 			/*
670 			 * We use a "recursive" mutex so that we
671 			 * can call dprintf_ds() with ds_lock held.
672 			 */
673 			if (!MUTEX_HELD(&ds->ds_lock)) {
674 				mutex_enter(&ds->ds_lock);
675 				VERIFY3U(strlen(name) +
676 					 strlen(ds->ds_snapname) + 1, <=,
677 					 ZFS_MAXNAMELEN);
678 				(void) strcat(name, ds->ds_snapname);
679 				mutex_exit(&ds->ds_lock);
680 			} else {
681 				VERIFY3U(strlen(name) +
682 					 strlen(ds->ds_snapname) + 1, <=,
683 					 ZFS_MAXNAMELEN);
684 				(void) strcat(name, ds->ds_snapname);
685 			}
686 		}
687 	}
688 }
689 
690 void
691 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
692 {
693 	dmu_buf_rele(ds->ds_dbuf, tag);
694 }
695 
696 void
697 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
698 {
699 	ASSERT3P(ds->ds_owner, ==, tag);
700 	ASSERT(ds->ds_dbuf != NULL);
701 
702 	mutex_enter(&ds->ds_lock);
703 	ds->ds_owner = NULL;
704 	mutex_exit(&ds->ds_lock);
705 	dsl_dataset_long_rele(ds, tag);
706 	dsl_dataset_rele(ds, tag);
707 }
708 
709 boolean_t
710 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
711 {
712 	boolean_t gotit = FALSE;
713 
714 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
715 	mutex_enter(&ds->ds_lock);
716 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
717 		ds->ds_owner = tag;
718 		dsl_dataset_long_hold(ds, tag);
719 		gotit = TRUE;
720 	}
721 	mutex_exit(&ds->ds_lock);
722 	return (gotit);
723 }
724 
725 boolean_t
726 dsl_dataset_has_owner(dsl_dataset_t *ds)
727 {
728 	boolean_t rv;
729 	mutex_enter(&ds->ds_lock);
730 	rv = (ds->ds_owner != NULL);
731 	mutex_exit(&ds->ds_lock);
732 	return (rv);
733 }
734 
735 static void
736 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
737 {
738 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
739 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
740 	uint64_t zero = 0;
741 
742 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
743 
744 	spa_feature_incr(spa, f, tx);
745 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
746 
747 	VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
748 	    sizeof (zero), 1, &zero, tx));
749 }
750 
751 void
752 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
753 {
754 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
755 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
756 
757 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
758 
759 	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
760 	spa_feature_decr(spa, f, tx);
761 }
762 
763 uint64_t
764 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
765     uint64_t flags, dmu_tx_t *tx)
766 {
767 	dsl_pool_t *dp = dd->dd_pool;
768 	dmu_buf_t *dbuf;
769 	dsl_dataset_phys_t *dsphys;
770 	uint64_t dsobj;
771 	objset_t *mos = dp->dp_meta_objset;
772 
773 	if (origin == NULL)
774 		origin = dp->dp_origin_snap;
775 
776 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
777 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
778 	ASSERT(dmu_tx_is_syncing(tx));
779 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
780 
781 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
782 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
783 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
784 	dmu_buf_will_dirty(dbuf, tx);
785 	dsphys = dbuf->db_data;
786 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
787 	dsphys->ds_dir_obj = dd->dd_object;
788 	dsphys->ds_flags = flags;
789 	dsphys->ds_fsid_guid = unique_create();
790 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
791 	    sizeof (dsphys->ds_guid));
792 	dsphys->ds_snapnames_zapobj =
793 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
794 	    DMU_OT_NONE, 0, tx);
795 	dsphys->ds_creation_time = gethrestime_sec();
796 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
797 
798 	if (origin == NULL) {
799 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
800 	} else {
801 		dsl_dataset_t *ohds; /* head of the origin snapshot */
802 
803 		dsphys->ds_prev_snap_obj = origin->ds_object;
804 		dsphys->ds_prev_snap_txg =
805 		    dsl_dataset_phys(origin)->ds_creation_txg;
806 		dsphys->ds_referenced_bytes =
807 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
808 		dsphys->ds_compressed_bytes =
809 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
810 		dsphys->ds_uncompressed_bytes =
811 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
812 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
813 
814 		/*
815 		 * Inherit flags that describe the dataset's contents
816 		 * (INCONSISTENT) or properties (Case Insensitive).
817 		 */
818 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
819 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
820 
821 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
822 			if (origin->ds_feature_inuse[f])
823 				dsl_dataset_activate_feature(dsobj, f, tx);
824 		}
825 
826 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
827 		dsl_dataset_phys(origin)->ds_num_children++;
828 
829 		VERIFY0(dsl_dataset_hold_obj(dp,
830 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
831 		    FTAG, &ohds));
832 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
833 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
834 		dsl_dataset_rele(ohds, FTAG);
835 
836 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
837 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
838 				dsl_dataset_phys(origin)->ds_next_clones_obj =
839 				    zap_create(mos,
840 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
841 			}
842 			VERIFY0(zap_add_int(mos,
843 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
844 			    dsobj, tx));
845 		}
846 
847 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
848 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
849 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
850 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
851 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
852 				dsl_dir_phys(origin->ds_dir)->dd_clones =
853 				    zap_create(mos,
854 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
855 			}
856 			VERIFY0(zap_add_int(mos,
857 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
858 			    dsobj, tx));
859 		}
860 	}
861 
862 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
863 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
864 
865 	dmu_buf_rele(dbuf, FTAG);
866 
867 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
868 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
869 
870 	return (dsobj);
871 }
872 
873 static void
874 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
875 {
876 	objset_t *os;
877 
878 	VERIFY0(dmu_objset_from_ds(ds, &os));
879 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
880 	dsl_dataset_dirty(ds, tx);
881 }
882 
883 uint64_t
884 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
885     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
886 {
887 	dsl_pool_t *dp = pdd->dd_pool;
888 	uint64_t dsobj, ddobj;
889 	dsl_dir_t *dd;
890 
891 	ASSERT(dmu_tx_is_syncing(tx));
892 	ASSERT(lastname[0] != '@');
893 
894 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
895 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
896 
897 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
898 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
899 
900 	dsl_deleg_set_create_perms(dd, tx, cr);
901 
902 	/*
903 	 * Since we're creating a new node we know it's a leaf, so we can
904 	 * initialize the counts if the limit feature is active.
905 	 */
906 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
907 		uint64_t cnt = 0;
908 		objset_t *os = dd->dd_pool->dp_meta_objset;
909 
910 		dsl_dir_zapify(dd, tx);
911 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
912 		    sizeof (cnt), 1, &cnt, tx));
913 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
914 		    sizeof (cnt), 1, &cnt, tx));
915 	}
916 
917 	dsl_dir_rele(dd, FTAG);
918 
919 	/*
920 	 * If we are creating a clone, make sure we zero out any stale
921 	 * data from the origin snapshots zil header.
922 	 */
923 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
924 		dsl_dataset_t *ds;
925 
926 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
927 		dsl_dataset_zero_zil(ds, tx);
928 		dsl_dataset_rele(ds, FTAG);
929 	}
930 
931 	return (dsobj);
932 }
933 
934 /*
935  * The unique space in the head dataset can be calculated by subtracting
936  * the space used in the most recent snapshot, that is still being used
937  * in this file system, from the space currently in use.  To figure out
938  * the space in the most recent snapshot still in use, we need to take
939  * the total space used in the snapshot and subtract out the space that
940  * has been freed up since the snapshot was taken.
941  */
942 void
943 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
944 {
945 	uint64_t mrs_used;
946 	uint64_t dlused, dlcomp, dluncomp;
947 
948 	ASSERT(!ds->ds_is_snapshot);
949 
950 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
951 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
952 	else
953 		mrs_used = 0;
954 
955 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
956 
957 	ASSERT3U(dlused, <=, mrs_used);
958 	dsl_dataset_phys(ds)->ds_unique_bytes =
959 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
960 
961 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
962 	    SPA_VERSION_UNIQUE_ACCURATE)
963 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
964 }
965 
966 void
967 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
968     dmu_tx_t *tx)
969 {
970 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
971 	uint64_t count;
972 	int err;
973 
974 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
975 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
976 	    obj, tx);
977 	/*
978 	 * The err should not be ENOENT, but a bug in a previous version
979 	 * of the code could cause upgrade_clones_cb() to not set
980 	 * ds_next_snap_obj when it should, leading to a missing entry.
981 	 * If we knew that the pool was created after
982 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
983 	 * ENOENT.  However, at least we can check that we don't have
984 	 * too many entries in the next_clones_obj even after failing to
985 	 * remove this one.
986 	 */
987 	if (err != ENOENT)
988 		VERIFY0(err);
989 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
990 	    &count));
991 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
992 }
993 
994 
995 blkptr_t *
996 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
997 {
998 	return (&dsl_dataset_phys(ds)->ds_bp);
999 }
1000 
1001 void
1002 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1003 {
1004 	ASSERT(dmu_tx_is_syncing(tx));
1005 	/* If it's the meta-objset, set dp_meta_rootbp */
1006 	if (ds == NULL) {
1007 		tx->tx_pool->dp_meta_rootbp = *bp;
1008 	} else {
1009 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1010 		dsl_dataset_phys(ds)->ds_bp = *bp;
1011 	}
1012 }
1013 
1014 spa_t *
1015 dsl_dataset_get_spa(dsl_dataset_t *ds)
1016 {
1017 	return (ds->ds_dir->dd_pool->dp_spa);
1018 }
1019 
1020 void
1021 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1022 {
1023 	dsl_pool_t *dp;
1024 
1025 	if (ds == NULL) /* this is the meta-objset */
1026 		return;
1027 
1028 	ASSERT(ds->ds_objset != NULL);
1029 
1030 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1031 		panic("dirtying snapshot!");
1032 
1033 	dp = ds->ds_dir->dd_pool;
1034 
1035 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1036 		/* up the hold count until we can be written out */
1037 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1038 	}
1039 }
1040 
1041 boolean_t
1042 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1043 {
1044 	for (int t = 0; t < TXG_SIZE; t++) {
1045 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1046 		    ds, t))
1047 			return (B_TRUE);
1048 	}
1049 	return (B_FALSE);
1050 }
1051 
1052 static int
1053 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1054 {
1055 	uint64_t asize;
1056 
1057 	if (!dmu_tx_is_syncing(tx))
1058 		return (0);
1059 
1060 	/*
1061 	 * If there's an fs-only reservation, any blocks that might become
1062 	 * owned by the snapshot dataset must be accommodated by space
1063 	 * outside of the reservation.
