xref: /titanic_51/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 0c3cd038dba1c3a4dcfa2ecc9d7da7c44b11730f)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
25  * Copyright (c) 2014 RackTop Systems.
26  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27  */
28 
29 #include <sys/dmu_objset.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dmu_traverse.h>
35 #include <sys/dmu_impl.h>
36 #include <sys/dmu_tx.h>
37 #include <sys/arc.h>
38 #include <sys/zio.h>
39 #include <sys/zap.h>
40 #include <sys/zfeature.h>
41 #include <sys/unique.h>
42 #include <sys/zfs_context.h>
43 #include <sys/zfs_ioctl.h>
44 #include <sys/spa.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/zfs_onexit.h>
47 #include <sys/zvol.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/dsl_deadlist.h>
50 #include <sys/dsl_destroy.h>
51 #include <sys/dsl_userhold.h>
52 #include <sys/dsl_bookmark.h>
53 #include <sys/dmu_send.h>
54 #include <sys/zio_checksum.h>
55 
56 /*
57  * The SPA supports block sizes up to 16MB.  However, very large blocks
58  * can have an impact on i/o latency (e.g. tying up a spinning disk for
59  * ~300ms), and also potentially on the memory allocator.  Therefore,
60  * we do not allow the recordsize to be set larger than zfs_max_recordsize
61  * (default 1MB).  Larger blocks can be created by changing this tunable,
62  * and pools with larger blocks can always be imported and used, regardless
63  * of this setting.
64  */
65 int zfs_max_recordsize = 1 * 1024 * 1024;
66 
67 #define	SWITCH64(x, y) \
68 	{ \
69 		uint64_t __tmp = (x); \
70 		(x) = (y); \
71 		(y) = __tmp; \
72 	}
73 
74 #define	DS_REF_MAX	(1ULL << 62)
75 
76 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
77 
78 /*
79  * Figure out how much of this delta should be propogated to the dsl_dir
80  * layer.  If there's a refreservation, that space has already been
81  * partially accounted for in our ancestors.
82  */
83 static int64_t
84 parent_delta(dsl_dataset_t *ds, int64_t delta)
85 {
86 	dsl_dataset_phys_t *ds_phys;
87 	uint64_t old_bytes, new_bytes;
88 
89 	if (ds->ds_reserved == 0)
90 		return (delta);
91 
92 	ds_phys = dsl_dataset_phys(ds);
93 	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
94 	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
95 
96 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
97 	return (new_bytes - old_bytes);
98 }
99 
100 void
101 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
102 {
103 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
104 	int compressed = BP_GET_PSIZE(bp);
105 	int uncompressed = BP_GET_UCSIZE(bp);
106 	int64_t delta;
107 
108 	dprintf_bp(bp, "ds=%p", ds);
109 
110 	ASSERT(dmu_tx_is_syncing(tx));
111 	/* It could have been compressed away to nothing */
112 	if (BP_IS_HOLE(bp))
113 		return;
114 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
115 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
116 	if (ds == NULL) {
117 		dsl_pool_mos_diduse_space(tx->tx_pool,
118 		    used, compressed, uncompressed);
119 		return;
120 	}
121 
122 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
123 	mutex_enter(&ds->ds_lock);
124 	delta = parent_delta(ds, used);
125 	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
126 	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
127 	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
128 	dsl_dataset_phys(ds)->ds_unique_bytes += used;
129 
130 	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
131 		ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
132 		    B_TRUE;
133 	}
134 
135 	spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
136 	if (f != SPA_FEATURE_NONE)
137 		ds->ds_feature_activation_needed[f] = B_TRUE;
138 
139 	mutex_exit(&ds->ds_lock);
140 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
141 	    compressed, uncompressed, tx);
142 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
143 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
144 }
145 
146 int
147 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
148     boolean_t async)
149 {
150 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
151 	int compressed = BP_GET_PSIZE(bp);
152 	int uncompressed = BP_GET_UCSIZE(bp);
153 
154 	if (BP_IS_HOLE(bp))
155 		return (0);
156 
157 	ASSERT(dmu_tx_is_syncing(tx));
158 	ASSERT(bp->blk_birth <= tx->tx_txg);
159 
160 	if (ds == NULL) {
161 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
162 		dsl_pool_mos_diduse_space(tx->tx_pool,
163 		    -used, -compressed, -uncompressed);
164 		return (used);
165 	}
166 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
167 
168 	ASSERT(!ds->ds_is_snapshot);
169 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
170 
171 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
172 		int64_t delta;
173 
174 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
175 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
176 
177 		mutex_enter(&ds->ds_lock);
178 		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
179 		    !DS_UNIQUE_IS_ACCURATE(ds));
180 		delta = parent_delta(ds, -used);
181 		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
182 		mutex_exit(&ds->ds_lock);
183 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
184 		    delta, -compressed, -uncompressed, tx);
185 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
186 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
187 	} else {
188 		dprintf_bp(bp, "putting on dead list: %s", "");
189 		if (async) {
190 			/*
191 			 * We are here as part of zio's write done callback,
192 			 * which means we're a zio interrupt thread.  We can't
193 			 * call dsl_deadlist_insert() now because it may block
194 			 * waiting for I/O.  Instead, put bp on the deferred
195 			 * queue and let dsl_pool_sync() finish the job.
