xref: /titanic_51/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision e6546372a86065934a1b311176ecd1b1c543a222)
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, 2016 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 /*
273  * We have to release the fsid syncronously or we risk that a subsequent
274  * mount of the same dataset will fail to unique_insert the fsid.  This
275  * failure would manifest itself as the fsid of this dataset changing
276  * between mounts which makes NFS clients quite unhappy.
277  */
278 static void
279 dsl_dataset_evict_sync(void *dbu)
280 {
281 	dsl_dataset_t *ds = dbu;
282 
283 	ASSERT(ds->ds_owner == NULL);
284 
285 	unique_remove(ds->ds_fsid_guid);
286 }
287 
288 static void
289 dsl_dataset_evict_async(void *dbu)
290 {
291 	dsl_dataset_t *ds = dbu;
292 
293 	ASSERT(ds->ds_owner == NULL);
294 
295 	ds->ds_dbuf = NULL;
296 
297 	if (ds->ds_objset != NULL)
298 		dmu_objset_evict(ds->ds_objset);
299 
300 	if (ds->ds_prev) {
301 		dsl_dataset_rele(ds->ds_prev, ds);
302 		ds->ds_prev = NULL;
303 	}
304 
305 	bplist_destroy(&ds->ds_pending_deadlist);
306 	if (ds->ds_deadlist.dl_os != NULL)
307 		dsl_deadlist_close(&ds->ds_deadlist);
308 	if (ds->ds_dir)
309 		dsl_dir_async_rele(ds->ds_dir, ds);
310 
311 	ASSERT(!list_link_active(&ds->ds_synced_link));
312 
313 	list_destroy(&ds->ds_prop_cbs);
314 	mutex_destroy(&ds->ds_lock);
315 	mutex_destroy(&ds->ds_opening_lock);
316 	mutex_destroy(&ds->ds_sendstream_lock);
317 	refcount_destroy(&ds->ds_longholds);
318 
319 	kmem_free(ds, sizeof (dsl_dataset_t));
320 }
321 
322 int
323 dsl_dataset_get_snapname(dsl_dataset_t *ds)
324 {
325 	dsl_dataset_phys_t *headphys;
326 	int err;
327 	dmu_buf_t *headdbuf;
328 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
329 	objset_t *mos = dp->dp_meta_objset;
330 
331 	if (ds->ds_snapname[0])
332 		return (0);
333 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
334 		return (0);
335 
336 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
337 	    FTAG, &headdbuf);
338 	if (err != 0)
339 		return (err);
340 	headphys = headdbuf->db_data;
341 	err = zap_value_search(dp->dp_meta_objset,
342 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
343 	dmu_buf_rele(headdbuf, FTAG);
344 	return (err);
345 }
346 
347 int
348 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
349 {
350 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
351 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
352 	matchtype_t mt;
353 	int err;
354 
355 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
356 		mt = MT_FIRST;
357 	else
358 		mt = MT_EXACT;
359 
360 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
361 	    value, mt, NULL, 0, NULL);
362 	if (err == ENOTSUP && mt == MT_FIRST)
363 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
364 	return (err);
365 }
366 
367 int
368 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
369     boolean_t adj_cnt)
370 {
371 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
372 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
373 	matchtype_t mt;
374 	int err;
375 
376 	dsl_dir_snap_cmtime_update(ds->ds_dir);
377 
378 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
379 		mt = MT_FIRST;
380 	else
381 		mt = MT_EXACT;
382 
383 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
384 	if (err == ENOTSUP && mt == MT_FIRST)
385 		err = zap_remove(mos, snapobj, name, tx);
386 
387 	if (err == 0 && adj_cnt)
388 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
389 		    DD_FIELD_SNAPSHOT_COUNT, tx);
390 
391 	return (err);
392 }
393 
394 boolean_t
395 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
396 {
397 	dmu_buf_t *dbuf = ds->ds_dbuf;
398 	boolean_t result = B_FALSE;
399 
400 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
401 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
402 
403 		if (ds == dmu_buf_get_user(dbuf))
404 			result = B_TRUE;
405 		else
406 			dmu_buf_rele(dbuf, tag);
407 	}
408 
409 	return (result);
410 }
411 
412 int
413 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
414     dsl_dataset_t **dsp)
415 {
416 	objset_t *mos = dp->dp_meta_objset;
417 	dmu_buf_t *dbuf;
418 	dsl_dataset_t *ds;
419 	int err;
420 	dmu_object_info_t doi;
421 
422 	ASSERT(dsl_pool_config_held(dp));
423 
424 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
425 	if (err != 0)
426 		return (err);
427 
428 	/* Make sure dsobj has the correct object type. */
429 	dmu_object_info_from_db(dbuf, &doi);
430 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
431 		dmu_buf_rele(dbuf, tag);
432 		return (SET_ERROR(EINVAL));
433 	}
434 
435 	ds = dmu_buf_get_user(dbuf);
436 	if (ds == NULL) {
437 		dsl_dataset_t *winner = NULL;
438 
439 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
440 		ds->ds_dbuf = dbuf;
441 		ds->ds_object = dsobj;
442 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
443 
444 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
445 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
446 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
447 		refcount_create(&ds->ds_longholds);
448 
449 		bplist_create(&ds->ds_pending_deadlist);
450 		dsl_deadlist_open(&ds->ds_deadlist,
451 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
452 
453 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
454 		    offsetof(dmu_sendarg_t, dsa_link));
455 
456 		list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
457 		    offsetof(dsl_prop_cb_record_t, cbr_ds_node));
458 
459 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
460 			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
461 				if (!(spa_feature_table[f].fi_flags &
462 				    ZFEATURE_FLAG_PER_DATASET))
463 					continue;
464 				err = zap_contains(mos, dsobj,
465 				    spa_feature_table[f].fi_guid);
466 				if (err == 0) {
467 					ds->ds_feature_inuse[f] = B_TRUE;
468 				} else {
469 					ASSERT3U(err, ==, ENOENT);
470 					err = 0;
471 				}
472 			}
473 		}
474 
475 		err = dsl_dir_hold_obj(dp,
476 		    dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
477 		if (err != 0) {
478 			mutex_destroy(&ds->ds_lock);
479 			mutex_destroy(&ds->ds_opening_lock);
480 			mutex_destroy(&ds->ds_sendstream_lock);
481 			refcount_destroy(&ds->ds_longholds);
482 			bplist_destroy(&ds->ds_pending_deadlist);
483 			dsl_deadlist_close(&ds->ds_deadlist);
484 			kmem_free(ds, sizeof (dsl_dataset_t));
485 			dmu_buf_rele(dbuf, tag);
486 			return (err);
487 		}
488 
489 		if (!ds->ds_is_snapshot) {
490 			ds->ds_snapname[0] = '\0';
491 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
492 				err = dsl_dataset_hold_obj(dp,
493 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
494 				    ds, &ds->ds_prev);
495 			}
496 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
497 				int zaperr = zap_lookup(mos, ds->ds_object,
498 				    DS_FIELD_BOOKMARK_NAMES,
499 				    sizeof (ds->ds_bookmarks), 1,
500 				    &ds->ds_bookmarks);
501 				if (zaperr != ENOENT)
502 					VERIFY0(zaperr);
503 			}
504 		} else {
505 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
506 				err = dsl_dataset_get_snapname(ds);
507 			if (err == 0 &&
508 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
509 				err = zap_count(
510 				    ds->ds_dir->dd_pool->dp_meta_objset,
511 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
512 				    &ds->ds_userrefs);
513 			}
514 		}
515 
516 		if (err == 0 && !ds->ds_is_snapshot) {
517 			err = dsl_prop_get_int_ds(ds,
518 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
519 			    &ds->ds_reserved);
520 			if (err == 0) {
521 				err = dsl_prop_get_int_ds(ds,
522 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
523 				    &ds->ds_quota);
524 			}
525 		} else {
526 			ds->ds_reserved = ds->ds_quota = 0;
527 		}
528 
529 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
530 		    dsl_dataset_evict_async, &ds->ds_dbuf);
531 		if (err == 0)
532 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
533 
534 		if (err != 0 || winner != NULL) {
535 			bplist_destroy(&ds->ds_pending_deadlist);
536 			dsl_deadlist_close(&ds->ds_deadlist);
537 			if (ds->ds_prev)
538 				dsl_dataset_rele(ds->ds_prev, ds);
539 			dsl_dir_rele(ds->ds_dir, ds);
540 			mutex_destroy(&ds->ds_lock);
541 			mutex_destroy(&ds->ds_opening_lock);
542 			mutex_destroy(&ds->ds_sendstream_lock);
543 			refcount_destroy(&ds->ds_longholds);
544 			kmem_free(ds, sizeof (dsl_dataset_t));
545 			if (err != 0) {
546 				dmu_buf_rele(dbuf, tag);
547 				return (err);
548 			}
549 			ds = winner;
550 		} else {
551 			ds->ds_fsid_guid =
552 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
553 			if (ds->ds_fsid_guid !=
554 			    dsl_dataset_phys(ds)->ds_fsid_guid) {
555 				zfs_dbgmsg("ds_fsid_guid changed from "
556 				    "%llx to %llx for pool %s dataset id %llu",
557 				    (long long)
558 				    dsl_dataset_phys(ds)->ds_fsid_guid,
559 				    (long long)ds->ds_fsid_guid,
560 				    spa_name(dp->dp_spa),
561 				    dsobj);
562 			}
563 		}
564 	}
565 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
566 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
567 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
568 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
569 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
570 	*dsp = ds;
571 	return (0);
572 }
573 
574 int
575 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
576     void *tag, dsl_dataset_t **dsp)
577 {
578 	dsl_dir_t *dd;
579 	const char *snapname;
580 	uint64_t obj;
581 	int err = 0;
582 	dsl_dataset_t *ds;
583 
584 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
585 	if (err != 0)
586 		return (err);
587 
588 	ASSERT(dsl_pool_config_held(dp));
589 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
590 	if (obj != 0)
591 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
592 	else
593 		err = SET_ERROR(ENOENT);
594 
595 	/* we may be looking for a snapshot */
596 	if (err == 0 && snapname != NULL) {
597 		dsl_dataset_t *snap_ds;
598 
599 		if (*snapname++ != '@') {
600 			dsl_dataset_rele(ds, tag);
601 			dsl_dir_rele(dd, FTAG);
602 			return (SET_ERROR(ENOENT));
603 		}
604 
605 		dprintf("looking for snapshot '%s'\n", snapname);
606 		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
607 		if (err == 0)
608 			err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
609 		dsl_dataset_rele(ds, tag);
610 
611 		if (err == 0) {
612 			mutex_enter(&snap_ds->ds_lock);
613 			if (snap_ds->ds_snapname[0] == 0)
614 				(void) strlcpy(snap_ds->ds_snapname, snapname,
615 				    sizeof (snap_ds->ds_snapname));
616 			mutex_exit(&snap_ds->ds_lock);
617 			ds = snap_ds;
618 		}
619 	}
620 	if (err == 0)
621 		*dsp = ds;
622 	dsl_dir_rele(dd, FTAG);
623 	return (err);
624 }
625 
626 int
627 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
628     void *tag, dsl_dataset_t **dsp)
629 {
630 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
631 	if (err != 0)
632 		return (err);
633 	if (!dsl_dataset_tryown(*dsp, tag)) {
634 		dsl_dataset_rele(*dsp, tag);
635 		*dsp = NULL;
636 		return (SET_ERROR(EBUSY));
637 	}
638 	return (0);
639 }
640 
641 int
642 dsl_dataset_own(dsl_pool_t *dp, const char *name,
643     void *tag, dsl_dataset_t **dsp)
644 {
645 	int err = dsl_dataset_hold(dp, name, tag, dsp);
646 	if (err != 0)
647 		return (err);
648 	if (!dsl_dataset_tryown(*dsp, tag)) {
649 		dsl_dataset_rele(*dsp, tag);
650 		return (SET_ERROR(EBUSY));
651 	}
652 	return (0);
653 }
654 
655 /*
656  * See the comment above dsl_pool_hold() for details.  In summary, a long
657  * hold is used to prevent destruction of a dataset while the pool hold
658  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
659  *
660  * The dataset and pool must be held when this function is called.  After it
661  * is called, the pool hold may be released while the dataset is still held
662  * and accessed.