1064 	 */
1065 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1066 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1067 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1068 		return (SET_ERROR(ENOSPC));
1069 
1070 	/*
1071 	 * Propagate any reserved space for this snapshot to other
1072 	 * snapshot checks in this sync group.
1073 	 */
1074 	if (asize > 0)
1075 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1076 
1077 	return (0);
1078 }
1079 
1080 typedef struct dsl_dataset_snapshot_arg {
1081 	nvlist_t *ddsa_snaps;
1082 	nvlist_t *ddsa_props;
1083 	nvlist_t *ddsa_errors;
1084 	cred_t *ddsa_cr;
1085 } dsl_dataset_snapshot_arg_t;
1086 
1087 int
1088 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1089     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1090 {
1091 	int error;
1092 	uint64_t value;
1093 
1094 	ds->ds_trysnap_txg = tx->tx_txg;
1095 
1096 	if (!dmu_tx_is_syncing(tx))
1097 		return (0);
1098 
1099 	/*
1100 	 * We don't allow multiple snapshots of the same txg.  If there
1101 	 * is already one, try again.
1102 	 */
1103 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1104 		return (SET_ERROR(EAGAIN));
1105 
1106 	/*
1107 	 * Check for conflicting snapshot name.
1108 	 */
1109 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1110 	if (error == 0)
1111 		return (SET_ERROR(EEXIST));
1112 	if (error != ENOENT)
1113 		return (error);
1114 
1115 	/*
1116 	 * We don't allow taking snapshots of inconsistent datasets, such as
1117 	 * those into which we are currently receiving.  However, if we are
1118 	 * creating this snapshot as part of a receive, this check will be
1119 	 * executed atomically with respect to the completion of the receive
1120 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1121 	 * case we ignore this, knowing it will be fixed up for us shortly in
1122 	 * dmu_recv_end_sync().
1123 	 */
1124 	if (!recv && DS_IS_INCONSISTENT(ds))
1125 		return (SET_ERROR(EBUSY));
1126 
1127 	/*
1128 	 * Skip the check for temporary snapshots or if we have already checked
1129 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1130 	 * check the count here when we're receiving a stream.
1131 	 */
1132 	if (cnt != 0 && cr != NULL) {
1133 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1134 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1135 		if (error != 0)
1136 			return (error);
1137 	}
1138 
1139 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1140 	if (error != 0)
1141 		return (error);
1142 
1143 	return (0);
1144 }
1145 
1146 static int
1147 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1148 {
1149 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1150 	dsl_pool_t *dp = dmu_tx_pool(tx);
1151 	nvpair_t *pair;
1152 	int rv = 0;
1153 
1154 	/*
1155 	 * Pre-compute how many total new snapshots will be created for each
1156 	 * level in the tree and below. This is needed for validating the
1157 	 * snapshot limit when either taking a recursive snapshot or when
1158 	 * taking multiple snapshots.
1159 	 *
1160 	 * The problem is that the counts are not actually adjusted when
1161 	 * we are checking, only when we finally sync. For a single snapshot,
1162 	 * this is easy, the count will increase by 1 at each node up the tree,
1163 	 * but its more complicated for the recursive/multiple snapshot case.
1164 	 *
1165 	 * The dsl_fs_ss_limit_check function does recursively check the count
1166 	 * at each level up the tree but since it is validating each snapshot
1167 	 * independently we need to be sure that we are validating the complete
1168 	 * count for the entire set of snapshots. We do this by rolling up the
1169 	 * counts for each component of the name into an nvlist and then
1170 	 * checking each of those cases with the aggregated count.
1171 	 *
1172 	 * This approach properly handles not only the recursive snapshot
1173 	 * case (where we get all of those on the ddsa_snaps list) but also
1174 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1175 	 * validate the limit on 'a' using a count of 2).
1176 	 *
1177 	 * We validate the snapshot names in the third loop and only report
1178 	 * name errors once.
1179 	 */
1180 	if (dmu_tx_is_syncing(tx)) {
1181 		nvlist_t *cnt_track = NULL;
1182 		cnt_track = fnvlist_alloc();
1183 
1184 		/* Rollup aggregated counts into the cnt_track list */
1185 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1186 		    pair != NULL;
1187 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1188 			char *pdelim;
1189 			uint64_t val;
1190 			char nm[MAXPATHLEN];
1191 
1192 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1193 			pdelim = strchr(nm, '@');
1194 			if (pdelim == NULL)
1195 				continue;
1196 			*pdelim = '\0';
1197 
1198 			do {
1199 				if (nvlist_lookup_uint64(cnt_track, nm,
1200 				    &val) == 0) {
1201 					/* update existing entry */
1202 					fnvlist_add_uint64(cnt_track, nm,
1203 					    val + 1);
1204 				} else {
1205 					/* add to list */
1206 					fnvlist_add_uint64(cnt_track, nm, 1);
1207 				}
1208 
1209 				pdelim = strrchr(nm, '/');
1210 				if (pdelim != NULL)
1211 					*pdelim = '\0';
1212 			} while (pdelim != NULL);
1213 		}
1214 
1215 		/* Check aggregated counts at each level */
1216 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1217 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1218 			int error = 0;
1219 			char *name;
1220 			uint64_t cnt = 0;
1221 			dsl_dataset_t *ds;
1222 
1223 			name = nvpair_name(pair);
1224 			cnt = fnvpair_value_uint64(pair);
1225 			ASSERT(cnt > 0);
1226 
1227 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1228 			if (error == 0) {
1229 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1230 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1231 				    ddsa->ddsa_cr);
1232 				dsl_dataset_rele(ds, FTAG);
1233 			}
1234 
1235 			if (error != 0) {
1236 				if (ddsa->ddsa_errors != NULL)
1237 					fnvlist_add_int32(ddsa->ddsa_errors,
1238 					    name, error);
1239 				rv = error;
1240 				/* only report one error for this check */
1241 				break;
1242 			}
1243 		}
1244 		nvlist_free(cnt_track);
1245 	}
1246 
1247 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1248 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1249 		int error = 0;
1250 		dsl_dataset_t *ds;
1251 		char *name, *atp;
1252 		char dsname[MAXNAMELEN];
1253 
1254 		name = nvpair_name(pair);
1255 		if (strlen(name) >= MAXNAMELEN)
1256 			error = SET_ERROR(ENAMETOOLONG);
1257 		if (error == 0) {
1258 			atp = strchr(name, '@');
1259 			if (atp == NULL)
1260 				error = SET_ERROR(EINVAL);
1261 			if (error == 0)
1262 				(void) strlcpy(dsname, name, atp - name + 1);
1263 		}
1264 		if (error == 0)
1265 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1266 		if (error == 0) {
1267 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1268 			error = dsl_dataset_snapshot_check_impl(ds,
1269 			    atp + 1, tx, B_FALSE, 0, NULL);
1270 			dsl_dataset_rele(ds, FTAG);
1271 		}
1272 
1273 		if (error != 0) {
1274 			if (ddsa->ddsa_errors != NULL) {
1275 				fnvlist_add_int32(ddsa->ddsa_errors,
1276 				    name, error);
1277 			}
1278 			rv = error;
1279 		}
1280 	}
1281 
1282 	return (rv);
1283 }
1284 
1285 void
1286 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1287     dmu_tx_t *tx)
1288 {
1289 	static zil_header_t zero_zil;
1290 
1291 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1292 	dmu_buf_t *dbuf;
1293 	dsl_dataset_phys_t *dsphys;
1294 	uint64_t dsobj, crtxg;
1295 	objset_t *mos = dp->dp_meta_objset;
1296 	objset_t *os;
1297 
1298 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1299 
1300 	/*
1301 	 * If we are on an old pool, the zil must not be active, in which
1302 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1303 	 */
1304 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1305 	    dmu_objset_from_ds(ds, &os) != 0 ||
1306 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1307 	    sizeof (zero_zil)) == 0);
1308 
1309 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1310 
1311 	/*
1312 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1313 	 */
1314 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1315 		crtxg = 1;
1316 	else
1317 		crtxg = tx->tx_txg;
1318 
1319 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1320 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1321 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1322 	dmu_buf_will_dirty(dbuf, tx);
1323 	dsphys = dbuf->db_data;
1324 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1325 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1326 	dsphys->ds_fsid_guid = unique_create();
1327 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1328 	    sizeof (dsphys->ds_guid));
1329 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1330 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1331 	dsphys->ds_next_snap_obj = ds->ds_object;
1332 	dsphys->ds_num_children = 1;
1333 	dsphys->ds_creation_time = gethrestime_sec();
1334 	dsphys->ds_creation_txg = crtxg;
1335 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1336 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1337 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1338 	dsphys->ds_uncompressed_bytes =
1339 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1340 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1341 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1342 	dmu_buf_rele(dbuf, FTAG);
1343 
1344 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1345 		if (ds->ds_feature_inuse[f])
1346 			dsl_dataset_activate_feature(dsobj, f, tx);
1347 	}
1348 
1349 	ASSERT3U(ds->ds_prev != 0, ==,
1350 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1351 	if (ds->ds_prev) {
1352 		uint64_t next_clones_obj =
1353 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1354 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1355 		    ds->ds_object ||
1356 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1357 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1358 		    ds->ds_object) {
1359 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1360 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1361 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1362 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1363 		} else if (next_clones_obj != 0) {
1364 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1365 			    dsphys->ds_next_snap_obj, tx);
1366 			VERIFY0(zap_add_int(mos,
1367 			    next_clones_obj, dsobj, tx));
1368 		}
1369 	}
1370 
1371 	/*
1372 	 * If we have a reference-reservation on this dataset, we will
1373 	 * need to increase the amount of refreservation being charged
1374 	 * since our unique space is going to zero.