196 			 */
197 			bplist_append(&ds->ds_pending_deadlist, bp);
198 		} else {
199 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
200 		}
201 		ASSERT3U(ds->ds_prev->ds_object, ==,
202 		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
203 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
204 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
205 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
206 		    ds->ds_object && bp->blk_birth >
207 		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
208 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
209 			mutex_enter(&ds->ds_prev->ds_lock);
210 			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
211 			mutex_exit(&ds->ds_prev->ds_lock);
212 		}
213 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
214 			dsl_dir_transfer_space(ds->ds_dir, used,
215 			    DD_USED_HEAD, DD_USED_SNAP, tx);
216 		}
217 	}
218 	mutex_enter(&ds->ds_lock);
219 	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
220 	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
221 	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
222 	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
223 	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
224 	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
225 	mutex_exit(&ds->ds_lock);
226 
227 	return (used);
228 }
229 
230 uint64_t
231 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
232 {
233 	uint64_t trysnap = 0;
234 
235 	if (ds == NULL)
236 		return (0);
237 	/*
238 	 * The snapshot creation could fail, but that would cause an
239 	 * incorrect FALSE return, which would only result in an
240 	 * overestimation of the amount of space that an operation would
241 	 * consume, which is OK.
242 	 *
243 	 * There's also a small window where we could miss a pending
244 	 * snapshot, because we could set the sync task in the quiescing
245 	 * phase.  So this should only be used as a guess.
246 	 */
247 	if (ds->ds_trysnap_txg >
248 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
249 		trysnap = ds->ds_trysnap_txg;
250 	return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
251 }
252 
253 boolean_t
254 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
255     uint64_t blk_birth)
256 {
257 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
258 	    (bp != NULL && BP_IS_HOLE(bp)))
259 		return (B_FALSE);
260 
261 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
262 
263 	return (B_TRUE);
264 }
265 
266 static void
267 dsl_dataset_evict(void *dbu)
268 {
269 	dsl_dataset_t *ds = dbu;
270 
271 	ASSERT(ds->ds_owner == NULL);
272 
273 	ds->ds_dbuf = NULL;
274 
275 	unique_remove(ds->ds_fsid_guid);
276 
277 	if (ds->ds_objset != NULL)
278 		dmu_objset_evict(ds->ds_objset);
279 
280 	if (ds->ds_prev) {
281 		dsl_dataset_rele(ds->ds_prev, ds);
282 		ds->ds_prev = NULL;
283 	}
284 
285 	bplist_destroy(&ds->ds_pending_deadlist);
286 	if (ds->ds_deadlist.dl_os != NULL)
287 		dsl_deadlist_close(&ds->ds_deadlist);
288 	if (ds->ds_dir)
289 		dsl_dir_async_rele(ds->ds_dir, ds);
290 
291 	ASSERT(!list_link_active(&ds->ds_synced_link));
292 
293 	list_destroy(&ds->ds_prop_cbs);
294 	mutex_destroy(&ds->ds_lock);
295 	mutex_destroy(&ds->ds_opening_lock);
296 	mutex_destroy(&ds->ds_sendstream_lock);
297 	refcount_destroy(&ds->ds_longholds);
298 
299 	kmem_free(ds, sizeof (dsl_dataset_t));
300 }
301 
302 int
303 dsl_dataset_get_snapname(dsl_dataset_t *ds)
304 {
305 	dsl_dataset_phys_t *headphys;
306 	int err;
307 	dmu_buf_t *headdbuf;
308 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