663  */
664 void
665 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
666 {
667 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
668 	(void) refcount_add(&ds->ds_longholds, tag);
669 }
670 
671 void
672 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
673 {
674 	(void) refcount_remove(&ds->ds_longholds, tag);
675 }
676 
677 /* Return B_TRUE if there are any long holds on this dataset. */
678 boolean_t
679 dsl_dataset_long_held(dsl_dataset_t *ds)
680 {
681 	return (!refcount_is_zero(&ds->ds_longholds));
682 }
683 
684 void
685 dsl_dataset_name(dsl_dataset_t *ds, char *name)
686 {
687 	if (ds == NULL) {
688 		(void) strcpy(name, "mos");
689 	} else {
690 		dsl_dir_name(ds->ds_dir, name);
691 		VERIFY0(dsl_dataset_get_snapname(ds));
692 		if (ds->ds_snapname[0]) {
693 			VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
694 			    <, ZFS_MAX_DATASET_NAME_LEN);
695 			/*
696 			 * We use a "recursive" mutex so that we
697 			 * can call dprintf_ds() with ds_lock held.
698 			 */
699 			if (!MUTEX_HELD(&ds->ds_lock)) {
700 				mutex_enter(&ds->ds_lock);
701 				VERIFY3U(strlcat(name, ds->ds_snapname,
702 				    ZFS_MAX_DATASET_NAME_LEN), <,
703 				    ZFS_MAX_DATASET_NAME_LEN);
704 				mutex_exit(&ds->ds_lock);
705 			} else {
706 				VERIFY3U(strlcat(name, ds->ds_snapname,
707 				    ZFS_MAX_DATASET_NAME_LEN), <,
708 				    ZFS_MAX_DATASET_NAME_LEN);
709 			}
710 		}
711 	}
712 }
713 
714 int
715 dsl_dataset_namelen(dsl_dataset_t *ds)
716 {
717 	VERIFY0(dsl_dataset_get_snapname(ds));
718 	mutex_enter(&ds->ds_lock);
719 	int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
720 	mutex_exit(&ds->ds_lock);
721 	return (len);
722 }
723 
724 void
725 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
726 {
727 	dmu_buf_rele(ds->ds_dbuf, tag);
728 }
729 
730 void
731 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
732 {
733 	ASSERT3P(ds->ds_owner, ==, tag);
734 	ASSERT(ds->ds_dbuf != NULL);
735 
736 	mutex_enter(&ds->ds_lock);
737 	ds->ds_owner = NULL;
738 	mutex_exit(&ds->ds_lock);
739 	dsl_dataset_long_rele(ds, tag);
740 	dsl_dataset_rele(ds, tag);
741 }
742 
743 boolean_t
744 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
745 {
746 	boolean_t gotit = FALSE;
747 
748 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
749 	mutex_enter(&ds->ds_lock);
750 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
751 		ds->ds_owner = tag;
752 		dsl_dataset_long_hold(ds, tag);
753 		gotit = TRUE;
754 	}
755 	mutex_exit(&ds->ds_lock);
756 	return (gotit);
757 }
758 
759 boolean_t
760 dsl_dataset_has_owner(dsl_dataset_t *ds)
761 {
762 	boolean_t rv;
763 	mutex_enter(&ds->ds_lock);
764 	rv = (ds->ds_owner != NULL);
765 	mutex_exit(&ds->ds_lock);
766 	return (rv);
767 }
768 
769 static void
770 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
771 {
772 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
773 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
774 	uint64_t zero = 0;
775 
776 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
777 
778 	spa_feature_incr(spa, f, tx);
779 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
780 
781 	VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
782 	    sizeof (zero), 1, &zero, tx));
783 }
784 
785 void
786 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
787 {
788 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
789 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
790 
791 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
792 
793 	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
794 	spa_feature_decr(spa, f, tx);
795 }
796 
797 uint64_t
798 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
799     uint64_t flags, dmu_tx_t *tx)
800 {
801 	dsl_pool_t *dp = dd->dd_pool;
802 	dmu_buf_t *dbuf;
803 	dsl_dataset_phys_t *dsphys;
804 	uint64_t dsobj;
805 	objset_t *mos = dp->dp_meta_objset;
806 
807 	if (origin == NULL)
808 		origin = dp->dp_origin_snap;
809 
810 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
811 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
812 	ASSERT(dmu_tx_is_syncing(tx));
813 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
814 
815 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
816 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
817 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
818 	dmu_buf_will_dirty(dbuf, tx);
819 	dsphys = dbuf->db_data;
820 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
821 	dsphys->ds_dir_obj = dd->dd_object;
822 	dsphys->ds_flags = flags;
823 	dsphys->ds_fsid_guid = unique_create();
824 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
825 	    sizeof (dsphys->ds_guid));
826 	dsphys->ds_snapnames_zapobj =
827 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
828 	    DMU_OT_NONE, 0, tx);
829 	dsphys->ds_creation_time = gethrestime_sec();
830 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
831 
832 	if (origin == NULL) {
833 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
834 	} else {
835 		dsl_dataset_t *ohds; /* head of the origin snapshot */
836 
837 		dsphys->ds_prev_snap_obj = origin->ds_object;
838 		dsphys->ds_prev_snap_txg =
839 		    dsl_dataset_phys(origin)->ds_creation_txg;
840 		dsphys->ds_referenced_bytes =
841 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
842 		dsphys->ds_compressed_bytes =
843 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
844 		dsphys->ds_uncompressed_bytes =
845 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
846 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
847 
848 		/*
849 		 * Inherit flags that describe the dataset's contents
850 		 * (INCONSISTENT) or properties (Case Insensitive).
851 		 */
852 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
853 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
854 
855 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
856 			if (origin->ds_feature_inuse[f])
857 				dsl_dataset_activate_feature(dsobj, f, tx);
858 		}
859 
860 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
861 		dsl_dataset_phys(origin)->ds_num_children++;
862 
863 		VERIFY0(dsl_dataset_hold_obj(dp,
864 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
865 		    FTAG, &ohds));
866 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
867 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
868 		dsl_dataset_rele(ohds, FTAG);
869 
870 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
871 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
872 				dsl_dataset_phys(origin)->ds_next_clones_obj =
873 				    zap_create(mos,
874 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
875 			}
876 			VERIFY0(zap_add_int(mos,
877 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
878 			    dsobj, tx));
879 		}
880 
881 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
882 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
883 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
884 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
885 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
886 				dsl_dir_phys(origin->ds_dir)->dd_clones =
887 				    zap_create(mos,
888 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
889 			}
890 			VERIFY0(zap_add_int(mos,
891 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
892 			    dsobj, tx));
893 		}
894 	}
895 
896 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
897 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
898 
899 	dmu_buf_rele(dbuf, FTAG);
900 
901 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
902 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
903 
904 	return (dsobj);
905 }
906 
907 static void
908 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
909 {
910 	objset_t *os;
911 
912 	VERIFY0(dmu_objset_from_ds(ds, &os));
913 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
914 	dsl_dataset_dirty(ds, tx);
915 }
916 
917 uint64_t
918 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
919     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
920 {
921 	dsl_pool_t *dp = pdd->dd_pool;
922 	uint64_t dsobj, ddobj;
923 	dsl_dir_t *dd;
924 
925 	ASSERT(dmu_tx_is_syncing(tx));
926 	ASSERT(lastname[0] != '@');
927 
928 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
929 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
930 
931 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
932 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
933 
934 	dsl_deleg_set_create_perms(dd, tx, cr);
935 
936 	/*
937 	 * Since we're creating a new node we know it's a leaf, so we can
938 	 * initialize the counts if the limit feature is active.
939 	 */
940 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
941 		uint64_t cnt = 0;
942 		objset_t *os = dd->dd_pool->dp_meta_objset;
943 
944 		dsl_dir_zapify(dd, tx);
945 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
946 		    sizeof (cnt), 1, &cnt, tx));
947 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
948 		    sizeof (cnt), 1, &cnt, tx));
949 	}
950 
951 	dsl_dir_rele(dd, FTAG);
952 
953 	/*
954 	 * If we are creating a clone, make sure we zero out any stale
955 	 * data from the origin snapshots zil header.
956 	 */
957 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
958 		dsl_dataset_t *ds;
959 
960 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
961 		dsl_dataset_zero_zil(ds, tx);
962 		dsl_dataset_rele(ds, FTAG);
963 	}
964 
965 	return (dsobj);
966 }
967 
968 /*
969  * The unique space in the head dataset can be calculated by subtracting
970  * the space used in the most recent snapshot, that is still being used
971  * in this file system, from the space currently in use.  To figure out
972  * the space in the most recent snapshot still in use, we need to take
973  * the total space used in the snapshot and subtract out the space that
974  * has been freed up since the snapshot was taken.
975  */
976 void
977 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
978 {
979 	uint64_t mrs_used;
980 	uint64_t dlused, dlcomp, dluncomp;
981 
982 	ASSERT(!ds->ds_is_snapshot);
983 
984 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
985 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
986 	else
987 		mrs_used = 0;
988 
989 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
990 
991 	ASSERT3U(dlused, <=, mrs_used);
992 	dsl_dataset_phys(ds)->ds_unique_bytes =
993 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
994 
995 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
996 	    SPA_VERSION_UNIQUE_ACCURATE)
997 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
998 }
999 
1000 void
1001 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1002     dmu_tx_t *tx)
1003 {
1004 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1005 	uint64_t count;
1006 	int err;
1007 
1008 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1009 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1010 	    obj, tx);
1011 	/*
1012 	 * The err should not be ENOENT, but a bug in a previous version
1013 	 * of the code could cause upgrade_clones_cb() to not set
1014 	 * ds_next_snap_obj when it should, leading to a missing entry.
1015 	 * If we knew that the pool was created after
1016 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1017 	 * ENOENT.  However, at least we can check that we don't have
1018 	 * too many entries in the next_clones_obj even after failing to
1019 	 * remove this one.
1020 	 */
1021 	if (err != ENOENT)
1022 		VERIFY0(err);
1023 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1024 	    &count));
1025 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1026 }
1027 
1028 
1029 blkptr_t *
1030 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1031 {
1032 	return (&dsl_dataset_phys(ds)->ds_bp);
1033 }
1034 
1035 void
1036 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1037 {
1038 	ASSERT(dmu_tx_is_syncing(tx));
1039 	/* If it's the meta-objset, set dp_meta_rootbp */
1040 	if (ds == NULL) {
1041 		tx->tx_pool->dp_meta_rootbp = *bp;
1042 	} else {
1043 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1044 		dsl_dataset_phys(ds)->ds_bp = *bp;
1045 	}
1046 }
1047 
1048 spa_t *
1049 dsl_dataset_get_spa(dsl_dataset_t *ds)
1050 {
1051 	return (ds->ds_dir->dd_pool->dp_spa);
1052 }
1053 
1054 void
1055 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1056 {
1057 	dsl_pool_t *dp;
1058 
1059 	if (ds == NULL) /* this is the meta-objset */
1060 		return;
1061 
1062 	ASSERT(ds->ds_objset != NULL);
1063 
1064 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1065 		panic("dirtying snapshot!");
1066 
1067 	dp = ds->ds_dir->dd_pool;
1068 
1069 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1070 		/* up the hold count until we can be written out */
1071 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1072 	}
1073 }
1074 
1075 boolean_t
1076 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1077 {
1078 	for (int t = 0; t < TXG_SIZE; t++) {
1079 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1080 		    ds, t))
1081 			return (B_TRUE);
1082 	}
1083 	return (B_FALSE);
1084 }
1085 
1086 static int
1087 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1088 {
1089 	uint64_t asize;
1090 
1091 	if (!dmu_tx_is_syncing(tx))
1092 		return (0);
1093 
1094 	/*
1095 	 * If there's an fs-only reservation, any blocks that might become
1096 	 * owned by the snapshot dataset must be accommodated by space
1097 	 * outside of the reservation.
1098 	 */
1099 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1100 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1101 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1102 		return (SET_ERROR(ENOSPC));
1103 
1104 	/*
1105 	 * Propagate any reserved space for this snapshot to other
1106 	 * snapshot checks in this sync group.