1375 	 */
1376 	if (ds->ds_reserved) {
1377 		int64_t delta;
1378 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1379 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1380 		    ds->ds_reserved);
1381 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1382 		    delta, 0, 0, tx);
1383 	}
1384 
1385 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1386 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1387 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1388 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1389 	dsl_deadlist_close(&ds->ds_deadlist);
1390 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1391 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1392 	dsl_deadlist_add_key(&ds->ds_deadlist,
1393 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1394 
1395 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1396 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1397 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1398 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1399 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1400 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1401 
1402 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1403 	    snapname, 8, 1, &dsobj, tx));
1404 
1405 	if (ds->ds_prev)
1406 		dsl_dataset_rele(ds->ds_prev, ds);
1407 	VERIFY0(dsl_dataset_hold_obj(dp,
1408 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1409 
1410 	dsl_scan_ds_snapshotted(ds, tx);
1411 
1412 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1413 
1414 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1415 }
1416 
1417 static void
1418 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1419 {
1420 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1421 	dsl_pool_t *dp = dmu_tx_pool(tx);
1422 	nvpair_t *pair;
1423 
1424 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1425 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1426 		dsl_dataset_t *ds;
1427 		char *name, *atp;
1428 		char dsname[MAXNAMELEN];
1429 
1430 		name = nvpair_name(pair);
1431 		atp = strchr(name, '@');
1432 		(void) strlcpy(dsname, name, atp - name + 1);
1433 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1434 
1435 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1436 		if (ddsa->ddsa_props != NULL) {
1437 			dsl_props_set_sync_impl(ds->ds_prev,
1438 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1439 		}
1440 		dsl_dataset_rele(ds, FTAG);
1441 	}
1442 }
1443 
1444 /*
1445  * The snapshots must all be in the same pool.
1446  * All-or-nothing: if there are any failures, nothing will be modified.
1447  */
1448 int
1449 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1450 {
1451 	dsl_dataset_snapshot_arg_t ddsa;
1452 	nvpair_t *pair;
1453 	boolean_t needsuspend;
1454 	int error;
1455 	spa_t *spa;
1456 	char *firstname;
1457 	nvlist_t *suspended = NULL;
1458 
1459 	pair = nvlist_next_nvpair(snaps, NULL);
1460 	if (pair == NULL)
1461 		return (0);
1462 	firstname = nvpair_name(pair);
1463 
1464 	error = spa_open(firstname, &spa, FTAG);
1465 	if (error != 0)
1466 		return (error);
1467 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1468 	spa_close(spa, FTAG);
1469 
1470 	if (needsuspend) {
1471 		suspended = fnvlist_alloc();
1472 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1473 		    pair = nvlist_next_nvpair(snaps, pair)) {
1474 			char fsname[MAXNAMELEN];
1475 			char *snapname = nvpair_name(pair);
1476 			char *atp;
1477 			void *cookie;
1478 
1479 			atp = strchr(snapname, '@');
1480 			if (atp == NULL) {
1481 				error = SET_ERROR(EINVAL);
1482 				break;
1483 			}
1484 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1485 
1486 			error = zil_suspend(fsname, &cookie);
1487 			if (error != 0)
1488 				break;
1489 			fnvlist_add_uint64(suspended, fsname,
1490 			    (uintptr_t)cookie);
1491 		}
1492 	}
1493 
1494 	ddsa.ddsa_snaps = snaps;
1495 	ddsa.ddsa_props = props;
1496 	ddsa.ddsa_errors = errors;
1497 	ddsa.ddsa_cr = CRED();
1498 
1499 	if (error == 0) {
1500 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1501 		    dsl_dataset_snapshot_sync, &ddsa,
1502 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1503 	}
1504 
1505 	if (suspended != NULL) {
1506 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1507 		    pair = nvlist_next_nvpair(suspended, pair)) {
1508 			zil_resume((void *)(uintptr_t)
1509 			    fnvpair_value_uint64(pair));
1510 		}
1511 		fnvlist_free(suspended);
1512 	}
1513 
1514 	return (error);
1515 }
1516 
1517 typedef struct dsl_dataset_snapshot_tmp_arg {
1518 	const char *ddsta_fsname;
1519 	const char *ddsta_snapname;
1520 	minor_t ddsta_cleanup_minor;
1521 	const char *ddsta_htag;
1522 } dsl_dataset_snapshot_tmp_arg_t;
1523 
1524 static int
1525 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1526 {
1527 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1528 	dsl_pool_t *dp = dmu_tx_pool(tx);
1529 	dsl_dataset_t *ds;
1530 	int error;
1531 
1532 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1533 	if (error != 0)
1534 		return (error);
1535 
1536 	/* NULL cred means no limit check for tmp snapshot */
1537 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1538 	    tx, B_FALSE, 0, NULL);
1539 	if (error != 0) {
1540 		dsl_dataset_rele(ds, FTAG);
1541 		return (error);
1542 	}
1543 
1544 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1545 		dsl_dataset_rele(ds, FTAG);
1546 		return (SET_ERROR(ENOTSUP));
1547 	}
1548 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1549 	    B_TRUE, tx);
1550 	if (error != 0) {
1551 		dsl_dataset_rele(ds, FTAG);
1552 		return (error);
1553 	}
1554 
1555 	dsl_dataset_rele(ds, FTAG);
1556 	return (0);
1557 }
1558 
1559 static void
1560 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1561 {
1562 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1563 	dsl_pool_t *dp = dmu_tx_pool(tx);
1564 	dsl_dataset_t *ds;
1565 
1566 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1567 
1568 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1569 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1570 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1571 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1572 
1573 	dsl_dataset_rele(ds, FTAG);
1574 }
1575 
1576 int
1577 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1578     minor_t cleanup_minor, const char *htag)
1579 {
1580 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1581 	int error;
1582 	spa_t *spa;
1583 	boolean_t needsuspend;
1584 	void *cookie;
1585 
1586 	ddsta.ddsta_fsname = fsname;
1587 	ddsta.ddsta_snapname = snapname;
1588 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1589 	ddsta.ddsta_htag = htag;
1590 
1591 	error = spa_open(fsname, &spa, FTAG);
1592 	if (error != 0)
1593 		return (error);
1594 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1595 	spa_close(spa, FTAG);
1596 
1597 	if (needsuspend) {
1598 		error = zil_suspend(fsname, &cookie);
1599 		if (error != 0)
1600 			return (error);
1601 	}
1602 
1603 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1604 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1605 
1606 	if (needsuspend)
1607 		zil_resume(cookie);
1608 	return (error);
1609 }
1610 
1611 
1612 void
1613 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1614 {
1615 	ASSERT(dmu_tx_is_syncing(tx));
1616 	ASSERT(ds->ds_objset != NULL);
1617 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1618 
1619 	/*
1620 	 * in case we had to change ds_fsid_guid when we opened it,
1621 	 * sync it out now.
1622 	 */
1623 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1624 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1625 
1626 	if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1627 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1628 		    ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1629 		    &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1630 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1631 		    ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1632 		    &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1633 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1634 		    ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1635 		    &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1636 		ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1637 		ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1638 		ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1639 	}
1640 
1641 	dmu_objset_sync(ds->ds_objset, zio, tx);
1642 
1643 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1644 		if (ds->ds_feature_activation_needed[f]) {
1645 			if (ds->ds_feature_inuse[f])
1646 				continue;
1647 			dsl_dataset_activate_feature(ds->ds_object, f, tx);
1648 			ds->ds_feature_inuse[f] = B_TRUE;
1649 		}
1650 	}
1651 }
1652 
1653 static void
1654 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1655 {
1656 	uint64_t count = 0;
1657 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1658 	zap_cursor_t zc;
1659 	zap_attribute_t za;
1660 	nvlist_t *propval = fnvlist_alloc();
1661 	nvlist_t *val = fnvlist_alloc();
1662 
1663 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1664 
1665 	/*
1666 	 * There may be missing entries in ds_next_clones_obj
1667 	 * due to a bug in a previous version of the code.
1668 	 * Only trust it if it has the right number of entries.
1669 	 */
1670 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1671 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1672 		    &count));
1673 	}
1674 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1675 		goto fail;
1676 	for (zap_cursor_init(&zc, mos,
1677 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1678 	    zap_cursor_retrieve(&zc, &za) == 0;
1679 	    zap_cursor_advance(&zc)) {
1680 		dsl_dataset_t *clone;
1681 		char buf[ZFS_MAXNAMELEN];
1682 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1683 		    za.za_first_integer, FTAG, &clone));
1684 		dsl_dir_name(clone->ds_dir, buf);
1685 		fnvlist_add_boolean(val, buf);
1686 		dsl_dataset_rele(clone, FTAG);
1687 	}
1688 	zap_cursor_fini(&zc);
1689 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1690 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1691 fail:
1692 	nvlist_free(val);
1693 	nvlist_free(propval);
1694 }
1695 
1696 static void
1697 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1698 {
1699 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1700 
1701 	if (dsl_dataset_has_resume_receive_state(ds)) {
1702 		char *str;
1703 		void *packed;
1704 		uint8_t *compressed;
1705 		uint64_t val;
1706 		nvlist_t *token_nv = fnvlist_alloc();
1707 		size_t packed_size, compressed_size;
1708 
1709 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1710 		    DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1711 			fnvlist_add_uint64(token_nv, "fromguid", val);
1712 		}
1713 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1714 		    DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1715 			fnvlist_add_uint64(token_nv, "object", val);
1716 		}
1717 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1718 		    DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1719 			fnvlist_add_uint64(token_nv, "offset", val);
1720 		}
1721 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1722 		    DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1723 			fnvlist_add_uint64(token_nv, "bytes", val);
1724 		}
1725 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1726 		    DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1727 			fnvlist_add_uint64(token_nv, "toguid", val);
1728 		}
1729 		char buf[256];
1730 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1731 		    DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1732 			fnvlist_add_string(token_nv, "toname", buf);
1733 		}
1734 		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1735 		    DS_FIELD_RESUME_EMBEDOK) == 0) {
1736 			fnvlist_add_boolean(token_nv, "embedok");
1737 		}
1738 		packed = fnvlist_pack(token_nv, &packed_size);
1739 		fnvlist_free(token_nv);
1740 		compressed = kmem_alloc(packed_size, KM_SLEEP);
1741 
1742 		compressed_size = gzip_compress(packed, compressed,
1743 		    packed_size, packed_size, 6);
1744 
1745 		zio_cksum_t cksum;
1746 		fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1747 
1748 		str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1749 		for (int i = 0; i < compressed_size; i++) {
1750 			(void) sprintf(str + i * 2, "%02x", compressed[i]);
1751 		}
1752 		str[compressed_size * 2] = '\0';
1753 		char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1754 		    ZFS_SEND_RESUME_TOKEN_VERSION,
1755 		    (longlong_t)cksum.zc_word[0],
1756 		    (longlong_t)packed_size, str);
1757 		dsl_prop_nvlist_add_string(nv,
1758 		    ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1759 		kmem_free(packed, packed_size);
1760 		kmem_free(str, compressed_size * 2 + 1);
1761 		kmem_free(compressed, packed_size);
1762 		strfree(propval);
1763 	}
1764 }
1765 
1766 void
1767 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1768 {
1769 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1770 	uint64_t refd, avail, uobjs, aobjs, ratio;
1771 
1772 	ASSERT(dsl_pool_config_held(dp));
1773 
1774 	ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1775 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1776 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1777 
1778 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1779 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1780 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1781 
1782 	if (ds->ds_is_snapshot) {
1783 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1784 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1785 		    dsl_dataset_phys(ds)->ds_unique_bytes);
1786 		get_clones_stat(ds, nv);
1787 	} else {
1788 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1789 			char buf[MAXNAMELEN];
1790 			dsl_dataset_name(ds->ds_prev, buf);
1791 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1792 		}
1793 
1794 		dsl_dir_stats(ds->ds_dir, nv);
1795 	}
1796 
1797 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1798 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1799 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1800 
1801 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1802 	    dsl_dataset_phys(ds)->ds_creation_time);
1803 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1804 	    dsl_dataset_phys(ds)->ds_creation_txg);
1805 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1806 	    ds->ds_quota);
1807 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1808 	    ds->ds_reserved);
1809 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1810 	    dsl_dataset_phys(ds)->ds_guid);
1811 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1812 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1813 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1814 	    ds->ds_object);
1815 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1816 	    ds->ds_userrefs);
1817 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1818 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1819 
1820 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1821 		uint64_t written, comp, uncomp;
1822 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1823 		dsl_dataset_t *prev;
1824 
1825 		int err = dsl_dataset_hold_obj(dp,
1826 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1827 		if (err == 0) {
1828 			err = dsl_dataset_space_written(prev, ds, &written,
1829 			    &comp, &uncomp);
1830 			dsl_dataset_rele(prev, FTAG);
1831 			if (err == 0) {
1832 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1833 				    written);
1834 			}
1835 		}
1836 	}
1837 
1838 	if (!dsl_dataset_is_snapshot(ds)) {
1839 		/*
1840 		 * A failed "newfs" (e.g. full) resumable receive leaves
1841 		 * the stats set on this dataset.  Check here for the prop.