309 	objset_t *mos = dp->dp_meta_objset;
310 
311 	if (ds->ds_snapname[0])
312 		return (0);
313 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
314 		return (0);
315 
316 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
317 	    FTAG, &headdbuf);
318 	if (err != 0)
319 		return (err);
320 	headphys = headdbuf->db_data;
321 	err = zap_value_search(dp->dp_meta_objset,
322 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
323 	dmu_buf_rele(headdbuf, FTAG);
324 	return (err);
325 }
326 
327 int
328 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
329 {
330 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
331 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
332 	matchtype_t mt;
333 	int err;
334 
335 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
336 		mt = MT_FIRST;
337 	else
338 		mt = MT_EXACT;
339 
340 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
341 	    value, mt, NULL, 0, NULL);
342 	if (err == ENOTSUP && mt == MT_FIRST)
343 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
344 	return (err);
345 }
346 
347 int
348 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
349     boolean_t adj_cnt)
350 {
351 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
352 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
353 	matchtype_t mt;
354 	int err;
355 
356 	dsl_dir_snap_cmtime_update(ds->ds_dir);
357 
358 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
359 		mt = MT_FIRST;
360 	else
361 		mt = MT_EXACT;
362 
363 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
364 	if (err == ENOTSUP && mt == MT_FIRST)
365 		err = zap_remove(mos, snapobj, name, tx);
366 
367 	if (err == 0 && adj_cnt)
368 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
369 		    DD_FIELD_SNAPSHOT_COUNT, tx);
370 
371 	return (err);
372 }
373 
374 boolean_t
375 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
376 {
377 	dmu_buf_t *dbuf = ds->ds_dbuf;
378 	boolean_t result = B_FALSE;
379 
380 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
381 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
382 
383 		if (ds == dmu_buf_get_user(dbuf))
384 			result = B_TRUE;
385 		else
386 			dmu_buf_rele(dbuf, tag);
387 	}
388 
389 	return (result);
390 }
391 
392 int
393 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
394     dsl_dataset_t **dsp)
395 {
396 	objset_t *mos = dp->dp_meta_objset;
397 	dmu_buf_t *dbuf;
398 	dsl_dataset_t *ds;
399 	int err;
400 	dmu_object_info_t doi;
401 
402 	ASSERT(dsl_pool_config_held(dp));
403 
404 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
405 	if (err != 0)
406 		return (err);
407 
408 	/* Make sure dsobj has the correct object type. */
409 	dmu_object_info_from_db(dbuf, &doi);
410 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
411 		dmu_buf_rele(dbuf, tag);
412 		return (SET_ERROR(EINVAL));
413 	}
414 
415 	ds = dmu_buf_get_user(dbuf);
416 	if (ds == NULL) {
417 		dsl_dataset_t *winner = NULL;
418 
419 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
420 		ds->ds_dbuf = dbuf;
421 		ds->ds_object = dsobj;
422 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
423 
424 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
425 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
426 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
427 		refcount_create(&ds->ds_longholds);
428 
429 		bplist_create(&ds->ds_pending_deadlist);
430 		dsl_deadlist_open(&ds->ds_deadlist,
431 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
432 
433 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