1107 	 */
1108 	if (asize > 0)
1109 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1110 
1111 	return (0);
1112 }
1113 
1114 typedef struct dsl_dataset_snapshot_arg {
1115 	nvlist_t *ddsa_snaps;
1116 	nvlist_t *ddsa_props;
1117 	nvlist_t *ddsa_errors;
1118 	cred_t *ddsa_cr;
1119 } dsl_dataset_snapshot_arg_t;
1120 
1121 int
1122 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1123     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1124 {
1125 	int error;
1126 	uint64_t value;
1127 
1128 	ds->ds_trysnap_txg = tx->tx_txg;
1129 
1130 	if (!dmu_tx_is_syncing(tx))
1131 		return (0);
1132 
1133 	/*
1134 	 * We don't allow multiple snapshots of the same txg.  If there
1135 	 * is already one, try again.
1136 	 */
1137 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1138 		return (SET_ERROR(EAGAIN));
1139 
1140 	/*
1141 	 * Check for conflicting snapshot name.
1142 	 */
1143 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1144 	if (error == 0)
1145 		return (SET_ERROR(EEXIST));
1146 	if (error != ENOENT)
1147 		return (error);
1148 
1149 	/*
1150 	 * We don't allow taking snapshots of inconsistent datasets, such as
1151 	 * those into which we are currently receiving.  However, if we are
1152 	 * creating this snapshot as part of a receive, this check will be
1153 	 * executed atomically with respect to the completion of the receive
1154 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1155 	 * case we ignore this, knowing it will be fixed up for us shortly in
1156 	 * dmu_recv_end_sync().
1157 	 */
1158 	if (!recv && DS_IS_INCONSISTENT(ds))
1159 		return (SET_ERROR(EBUSY));
1160 
1161 	/*
1162 	 * Skip the check for temporary snapshots or if we have already checked
1163 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1164 	 * check the count here when we're receiving a stream.
1165 	 */
1166 	if (cnt != 0 && cr != NULL) {
1167 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1168 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1169 		if (error != 0)
1170 			return (error);
1171 	}
1172 
1173 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1174 	if (error != 0)
1175 		return (error);
1176 
1177 	return (0);
1178 }
1179 
1180 static int
1181 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1182 {
1183 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1184 	dsl_pool_t *dp = dmu_tx_pool(tx);
1185 	nvpair_t *pair;
1186 	int rv = 0;
1187 
1188 	/*
1189 	 * Pre-compute how many total new snapshots will be created for each
1190 	 * level in the tree and below. This is needed for validating the
1191 	 * snapshot limit when either taking a recursive snapshot or when
1192 	 * taking multiple snapshots.
1193 	 *
1194 	 * The problem is that the counts are not actually adjusted when
1195 	 * we are checking, only when we finally sync. For a single snapshot,
1196 	 * this is easy, the count will increase by 1 at each node up the tree,
1197 	 * but its more complicated for the recursive/multiple snapshot case.
1198 	 *
1199 	 * The dsl_fs_ss_limit_check function does recursively check the count
1200 	 * at each level up the tree but since it is validating each snapshot
1201 	 * independently we need to be sure that we are validating the complete
1202 	 * count for the entire set of snapshots. We do this by rolling up the
1203 	 * counts for each component of the name into an nvlist and then
1204 	 * checking each of those cases with the aggregated count.
1205 	 *
1206 	 * This approach properly handles not only the recursive snapshot
1207 	 * case (where we get all of those on the ddsa_snaps list) but also
1208 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1209 	 * validate the limit on 'a' using a count of 2).
1210 	 *
1211 	 * We validate the snapshot names in the third loop and only report
1212 	 * name errors once.
1213 	 */
1214 	if (dmu_tx_is_syncing(tx)) {
1215 		nvlist_t *cnt_track = NULL;
1216 		cnt_track = fnvlist_alloc();
1217 
1218 		/* Rollup aggregated counts into the cnt_track list */
1219 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1220 		    pair != NULL;
1221 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1222 			char *pdelim;
1223 			uint64_t val;
1224 			char nm[MAXPATHLEN];
1225 
1226 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1227 			pdelim = strchr(nm, '@');
1228 			if (pdelim == NULL)
1229 				continue;
1230 			*pdelim = '\0';
1231 
1232 			do {
1233 				if (nvlist_lookup_uint64(cnt_track, nm,
1234 				    &val) == 0) {
1235 					/* update existing entry */
1236 					fnvlist_add_uint64(cnt_track, nm,
1237 					    val + 1);
1238 				} else {
1239 					/* add to list */
1240 					fnvlist_add_uint64(cnt_track, nm, 1);
1241 				}
1242 
1243 				pdelim = strrchr(nm, '/');
1244 				if (pdelim != NULL)
1245 					*pdelim = '\0';
1246 			} while (pdelim != NULL);
1247 		}
1248 
1249 		/* Check aggregated counts at each level */
1250 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1251 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1252 			int error = 0;
1253 			char *name;
1254 			uint64_t cnt = 0;
1255 			dsl_dataset_t *ds;
1256 
1257 			name = nvpair_name(pair);
1258 			cnt = fnvpair_value_uint64(pair);
1259 			ASSERT(cnt > 0);
1260 
1261 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1262 			if (error == 0) {
1263 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1264 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1265 				    ddsa->ddsa_cr);
1266 				dsl_dataset_rele(ds, FTAG);
1267 			}
1268 
1269 			if (error != 0) {
1270 				if (ddsa->ddsa_errors != NULL)
1271 					fnvlist_add_int32(ddsa->ddsa_errors,
1272 					    name, error);
1273 				rv = error;
1274 				/* only report one error for this check */
1275 				break;
1276 			}
1277 		}
1278 		nvlist_free(cnt_track);
1279 	}
1280 
1281 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1282 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1283 		int error = 0;
1284 		dsl_dataset_t *ds;
1285 		char *name, *atp;
1286 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1287 
1288 		name = nvpair_name(pair);
1289 		if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1290 			error = SET_ERROR(ENAMETOOLONG);
1291 		if (error == 0) {
1292 			atp = strchr(name, '@');
1293 			if (atp == NULL)
1294 				error = SET_ERROR(EINVAL);
1295 			if (error == 0)
1296 				(void) strlcpy(dsname, name, atp - name + 1);
1297 		}
1298 		if (error == 0)
1299 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1300 		if (error == 0) {
1301 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1302 			error = dsl_dataset_snapshot_check_impl(ds,
1303 			    atp + 1, tx, B_FALSE, 0, NULL);
1304 			dsl_dataset_rele(ds, FTAG);
1305 		}
1306 
1307 		if (error != 0) {
1308 			if (ddsa->ddsa_errors != NULL) {
1309 				fnvlist_add_int32(ddsa->ddsa_errors,
1310 				    name, error);
1311 			}
1312 			rv = error;
1313 		}
1314 	}
1315 
1316 	return (rv);
1317 }
1318 
1319 void
1320 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1321     dmu_tx_t *tx)
1322 {
1323 	static zil_header_t zero_zil;
1324 
1325 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1326 	dmu_buf_t *dbuf;
1327 	dsl_dataset_phys_t *dsphys;
1328 	uint64_t dsobj, crtxg;
1329 	objset_t *mos = dp->dp_meta_objset;
1330 	objset_t *os;
1331 
1332 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1333 
1334 	/*
1335 	 * If we are on an old pool, the zil must not be active, in which
1336 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1337 	 */
1338 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1339 	    dmu_objset_from_ds(ds, &os) != 0 ||
1340 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1341 	    sizeof (zero_zil)) == 0);
1342 
1343 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1344 
1345 	/*
1346 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1347 	 */
1348 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1349 		crtxg = 1;
1350 	else
1351 		crtxg = tx->tx_txg;
1352 
1353 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1354 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1355 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1356 	dmu_buf_will_dirty(dbuf, tx);
1357 	dsphys = dbuf->db_data;
1358 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1359 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1360 	dsphys->ds_fsid_guid = unique_create();
1361 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1362 	    sizeof (dsphys->ds_guid));
1363 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1364 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1365 	dsphys->ds_next_snap_obj = ds->ds_object;
1366 	dsphys->ds_num_children = 1;
1367 	dsphys->ds_creation_time = gethrestime_sec();
1368 	dsphys->ds_creation_txg = crtxg;
1369 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1370 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1371 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1372 	dsphys->ds_uncompressed_bytes =
1373 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1374 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1375 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1376 	dmu_buf_rele(dbuf, FTAG);
1377 
1378 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1379 		if (ds->ds_feature_inuse[f])
1380 			dsl_dataset_activate_feature(dsobj, f, tx);
1381 	}
1382 
1383 	ASSERT3U(ds->ds_prev != 0, ==,
1384 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1385 	if (ds->ds_prev) {
1386 		uint64_t next_clones_obj =
1387 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1388 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1389 		    ds->ds_object ||
1390 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1391 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1392 		    ds->ds_object) {
1393 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1394 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1395 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1396 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1397 		} else if (next_clones_obj != 0) {
1398 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1399 			    dsphys->ds_next_snap_obj, tx);
1400 			VERIFY0(zap_add_int(mos,
1401 			    next_clones_obj, dsobj, tx));
1402 		}
1403 	}
1404 
1405 	/*
1406 	 * If we have a reference-reservation on this dataset, we will
1407 	 * need to increase the amount of refreservation being charged
1408 	 * since our unique space is going to zero.
1409 	 */
1410 	if (ds->ds_reserved) {
1411 		int64_t delta;
1412 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1413 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1414 		    ds->ds_reserved);
1415 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1416 		    delta, 0, 0, tx);
1417 	}
1418 
1419 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1420 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1421 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1422 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1423 	dsl_deadlist_close(&ds->ds_deadlist);
1424 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1425 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1426 	dsl_deadlist_add_key(&ds->ds_deadlist,
1427 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1428 
1429 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1430 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1431 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1432 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1433 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1434 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1435 
1436 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1437 	    snapname, 8, 1, &dsobj, tx));
1438 
1439 	if (ds->ds_prev)
1440 		dsl_dataset_rele(ds->ds_prev, ds);
1441 	VERIFY0(dsl_dataset_hold_obj(dp,
1442 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1443 
1444 	dsl_scan_ds_snapshotted(ds, tx);
1445 
1446 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1447 
1448 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1449 }
1450 
1451 static void
1452 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1453 {
1454 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1455 	dsl_pool_t *dp = dmu_tx_pool(tx);
1456 	nvpair_t *pair;
1457 
1458 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1459 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1460 		dsl_dataset_t *ds;
1461 		char *name, *atp;
1462 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1463 
1464 		name = nvpair_name(pair);
1465 		atp = strchr(name, '@');
1466 		(void) strlcpy(dsname, name, atp - name + 1);
1467 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1468 
1469 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1470 		if (ddsa->ddsa_props != NULL) {
1471 			dsl_props_set_sync_impl(ds->ds_prev,
1472 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1473 		}
1474 		dsl_dataset_rele(ds, FTAG);
1475 	}
1476 }
1477 
1478 /*
1479  * The snapshots must all be in the same pool.
1480  * All-or-nothing: if there are any failures, nothing will be modified.