1842 		 */
1843 		get_receive_resume_stats(ds, nv);
1844 
1845 		/*
1846 		 * A failed incremental resumable receive leaves the
1847 		 * stats set on our child named "%recv".  Check the child
1848 		 * for the prop.
1849 		 */
1850 		char recvname[ZFS_MAXNAMELEN];
1851 		dsl_dataset_t *recv_ds;
1852 		dsl_dataset_name(ds, recvname);
1853 		(void) strcat(recvname, "/");
1854 		(void) strcat(recvname, recv_clone_name);
1855 		if (dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1856 			get_receive_resume_stats(recv_ds, nv);
1857 			dsl_dataset_rele(recv_ds, FTAG);
1858 		}
1859 	}
1860 }
1861 
1862 void
1863 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1864 {
1865 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1866 	ASSERT(dsl_pool_config_held(dp));
1867 
1868 	stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1869 	stat->dds_inconsistent =
1870 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1871 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1872 	stat->dds_origin[0] = '\0';
1873 	if (ds->ds_is_snapshot) {
1874 		stat->dds_is_snapshot = B_TRUE;
1875 		stat->dds_num_clones =
1876 		    dsl_dataset_phys(ds)->ds_num_children - 1;
1877 	} else {
1878 		stat->dds_is_snapshot = B_FALSE;
1879 		stat->dds_num_clones = 0;
1880 
1881 		if (dsl_dir_is_clone(ds->ds_dir)) {
1882 			dsl_dataset_t *ods;
1883 
1884 			VERIFY0(dsl_dataset_hold_obj(dp,
1885 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1886 			    FTAG, &ods));
1887 			dsl_dataset_name(ods, stat->dds_origin);
1888 			dsl_dataset_rele(ods, FTAG);
1889 		}
1890 	}
1891 }
1892 
1893 uint64_t
1894 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1895 {
1896 	return (ds->ds_fsid_guid);
1897 }
1898 
1899 void
1900 dsl_dataset_space(dsl_dataset_t *ds,
1901     uint64_t *refdbytesp, uint64_t *availbytesp,
1902     uint64_t *usedobjsp, uint64_t *availobjsp)
1903 {
1904 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1905 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1906 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1907 		*availbytesp +=
1908 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1909 	if (ds->ds_quota != 0) {
1910 		/*
1911 		 * Adjust available bytes according to refquota
1912 		 */
1913 		if (*refdbytesp < ds->ds_quota)
1914 			*availbytesp = MIN(*availbytesp,
1915 			    ds->ds_quota - *refdbytesp);
1916 		else
1917 			*availbytesp = 0;
1918 	}
1919 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1920 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1921 }
1922 
1923 boolean_t
1924 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1925 {
1926 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1927 
1928 	ASSERT(dsl_pool_config_held(dp));
1929 	if (snap == NULL)
1930 		return (B_FALSE);
1931 	if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1932 	    dsl_dataset_phys(snap)->ds_creation_txg) {
1933 		objset_t *os, *os_snap;
1934 		/*
1935 		 * It may be that only the ZIL differs, because it was
1936 		 * reset in the head.  Don't count that as being
1937 		 * modified.
1938 		 */
1939 		if (dmu_objset_from_ds(ds, &os) != 0)
1940 			return (B_TRUE);
1941 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1942 			return (B_TRUE);
1943 		return (bcmp(&os->os_phys->os_meta_dnode,
1944 		    &os_snap->os_phys->os_meta_dnode,
1945 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1946 	}
1947 	return (B_FALSE);
1948 }
1949 
1950 typedef struct dsl_dataset_rename_snapshot_arg {
1951 	const char *ddrsa_fsname;
1952 	const char *ddrsa_oldsnapname;
1953 	const char *ddrsa_newsnapname;
1954 	boolean_t ddrsa_recursive;
1955 	dmu_tx_t *ddrsa_tx;
1956 } dsl_dataset_rename_snapshot_arg_t;
1957 
1958 /* ARGSUSED */
1959 static int
1960 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1961     dsl_dataset_t *hds, void *arg)
1962 {
1963 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1964 	int error;
1965 	uint64_t val;
1966 
1967 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1968 	if (error != 0) {
1969 		/* ignore nonexistent snapshots */
1970 		return (error == ENOENT ? 0 : error);
1971 	}
1972 
1973 	/* new name should not exist */
1974 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1975 	if (error == 0)
1976 		error = SET_ERROR(EEXIST);
1977 	else if (error == ENOENT)
1978 		error = 0;
1979 
1980 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1981 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1982 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1983 		error = SET_ERROR(ENAMETOOLONG);
1984 
1985 	return (error);
1986 }
1987 
1988 static int
1989 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1990 {
1991 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1992 	dsl_pool_t *dp = dmu_tx_pool(tx);
1993 	dsl_dataset_t *hds;
1994 	int error;
1995 
1996 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1997 	if (error != 0)
1998 		return (error);
1999 
2000 	if (ddrsa->ddrsa_recursive) {
2001 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2002 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
2003 		    DS_FIND_CHILDREN);
2004 	} else {
2005 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2006 	}
2007 	dsl_dataset_rele(hds, FTAG);
2008 	return (error);
2009 }
2010 
2011 static int
2012 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2013     dsl_dataset_t *hds, void *arg)
2014 {
2015 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2016 	dsl_dataset_t *ds;
2017 	uint64_t val;
2018 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
2019 	int error;
2020 
2021 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2022 	ASSERT(error == 0 || error == ENOENT);
2023 	if (error == ENOENT) {
2024 		/* ignore nonexistent snapshots */
2025 		return (0);
2026 	}
2027 
2028 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2029 
2030 	/* log before we change the name */
2031 	spa_history_log_internal_ds(ds, "rename", tx,
2032 	    "-> @%s", ddrsa->ddrsa_newsnapname);
2033 
2034 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2035 	    B_FALSE));
2036 	mutex_enter(&ds->ds_lock);
2037 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2038 	mutex_exit(&ds->ds_lock);
2039 	VERIFY0(zap_add(dp->dp_meta_objset,
2040 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2041 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2042 
2043 	dsl_dataset_rele(ds, FTAG);
2044 	return (0);
2045 }
2046 
2047 static void
2048 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2049 {
2050 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2051 	dsl_pool_t *dp = dmu_tx_pool(tx);
2052 	dsl_dataset_t *hds;
2053 
2054 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2055 	ddrsa->ddrsa_tx = tx;
2056 	if (ddrsa->ddrsa_recursive) {
2057 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2058 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2059 		    DS_FIND_CHILDREN));
2060 	} else {
2061 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2062 	}
2063 	dsl_dataset_rele(hds, FTAG);
2064 }
2065 
2066 int
2067 dsl_dataset_rename_snapshot(const char *fsname,
2068     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2069 {
2070 	dsl_dataset_rename_snapshot_arg_t ddrsa;
2071 
2072 	ddrsa.ddrsa_fsname = fsname;
2073 	ddrsa.ddrsa_oldsnapname = oldsnapname;
2074 	ddrsa.ddrsa_newsnapname = newsnapname;
2075 	ddrsa.ddrsa_recursive = recursive;
2076 
2077 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2078 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
2079 	    1, ZFS_SPACE_CHECK_RESERVED));
2080 }
2081 
2082 /*
2083  * If we're doing an ownership handoff, we need to make sure that there is
2084  * only one long hold on the dataset.  We're not allowed to change anything here
2085  * so we don't permanently release the long hold or regular hold here.  We want
2086  * to do this only when syncing to avoid the dataset unexpectedly going away
2087  * when we release the long hold.