434 		    offsetof(dmu_sendarg_t, dsa_link));
435 
436 		list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
437 		    offsetof(dsl_prop_cb_record_t, cbr_ds_node));
438 
439 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
440 			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
441 				if (!(spa_feature_table[f].fi_flags &
442 				    ZFEATURE_FLAG_PER_DATASET))
443 					continue;
444 				err = zap_contains(mos, dsobj,
445 				    spa_feature_table[f].fi_guid);
446 				if (err == 0) {
447 					ds->ds_feature_inuse[f] = B_TRUE;
448 				} else {
449 					ASSERT3U(err, ==, ENOENT);
450 					err = 0;
451 				}
452 			}
453 		}
454 
455 		err = dsl_dir_hold_obj(dp,
456 		    dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
457 		if (err != 0) {
458 			mutex_destroy(&ds->ds_lock);
459 			mutex_destroy(&ds->ds_opening_lock);
460 			mutex_destroy(&ds->ds_sendstream_lock);
461 			refcount_destroy(&ds->ds_longholds);
462 			bplist_destroy(&ds->ds_pending_deadlist);
463 			dsl_deadlist_close(&ds->ds_deadlist);
464 			kmem_free(ds, sizeof (dsl_dataset_t));
465 			dmu_buf_rele(dbuf, tag);
466 			return (err);
467 		}
468 
469 		if (!ds->ds_is_snapshot) {
470 			ds->ds_snapname[0] = '\0';
471 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
472 				err = dsl_dataset_hold_obj(dp,
473 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
474 				    ds, &ds->ds_prev);
475 			}
476 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
477 				int zaperr = zap_lookup(mos, ds->ds_object,
478 				    DS_FIELD_BOOKMARK_NAMES,
479 				    sizeof (ds->ds_bookmarks), 1,
480 				    &ds->ds_bookmarks);
481 				if (zaperr != ENOENT)
482 					VERIFY0(zaperr);
483 			}
484 		} else {
485 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
486 				err = dsl_dataset_get_snapname(ds);
487 			if (err == 0 &&
488 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
489 				err = zap_count(
490 				    ds->ds_dir->dd_pool->dp_meta_objset,
491 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
492 				    &ds->ds_userrefs);
493 			}
494 		}
495 
496 		if (err == 0 && !ds->ds_is_snapshot) {
497 			err = dsl_prop_get_int_ds(ds,
498 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
499 			    &ds->ds_reserved);
500 			if (err == 0) {
501 				err = dsl_prop_get_int_ds(ds,
502 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
503 				    &ds->ds_quota);
504 			}
505 		} else {
506 			ds->ds_reserved = ds->ds_quota = 0;
507 		}
508 
509 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
510 		if (err == 0)
511 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
512 
513 		if (err != 0 || winner != NULL) {
514 			bplist_destroy(&ds->ds_pending_deadlist);
515 			dsl_deadlist_close(&ds->ds_deadlist);
516 			if (ds->ds_prev)
517 				dsl_dataset_rele(ds->ds_prev, ds);
518 			dsl_dir_rele(ds->ds_dir, ds);
519 			mutex_destroy(&ds->ds_lock);
520 			mutex_destroy(&ds->ds_opening_lock);
521 			mutex_destroy(&ds->ds_sendstream_lock);
522 			refcount_destroy(&ds->ds_longholds);
523 			kmem_free(ds, sizeof (dsl_dataset_t));
524 			if (err != 0) {
525 				dmu_buf_rele(dbuf, tag);
526 				return (err);
527 			}
528 			ds = winner;
529 		} else {
530 			ds->ds_fsid_guid =
531 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
532 		}
533 	}
534 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
535 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
536 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
537 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