1481  */
1482 int
1483 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1484 {
1485 	dsl_dataset_snapshot_arg_t ddsa;
1486 	nvpair_t *pair;
1487 	boolean_t needsuspend;
1488 	int error;
1489 	spa_t *spa;
1490 	char *firstname;
1491 	nvlist_t *suspended = NULL;
1492 
1493 	pair = nvlist_next_nvpair(snaps, NULL);
1494 	if (pair == NULL)
1495 		return (0);
1496 	firstname = nvpair_name(pair);
1497 
1498 	error = spa_open(firstname, &spa, FTAG);
1499 	if (error != 0)
1500 		return (error);
1501 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1502 	spa_close(spa, FTAG);
1503 
1504 	if (needsuspend) {
1505 		suspended = fnvlist_alloc();
1506 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1507 		    pair = nvlist_next_nvpair(snaps, pair)) {
1508 			char fsname[ZFS_MAX_DATASET_NAME_LEN];
1509 			char *snapname = nvpair_name(pair);
1510 			char *atp;
1511 			void *cookie;
1512 
1513 			atp = strchr(snapname, '@');
1514 			if (atp == NULL) {
1515 				error = SET_ERROR(EINVAL);
1516 				break;
1517 			}
1518 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1519 
1520 			error = zil_suspend(fsname, &cookie);
1521 			if (error != 0)
1522 				break;
1523 			fnvlist_add_uint64(suspended, fsname,
1524 			    (uintptr_t)cookie);
1525 		}
1526 	}
1527 
1528 	ddsa.ddsa_snaps = snaps;
1529 	ddsa.ddsa_props = props;
1530 	ddsa.ddsa_errors = errors;
1531 	ddsa.ddsa_cr = CRED();
1532 
1533 	if (error == 0) {
1534 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1535 		    dsl_dataset_snapshot_sync, &ddsa,
1536 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1537 	}
1538 
1539 	if (suspended != NULL) {
1540 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1541 		    pair = nvlist_next_nvpair(suspended, pair)) {
1542 			zil_resume((void *)(uintptr_t)
1543 			    fnvpair_value_uint64(pair));
1544 		}
1545 		fnvlist_free(suspended);
1546 	}
1547 
1548 	return (error);
1549 }
1550 
1551 typedef struct dsl_dataset_snapshot_tmp_arg {
1552 	const char *ddsta_fsname;
1553 	const char *ddsta_snapname;
1554 	minor_t ddsta_cleanup_minor;
1555 	const char *ddsta_htag;
1556 } dsl_dataset_snapshot_tmp_arg_t;
1557 
1558 static int
1559 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1560 {
1561 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1562 	dsl_pool_t *dp = dmu_tx_pool(tx);
1563 	dsl_dataset_t *ds;
1564 	int error;
1565 
1566 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1567 	if (error != 0)
1568 		return (error);
1569 
1570 	/* NULL cred means no limit check for tmp snapshot */
1571 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1572 	    tx, B_FALSE, 0, NULL);
1573 	if (error != 0) {
1574 		dsl_dataset_rele(ds, FTAG);
1575 		return (error);
1576 	}
1577 
1578 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1579 		dsl_dataset_rele(ds, FTAG);
1580 		return (SET_ERROR(ENOTSUP));
1581 	}
1582 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1583 	    B_TRUE, tx);
1584 	if (error != 0) {
1585 		dsl_dataset_rele(ds, FTAG);
1586 		return (error);
1587 	}
1588 
1589 	dsl_dataset_rele(ds, FTAG);
1590 	return (0);
1591 }
1592 
1593 static void
1594 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1595 {
1596 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1597 	dsl_pool_t *dp = dmu_tx_pool(tx);
1598 	dsl_dataset_t *ds;
1599 
1600 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1601 
1602 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1603 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1604 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1605 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1606 
1607 	dsl_dataset_rele(ds, FTAG);
1608 }
1609 
1610 int
1611 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1612     minor_t cleanup_minor, const char *htag)
1613 {
1614 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1615 	int error;
1616 	spa_t *spa;
1617 	boolean_t needsuspend;
1618 	void *cookie;
1619 
1620 	ddsta.ddsta_fsname = fsname;
1621 	ddsta.ddsta_snapname = snapname;
1622 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1623 	ddsta.ddsta_htag = htag;
1624 
1625 	error = spa_open(fsname, &spa, FTAG);
1626 	if (error != 0)
1627 		return (error);
1628 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1629 	spa_close(spa, FTAG);
1630 
1631 	if (needsuspend) {
1632 		error = zil_suspend(fsname, &cookie);
1633 		if (error != 0)
1634 			return (error);
1635 	}
1636 
1637 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1638 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1639 
1640 	if (needsuspend)
1641 		zil_resume(cookie);
1642 	return (error);
1643 }
1644 
1645 
1646 void
1647 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1648 {
1649 	ASSERT(dmu_tx_is_syncing(tx));
1650 	ASSERT(ds->ds_objset != NULL);
1651 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1652 
1653 	/*
1654 	 * in case we had to change ds_fsid_guid when we opened it,
1655 	 * sync it out now.
1656 	 */
1657 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1658 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1659 
1660 	if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1661 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1662 		    ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1663 		    &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1664 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1665 		    ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1666 		    &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1667 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1668 		    ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1669 		    &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1670 		ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1671 		ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1672 		ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1673 	}
1674 
1675 	dmu_objset_sync(ds->ds_objset, zio, tx);
1676 
1677 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1678 		if (ds->ds_feature_activation_needed[f]) {
1679 			if (ds->ds_feature_inuse[f])
1680 				continue;
1681 			dsl_dataset_activate_feature(ds->ds_object, f, tx);
1682 			ds->ds_feature_inuse[f] = B_TRUE;
1683 		}
1684 	}
1685 }
1686 
1687 static void
1688 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1689 {
1690 	uint64_t count = 0;
1691 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1692 	zap_cursor_t zc;
1693 	zap_attribute_t za;
1694 	nvlist_t *propval = fnvlist_alloc();
1695 	nvlist_t *val = fnvlist_alloc();
1696 
1697 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1698 
1699 	/*
1700 	 * There may be missing entries in ds_next_clones_obj
1701 	 * due to a bug in a previous version of the code.
1702 	 * Only trust it if it has the right number of entries.
1703 	 */
1704 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1705 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1706 		    &count));
1707 	}
1708 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1709 		goto fail;
1710 	for (zap_cursor_init(&zc, mos,
1711 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1712 	    zap_cursor_retrieve(&zc, &za) == 0;
1713 	    zap_cursor_advance(&zc)) {
1714 		dsl_dataset_t *clone;
1715 		char buf[ZFS_MAX_DATASET_NAME_LEN];
1716 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1717 		    za.za_first_integer, FTAG, &clone));
1718 		dsl_dir_name(clone->ds_dir, buf);
1719 		fnvlist_add_boolean(val, buf);
1720 		dsl_dataset_rele(clone, FTAG);
1721 	}
1722 	zap_cursor_fini(&zc);
1723 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1724 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1725 fail:
1726 	nvlist_free(val);
1727 	nvlist_free(propval);
1728 }
1729 
1730 static void
1731 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1732 {
1733 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1734 
1735 	if (dsl_dataset_has_resume_receive_state(ds)) {
1736 		char *str;
1737 		void *packed;
1738 		uint8_t *compressed;
1739 		uint64_t val;
1740 		nvlist_t *token_nv = fnvlist_alloc();
1741 		size_t packed_size, compressed_size;
1742 
1743 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1744 		    DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1745 			fnvlist_add_uint64(token_nv, "fromguid", val);
1746 		}
1747 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1748 		    DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1749 			fnvlist_add_uint64(token_nv, "object", val);
1750 		}
1751 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1752 		    DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1753 			fnvlist_add_uint64(token_nv, "offset", val);
1754 		}
1755 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1756 		    DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1757 			fnvlist_add_uint64(token_nv, "bytes", val);
1758 		}
1759 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1760 		    DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1761 			fnvlist_add_uint64(token_nv, "toguid", val);
1762 		}
1763 		char buf[256];
1764 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1765 		    DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1766 			fnvlist_add_string(token_nv, "toname", buf);
1767 		}
1768 		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1769 		    DS_FIELD_RESUME_EMBEDOK) == 0) {
1770 			fnvlist_add_boolean(token_nv, "embedok");
1771 		}
1772 		packed = fnvlist_pack(token_nv, &packed_size);
1773 		fnvlist_free(token_nv);
1774 		compressed = kmem_alloc(packed_size, KM_SLEEP);
1775 
1776 		compressed_size = gzip_compress(packed, compressed,
1777 		    packed_size, packed_size, 6);
1778 
1779 		zio_cksum_t cksum;
1780 		fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1781 
1782 		str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1783 		for (int i = 0; i < compressed_size; i++) {
1784 			(void) sprintf(str + i * 2, "%02x", compressed[i]);
1785 		}
1786 		str[compressed_size * 2] = '\0';
1787 		char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1788 		    ZFS_SEND_RESUME_TOKEN_VERSION,
1789 		    (longlong_t)cksum.zc_word[0],
1790 		    (longlong_t)packed_size, str);
1791 		dsl_prop_nvlist_add_string(nv,
1792 		    ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1793 		kmem_free(packed, packed_size);
1794 		kmem_free(str, compressed_size * 2 + 1);
1795 		kmem_free(compressed, packed_size);
1796 		strfree(propval);
1797 	}
1798 }
1799 
1800 void
1801 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1802 {
1803 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1804 	uint64_t refd, avail, uobjs, aobjs, ratio;
1805 
1806 	ASSERT(dsl_pool_config_held(dp));
1807 
1808 	ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1809 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1810 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1811 
1812 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1813 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1814 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1815 
1816 	if (ds->ds_is_snapshot) {
1817 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1818 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1819 		    dsl_dataset_phys(ds)->ds_unique_bytes);
1820 		get_clones_stat(ds, nv);
1821 	} else {
1822 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1823 			char buf[ZFS_MAX_DATASET_NAME_LEN];
1824 			dsl_dataset_name(ds->ds_prev, buf);
1825 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1826 		}
1827 
1828 		dsl_dir_stats(ds->ds_dir, nv);
1829 	}
1830 
1831 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1832 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1833 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1834 
1835 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1836 	    dsl_dataset_phys(ds)->ds_creation_time);
1837 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1838 	    dsl_dataset_phys(ds)->ds_creation_txg);
1839 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1840 	    ds->ds_quota);
1841 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1842 	    ds->ds_reserved);
1843 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1844 	    dsl_dataset_phys(ds)->ds_guid);
1845 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1846 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1847 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1848 	    ds->ds_object);
1849 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1850 	    ds->ds_userrefs);
1851 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1852 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1853 
1854 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1855 		uint64_t written, comp, uncomp;
1856 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1857 		dsl_dataset_t *prev;
1858 
1859 		int err = dsl_dataset_hold_obj(dp,
1860 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1861 		if (err == 0) {
1862 			err = dsl_dataset_space_written(prev, ds, &written,
1863 			    &comp, &uncomp);
1864 			dsl_dataset_rele(prev, FTAG);
1865 			if (err == 0) {
1866 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1867 				    written);
1868 			}
1869 		}
1870 	}
1871 
1872 	if (!dsl_dataset_is_snapshot(ds)) {
1873 		/*
1874 		 * A failed "newfs" (e.g. full) resumable receive leaves
1875 		 * the stats set on this dataset.  Check here for the prop.
1876 		 */
1877 		get_receive_resume_stats(ds, nv);
1878 
1879 		/*
1880 		 * A failed incremental resumable receive leaves the
1881 		 * stats set on our child named "%recv".  Check the child
1882 		 * for the prop.
1883 		 */
1884 		/* 6 extra bytes for /%recv */
1885 		char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1886 		dsl_dataset_t *recv_ds;
1887 		dsl_dataset_name(ds, recvname);
1888 		if (strlcat(recvname, "/", sizeof (recvname)) <
1889 		    sizeof (recvname) &&
1890 		    strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1891 		    sizeof (recvname) &&
1892 		    dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1893 			get_receive_resume_stats(recv_ds, nv);
1894 			dsl_dataset_rele(recv_ds, FTAG);
1895 		}
1896 	}
1897 }
1898 
1899 void
1900 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1901 {
1902 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1903 	ASSERT(dsl_pool_config_held(dp));
1904 
1905 	stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1906 	stat->dds_inconsistent =
1907 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1908 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1909 	stat->dds_origin[0] = '\0';
1910 	if (ds->ds_is_snapshot) {
1911 		stat->dds_is_snapshot = B_TRUE;
1912 		stat->dds_num_clones =
1913 		    dsl_dataset_phys(ds)->ds_num_children - 1;
1914 	} else {
1915 		stat->dds_is_snapshot = B_FALSE;
1916 		stat->dds_num_clones = 0;
1917 
1918 		if (dsl_dir_is_clone(ds->ds_dir)) {
1919 			dsl_dataset_t *ods;
1920 
1921 			VERIFY0(dsl_dataset_hold_obj(dp,
1922 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1923 			    FTAG, &ods));
1924 			dsl_dataset_name(ods, stat->dds_origin);
1925 			dsl_dataset_rele(ods, FTAG);
1926 		}
1927 	}
1928 }
1929 
1930 uint64_t
1931 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1932 {
1933 	return (ds->ds_fsid_guid);
1934 }
1935 
1936 void
1937 dsl_dataset_space(dsl_dataset_t *ds,
1938     uint64_t *refdbytesp, uint64_t *availbytesp,
1939     uint64_t *usedobjsp, uint64_t *availobjsp)
1940 {
1941 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1942 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1943 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1944 		*availbytesp +=
1945 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1946 	if (ds->ds_quota != 0) {
1947 		/*
1948 		 * Adjust available bytes according to refquota
1949 		 */
1950 		if (*refdbytesp < ds->ds_quota)
1951 			*availbytesp = MIN(*availbytesp,
1952 			    ds->ds_quota - *refdbytesp);
1953 		else
1954 			*availbytesp = 0;
1955 	}
1956 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1957 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1958 }
1959 
1960 boolean_t
1961 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1962 {
1963 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1964 
1965 	ASSERT(dsl_pool_config_held(dp));
1966 	if (snap == NULL)
1967 		return (B_FALSE);
1968 	if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1969 	    dsl_dataset_phys(snap)->ds_creation_txg) {
1970 		objset_t *os, *os_snap;
1971 		/*
1972 		 * It may be that only the ZIL differs, because it was
1973 		 * reset in the head.  Don't count that as being
1974 		 * modified.