2088  */
2089 static int
2090 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2091 {
2092 	boolean_t held;
2093 
2094 	if (!dmu_tx_is_syncing(tx))
2095 		return (0);
2096 
2097 	if (owner != NULL) {
2098 		VERIFY3P(ds->ds_owner, ==, owner);
2099 		dsl_dataset_long_rele(ds, owner);
2100 	}
2101 
2102 	held = dsl_dataset_long_held(ds);
2103 
2104 	if (owner != NULL)
2105 		dsl_dataset_long_hold(ds, owner);
2106 
2107 	if (held)
2108 		return (SET_ERROR(EBUSY));
2109 
2110 	return (0);
2111 }
2112 
2113 typedef struct dsl_dataset_rollback_arg {
2114 	const char *ddra_fsname;
2115 	void *ddra_owner;
2116 	nvlist_t *ddra_result;
2117 } dsl_dataset_rollback_arg_t;
2118 
2119 static int
2120 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2121 {
2122 	dsl_dataset_rollback_arg_t *ddra = arg;
2123 	dsl_pool_t *dp = dmu_tx_pool(tx);
2124 	dsl_dataset_t *ds;
2125 	int64_t unused_refres_delta;
2126 	int error;
2127 
2128 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2129 	if (error != 0)
2130 		return (error);
2131 
2132 	/* must not be a snapshot */
2133 	if (ds->ds_is_snapshot) {
2134 		dsl_dataset_rele(ds, FTAG);
2135 		return (SET_ERROR(EINVAL));
2136 	}
2137 
2138 	/* must have a most recent snapshot */
2139 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2140 		dsl_dataset_rele(ds, FTAG);
2141 		return (SET_ERROR(EINVAL));
2142 	}
2143 
2144 	/* must not have any bookmarks after the most recent snapshot */
2145 	nvlist_t *proprequest = fnvlist_alloc();
2146 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2147 	nvlist_t *bookmarks = fnvlist_alloc();
2148 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2149 	fnvlist_free(proprequest);
2150 	if (error != 0)
2151 		return (error);
2152 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2153 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2154 		nvlist_t *valuenv =
2155 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2156 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
2157 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2158 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2159 			fnvlist_free(bookmarks);
2160 			dsl_dataset_rele(ds, FTAG);
2161 			return (SET_ERROR(EEXIST));
2162 		}
2163 	}
2164 	fnvlist_free(bookmarks);
2165 
2166 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2167 	if (error != 0) {
2168 		dsl_dataset_rele(ds, FTAG);
2169 		return (error);
2170 	}
2171 
2172 	/*
2173 	 * Check if the snap we are rolling back to uses more than
2174 	 * the refquota.
2175 	 */
2176 	if (ds->ds_quota != 0 &&
2177 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2178 		dsl_dataset_rele(ds, FTAG);
2179 		return (SET_ERROR(EDQUOT));
2180 	}
2181 
2182 	/*
2183 	 * When we do the clone swap, we will temporarily use more space
2184 	 * due to the refreservation (the head will no longer have any
2185 	 * unique space, so the entire amount of the refreservation will need
2186 	 * to be free).  We will immediately destroy the clone, freeing
2187 	 * this space, but the freeing happens over many txg's.
2188 	 */
2189 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2190 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2191 
2192 	if (unused_refres_delta > 0 &&
2193 	    unused_refres_delta >
2194 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2195 		dsl_dataset_rele(ds, FTAG);
2196 		return (SET_ERROR(ENOSPC));
2197 	}
2198 
2199 	dsl_dataset_rele(ds, FTAG);
2200 	return (0);
2201 }
2202 
2203 static void
2204 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2205 {
2206 	dsl_dataset_rollback_arg_t *ddra = arg;
2207 	dsl_pool_t *dp = dmu_tx_pool(tx);
2208 	dsl_dataset_t *ds, *clone;
2209 	uint64_t cloneobj;
2210 	char namebuf[ZFS_MAXNAMELEN];
2211 
2212 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2213 
2214 	dsl_dataset_name(ds->ds_prev, namebuf);
2215 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2216 
2217 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2218 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2219 
2220 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2221 
2222 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2223 	dsl_dataset_zero_zil(ds, tx);
2224 
2225 	dsl_destroy_head_sync_impl(clone, tx);
2226 
2227 	dsl_dataset_rele(clone, FTAG);
2228 	dsl_dataset_rele(ds, FTAG);
2229 }
2230 
2231 /*
2232  * Rolls back the given filesystem or volume to the most recent snapshot.
2233  * The name of the most recent snapshot will be returned under key "target"
2234  * in the result nvlist.
2235  *
2236  * If owner != NULL:
2237  * - The existing dataset MUST be owned by the specified owner at entry
2238  * - Upon return, dataset will still be held by the same owner, whether we
2239  *   succeed or not.
2240  *
2241  * This mode is required any time the existing filesystem is mounted.  See
2242  * notes above zfs_suspend_fs() for further details.
2243  */
2244 int
2245 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2246 {
2247 	dsl_dataset_rollback_arg_t ddra;
2248 
2249 	ddra.ddra_fsname = fsname;
2250 	ddra.ddra_owner = owner;
2251 	ddra.ddra_result = result;
2252 
2253 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2254 	    dsl_dataset_rollback_sync, &ddra,
2255 	    1, ZFS_SPACE_CHECK_RESERVED));
2256 }
2257 
2258 struct promotenode {
2259 	list_node_t link;
2260 	dsl_dataset_t *ds;
2261 };
2262 
2263 typedef struct dsl_dataset_promote_arg {
2264 	const char *ddpa_clonename;
2265 	dsl_dataset_t *ddpa_clone;
2266 	list_t shared_snaps, origin_snaps, clone_snaps;
2267 	dsl_dataset_t *origin_origin; /* origin of the origin */
2268 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2269 	char *err_ds;
2270 	cred_t *cr;
2271 } dsl_dataset_promote_arg_t;
2272 
2273 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2274 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2275     void *tag);
2276 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2277 
2278 static int
2279 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2280 {
2281 	dsl_dataset_promote_arg_t *ddpa = arg;
2282 	dsl_pool_t *dp = dmu_tx_pool(tx);
2283 	dsl_dataset_t *hds;
2284 	struct promotenode *snap;
2285 	dsl_dataset_t *origin_ds;
2286 	int err;
2287 	uint64_t unused;
2288 	uint64_t ss_mv_cnt;
2289 	size_t max_snap_len;
2290 
2291 	err = promote_hold(ddpa, dp, FTAG);
2292 	if (err != 0)
2293 		return (err);
2294 
2295 	hds = ddpa->ddpa_clone;
2296 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2297 
2298 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2299 		promote_rele(ddpa, FTAG);
2300 		return (SET_ERROR(EXDEV));
2301 	}
2302 
2303 	/*
2304 	 * Compute and check the amount of space to transfer.  Since this is
2305 	 * so expensive, don't do the preliminary check.
2306 	 */
2307 	if (!dmu_tx_is_syncing(tx)) {
2308 		promote_rele(ddpa, FTAG);
2309 		return (0);
2310 	}
2311 
2312 	snap = list_head(&ddpa->shared_snaps);
2313 	origin_ds = snap->ds;
2314 
2315 	/* compute origin's new unique space */
2316 	snap = list_tail(&ddpa->clone_snaps);
2317 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2318 	    origin_ds->ds_object);
2319 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2320 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2321 	    &ddpa->unique, &unused, &unused);
2322 
2323 	/*
2324 	 * Walk the snapshots that we are moving
2325 	 *
2326 	 * Compute space to transfer.  Consider the incremental changes
2327 	 * to used by each snapshot:
2328 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2329 	 * So each snapshot gave birth to:
2330 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2331 	 * So a sequence would look like:
2332 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2333 	 * Which simplifies to:
2334 	 * uN + kN + kN-1 + ... + k1 + k0
2335 	 * Note however, if we stop before we reach the ORIGIN we get:
2336 	 * uN + kN + kN-1 + ... + kM - uM-1
2337 	 */
2338 	ss_mv_cnt = 0;
2339 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2340 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2341 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2342 	for (snap = list_head(&ddpa->shared_snaps); snap;
2343 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2344 		uint64_t val, dlused, dlcomp, dluncomp;
2345 		dsl_dataset_t *ds = snap->ds;
2346 
2347 		ss_mv_cnt++;
2348 
2349 		/*
2350 		 * If there are long holds, we won't be able to evict
2351 		 * the objset.
2352 		 */
2353 		if (dsl_dataset_long_held(ds)) {
2354 			err = SET_ERROR(EBUSY);
2355 			goto out;
2356 		}
2357 
2358 		/* Check that the snapshot name does not conflict */
2359 		VERIFY0(dsl_dataset_get_snapname(ds));
2360 		if (strlen(ds->ds_snapname) >= max_snap_len) {
2361 			err = SET_ERROR(ENAMETOOLONG);
2362 			goto out;
2363 		}
2364 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2365 		if (err == 0) {
2366 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2367 			err = SET_ERROR(EEXIST);
2368 			goto out;
2369 		}
2370 		if (err != ENOENT)
2371 			goto out;
2372 
2373 		/* The very first snapshot does not have a deadlist */
2374 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2375 			continue;
2376 
2377 		dsl_deadlist_space(&ds->ds_deadlist,
2378 		    &dlused, &dlcomp, &dluncomp);
2379 		ddpa->used += dlused;
2380 		ddpa->comp += dlcomp;
2381 		ddpa->uncomp += dluncomp;
2382 	}
2383 
2384 	/*
2385 	 * If we are a clone of a clone then we never reached ORIGIN,
2386 	 * so we need to subtract out the clone origin's used space.
2387 	 */
2388 	if (ddpa->origin_origin) {
2389 		ddpa->used -=
2390 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2391 		ddpa->comp -=
2392 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2393 		ddpa->uncomp -=
2394 		    dsl_dataset_phys(ddpa->origin_origin)->
2395 		    ds_uncompressed_bytes;
2396 	}
2397 
2398 	/* Check that there is enough space and limit headroom here */
2399 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2400 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2401 	if (err != 0)
2402 		goto out;
2403 
2404 	/*
2405 	 * Compute the amounts of space that will be used by snapshots
2406 	 * after the promotion (for both origin and clone).  For each,
2407 	 * it is the amount of space that will be on all of their
2408 	 * deadlists (that was not born before their new origin).
2409 	 */
2410 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2411 		uint64_t space;
2412 
2413 		/*
2414 		 * Note, typically this will not be a clone of a clone,
2415 		 * so dd_origin_txg will be < TXG_INITIAL, so
2416 		 * these snaplist_space() -> dsl_deadlist_space_range()
2417 		 * calls will be fast because they do not have to
2418 		 * iterate over all bps.
2419 		 */
2420 		snap = list_head(&ddpa->origin_snaps);
2421 		err = snaplist_space(&ddpa->shared_snaps,
2422 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2423 		if (err != 0)
2424 			goto out;
2425 
2426 		err = snaplist_space(&ddpa->clone_snaps,
2427 		    snap->ds->ds_dir->dd_origin_txg, &space);
2428 		if (err != 0)
2429 			goto out;
2430 		ddpa->cloneusedsnap += space;
2431 	}
2432 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2433 	    DD_FLAG_USED_BREAKDOWN) {
2434 		err = snaplist_space(&ddpa->origin_snaps,
2435 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2436 		    &ddpa->originusedsnap);
2437 		if (err != 0)
2438 			goto out;
2439 	}
2440 
2441 out:
2442 	promote_rele(ddpa, FTAG);
2443 	return (err);
2444 }
2445 
2446 static void
2447 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2448 {
2449 	dsl_dataset_promote_arg_t *ddpa = arg;
2450 	dsl_pool_t *dp = dmu_tx_pool(tx);
2451 	dsl_dataset_t *hds;
2452 	struct promotenode *snap;
2453 	dsl_dataset_t *origin_ds;
2454 	dsl_dataset_t *origin_head;
2455 	dsl_dir_t *dd;
2456 	dsl_dir_t *odd = NULL;
2457 	uint64_t oldnext_obj;
2458 	int64_t delta;
2459 
2460 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2461 	hds = ddpa->ddpa_clone;
2462 
2463 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2464 
2465 	snap = list_head(&ddpa->shared_snaps);
2466 	origin_ds = snap->ds;
2467 	dd = hds->ds_dir;
2468 
2469 	snap = list_head(&ddpa->origin_snaps);
2470 	origin_head = snap->ds;
2471 
2472 	/*
2473 	 * We need to explicitly open odd, since origin_ds's dd will be
2474 	 * changing.