538 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
539 	*dsp = ds;
540 	return (0);
541 }
542 
543 int
544 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
545     void *tag, dsl_dataset_t **dsp)
546 {
547 	dsl_dir_t *dd;
548 	const char *snapname;
549 	uint64_t obj;
550 	int err = 0;
551 	dsl_dataset_t *ds;
552 
553 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
554 	if (err != 0)
555 		return (err);
556 
557 	ASSERT(dsl_pool_config_held(dp));
558 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
559 	if (obj != 0)
560 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
561 	else
562 		err = SET_ERROR(ENOENT);
563 
564 	/* we may be looking for a snapshot */
565 	if (err == 0 && snapname != NULL) {
566 		dsl_dataset_t *snap_ds;
567 
568 		if (*snapname++ != '@') {
569 			dsl_dataset_rele(ds, tag);
570 			dsl_dir_rele(dd, FTAG);
571 			return (SET_ERROR(ENOENT));
572 		}
573 
574 		dprintf("looking for snapshot '%s'\n", snapname);
575 		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
576 		if (err == 0)
577 			err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
578 		dsl_dataset_rele(ds, tag);
579 
580 		if (err == 0) {
581 			mutex_enter(&snap_ds->ds_lock);
582 			if (snap_ds->ds_snapname[0] == 0)
583 				(void) strlcpy(snap_ds->ds_snapname, snapname,
584 				    sizeof (snap_ds->ds_snapname));
585 			mutex_exit(&snap_ds->ds_lock);
586 			ds = snap_ds;
587 		}
588 	}
589 	if (err == 0)
590 		*dsp = ds;
591 	dsl_dir_rele(dd, FTAG);
592 	return (err);
593 }
594 
595 int
596 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
597     void *tag, dsl_dataset_t **dsp)
598 {
599 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
600 	if (err != 0)
601 		return (err);
602 	if (!dsl_dataset_tryown(*dsp, tag)) {
603 		dsl_dataset_rele(*dsp, tag);
604 		*dsp = NULL;
605 		return (SET_ERROR(EBUSY));
606 	}
607 	return (0);
608 }
609 
610 int
611 dsl_dataset_own(dsl_pool_t *dp, const char *name,
612     void *tag, dsl_dataset_t **dsp)
613 {
614 	int err = dsl_dataset_hold(dp, name, tag, dsp);
615 	if (err != 0)
616 		return (err);
617 	if (!dsl_dataset_tryown(*dsp, tag)) {
618 		dsl_dataset_rele(*dsp, tag);
619 		return (SET_ERROR(EBUSY));
620 	}
621 	return (0);
622 }
623 
624 /*
625  * See the comment above dsl_pool_hold() for details.  In summary, a long
626  * hold is used to prevent destruction of a dataset while the pool hold
627  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
628  *
629  * The dataset and pool must be held when this function is called.  After it
630  * is called, the pool hold may be released while the dataset is still held
631  * and accessed.
632  */
633 void
634 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
635 {
636 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
637 	(void) refcount_add(&ds->ds_longholds, tag);
638 }
639 
640 void
641 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
642 {
643 	(void) refcount_remove(&ds->ds_longholds, tag);
644 }
645 
646 /* Return B_TRUE if there are any long holds on this dataset. */
647 boolean_t
648 dsl_dataset_long_held(dsl_dataset_t *ds)
649 {
650 	return (!refcount_is_zero(&ds->ds_longholds));
651 }
652 
653 void
654 dsl_dataset_name(dsl_dataset_t *ds, char *name)
655 {
656 	if (ds == NULL) {
657 		(void) strcpy(name, "mos");
658 	} else {
659 		dsl_dir_name(ds->ds_dir, name);
660 		VERIFY0(dsl_dataset_get_snapname(ds));
661 		if (ds->ds_snapname[0]) {
662 			(void) strcat(name, "@");
663 			/*
664 			 * We use a "recursive" mutex so that we
665 			 * can call dprintf_ds() with ds_lock held.