1975 		 */
1976 		if (dmu_objset_from_ds(ds, &os) != 0)
1977 			return (B_TRUE);
1978 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1979 			return (B_TRUE);
1980 		return (bcmp(&os->os_phys->os_meta_dnode,
1981 		    &os_snap->os_phys->os_meta_dnode,
1982 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1983 	}
1984 	return (B_FALSE);
1985 }
1986 
1987 typedef struct dsl_dataset_rename_snapshot_arg {
1988 	const char *ddrsa_fsname;
1989 	const char *ddrsa_oldsnapname;
1990 	const char *ddrsa_newsnapname;
1991 	boolean_t ddrsa_recursive;
1992 	dmu_tx_t *ddrsa_tx;
1993 } dsl_dataset_rename_snapshot_arg_t;
1994 
1995 /* ARGSUSED */
1996 static int
1997 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1998     dsl_dataset_t *hds, void *arg)
1999 {
2000 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2001 	int error;
2002 	uint64_t val;
2003 
2004 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2005 	if (error != 0) {
2006 		/* ignore nonexistent snapshots */
2007 		return (error == ENOENT ? 0 : error);
2008 	}
2009 
2010 	/* new name should not exist */
2011 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2012 	if (error == 0)
2013 		error = SET_ERROR(EEXIST);
2014 	else if (error == ENOENT)
2015 		error = 0;
2016 
2017 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2018 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
2019 	    strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2020 		error = SET_ERROR(ENAMETOOLONG);
2021 
2022 	return (error);
2023 }
2024 
2025 static int
2026 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2027 {
2028 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2029 	dsl_pool_t *dp = dmu_tx_pool(tx);
2030 	dsl_dataset_t *hds;
2031 	int error;
2032 
2033 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2034 	if (error != 0)
2035 		return (error);
2036 
2037 	if (ddrsa->ddrsa_recursive) {
2038 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2039 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
2040 		    DS_FIND_CHILDREN);
2041 	} else {
2042 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2043 	}
2044 	dsl_dataset_rele(hds, FTAG);
2045 	return (error);
2046 }
2047 
2048 static int
2049 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2050     dsl_dataset_t *hds, void *arg)
2051 {
2052 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2053 	dsl_dataset_t *ds;
2054 	uint64_t val;
2055 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
2056 	int error;
2057 
2058 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2059 	ASSERT(error == 0 || error == ENOENT);
2060 	if (error == ENOENT) {
2061 		/* ignore nonexistent snapshots */
2062 		return (0);
2063 	}
2064 
2065 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2066 
2067 	/* log before we change the name */
2068 	spa_history_log_internal_ds(ds, "rename", tx,
2069 	    "-> @%s", ddrsa->ddrsa_newsnapname);
2070 
2071 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2072 	    B_FALSE));
2073 	mutex_enter(&ds->ds_lock);
2074 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2075 	mutex_exit(&ds->ds_lock);
2076 	VERIFY0(zap_add(dp->dp_meta_objset,
2077 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2078 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2079 
2080 	dsl_dataset_rele(ds, FTAG);
2081 	return (0);
2082 }
2083 
2084 static void
2085 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2086 {
2087 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2088 	dsl_pool_t *dp = dmu_tx_pool(tx);
2089 	dsl_dataset_t *hds;
2090 
2091 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2092 	ddrsa->ddrsa_tx = tx;
2093 	if (ddrsa->ddrsa_recursive) {
2094 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2095 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2096 		    DS_FIND_CHILDREN));
2097 	} else {
2098 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2099 	}
2100 	dsl_dataset_rele(hds, FTAG);
2101 }
2102 
2103 int
2104 dsl_dataset_rename_snapshot(const char *fsname,
2105     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2106 {
2107 	dsl_dataset_rename_snapshot_arg_t ddrsa;
2108 
2109 	ddrsa.ddrsa_fsname = fsname;
2110 	ddrsa.ddrsa_oldsnapname = oldsnapname;
2111 	ddrsa.ddrsa_newsnapname = newsnapname;
2112 	ddrsa.ddrsa_recursive = recursive;
2113 
2114 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2115 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
2116 	    1, ZFS_SPACE_CHECK_RESERVED));
2117 }
2118 
2119 /*
2120  * If we're doing an ownership handoff, we need to make sure that there is
2121  * only one long hold on the dataset.  We're not allowed to change anything here
2122  * so we don't permanently release the long hold or regular hold here.  We want
2123  * to do this only when syncing to avoid the dataset unexpectedly going away
2124  * when we release the long hold.
2125  */
2126 static int
2127 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2128 {
2129 	boolean_t held;
2130 
2131 	if (!dmu_tx_is_syncing(tx))
2132 		return (0);
2133 
2134 	if (owner != NULL) {
2135 		VERIFY3P(ds->ds_owner, ==, owner);
2136 		dsl_dataset_long_rele(ds, owner);
2137 	}
2138 
2139 	held = dsl_dataset_long_held(ds);
2140 
2141 	if (owner != NULL)
2142 		dsl_dataset_long_hold(ds, owner);
2143 
2144 	if (held)
2145 		return (SET_ERROR(EBUSY));
2146 
2147 	return (0);
2148 }
2149 
2150 typedef struct dsl_dataset_rollback_arg {
2151 	const char *ddra_fsname;
2152 	void *ddra_owner;
2153 	nvlist_t *ddra_result;
2154 } dsl_dataset_rollback_arg_t;
2155 
2156 static int
2157 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2158 {
2159 	dsl_dataset_rollback_arg_t *ddra = arg;
2160 	dsl_pool_t *dp = dmu_tx_pool(tx);
2161 	dsl_dataset_t *ds;
2162 	int64_t unused_refres_delta;
2163 	int error;
2164 
2165 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2166 	if (error != 0)
2167 		return (error);
2168 
2169 	/* must not be a snapshot */
2170 	if (ds->ds_is_snapshot) {
2171 		dsl_dataset_rele(ds, FTAG);
2172 		return (SET_ERROR(EINVAL));
2173 	}
2174 
2175 	/* must have a most recent snapshot */
2176 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2177 		dsl_dataset_rele(ds, FTAG);
2178 		return (SET_ERROR(EINVAL));
2179 	}
2180 
2181 	/* must not have any bookmarks after the most recent snapshot */
2182 	nvlist_t *proprequest = fnvlist_alloc();
2183 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2184 	nvlist_t *bookmarks = fnvlist_alloc();
2185 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2186 	fnvlist_free(proprequest);
2187 	if (error != 0)
2188 		return (error);
2189 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2190 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2191 		nvlist_t *valuenv =
2192 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2193 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
2194 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2195 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2196 			fnvlist_free(bookmarks);
2197 			dsl_dataset_rele(ds, FTAG);
2198 			return (SET_ERROR(EEXIST));
2199 		}
2200 	}
2201 	fnvlist_free(bookmarks);
2202 
2203 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2204 	if (error != 0) {
2205 		dsl_dataset_rele(ds, FTAG);
2206 		return (error);
2207 	}
2208 
2209 	/*
2210 	 * Check if the snap we are rolling back to uses more than
2211 	 * the refquota.
2212 	 */
2213 	if (ds->ds_quota != 0 &&
2214 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2215 		dsl_dataset_rele(ds, FTAG);
2216 		return (SET_ERROR(EDQUOT));
2217 	}
2218 
2219 	/*
2220 	 * When we do the clone swap, we will temporarily use more space
2221 	 * due to the refreservation (the head will no longer have any
2222 	 * unique space, so the entire amount of the refreservation will need
2223 	 * to be free).  We will immediately destroy the clone, freeing
2224 	 * this space, but the freeing happens over many txg's.
2225 	 */
2226 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2227 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2228 
2229 	if (unused_refres_delta > 0 &&
2230 	    unused_refres_delta >
2231 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2232 		dsl_dataset_rele(ds, FTAG);
2233 		return (SET_ERROR(ENOSPC));
2234 	}
2235 
2236 	dsl_dataset_rele(ds, FTAG);
2237 	return (0);
2238 }
2239 
2240 static void
2241 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2242 {
2243 	dsl_dataset_rollback_arg_t *ddra = arg;
2244 	dsl_pool_t *dp = dmu_tx_pool(tx);
2245 	dsl_dataset_t *ds, *clone;
2246 	uint64_t cloneobj;
2247 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2248 
2249 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2250 
2251 	dsl_dataset_name(ds->ds_prev, namebuf);
2252 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2253 
2254 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2255 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2256 
2257 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2258 
2259 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2260 	dsl_dataset_zero_zil(ds, tx);
2261 
2262 	dsl_destroy_head_sync_impl(clone, tx);
2263 
2264 	dsl_dataset_rele(clone, FTAG);
2265 	dsl_dataset_rele(ds, FTAG);
2266 }
2267 
2268 /*
2269  * Rolls back the given filesystem or volume to the most recent snapshot.
2270  * The name of the most recent snapshot will be returned under key "target"
2271  * in the result nvlist.
2272  *
2273  * If owner != NULL:
2274  * - The existing dataset MUST be owned by the specified owner at entry
2275  * - Upon return, dataset will still be held by the same owner, whether we
2276  *   succeed or not.
2277  *
2278  * This mode is required any time the existing filesystem is mounted.  See
2279  * notes above zfs_suspend_fs() for further details.
2280  */
2281 int
2282 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2283 {
2284 	dsl_dataset_rollback_arg_t ddra;
2285 
2286 	ddra.ddra_fsname = fsname;
2287 	ddra.ddra_owner = owner;
2288 	ddra.ddra_result = result;
2289 
2290 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2291 	    dsl_dataset_rollback_sync, &ddra,
2292 	    1, ZFS_SPACE_CHECK_RESERVED));
2293 }
2294 
2295 struct promotenode {
2296 	list_node_t link;
2297 	dsl_dataset_t *ds;
2298 };
2299 
2300 typedef struct dsl_dataset_promote_arg {
2301 	const char *ddpa_clonename;
2302 	dsl_dataset_t *ddpa_clone;
2303 	list_t shared_snaps, origin_snaps, clone_snaps;
2304 	dsl_dataset_t *origin_origin; /* origin of the origin */
2305 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2306 	char *err_ds;
2307 	cred_t *cr;
2308 } dsl_dataset_promote_arg_t;
2309 
2310 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2311 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2312     void *tag);
2313 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2314 
2315 static int
2316 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2317 {
2318 	dsl_dataset_promote_arg_t *ddpa = arg;
2319 	dsl_pool_t *dp = dmu_tx_pool(tx);
2320 	dsl_dataset_t *hds;
2321 	struct promotenode *snap;
2322 	dsl_dataset_t *origin_ds;
2323 	int err;
2324 	uint64_t unused;
2325 	uint64_t ss_mv_cnt;
2326 	size_t max_snap_len;
2327 
2328 	err = promote_hold(ddpa, dp, FTAG);
2329 	if (err != 0)
2330 		return (err);
2331 
2332 	hds = ddpa->ddpa_clone;
2333 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2334 
2335 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2336 		promote_rele(ddpa, FTAG);
2337 		return (SET_ERROR(EXDEV));
2338 	}
2339 
2340 	/*
2341 	 * Compute and check the amount of space to transfer.  Since this is
2342 	 * so expensive, don't do the preliminary check.