2475 	 */
2476 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2477 	    NULL, FTAG, &odd));
2478 
2479 	/* change origin's next snap */
2480 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2481 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2482 	snap = list_tail(&ddpa->clone_snaps);
2483 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2484 	    origin_ds->ds_object);
2485 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2486 
2487 	/* change the origin's next clone */
2488 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2489 		dsl_dataset_remove_from_next_clones(origin_ds,
2490 		    snap->ds->ds_object, tx);
2491 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2492 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2493 		    oldnext_obj, tx));
2494 	}
2495 
2496 	/* change origin */
2497 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2498 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2499 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2500 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2501 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2502 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2503 	origin_head->ds_dir->dd_origin_txg =
2504 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2505 
2506 	/* change dd_clone entries */
2507 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2508 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2509 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2510 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2511 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2512 		    hds->ds_object, tx));
2513 
2514 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2515 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2516 		    origin_head->ds_object, tx));
2517 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2518 			dsl_dir_phys(dd)->dd_clones =
2519 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2520 			    DMU_OT_NONE, 0, tx);
2521 		}
2522 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2523 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2524 	}
2525 
2526 	/* move snapshots to this dir */
2527 	for (snap = list_head(&ddpa->shared_snaps); snap;
2528 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2529 		dsl_dataset_t *ds = snap->ds;
2530 
2531 		/*
2532 		 * Property callbacks are registered to a particular
2533 		 * dsl_dir.  Since ours is changing, evict the objset
2534 		 * so that they will be unregistered from the old dsl_dir.
2535 		 */
2536 		if (ds->ds_objset) {
2537 			dmu_objset_evict(ds->ds_objset);
2538 			ds->ds_objset = NULL;
2539 		}
2540 
2541 		/* move snap name entry */
2542 		VERIFY0(dsl_dataset_get_snapname(ds));
2543 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2544 		    ds->ds_snapname, tx, B_TRUE));
2545 		VERIFY0(zap_add(dp->dp_meta_objset,
2546 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2547 		    8, 1, &ds->ds_object, tx));
2548 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2549 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2550 
2551 		/* change containing dsl_dir */
2552 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2553 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2554 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2555 		ASSERT3P(ds->ds_dir, ==, odd);
2556 		dsl_dir_rele(ds->ds_dir, ds);
2557 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2558 		    NULL, ds, &ds->ds_dir));
2559 
2560 		/* move any clone references */
2561 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2562 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2563 			zap_cursor_t zc;
2564 			zap_attribute_t za;
2565 
2566 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2567 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
2568 			    zap_cursor_retrieve(&zc, &za) == 0;
2569 			    zap_cursor_advance(&zc)) {
2570 				dsl_dataset_t *cnds;
2571 				uint64_t o;
2572 
2573 				if (za.za_first_integer == oldnext_obj) {
2574 					/*
2575 					 * We've already moved the
2576 					 * origin's reference.
2577 					 */
2578 					continue;
2579 				}
2580 
2581 				VERIFY0(dsl_dataset_hold_obj(dp,
2582 				    za.za_first_integer, FTAG, &cnds));
2583 				o = dsl_dir_phys(cnds->ds_dir)->
2584 				    dd_head_dataset_obj;
2585 
2586 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2587 				    dsl_dir_phys(odd)->dd_clones, o, tx));
2588 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2589 				    dsl_dir_phys(dd)->dd_clones, o, tx));
2590 				dsl_dataset_rele(cnds, FTAG);
2591 			}
2592 			zap_cursor_fini(&zc);
2593 		}
2594 
2595 		ASSERT(!dsl_prop_hascb(ds));
2596 	}
2597 
2598 	/*
2599 	 * Change space accounting.
2600 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2601 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2602 	 * is true for each of {clone,origin} independently.
2603 	 */
2604 
2605 	delta = ddpa->cloneusedsnap -
2606 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2607 	ASSERT3S(delta, >=, 0);
2608 	ASSERT3U(ddpa->used, >=, delta);
2609 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2610 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2611 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2612 
2613 	delta = ddpa->originusedsnap -
2614 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2615 	ASSERT3S(delta, <=, 0);
2616 	ASSERT3U(ddpa->used, >=, -delta);
2617 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2618 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2619 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2620 
2621 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2622 
2623 	/* log history record */
2624 	spa_history_log_internal_ds(hds, "promote", tx, "");
2625 
2626 	dsl_dir_rele(odd, FTAG);
2627 	promote_rele(ddpa, FTAG);
2628 }
2629 
2630 /*
2631  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2632  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2633  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2634  * snapshots back to this dataset's origin.
2635  */
2636 static int
2637 snaplist_make(dsl_pool_t *dp,
2638     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2639 {
2640 	uint64_t obj = last_obj;
2641 
2642 	list_create(l, sizeof (struct promotenode),
2643 	    offsetof(struct promotenode, link));
2644 
2645 	while (obj != first_obj) {
2646 		dsl_dataset_t *ds;
2647 		struct promotenode *snap;
2648 		int err;
2649 
2650 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2651 		ASSERT(err != ENOENT);
2652 		if (err != 0)
2653 			return (err);
2654 
2655 		if (first_obj == 0)
2656 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2657 
2658 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2659 		snap->ds = ds;
2660 		list_insert_tail(l, snap);
2661 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2662 	}
2663 
2664 	return (0);
2665 }
2666 
2667 static int
2668 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2669 {
2670 	struct promotenode *snap;
2671 
2672 	*spacep = 0;
2673 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2674 		uint64_t used, comp, uncomp;
2675 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2676 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2677 		*spacep += used;
2678 	}
2679 	return (0);
2680 }
2681 
2682 static void
2683 snaplist_destroy(list_t *l, void *tag)
2684 {
2685 	struct promotenode *snap;
2686 
2687 	if (l == NULL || !list_link_active(&l->list_head))
2688 		return;
2689 
2690 	while ((snap = list_tail(l)) != NULL) {
2691 		list_remove(l, snap);
2692 		dsl_dataset_rele(snap->ds, tag);
2693 		kmem_free(snap, sizeof (*snap));
2694 	}
2695 	list_destroy(l);
2696 }
2697 
2698 static int
2699 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2700 {
2701 	int error;
2702 	dsl_dir_t *dd;
2703 	struct promotenode *snap;
2704 
2705 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2706 	    &ddpa->ddpa_clone);
2707 	if (error != 0)
2708 		return (error);
2709 	dd = ddpa->ddpa_clone->ds_dir;
2710 
2711 	if (ddpa->ddpa_clone->ds_is_snapshot ||
2712 	    !dsl_dir_is_clone(dd)) {
2713 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2714 		return (SET_ERROR(EINVAL));
2715 	}
2716 
2717 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2718 	    &ddpa->shared_snaps, tag);
2719 	if (error != 0)
2720 		goto out;
2721 
2722 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2723 	    &ddpa->clone_snaps, tag);
2724 	if (error != 0)
2725 		goto out;
2726 
2727 	snap = list_head(&ddpa->shared_snaps);
2728 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2729 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2730 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2731 	    &ddpa->origin_snaps, tag);
2732 	if (error != 0)
2733 		goto out;
2734 
2735 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2736 		error = dsl_dataset_hold_obj(dp,
2737 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2738 		    tag, &ddpa->origin_origin);
2739 		if (error != 0)
2740 			goto out;
2741 	}
2742 out:
2743 	if (error != 0)
2744 		promote_rele(ddpa, tag);
2745 	return (error);
2746 }
2747 
2748 static void
2749 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2750 {
2751 	snaplist_destroy(&ddpa->shared_snaps, tag);
2752 	snaplist_destroy(&ddpa->clone_snaps, tag);
2753 	snaplist_destroy(&ddpa->origin_snaps, tag);
2754 	if (ddpa->origin_origin != NULL)
2755 		dsl_dataset_rele(ddpa->origin_origin, tag);
2756 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2757 }
2758 
2759 /*
2760  * Promote a clone.
2761  *
2762  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2763  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2764  */
2765 int
2766 dsl_dataset_promote(const char *name, char *conflsnap)
2767 {
2768 	dsl_dataset_promote_arg_t ddpa = { 0 };
2769 	uint64_t numsnaps;
2770 	int error;
2771 	objset_t *os;
2772 
2773 	/*
2774 	 * We will modify space proportional to the number of
2775 	 * snapshots.  Compute numsnaps.
2776 	 */
2777 	error = dmu_objset_hold(name, FTAG, &os);
2778 	if (error != 0)
2779 		return (error);
2780 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2781 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2782 	    &numsnaps);
2783 	dmu_objset_rele(os, FTAG);
2784 	if (error != 0)
2785 		return (error);
2786 
2787 	ddpa.ddpa_clonename = name;
2788 	ddpa.err_ds = conflsnap;
2789 	ddpa.cr = CRED();
2790 
2791 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2792 	    dsl_dataset_promote_sync, &ddpa,
2793 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2794 }
2795 
2796 int
2797 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2798     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2799 {
2800 	/*
2801 	 * "slack" factor for received datasets with refquota set on them.
2802 	 * See the bottom of this function for details on its use.