666 			 */
667 			if (!MUTEX_HELD(&ds->ds_lock)) {
668 				mutex_enter(&ds->ds_lock);
669 				(void) strcat(name, ds->ds_snapname);
670 				mutex_exit(&ds->ds_lock);
671 			} else {
672 				(void) strcat(name, ds->ds_snapname);
673 			}
674 		}
675 	}
676 }
677 
678 void
679 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
680 {
681 	dmu_buf_rele(ds->ds_dbuf, tag);
682 }
683 
684 void
685 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
686 {
687 	ASSERT3P(ds->ds_owner, ==, tag);
688 	ASSERT(ds->ds_dbuf != NULL);
689 
690 	mutex_enter(&ds->ds_lock);
691 	ds->ds_owner = NULL;
692 	mutex_exit(&ds->ds_lock);
693 	dsl_dataset_long_rele(ds, tag);
694 	dsl_dataset_rele(ds, tag);
695 }
696 
697 boolean_t
698 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
699 {
700 	boolean_t gotit = FALSE;
701 
702 	mutex_enter(&ds->ds_lock);
703 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
704 		ds->ds_owner = tag;
705 		dsl_dataset_long_hold(ds, tag);
706 		gotit = TRUE;
707 	}
708 	mutex_exit(&ds->ds_lock);
709 	return (gotit);
710 }
711 
712 static void
713 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
714 {
715 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
716 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
717 	uint64_t zero = 0;
718 
719 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
720 
721 	spa_feature_incr(spa, f, tx);
722 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
723 
724 	VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
725 	    sizeof (zero), 1, &zero, tx));
726 }
727 
728 void
729 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
730 {
731 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
732 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
733 
734 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
735 
736 	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
737 	spa_feature_decr(spa, f, tx);
738 }
739 
740 uint64_t
741 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
742     uint64_t flags, dmu_tx_t *tx)
743 {
744 	dsl_pool_t *dp = dd->dd_pool;
745 	dmu_buf_t *dbuf;
746 	dsl_dataset_phys_t *dsphys;
747 	uint64_t dsobj;
748 	objset_t *mos = dp->dp_meta_objset;
749 
750 	if (origin == NULL)
751 		origin = dp->dp_origin_snap;
752 
753 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
754 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
755 	ASSERT(dmu_tx_is_syncing(tx));
756 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
757 
758 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
759 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
760 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
761 	dmu_buf_will_dirty(dbuf, tx);
762 	dsphys = dbuf->db_data;
763 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
764 	dsphys->ds_dir_obj = dd->dd_object;
765 	dsphys->ds_flags = flags;
766 	dsphys->ds_fsid_guid = unique_create();
767 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
768 	    sizeof (dsphys->ds_guid));
769 	dsphys->ds_snapnames_zapobj =
770 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
771 	    DMU_OT_NONE, 0, tx);
772 	dsphys->ds_creation_time = gethrestime_sec();
773 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
774 
775 	if (origin == NULL) {
776 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
777 	} else {
778 		dsl_dataset_t *ohds; /* head of the origin snapshot */
779 
780 		dsphys->ds_prev_snap_obj = origin->ds_object;
781 		dsphys->ds_prev_snap_txg =
782 		    dsl_dataset_phys(origin)->ds_creation_txg;
783 		dsphys->ds_referenced_bytes =
784 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
785 		dsphys->ds_compressed_bytes =
786 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
787 		dsphys->ds_uncompressed_bytes =
788 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
789 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
790 
791 		/*
792 		 * Inherit flags that describe the dataset's contents
793 		 * (INCONSISTENT) or properties (Case Insensitive).
794 		 */
795 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
796 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
797 
798 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
799 			if (origin->ds_feature_inuse[f])
800 				dsl_dataset_activate_feature(dsobj, f, tx);
801 		}
802 
803 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
804 		dsl_dataset_phys(origin)->ds_num_children++;
805 
806 		VERIFY0(dsl_dataset_hold_obj(dp,
807 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
808 		    FTAG, &ohds));
809 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
810 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
811 		dsl_dataset_rele(ohds, FTAG);
812 
813 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
814 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
815 				dsl_dataset_phys(origin)->ds_next_clones_obj =
816 				    zap_create(mos,
817 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
818 			}
819 			VERIFY0(zap_add_int(mos,
820 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
821 			    dsobj, tx));
822 		}
823 
824 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
825 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
826 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
827 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
828 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
829 				dsl_dir_phys(origin->ds_dir)->dd_clones =
830 				    zap_create(mos,
831 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
832 			}
833 			VERIFY0(zap_add_int(mos,
834 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
835 			    dsobj, tx));
836 		}
837 	}
838 
839 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
840 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
841 
842 	dmu_buf_rele(dbuf, FTAG);
843 
844 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
845 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
846 
847 	return (dsobj);
848 }
849 
850 static void
851 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
852 {
853 	objset_t *os;
854 
855 	VERIFY0(dmu_objset_from_ds(ds, &os));
856 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
857 	dsl_dataset_dirty(ds, tx);
858 }
859 
860 uint64_t
861 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
862     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
863 {
864 	dsl_pool_t *dp = pdd->dd_pool;
865 	uint64_t dsobj, ddobj;
866 	dsl_dir_t *dd;
867 
868 	ASSERT(dmu_tx_is_syncing(tx));
869 	ASSERT(lastname[0] != '@');
870 
871 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
872 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
873 
874 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
875 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
876 
877 	dsl_deleg_set_create_perms(dd, tx, cr);
878 
879 	/*
880 	 * Since we're creating a new node we know it's a leaf, so we can
881 	 * initialize the counts if the limit feature is active.