2343 	 */
2344 	if (!dmu_tx_is_syncing(tx)) {
2345 		promote_rele(ddpa, FTAG);
2346 		return (0);
2347 	}
2348 
2349 	snap = list_head(&ddpa->shared_snaps);
2350 	origin_ds = snap->ds;
2351 
2352 	/* compute origin's new unique space */
2353 	snap = list_tail(&ddpa->clone_snaps);
2354 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2355 	    origin_ds->ds_object);
2356 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2357 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2358 	    &ddpa->unique, &unused, &unused);
2359 
2360 	/*
2361 	 * Walk the snapshots that we are moving
2362 	 *
2363 	 * Compute space to transfer.  Consider the incremental changes
2364 	 * to used by each snapshot:
2365 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2366 	 * So each snapshot gave birth to:
2367 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2368 	 * So a sequence would look like:
2369 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2370 	 * Which simplifies to:
2371 	 * uN + kN + kN-1 + ... + k1 + k0
2372 	 * Note however, if we stop before we reach the ORIGIN we get:
2373 	 * uN + kN + kN-1 + ... + kM - uM-1
2374 	 */
2375 	ss_mv_cnt = 0;
2376 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2377 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2378 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2379 	for (snap = list_head(&ddpa->shared_snaps); snap;
2380 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2381 		uint64_t val, dlused, dlcomp, dluncomp;
2382 		dsl_dataset_t *ds = snap->ds;
2383 
2384 		ss_mv_cnt++;
2385 
2386 		/*
2387 		 * If there are long holds, we won't be able to evict
2388 		 * the objset.
2389 		 */
2390 		if (dsl_dataset_long_held(ds)) {
2391 			err = SET_ERROR(EBUSY);
2392 			goto out;
2393 		}
2394 
2395 		/* Check that the snapshot name does not conflict */
2396 		VERIFY0(dsl_dataset_get_snapname(ds));
2397 		if (strlen(ds->ds_snapname) >= max_snap_len) {
2398 			err = SET_ERROR(ENAMETOOLONG);
2399 			goto out;
2400 		}
2401 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2402 		if (err == 0) {
2403 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2404 			err = SET_ERROR(EEXIST);
2405 			goto out;
2406 		}
2407 		if (err != ENOENT)
2408 			goto out;
2409 
2410 		/* The very first snapshot does not have a deadlist */
2411 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2412 			continue;
2413 
2414 		dsl_deadlist_space(&ds->ds_deadlist,
2415 		    &dlused, &dlcomp, &dluncomp);
2416 		ddpa->used += dlused;
2417 		ddpa->comp += dlcomp;
2418 		ddpa->uncomp += dluncomp;
2419 	}
2420 
2421 	/*
2422 	 * If we are a clone of a clone then we never reached ORIGIN,
2423 	 * so we need to subtract out the clone origin's used space.
2424 	 */
2425 	if (ddpa->origin_origin) {
2426 		ddpa->used -=
2427 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2428 		ddpa->comp -=
2429 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2430 		ddpa->uncomp -=
2431 		    dsl_dataset_phys(ddpa->origin_origin)->
2432 		    ds_uncompressed_bytes;
2433 	}
2434 
2435 	/* Check that there is enough space and limit headroom here */
2436 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2437 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2438 	if (err != 0)
2439 		goto out;
2440 
2441 	/*
2442 	 * Compute the amounts of space that will be used by snapshots
2443 	 * after the promotion (for both origin and clone).  For each,
2444 	 * it is the amount of space that will be on all of their
2445 	 * deadlists (that was not born before their new origin).
2446 	 */
2447 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2448 		uint64_t space;
2449 
2450 		/*
2451 		 * Note, typically this will not be a clone of a clone,
2452 		 * so dd_origin_txg will be < TXG_INITIAL, so
2453 		 * these snaplist_space() -> dsl_deadlist_space_range()
2454 		 * calls will be fast because they do not have to
2455 		 * iterate over all bps.
2456 		 */
2457 		snap = list_head(&ddpa->origin_snaps);
2458 		err = snaplist_space(&ddpa->shared_snaps,
2459 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2460 		if (err != 0)
2461 			goto out;
2462 
2463 		err = snaplist_space(&ddpa->clone_snaps,
2464 		    snap->ds->ds_dir->dd_origin_txg, &space);
2465 		if (err != 0)
2466 			goto out;
2467 		ddpa->cloneusedsnap += space;
2468 	}
2469 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2470 	    DD_FLAG_USED_BREAKDOWN) {
2471 		err = snaplist_space(&ddpa->origin_snaps,
2472 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2473 		    &ddpa->originusedsnap);
2474 		if (err != 0)
2475 			goto out;
2476 	}
2477 
2478 out:
2479 	promote_rele(ddpa, FTAG);
2480 	return (err);
2481 }
2482 
2483 static void
2484 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2485 {
2486 	dsl_dataset_promote_arg_t *ddpa = arg;
2487 	dsl_pool_t *dp = dmu_tx_pool(tx);
2488 	dsl_dataset_t *hds;
2489 	struct promotenode *snap;
2490 	dsl_dataset_t *origin_ds;
2491 	dsl_dataset_t *origin_head;
2492 	dsl_dir_t *dd;
2493 	dsl_dir_t *odd = NULL;
2494 	uint64_t oldnext_obj;
2495 	int64_t delta;
2496 
2497 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2498 	hds = ddpa->ddpa_clone;
2499 
2500 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2501 
2502 	snap = list_head(&ddpa->shared_snaps);
2503 	origin_ds = snap->ds;
2504 	dd = hds->ds_dir;
2505 
2506 	snap = list_head(&ddpa->origin_snaps);
2507 	origin_head = snap->ds;
2508 
2509 	/*
2510 	 * We need to explicitly open odd, since origin_ds's dd will be
2511 	 * changing.
2512 	 */
2513 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2514 	    NULL, FTAG, &odd));
2515 
2516 	/* change origin's next snap */
2517 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2518 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2519 	snap = list_tail(&ddpa->clone_snaps);
2520 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2521 	    origin_ds->ds_object);
2522 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2523 
2524 	/* change the origin's next clone */
2525 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2526 		dsl_dataset_remove_from_next_clones(origin_ds,
2527 		    snap->ds->ds_object, tx);
2528 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2529 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2530 		    oldnext_obj, tx));
2531 	}
2532 
2533 	/* change origin */
2534 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2535 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2536 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2537 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2538 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2539 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2540 	origin_head->ds_dir->dd_origin_txg =
2541 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2542 
2543 	/* change dd_clone entries */
2544 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2545 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2546 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2547 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2548 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2549 		    hds->ds_object, tx));
2550 
2551 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2552 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2553 		    origin_head->ds_object, tx));
2554 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2555 			dsl_dir_phys(dd)->dd_clones =
2556 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2557 			    DMU_OT_NONE, 0, tx);
2558 		}
2559 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2560 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2561 	}
2562 
2563 	/* move snapshots to this dir */
2564 	for (snap = list_head(&ddpa->shared_snaps); snap;
2565 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2566 		dsl_dataset_t *ds = snap->ds;
2567 
2568 		/*
2569 		 * Property callbacks are registered to a particular
2570 		 * dsl_dir.  Since ours is changing, evict the objset
2571 		 * so that they will be unregistered from the old dsl_dir.
2572 		 */
2573 		if (ds->ds_objset) {
2574 			dmu_objset_evict(ds->ds_objset);
2575 			ds->ds_objset = NULL;
2576 		}
2577 
2578 		/* move snap name entry */
2579 		VERIFY0(dsl_dataset_get_snapname(ds));
2580 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2581 		    ds->ds_snapname, tx, B_TRUE));
2582 		VERIFY0(zap_add(dp->dp_meta_objset,
2583 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2584 		    8, 1, &ds->ds_object, tx));
2585 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2586 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2587 
2588 		/* change containing dsl_dir */
2589 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2590 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2591 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2592 		ASSERT3P(ds->ds_dir, ==, odd);
2593 		dsl_dir_rele(ds->ds_dir, ds);
2594 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2595 		    NULL, ds, &ds->ds_dir));
2596 
2597 		/* move any clone references */
2598 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2599 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2600 			zap_cursor_t zc;
2601 			zap_attribute_t za;
2602 
2603 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2604 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
2605 			    zap_cursor_retrieve(&zc, &za) == 0;
2606 			    zap_cursor_advance(&zc)) {
2607 				dsl_dataset_t *cnds;
2608 				uint64_t o;
2609 
2610 				if (za.za_first_integer == oldnext_obj) {
2611 					/*
2612 					 * We've already moved the
2613 					 * origin's reference.
2614 					 */
2615 					continue;
2616 				}
2617 
2618 				VERIFY0(dsl_dataset_hold_obj(dp,
2619 				    za.za_first_integer, FTAG, &cnds));
2620 				o = dsl_dir_phys(cnds->ds_dir)->
2621 				    dd_head_dataset_obj;
2622 
2623 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2624 				    dsl_dir_phys(odd)->dd_clones, o, tx));
2625 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2626 				    dsl_dir_phys(dd)->dd_clones, o, tx));
2627 				dsl_dataset_rele(cnds, FTAG);
2628 			}
2629 			zap_cursor_fini(&zc);
2630 		}
2631 
2632 		ASSERT(!dsl_prop_hascb(ds));
2633 	}
2634 
2635 	/*
2636 	 * Change space accounting.
2637 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2638 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2639 	 * is true for each of {clone,origin} independently.
2640 	 */
2641 
2642 	delta = ddpa->cloneusedsnap -
2643 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2644 	ASSERT3S(delta, >=, 0);
2645 	ASSERT3U(ddpa->used, >=, delta);
2646 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2647 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2648 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2649 
2650 	delta = ddpa->originusedsnap -
2651 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2652 	ASSERT3S(delta, <=, 0);
2653 	ASSERT3U(ddpa->used, >=, -delta);
2654 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2655 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2656 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2657 
2658 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2659 
2660 	/* log history record */
2661 	spa_history_log_internal_ds(hds, "promote", tx, "");
2662 
2663 	dsl_dir_rele(odd, FTAG);
2664 	promote_rele(ddpa, FTAG);
2665 }
2666 
2667 /*
2668  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2669  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2670  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2671  * snapshots back to this dataset's origin.
2672  */
2673 static int
2674 snaplist_make(dsl_pool_t *dp,
2675     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2676 {
2677 	uint64_t obj = last_obj;
2678 
2679 	list_create(l, sizeof (struct promotenode),
2680 	    offsetof(struct promotenode, link));
2681 
2682 	while (obj != first_obj) {
2683 		dsl_dataset_t *ds;
2684 		struct promotenode *snap;
2685 		int err;
2686 
2687 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2688 		ASSERT(err != ENOENT);
2689 		if (err != 0)
2690 			return (err);
2691 
2692 		if (first_obj == 0)
2693 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2694 
2695 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2696 		snap->ds = ds;
2697 		list_insert_tail(l, snap);
2698 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2699 	}
2700 
2701 	return (0);
2702 }
2703 
2704 static int
2705 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2706 {
2707 	struct promotenode *snap;
2708 
2709 	*spacep = 0;
2710 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2711 		uint64_t used, comp, uncomp;
2712 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2713 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2714 		*spacep += used;
2715 	}
2716 	return (0);
2717 }
2718 
2719 static void
2720 snaplist_destroy(list_t *l, void *tag)
2721 {
2722 	struct promotenode *snap;
2723 
2724 	if (l == NULL || !list_link_active(&l->list_head))
2725 		return;
2726 
2727 	while ((snap = list_tail(l)) != NULL) {
2728 		list_remove(l, snap);
2729 		dsl_dataset_rele(snap->ds, tag);
2730 		kmem_free(snap, sizeof (*snap));
2731 	}
2732 	list_destroy(l);
2733 }
2734 
2735 static int
2736 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2737 {
2738 	int error;
2739 	dsl_dir_t *dd;
2740 	struct promotenode *snap;
2741 
2742 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2743 	    &ddpa->ddpa_clone);
2744 	if (error != 0)
2745 		return (error);
2746 	dd = ddpa->ddpa_clone->ds_dir;
2747 
2748 	if (ddpa->ddpa_clone->ds_is_snapshot ||
2749 	    !dsl_dir_is_clone(dd)) {
2750 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2751 		return (SET_ERROR(EINVAL));
2752 	}
2753 
2754 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2755 	    &ddpa->shared_snaps, tag);
2756 	if (error != 0)
2757 		goto out;
2758 
2759 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2760 	    &ddpa->clone_snaps, tag);
2761 	if (error != 0)
2762 		goto out;
2763 
2764 	snap = list_head(&ddpa->shared_snaps);
2765 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2766 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2767 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2768 	    &ddpa->origin_snaps, tag);
2769 	if (error != 0)
2770 		goto out;
2771 
2772 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2773 		error = dsl_dataset_hold_obj(dp,
2774 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2775 		    tag, &ddpa->origin_origin);
2776 		if (error != 0)
2777 			goto out;
2778 	}
2779 out:
2780 	if (error != 0)
2781 		promote_rele(ddpa, tag);
2782 	return (error);
2783 }
2784 
2785 static void
2786 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2787 {
2788 	snaplist_destroy(&ddpa->shared_snaps, tag);
2789 	snaplist_destroy(&ddpa->clone_snaps, tag);
2790 	snaplist_destroy(&ddpa->origin_snaps, tag);
2791 	if (ddpa->origin_origin != NULL)
2792 		dsl_dataset_rele(ddpa->origin_origin, tag);
2793 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2794 }
2795 
2796 /*
2797  * Promote a clone.