2803 	 */
2804 	uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
2805 	int64_t unused_refres_delta;
2806 
2807 	/* they should both be heads */
2808 	if (clone->ds_is_snapshot ||
2809 	    origin_head->ds_is_snapshot)
2810 		return (SET_ERROR(EINVAL));
2811 
2812 	/* if we are not forcing, the branch point should be just before them */
2813 	if (!force && clone->ds_prev != origin_head->ds_prev)
2814 		return (SET_ERROR(EINVAL));
2815 
2816 	/* clone should be the clone (unless they are unrelated) */
2817 	if (clone->ds_prev != NULL &&
2818 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2819 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2820 		return (SET_ERROR(EINVAL));
2821 
2822 	/* the clone should be a child of the origin */
2823 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2824 		return (SET_ERROR(EINVAL));
2825 
2826 	/* origin_head shouldn't be modified unless 'force' */
2827 	if (!force &&
2828 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2829 		return (SET_ERROR(ETXTBSY));
2830 
2831 	/* origin_head should have no long holds (e.g. is not mounted) */
2832 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2833 		return (SET_ERROR(EBUSY));
2834 
2835 	/* check amount of any unconsumed refreservation */
2836 	unused_refres_delta =
2837 	    (int64_t)MIN(origin_head->ds_reserved,
2838 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2839 	    (int64_t)MIN(origin_head->ds_reserved,
2840 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2841 
2842 	if (unused_refres_delta > 0 &&
2843 	    unused_refres_delta >
2844 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2845 		return (SET_ERROR(ENOSPC));
2846 
2847 	/*
2848 	 * The clone can't be too much over the head's refquota.
2849 	 *
2850 	 * To ensure that the entire refquota can be used, we allow one
2851 	 * transaction to exceed the the refquota.  Therefore, this check
2852 	 * needs to also allow for the space referenced to be more than the
2853 	 * refquota.  The maximum amount of space that one transaction can use
2854 	 * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
2855 	 * overage ensures that we are able to receive a filesystem that
2856 	 * exceeds the refquota on the source system.
2857 	 *
2858 	 * So that overage is the refquota_slack we use below.
2859 	 */
2860 	if (origin_head->ds_quota != 0 &&
2861 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
2862 	    origin_head->ds_quota + refquota_slack)
2863 		return (SET_ERROR(EDQUOT));
2864 
2865 	return (0);
2866 }
2867 
2868 void
2869 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2870     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2871 {
2872 	dsl_pool_t *dp = dmu_tx_pool(tx);
2873 	int64_t unused_refres_delta;
2874 
2875 	ASSERT(clone->ds_reserved == 0);
2876 	/*
2877 	 * NOTE: On DEBUG kernels there could be a race between this and
2878 	 * the check function if spa_asize_inflation is adjusted...
2879 	 */
2880 	ASSERT(origin_head->ds_quota == 0 ||
2881 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
2882 	    DMU_MAX_ACCESS * spa_asize_inflation);
2883 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2884 
2885 	/*
2886 	 * Swap per-dataset feature flags.
2887 	 */
2888 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2889 		if (!(spa_feature_table[f].fi_flags &
2890 		    ZFEATURE_FLAG_PER_DATASET)) {
2891 			ASSERT(!clone->ds_feature_inuse[f]);
2892 			ASSERT(!origin_head->ds_feature_inuse[f]);
2893 			continue;
2894 		}
2895 
2896 		boolean_t clone_inuse = clone->ds_feature_inuse[f];
2897 		boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2898 
2899 		if (clone_inuse) {
2900 			dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2901 			clone->ds_feature_inuse[f] = B_FALSE;
2902 		}
2903 		if (origin_head_inuse) {
2904 			dsl_dataset_deactivate_feature(origin_head->ds_object,
2905 			    f, tx);
2906 			origin_head->ds_feature_inuse[f] = B_FALSE;
2907 		}
2908 		if (clone_inuse) {
2909 			dsl_dataset_activate_feature(origin_head->ds_object,
2910 			    f, tx);
2911 			origin_head->ds_feature_inuse[f] = B_TRUE;
2912 		}
2913 		if (origin_head_inuse) {
2914 			dsl_dataset_activate_feature(clone->ds_object, f, tx);
2915 			clone->ds_feature_inuse[f] = B_TRUE;
2916 		}
2917 	}
2918 
2919 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2920 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2921 
2922 	if (clone->ds_objset != NULL) {
2923 		dmu_objset_evict(clone->ds_objset);
2924 		clone->ds_objset = NULL;
2925 	}
2926 
2927 	if (origin_head->ds_objset != NULL) {
2928 		dmu_objset_evict(origin_head->ds_objset);
2929 		origin_head->ds_objset = NULL;
2930 	}
2931 
2932 	unused_refres_delta =
2933 	    (int64_t)MIN(origin_head->ds_reserved,
2934 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2935 	    (int64_t)MIN(origin_head->ds_reserved,
2936 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2937 
2938 	/*
2939 	 * Reset origin's unique bytes, if it exists.
2940 	 */
2941 	if (clone->ds_prev) {
2942 		dsl_dataset_t *origin = clone->ds_prev;
2943 		uint64_t comp, uncomp;
2944 
2945 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2946 		dsl_deadlist_space_range(&clone->ds_deadlist,
2947 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2948 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2949 	}
2950 
2951 	/* swap blkptrs */
2952 	{
2953 		blkptr_t tmp;
2954 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
2955 		dsl_dataset_phys(origin_head)->ds_bp =
2956 		    dsl_dataset_phys(clone)->ds_bp;
2957 		dsl_dataset_phys(clone)->ds_bp = tmp;
2958 	}
2959 
2960 	/* set dd_*_bytes */
2961 	{
2962 		int64_t dused, dcomp, duncomp;
2963 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2964 		uint64_t odl_used, odl_comp, odl_uncomp;
2965 
2966 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2967 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2968 
2969 		dsl_deadlist_space(&clone->ds_deadlist,
2970 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2971 		dsl_deadlist_space(&origin_head->ds_deadlist,
2972 		    &odl_used, &odl_comp, &odl_uncomp);
2973 
2974 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2975 		    cdl_used -
2976 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2977 		    odl_used);
2978 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2979 		    cdl_comp -
2980 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2981 		    odl_comp);
2982 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2983 		    cdl_uncomp -
2984 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2985 		    odl_uncomp);
2986 
2987 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2988 		    dused, dcomp, duncomp, tx);
2989 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2990 		    -dused, -dcomp, -duncomp, tx);
2991 
2992 		/*
2993 		 * The difference in the space used by snapshots is the
2994 		 * difference in snapshot space due to the head's
2995 		 * deadlist (since that's the only thing that's
2996 		 * changing that affects the snapused).
2997 		 */
2998 		dsl_deadlist_space_range(&clone->ds_deadlist,
2999 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3000 		    &cdl_used, &cdl_comp, &cdl_uncomp);
3001 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
3002 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3003 		    &odl_used, &odl_comp, &odl_uncomp);
3004 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3005 		    DD_USED_HEAD, DD_USED_SNAP, tx);
3006 	}
3007 
3008 	/* swap ds_*_bytes */
3009 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3010 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
3011 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3012 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
3013 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3014 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3015 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3016 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3017 
3018 	/* apply any parent delta for change in unconsumed refreservation */
3019 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3020 	    unused_refres_delta, 0, 0, tx);
3021 
3022 	/*
3023 	 * Swap deadlists.
3024 	 */
3025 	dsl_deadlist_close(&clone->ds_deadlist);
3026 	dsl_deadlist_close(&origin_head->ds_deadlist);
3027 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3028 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3029 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3030 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3031 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3032 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3033 
3034 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3035 
3036 	spa_history_log_internal_ds(clone, "clone swap", tx,
3037 	    "parent=%s", origin_head->ds_dir->dd_myname);
3038 }
3039 
3040 /*
3041  * Given a pool name and a dataset object number in that pool,
3042  * return the name of that dataset.
3043  */
3044 int
3045 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3046 {
3047 	dsl_pool_t *dp;
3048 	dsl_dataset_t *ds;
3049 	int error;
3050 
3051 	error = dsl_pool_hold(pname, FTAG, &dp);
3052 	if (error != 0)
3053 		return (error);
3054 
3055 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3056 	if (error == 0) {
3057 		dsl_dataset_name(ds, buf);
3058 		dsl_dataset_rele(ds, FTAG);
3059 	}
3060 	dsl_pool_rele(dp, FTAG);
3061 
3062 	return (error);
3063 }
3064 
3065 int
3066 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3067     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3068 {
3069 	int error = 0;
3070 
3071 	ASSERT3S(asize, >, 0);
3072 
3073 	/*
3074 	 * *ref_rsrv is the portion of asize that will come from any
3075 	 * unconsumed refreservation space.
3076 	 */
3077 	*ref_rsrv = 0;
3078 
3079 	mutex_enter(&ds->ds_lock);
3080 	/*
3081 	 * Make a space adjustment for reserved bytes.
3082 	 */
3083 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3084 		ASSERT3U(*used, >=,
3085 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3086 		*used -=
3087 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3088 		*ref_rsrv =
3089 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3090 	}
3091 
3092 	if (!check_quota || ds->ds_quota == 0) {
3093 		mutex_exit(&ds->ds_lock);
3094 		return (0);
3095 	}
3096 	/*
3097 	 * If they are requesting more space, and our current estimate
3098 	 * is over quota, they get to try again unless the actual
3099 	 * on-disk is over quota and there are no pending changes (which
3100 	 * may free up space for us).