882 	 */
883 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
884 		uint64_t cnt = 0;
885 		objset_t *os = dd->dd_pool->dp_meta_objset;
886 
887 		dsl_dir_zapify(dd, tx);
888 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
889 		    sizeof (cnt), 1, &cnt, tx));
890 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
891 		    sizeof (cnt), 1, &cnt, tx));
892 	}
893 
894 	dsl_dir_rele(dd, FTAG);
895 
896 	/*
897 	 * If we are creating a clone, make sure we zero out any stale
898 	 * data from the origin snapshots zil header.
899 	 */
900 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
901 		dsl_dataset_t *ds;
902 
903 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
904 		dsl_dataset_zero_zil(ds, tx);
905 		dsl_dataset_rele(ds, FTAG);
906 	}
907 
908 	return (dsobj);
909 }
910 
911 /*
912  * The unique space in the head dataset can be calculated by subtracting
913  * the space used in the most recent snapshot, that is still being used
914  * in this file system, from the space currently in use.  To figure out
915  * the space in the most recent snapshot still in use, we need to take
916  * the total space used in the snapshot and subtract out the space that
917  * has been freed up since the snapshot was taken.
918  */
919 void
920 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
921 {
922 	uint64_t mrs_used;
923 	uint64_t dlused, dlcomp, dluncomp;
924 
925 	ASSERT(!ds->ds_is_snapshot);
926 
927 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
928 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
929 	else
930 		mrs_used = 0;
931 
932 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
933 
934 	ASSERT3U(dlused, <=, mrs_used);
935 	dsl_dataset_phys(ds)->ds_unique_bytes =
936 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
937 
938 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
939 	    SPA_VERSION_UNIQUE_ACCURATE)
940 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
941 }
942 
943 void
944 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
945     dmu_tx_t *tx)
946 {
947 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
948 	uint64_t count;
949 	int err;
950 
951 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
952 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
953 	    obj, tx);
954 	/*
955 	 * The err should not be ENOENT, but a bug in a previous version
956 	 * of the code could cause upgrade_clones_cb() to not set
957 	 * ds_next_snap_obj when it should, leading to a missing entry.
958 	 * If we knew that the pool was created after
959 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
960 	 * ENOENT.  However, at least we can check that we don't have
961 	 * too many entries in the next_clones_obj even after failing to
962 	 * remove this one.
963 	 */
964 	if (err != ENOENT)
965 		VERIFY0(err);
966 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
967 	    &count));
968 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
969 }
970 
971 
972 blkptr_t *
973 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
974 {
975 	return (&dsl_dataset_phys(ds)->ds_bp);
976 }
977 
978 void
979 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
980 {
981 	ASSERT(dmu_tx_is_syncing(tx));
982 	/* If it's the meta-objset, set dp_meta_rootbp */
983 	if (ds == NULL) {
984 		tx->tx_pool->dp_meta_rootbp = *bp;
985 	} else {
986 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
987 		dsl_dataset_phys(ds)->ds_bp = *bp;
988 	}
989 }
990 
991 spa_t *
992 dsl_dataset_get_spa(dsl_dataset_t *ds)
993 {
994 	return (ds->ds_dir->dd_pool->dp_spa);
995 }
996 
997 void
998 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
999 {
1000 	dsl_pool_t *dp;
1001 
1002 	if (ds == NULL) /* this is the meta-objset */
1003 		return;
1004 
1005 	ASSERT(ds->ds_objset != NULL);
1006 
1007 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1008 		panic("dirtying snapshot!");
1009 
1010 	dp = ds->ds_dir->dd_pool;
1011 
1012 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1013 		/* up the hold count until we can be written out */
1014 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1015 	}
1016 }
1017 
1018 boolean_t
1019 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1020 {
1021 	for (int t = 0; t < TXG_SIZE; t++) {
1022 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1023 		    ds, t))
1024 			return (B_TRUE);
1025 	}
1026 	return (B_FALSE);
1027 }
1028 
1029 static int
1030 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1031 {
1032 	uint64_t asize;
1033 
1034 	if (!dmu_tx_is_syncing(tx))
1035 		return (0);
1036 
1037 	/*
1038 	 * If there's an fs-only reservation, any blocks that might become
1039 	 * owned by the snapshot dataset must be accommodated by space
1040 	 * outside of the reservation.