2798  *
2799  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2800  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2801  */
2802 int
2803 dsl_dataset_promote(const char *name, char *conflsnap)
2804 {
2805 	dsl_dataset_promote_arg_t ddpa = { 0 };
2806 	uint64_t numsnaps;
2807 	int error;
2808 	objset_t *os;
2809 
2810 	/*
2811 	 * We will modify space proportional to the number of
2812 	 * snapshots.  Compute numsnaps.
2813 	 */
2814 	error = dmu_objset_hold(name, FTAG, &os);
2815 	if (error != 0)
2816 		return (error);
2817 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2818 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2819 	    &numsnaps);
2820 	dmu_objset_rele(os, FTAG);
2821 	if (error != 0)
2822 		return (error);
2823 
2824 	ddpa.ddpa_clonename = name;
2825 	ddpa.err_ds = conflsnap;
2826 	ddpa.cr = CRED();
2827 
2828 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2829 	    dsl_dataset_promote_sync, &ddpa,
2830 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2831 }
2832 
2833 int
2834 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2835     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2836 {
2837 	/*
2838 	 * "slack" factor for received datasets with refquota set on them.
2839 	 * See the bottom of this function for details on its use.
2840 	 */
2841 	uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
2842 	int64_t unused_refres_delta;
2843 
2844 	/* they should both be heads */
2845 	if (clone->ds_is_snapshot ||
2846 	    origin_head->ds_is_snapshot)
2847 		return (SET_ERROR(EINVAL));
2848 
2849 	/* if we are not forcing, the branch point should be just before them */
2850 	if (!force && clone->ds_prev != origin_head->ds_prev)
2851 		return (SET_ERROR(EINVAL));
2852 
2853 	/* clone should be the clone (unless they are unrelated) */
2854 	if (clone->ds_prev != NULL &&
2855 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2856 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2857 		return (SET_ERROR(EINVAL));
2858 
2859 	/* the clone should be a child of the origin */
2860 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2861 		return (SET_ERROR(EINVAL));
2862 
2863 	/* origin_head shouldn't be modified unless 'force' */
2864 	if (!force &&
2865 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2866 		return (SET_ERROR(ETXTBSY));
2867 
2868 	/* origin_head should have no long holds (e.g. is not mounted) */
2869 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2870 		return (SET_ERROR(EBUSY));
2871 
2872 	/* check amount of any unconsumed refreservation */
2873 	unused_refres_delta =
2874 	    (int64_t)MIN(origin_head->ds_reserved,
2875 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2876 	    (int64_t)MIN(origin_head->ds_reserved,
2877 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2878 
2879 	if (unused_refres_delta > 0 &&
2880 	    unused_refres_delta >
2881 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2882 		return (SET_ERROR(ENOSPC));
2883 
2884 	/*
2885 	 * The clone can't be too much over the head's refquota.
2886 	 *
2887 	 * To ensure that the entire refquota can be used, we allow one
2888 	 * transaction to exceed the the refquota.  Therefore, this check
2889 	 * needs to also allow for the space referenced to be more than the
2890 	 * refquota.  The maximum amount of space that one transaction can use
2891 	 * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
2892 	 * overage ensures that we are able to receive a filesystem that
2893 	 * exceeds the refquota on the source system.
2894 	 *
2895 	 * So that overage is the refquota_slack we use below.
2896 	 */
2897 	if (origin_head->ds_quota != 0 &&
2898 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
2899 	    origin_head->ds_quota + refquota_slack)
2900 		return (SET_ERROR(EDQUOT));
2901 
2902 	return (0);
2903 }
2904 
2905 void
2906 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2907     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2908 {
2909 	dsl_pool_t *dp = dmu_tx_pool(tx);
2910 	int64_t unused_refres_delta;
2911 
2912 	ASSERT(clone->ds_reserved == 0);
2913 	/*
2914 	 * NOTE: On DEBUG kernels there could be a race between this and
2915 	 * the check function if spa_asize_inflation is adjusted...
2916 	 */
2917 	ASSERT(origin_head->ds_quota == 0 ||
2918 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
2919 	    DMU_MAX_ACCESS * spa_asize_inflation);
2920 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2921 
2922 	/*
2923 	 * Swap per-dataset feature flags.
2924 	 */
2925 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2926 		if (!(spa_feature_table[f].fi_flags &
2927 		    ZFEATURE_FLAG_PER_DATASET)) {
2928 			ASSERT(!clone->ds_feature_inuse[f]);
2929 			ASSERT(!origin_head->ds_feature_inuse[f]);
2930 			continue;
2931 		}
2932 
2933 		boolean_t clone_inuse = clone->ds_feature_inuse[f];
2934 		boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2935 
2936 		if (clone_inuse) {
2937 			dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2938 			clone->ds_feature_inuse[f] = B_FALSE;
2939 		}
2940 		if (origin_head_inuse) {
2941 			dsl_dataset_deactivate_feature(origin_head->ds_object,
2942 			    f, tx);
2943 			origin_head->ds_feature_inuse[f] = B_FALSE;
2944 		}
2945 		if (clone_inuse) {
2946 			dsl_dataset_activate_feature(origin_head->ds_object,
2947 			    f, tx);
2948 			origin_head->ds_feature_inuse[f] = B_TRUE;
2949 		}
2950 		if (origin_head_inuse) {
2951 			dsl_dataset_activate_feature(clone->ds_object, f, tx);
2952 			clone->ds_feature_inuse[f] = B_TRUE;
2953 		}
2954 	}
2955 
2956 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2957 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2958 
2959 	if (clone->ds_objset != NULL) {
2960 		dmu_objset_evict(clone->ds_objset);
2961 		clone->ds_objset = NULL;
2962 	}
2963 
2964 	if (origin_head->ds_objset != NULL) {
2965 		dmu_objset_evict(origin_head->ds_objset);
2966 		origin_head->ds_objset = NULL;
2967 	}
2968 
2969 	unused_refres_delta =
2970 	    (int64_t)MIN(origin_head->ds_reserved,
2971 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2972 	    (int64_t)MIN(origin_head->ds_reserved,
2973 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2974 
2975 	/*
2976 	 * Reset origin's unique bytes, if it exists.
2977 	 */
2978 	if (clone->ds_prev) {
2979 		dsl_dataset_t *origin = clone->ds_prev;
2980 		uint64_t comp, uncomp;
2981 
2982 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2983 		dsl_deadlist_space_range(&clone->ds_deadlist,
2984 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2985 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2986 	}
2987 
2988 	/* swap blkptrs */
2989 	{
2990 		blkptr_t tmp;
2991 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
2992 		dsl_dataset_phys(origin_head)->ds_bp =
2993 		    dsl_dataset_phys(clone)->ds_bp;
2994 		dsl_dataset_phys(clone)->ds_bp = tmp;
2995 	}
2996 
2997 	/* set dd_*_bytes */
2998 	{
2999 		int64_t dused, dcomp, duncomp;
3000 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
3001 		uint64_t odl_used, odl_comp, odl_uncomp;
3002 
3003 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3004 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
3005 
3006 		dsl_deadlist_space(&clone->ds_deadlist,
3007 		    &cdl_used, &cdl_comp, &cdl_uncomp);
3008 		dsl_deadlist_space(&origin_head->ds_deadlist,
3009 		    &odl_used, &odl_comp, &odl_uncomp);
3010 
3011 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3012 		    cdl_used -
3013 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3014 		    odl_used);
3015 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3016 		    cdl_comp -
3017 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3018 		    odl_comp);
3019 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3020 		    cdl_uncomp -
3021 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3022 		    odl_uncomp);
3023 
3024 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3025 		    dused, dcomp, duncomp, tx);
3026 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3027 		    -dused, -dcomp, -duncomp, tx);
3028 
3029 		/*
3030 		 * The difference in the space used by snapshots is the
3031 		 * difference in snapshot space due to the head's
3032 		 * deadlist (since that's the only thing that's
3033 		 * changing that affects the snapused).
3034 		 */
3035 		dsl_deadlist_space_range(&clone->ds_deadlist,
3036 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3037 		    &cdl_used, &cdl_comp, &cdl_uncomp);
3038 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
3039 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3040 		    &odl_used, &odl_comp, &odl_uncomp);
3041 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3042 		    DD_USED_HEAD, DD_USED_SNAP, tx);
3043 	}
3044 
3045 	/* swap ds_*_bytes */
3046 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3047 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
3048 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3049 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
3050 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3051 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3052 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3053 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3054 
3055 	/* apply any parent delta for change in unconsumed refreservation */
3056 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3057 	    unused_refres_delta, 0, 0, tx);
3058 
3059 	/*
3060 	 * Swap deadlists.
3061 	 */
3062 	dsl_deadlist_close(&clone->ds_deadlist);
3063 	dsl_deadlist_close(&origin_head->ds_deadlist);
3064 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3065 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3066 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3067 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3068 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3069 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3070 
3071 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3072 
3073 	spa_history_log_internal_ds(clone, "clone swap", tx,
3074 	    "parent=%s", origin_head->ds_dir->dd_myname);
3075 }
3076 
3077 /*
3078  * Given a pool name and a dataset object number in that pool,
3079  * return the name of that dataset.
3080  */
3081 int
3082 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3083 {
3084 	dsl_pool_t *dp;
3085 	dsl_dataset_t *ds;
3086 	int error;
3087 
3088 	error = dsl_pool_hold(pname, FTAG, &dp);
3089 	if (error != 0)
3090 		return (error);
3091 
3092 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3093 	if (error == 0) {
3094 		dsl_dataset_name(ds, buf);
3095 		dsl_dataset_rele(ds, FTAG);
3096 	}
3097 	dsl_pool_rele(dp, FTAG);
3098 
3099 	return (error);
3100 }
3101 
3102 int
3103 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3104     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3105 {
3106 	int error = 0;
3107 
3108 	ASSERT3S(asize, >, 0);
3109 
3110 	/*
3111 	 * *ref_rsrv is the portion of asize that will come from any
3112 	 * unconsumed refreservation space.
3113 	 */
3114 	*ref_rsrv = 0;
3115 
3116 	mutex_enter(&ds->ds_lock);
3117 	/*
3118 	 * Make a space adjustment for reserved bytes.
3119 	 */
3120 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3121 		ASSERT3U(*used, >=,
3122 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3123 		*used -=
3124 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3125 		*ref_rsrv =
3126 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3127 	}
3128 
3129 	if (!check_quota || ds->ds_quota == 0) {
3130 		mutex_exit(&ds->ds_lock);
3131 		return (0);
3132 	}
3133 	/*
3134 	 * If they are requesting more space, and our current estimate
3135 	 * is over quota, they get to try again unless the actual
3136 	 * on-disk is over quota and there are no pending changes (which
3137 	 * may free up space for us).