3101 	 */
3102 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3103 	    ds->ds_quota) {
3104 		if (inflight > 0 ||
3105 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3106 			error = SET_ERROR(ERESTART);
3107 		else
3108 			error = SET_ERROR(EDQUOT);
3109 	}
3110 	mutex_exit(&ds->ds_lock);
3111 
3112 	return (error);
3113 }
3114 
3115 typedef struct dsl_dataset_set_qr_arg {
3116 	const char *ddsqra_name;
3117 	zprop_source_t ddsqra_source;
3118 	uint64_t ddsqra_value;
3119 } dsl_dataset_set_qr_arg_t;
3120 
3121 
3122 /* ARGSUSED */
3123 static int
3124 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3125 {
3126 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3127 	dsl_pool_t *dp = dmu_tx_pool(tx);
3128 	dsl_dataset_t *ds;
3129 	int error;
3130 	uint64_t newval;
3131 
3132 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3133 		return (SET_ERROR(ENOTSUP));
3134 
3135 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3136 	if (error != 0)
3137 		return (error);
3138 
3139 	if (ds->ds_is_snapshot) {
3140 		dsl_dataset_rele(ds, FTAG);
3141 		return (SET_ERROR(EINVAL));
3142 	}
3143 
3144 	error = dsl_prop_predict(ds->ds_dir,
3145 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3146 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3147 	if (error != 0) {
3148 		dsl_dataset_rele(ds, FTAG);
3149 		return (error);
3150 	}
3151 
3152 	if (newval == 0) {
3153 		dsl_dataset_rele(ds, FTAG);
3154 		return (0);
3155 	}
3156 
3157 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3158 	    newval < ds->ds_reserved) {
3159 		dsl_dataset_rele(ds, FTAG);
3160 		return (SET_ERROR(ENOSPC));
3161 	}
3162 
3163 	dsl_dataset_rele(ds, FTAG);
3164 	return (0);
3165 }
3166 
3167 static void
3168 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3169 {
3170 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3171 	dsl_pool_t *dp = dmu_tx_pool(tx);
3172 	dsl_dataset_t *ds;
3173 	uint64_t newval;
3174 
3175 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3176 
3177 	dsl_prop_set_sync_impl(ds,
3178 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3179 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3180 	    &ddsqra->ddsqra_value, tx);
3181 
3182 	VERIFY0(dsl_prop_get_int_ds(ds,
3183 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3184 
3185 	if (ds->ds_quota != newval) {
3186 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3187 		ds->ds_quota = newval;
3188 	}
3189 	dsl_dataset_rele(ds, FTAG);
3190 }
3191 
3192 int
3193 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3194     uint64_t refquota)
3195 {
3196 	dsl_dataset_set_qr_arg_t ddsqra;
3197 
3198 	ddsqra.ddsqra_name = dsname;
3199 	ddsqra.ddsqra_source = source;
3200 	ddsqra.ddsqra_value = refquota;
3201 
3202 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3203 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3204 }
3205 
3206 static int
3207 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3208 {
3209 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3210 	dsl_pool_t *dp = dmu_tx_pool(tx);
3211 	dsl_dataset_t *ds;
3212 	int error;
3213 	uint64_t newval, unique;
3214 
3215 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3216 		return (SET_ERROR(ENOTSUP));
3217 
3218 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3219 	if (error != 0)
3220 		return (error);
3221 
3222 	if (ds->ds_is_snapshot) {
3223 		dsl_dataset_rele(ds, FTAG);
3224 		return (SET_ERROR(EINVAL));
3225 	}
3226 
3227 	error = dsl_prop_predict(ds->ds_dir,
3228 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3229 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3230 	if (error != 0) {
3231 		dsl_dataset_rele(ds, FTAG);
3232 		return (error);
3233 	}
3234 
3235 	/*
3236 	 * If we are doing the preliminary check in open context, the
3237 	 * space estimates may be inaccurate.
3238 	 */
3239 	if (!dmu_tx_is_syncing(tx)) {
3240 		dsl_dataset_rele(ds, FTAG);
3241 		return (0);
3242 	}
3243 
3244 	mutex_enter(&ds->ds_lock);
3245 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3246 		dsl_dataset_recalc_head_uniq(ds);
3247 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3248 	mutex_exit(&ds->ds_lock);
3249 
3250 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3251 		uint64_t delta = MAX(unique, newval) -
3252 		    MAX(unique, ds->ds_reserved);
3253 
3254 		if (delta >
3255 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3256 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3257 			dsl_dataset_rele(ds, FTAG);
3258 			return (SET_ERROR(ENOSPC));
3259 		}
3260 	}
3261 
3262 	dsl_dataset_rele(ds, FTAG);
3263 	return (0);
3264 }
3265 
3266 void
3267 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3268     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3269 {
3270 	uint64_t newval;
3271 	uint64_t unique;
3272 	int64_t delta;
3273 
3274 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3275 	    source, sizeof (value), 1, &value, tx);
3276 
3277 	VERIFY0(dsl_prop_get_int_ds(ds,
3278 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3279 
3280 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3281 	mutex_enter(&ds->ds_dir->dd_lock);
3282 	mutex_enter(&ds->ds_lock);
3283 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3284 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3285 	delta = MAX(0, (int64_t)(newval - unique)) -
3286 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3287 	ds->ds_reserved = newval;
3288 	mutex_exit(&ds->ds_lock);
3289 
3290 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3291 	mutex_exit(&ds->ds_dir->dd_lock);
3292 }
3293 
3294 static void
3295 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3296 {
3297 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3298 	dsl_pool_t *dp = dmu_tx_pool(tx);
3299 	dsl_dataset_t *ds;
3300 
3301 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3302 	dsl_dataset_set_refreservation_sync_impl(ds,
3303 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3304 	dsl_dataset_rele(ds, FTAG);
3305 }
3306 
3307 int
3308 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3309     uint64_t refreservation)
3310 {
3311 	dsl_dataset_set_qr_arg_t ddsqra;
3312 
3313 	ddsqra.ddsqra_name = dsname;
3314 	ddsqra.ddsqra_source = source;
3315 	ddsqra.ddsqra_value = refreservation;
3316 
3317 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3318 	    dsl_dataset_set_refreservation_sync, &ddsqra,
3319 	    0, ZFS_SPACE_CHECK_NONE));
3320 }
3321 
3322 /*
3323  * Return (in *usedp) the amount of space written in new that is not
3324  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3325  * a snapshot before new, in new's filesystem (or its origin).  If not then
3326  * fail and return EINVAL.
3327  *
3328  * The written space is calculated by considering two components:  First, we
3329  * ignore any freed space, and calculate the written as new's used space
3330  * minus old's used space.  Next, we add in the amount of space that was freed
3331  * between the two snapshots, thus reducing new's used space relative to old's.
3332  * Specifically, this is the space that was born before old->ds_creation_txg,
3333  * and freed before new (ie. on new's deadlist or a previous deadlist).
3334  *
3335  * space freed                         [---------------------]
3336  * snapshots                       ---O-------O--------O-------O------
3337  *                                         oldsnap            new
3338  */
3339 int
3340 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3341     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3342 {
3343 	int err = 0;
3344 	uint64_t snapobj;
3345 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3346 
3347 	ASSERT(dsl_pool_config_held(dp));
3348 
3349 	*usedp = 0;
3350 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3351 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3352 
3353 	*compp = 0;
3354 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3355 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3356 
3357 	*uncompp = 0;
3358 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3359 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3360 
3361 	snapobj = new->ds_object;
3362 	while (snapobj != oldsnap->ds_object) {
3363 		dsl_dataset_t *snap;
3364 		uint64_t used, comp, uncomp;
3365 
3366 		if (snapobj == new->ds_object) {
3367 			snap = new;
3368 		} else {
3369 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3370 			if (err != 0)
3371 				break;
3372 		}
3373 
3374 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3375 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3376 			/*
3377 			 * The blocks in the deadlist can not be born after
3378 			 * ds_prev_snap_txg, so get the whole deadlist space,
3379 			 * which is more efficient (especially for old-format
3380 			 * deadlists).  Unfortunately the deadlist code
3381 			 * doesn't have enough information to make this
3382 			 * optimization itself.
3383 			 */
3384 			dsl_deadlist_space(&snap->ds_deadlist,
3385 			    &used, &comp, &uncomp);
3386 		} else {
3387 			dsl_deadlist_space_range(&snap->ds_deadlist,
3388 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3389 			    &used, &comp, &uncomp);
3390 		}
3391 		*usedp += used;
3392 		*compp += comp;
3393 		*uncompp += uncomp;
3394 
3395 		/*
3396 		 * If we get to the beginning of the chain of snapshots
3397 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3398 		 * was not a snapshot of/before new.
3399 		 */
3400 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3401 		if (snap != new)
3402 			dsl_dataset_rele(snap, FTAG);
3403 		if (snapobj == 0) {
3404 			err = SET_ERROR(EINVAL);
3405 			break;
3406 		}
3407 
3408 	}
3409 	return (err);
3410 }
3411 
3412 /*
3413  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3414  * lastsnap, and all snapshots in between are deleted.
3415  *
3416  * blocks that would be freed            [---------------------------]
3417  * snapshots                       ---O-------O--------O-------O--------O
3418  *                                        firstsnap        lastsnap
3419  *
3420  * This is the set of blocks that were born after the snap before firstsnap,
3421  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3422  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3423  * We calculate this by iterating over the relevant deadlists (from the snap
3424  * after lastsnap, backward to the snap after firstsnap), summing up the
3425  * space on the deadlist that was born after the snap before firstsnap.
3426  */
3427 int
3428 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3429     dsl_dataset_t *lastsnap,
3430     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3431 {
3432 	int err = 0;
3433 	uint64_t snapobj;
3434 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3435 
3436 	ASSERT(firstsnap->ds_is_snapshot);
3437 	ASSERT(lastsnap->ds_is_snapshot);
3438 
3439 	/*
3440 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3441 	 * is before lastsnap.
3442 	 */
3443 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3444 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3445 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3446 		return (SET_ERROR(EINVAL));
3447 
3448 	*usedp = *compp = *uncompp = 0;
3449 
3450 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3451 	while (snapobj != firstsnap->ds_object) {
3452 		dsl_dataset_t *ds;
3453 		uint64_t used, comp, uncomp;
3454 
3455 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3456 		if (err != 0)
3457 			break;
3458 
3459 		dsl_deadlist_space_range(&ds->ds_deadlist,
3460 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3461 		    &used, &comp, &uncomp);
3462 		*usedp += used;
3463 		*compp += comp;
3464 		*uncompp += uncomp;
3465 
3466 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3467 		ASSERT3U(snapobj, !=, 0);
3468 		dsl_dataset_rele(ds, FTAG);
3469 	}
3470 	return (err);
3471 }
3472 
3473 /*
3474  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3475  * For example, they could both be snapshots of the same filesystem, and
3476  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3477  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3478  * filesystem.  Or 'earlier' could be the origin's origin.
3479  *
3480  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3481  */
3482 boolean_t
3483 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3484     uint64_t earlier_txg)
3485 {
3486 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3487 	int error;
3488 	boolean_t ret;
3489 
3490 	ASSERT(dsl_pool_config_held(dp));
3491 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3492 
3493 	if (earlier_txg == 0)
3494 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3495 
3496 	if (later->ds_is_snapshot &&
3497 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3498 		return (B_FALSE);
3499 
3500 	if (later->ds_dir == earlier->ds_dir)
3501 		return (B_TRUE);
3502 	if (!dsl_dir_is_clone(later->ds_dir))
3503 		return (B_FALSE);
3504 
3505 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3506 		return (B_TRUE);
3507 	dsl_dataset_t *origin;
3508 	error = dsl_dataset_hold_obj(dp,
3509 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3510 	if (error != 0)
3511 		return (B_FALSE);
3512 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3513 	dsl_dataset_rele(origin, FTAG);
3514 	return (ret);
3515 }
3516 
3517 void
3518 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3519 {
3520 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3521 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3522 }
3523 
3524 boolean_t
3525 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3526 {
3527 	dmu_object_info_t doi;
3528 
3529 	dmu_object_info_from_db(ds->ds_dbuf, &doi);
3530 	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3531 }
3532 
3533 boolean_t
3534 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3535 {
3536 	return (dsl_dataset_is_zapified(ds) &&
3537 	    zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3538 	    ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3539 }
3540