3138 	 */
3139 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3140 	    ds->ds_quota) {
3141 		if (inflight > 0 ||
3142 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3143 			error = SET_ERROR(ERESTART);
3144 		else
3145 			error = SET_ERROR(EDQUOT);
3146 	}
3147 	mutex_exit(&ds->ds_lock);
3148 
3149 	return (error);
3150 }
3151 
3152 typedef struct dsl_dataset_set_qr_arg {
3153 	const char *ddsqra_name;
3154 	zprop_source_t ddsqra_source;
3155 	uint64_t ddsqra_value;
3156 } dsl_dataset_set_qr_arg_t;
3157 
3158 
3159 /* ARGSUSED */
3160 static int
3161 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3162 {
3163 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3164 	dsl_pool_t *dp = dmu_tx_pool(tx);
3165 	dsl_dataset_t *ds;
3166 	int error;
3167 	uint64_t newval;
3168 
3169 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3170 		return (SET_ERROR(ENOTSUP));
3171 
3172 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3173 	if (error != 0)
3174 		return (error);
3175 
3176 	if (ds->ds_is_snapshot) {
3177 		dsl_dataset_rele(ds, FTAG);
3178 		return (SET_ERROR(EINVAL));
3179 	}
3180 
3181 	error = dsl_prop_predict(ds->ds_dir,
3182 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3183 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3184 	if (error != 0) {
3185 		dsl_dataset_rele(ds, FTAG);
3186 		return (error);
3187 	}
3188 
3189 	if (newval == 0) {
3190 		dsl_dataset_rele(ds, FTAG);
3191 		return (0);
3192 	}
3193 
3194 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3195 	    newval < ds->ds_reserved) {
3196 		dsl_dataset_rele(ds, FTAG);
3197 		return (SET_ERROR(ENOSPC));
3198 	}
3199 
3200 	dsl_dataset_rele(ds, FTAG);
3201 	return (0);
3202 }
3203 
3204 static void
3205 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3206 {
3207 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3208 	dsl_pool_t *dp = dmu_tx_pool(tx);
3209 	dsl_dataset_t *ds;
3210 	uint64_t newval;
3211 
3212 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3213 
3214 	dsl_prop_set_sync_impl(ds,
3215 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3216 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3217 	    &ddsqra->ddsqra_value, tx);
3218 
3219 	VERIFY0(dsl_prop_get_int_ds(ds,
3220 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3221 
3222 	if (ds->ds_quota != newval) {
3223 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3224 		ds->ds_quota = newval;
3225 	}
3226 	dsl_dataset_rele(ds, FTAG);
3227 }
3228 
3229 int
3230 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3231     uint64_t refquota)
3232 {
3233 	dsl_dataset_set_qr_arg_t ddsqra;
3234 
3235 	ddsqra.ddsqra_name = dsname;
3236 	ddsqra.ddsqra_source = source;
3237 	ddsqra.ddsqra_value = refquota;
3238 
3239 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3240 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3241 }
3242 
3243 static int
3244 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3245 {
3246 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3247 	dsl_pool_t *dp = dmu_tx_pool(tx);
3248 	dsl_dataset_t *ds;
3249 	int error;
3250 	uint64_t newval, unique;
3251 
3252 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3253 		return (SET_ERROR(ENOTSUP));
3254 
3255 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3256 	if (error != 0)
3257 		return (error);
3258 
3259 	if (ds->ds_is_snapshot) {
3260 		dsl_dataset_rele(ds, FTAG);
3261 		return (SET_ERROR(EINVAL));
3262 	}
3263 
3264 	error = dsl_prop_predict(ds->ds_dir,
3265 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3266 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3267 	if (error != 0) {
3268 		dsl_dataset_rele(ds, FTAG);
3269 		return (error);
3270 	}
3271 
3272 	/*
3273 	 * If we are doing the preliminary check in open context, the
3274 	 * space estimates may be inaccurate.
3275 	 */
3276 	if (!dmu_tx_is_syncing(tx)) {
3277 		dsl_dataset_rele(ds, FTAG);
3278 		return (0);
3279 	}
3280 
3281 	mutex_enter(&ds->ds_lock);
3282 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3283 		dsl_dataset_recalc_head_uniq(ds);
3284 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3285 	mutex_exit(&ds->ds_lock);
3286 
3287 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3288 		uint64_t delta = MAX(unique, newval) -
3289 		    MAX(unique, ds->ds_reserved);
3290 
3291 		if (delta >
3292 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3293 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3294 			dsl_dataset_rele(ds, FTAG);
3295 			return (SET_ERROR(ENOSPC));
3296 		}
3297 	}
3298 
3299 	dsl_dataset_rele(ds, FTAG);
3300 	return (0);
3301 }
3302 
3303 void
3304 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3305     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3306 {
3307 	uint64_t newval;
3308 	uint64_t unique;
3309 	int64_t delta;
3310 
3311 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3312 	    source, sizeof (value), 1, &value, tx);
3313 
3314 	VERIFY0(dsl_prop_get_int_ds(ds,
3315 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3316 
3317 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3318 	mutex_enter(&ds->ds_dir->dd_lock);
3319 	mutex_enter(&ds->ds_lock);
3320 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3321 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3322 	delta = MAX(0, (int64_t)(newval - unique)) -
3323 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3324 	ds->ds_reserved = newval;
3325 	mutex_exit(&ds->ds_lock);
3326 
3327 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3328 	mutex_exit(&ds->ds_dir->dd_lock);
3329 }
3330 
3331 static void
3332 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3333 {
3334 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3335 	dsl_pool_t *dp = dmu_tx_pool(tx);
3336 	dsl_dataset_t *ds;
3337 
3338 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3339 	dsl_dataset_set_refreservation_sync_impl(ds,
3340 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3341 	dsl_dataset_rele(ds, FTAG);
3342 }
3343 
3344 int
3345 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3346     uint64_t refreservation)
3347 {
3348 	dsl_dataset_set_qr_arg_t ddsqra;
3349 
3350 	ddsqra.ddsqra_name = dsname;
3351 	ddsqra.ddsqra_source = source;
3352 	ddsqra.ddsqra_value = refreservation;
3353 
3354 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3355 	    dsl_dataset_set_refreservation_sync, &ddsqra,
3356 	    0, ZFS_SPACE_CHECK_NONE));
3357 }
3358 
3359 /*
3360  * Return (in *usedp) the amount of space written in new that is not
3361  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3362  * a snapshot before new, in new's filesystem (or its origin).  If not then
3363  * fail and return EINVAL.
3364  *
3365  * The written space is calculated by considering two components:  First, we
3366  * ignore any freed space, and calculate the written as new's used space
3367  * minus old's used space.  Next, we add in the amount of space that was freed
3368  * between the two snapshots, thus reducing new's used space relative to old's.
3369  * Specifically, this is the space that was born before old->ds_creation_txg,
3370  * and freed before new (ie. on new's deadlist or a previous deadlist).
3371  *
3372  * space freed                         [---------------------]
3373  * snapshots                       ---O-------O--------O-------O------
3374  *                                         oldsnap            new
3375  */
3376 int
3377 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3378     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3379 {
3380 	int err = 0;
3381 	uint64_t snapobj;
3382 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3383 
3384 	ASSERT(dsl_pool_config_held(dp));
3385 
3386 	*usedp = 0;
3387 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3388 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3389 
3390 	*compp = 0;
3391 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3392 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3393 
3394 	*uncompp = 0;
3395 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3396 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3397 
3398 	snapobj = new->ds_object;
3399 	while (snapobj != oldsnap->ds_object) {
3400 		dsl_dataset_t *snap;
3401 		uint64_t used, comp, uncomp;
3402 
3403 		if (snapobj == new->ds_object) {
3404 			snap = new;
3405 		} else {
3406 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3407 			if (err != 0)
3408 				break;
3409 		}
3410 
3411 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3412 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3413 			/*
3414 			 * The blocks in the deadlist can not be born after
3415 			 * ds_prev_snap_txg, so get the whole deadlist space,
3416 			 * which is more efficient (especially for old-format
3417 			 * deadlists).  Unfortunately the deadlist code
3418 			 * doesn't have enough information to make this
3419 			 * optimization itself.
3420 			 */
3421 			dsl_deadlist_space(&snap->ds_deadlist,
3422 			    &used, &comp, &uncomp);
3423 		} else {
3424 			dsl_deadlist_space_range(&snap->ds_deadlist,
3425 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3426 			    &used, &comp, &uncomp);
3427 		}
3428 		*usedp += used;
3429 		*compp += comp;
3430 		*uncompp += uncomp;
3431 
3432 		/*
3433 		 * If we get to the beginning of the chain of snapshots
3434 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3435 		 * was not a snapshot of/before new.
3436 		 */
3437 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3438 		if (snap != new)
3439 			dsl_dataset_rele(snap, FTAG);
3440 		if (snapobj == 0) {
3441 			err = SET_ERROR(EINVAL);
3442 			break;
3443 		}
3444 
3445 	}
3446 	return (err);
3447 }
3448 
3449 /*
3450  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3451  * lastsnap, and all snapshots in between are deleted.
3452  *
3453  * blocks that would be freed            [---------------------------]
3454  * snapshots                       ---O-------O--------O-------O--------O
3455  *                                        firstsnap        lastsnap
3456  *
3457  * This is the set of blocks that were born after the snap before firstsnap,
3458  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3459  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3460  * We calculate this by iterating over the relevant deadlists (from the snap
3461  * after lastsnap, backward to the snap after firstsnap), summing up the
3462  * space on the deadlist that was born after the snap before firstsnap.
3463  */
3464 int
3465 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3466     dsl_dataset_t *lastsnap,
3467     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3468 {
3469 	int err = 0;
3470 	uint64_t snapobj;
3471 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3472 
3473 	ASSERT(firstsnap->ds_is_snapshot);
3474 	ASSERT(lastsnap->ds_is_snapshot);
3475 
3476 	/*
3477 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3478 	 * is before lastsnap.
3479 	 */
3480 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3481 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3482 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3483 		return (SET_ERROR(EINVAL));
3484 
3485 	*usedp = *compp = *uncompp = 0;
3486 
3487 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3488 	while (snapobj != firstsnap->ds_object) {
3489 		dsl_dataset_t *ds;
3490 		uint64_t used, comp, uncomp;
3491 
3492 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3493 		if (err != 0)
3494 			break;
3495 
3496 		dsl_deadlist_space_range(&ds->ds_deadlist,
3497 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3498 		    &used, &comp, &uncomp);
3499 		*usedp += used;
3500 		*compp += comp;
3501 		*uncompp += uncomp;
3502 
3503 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3504 		ASSERT3U(snapobj, !=, 0);
3505 		dsl_dataset_rele(ds, FTAG);
3506 	}
3507 	return (err);
3508 }
3509 
3510 /*
3511  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3512  * For example, they could both be snapshots of the same filesystem, and
3513  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3514  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3515  * filesystem.  Or 'earlier' could be the origin's origin.
3516  *
3517  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3518  */
3519 boolean_t
3520 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3521     uint64_t earlier_txg)
3522 {
3523 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3524 	int error;
3525 	boolean_t ret;
3526 
3527 	ASSERT(dsl_pool_config_held(dp));
3528 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3529 
3530 	if (earlier_txg == 0)
3531 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3532 
3533 	if (later->ds_is_snapshot &&
3534 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3535 		return (B_FALSE);
3536 
3537 	if (later->ds_dir == earlier->ds_dir)
3538 		return (B_TRUE);
3539 	if (!dsl_dir_is_clone(later->ds_dir))
3540 		return (B_FALSE);
3541 
3542 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3543 		return (B_TRUE);
3544 	dsl_dataset_t *origin;
3545 	error = dsl_dataset_hold_obj(dp,
3546 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3547 	if (error != 0)
3548 		return (B_FALSE);
3549 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3550 	dsl_dataset_rele(origin, FTAG);
3551 	return (ret);
3552 }
3553 
3554 void
3555 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3556 {
3557 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3558 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3559 }
3560 
3561 boolean_t
3562 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3563 {
3564 	dmu_object_info_t doi;
3565 
3566 	dmu_object_info_from_db(ds->ds_dbuf, &doi);
3567 	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3568 }
3569 
3570 boolean_t
3571 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3572 {
3573 	return (dsl_dataset_is_zapified(ds) &&
3574 	    zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3575 	    ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3576 }
3577