1041 	 */
1042 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1043 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1044 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1045 		return (SET_ERROR(ENOSPC));
1046 
1047 	/*
1048 	 * Propagate any reserved space for this snapshot to other
1049 	 * snapshot checks in this sync group.
1050 	 */
1051 	if (asize > 0)
1052 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1053 
1054 	return (0);
1055 }
1056 
1057 typedef struct dsl_dataset_snapshot_arg {
1058 	nvlist_t *ddsa_snaps;
1059 	nvlist_t *ddsa_props;
1060 	nvlist_t *ddsa_errors;
1061 	cred_t *ddsa_cr;
1062 } dsl_dataset_snapshot_arg_t;
1063 
1064 int
1065 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1066     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1067 {
1068 	int error;
1069 	uint64_t value;
1070 
1071 	ds->ds_trysnap_txg = tx->tx_txg;
1072 
1073 	if (!dmu_tx_is_syncing(tx))
1074 		return (0);
1075 
1076 	/*
1077 	 * We don't allow multiple snapshots of the same txg.  If there
1078 	 * is already one, try again.
1079 	 */
1080 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1081 		return (SET_ERROR(EAGAIN));
1082 
1083 	/*
1084 	 * Check for conflicting snapshot name.
1085 	 */
1086 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1087 	if (error == 0)
1088 		return (SET_ERROR(EEXIST));
1089 	if (error != ENOENT)
1090 		return (error);
1091 
1092 	/*
1093 	 * We don't allow taking snapshots of inconsistent datasets, such as
1094 	 * those into which we are currently receiving.  However, if we are
1095 	 * creating this snapshot as part of a receive, this check will be
1096 	 * executed atomically with respect to the completion of the receive
1097 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1098 	 * case we ignore this, knowing it will be fixed up for us shortly in
1099 	 * dmu_recv_end_sync().
1100 	 */
1101 	if (!recv && DS_IS_INCONSISTENT(ds))
1102 		return (SET_ERROR(EBUSY));
1103 
1104 	/*
1105 	 * Skip the check for temporary snapshots or if we have already checked
1106 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1107 	 * check the count here when we're receiving a stream.
1108 	 */
1109 	if (cnt != 0 && cr != NULL) {
1110 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1111 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1112 		if (error != 0)
1113 			return (error);
1114 	}
1115 
1116 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1117 	if (error != 0)
1118 		return (error);
1119 
1120 	return (0);
1121 }
1122 
1123 static int
1124 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1125 {
1126 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1127 	dsl_pool_t *dp = dmu_tx_pool(tx);
1128 	nvpair_t *pair;
1129 	int rv = 0;
1130 
1131 	/*
1132 	 * Pre-compute how many total new snapshots will be created for each
1133 	 * level in the tree and below. This is needed for validating the
1134 	 * snapshot limit when either taking a recursive snapshot or when
1135 	 * taking multiple snapshots.
1136 	 *
1137 	 * The problem is that the counts are not actually adjusted when
1138 	 * we are checking, only when we finally sync. For a single snapshot,
1139 	 * this is easy, the count will increase by 1 at each node up the tree,
1140 	 * but its more complicated for the recursive/multiple snapshot case.
1141 	 *
1142 	 * The dsl_fs_ss_limit_check function does recursively check the count
1143 	 * at each level up the tree but since it is validating each snapshot
1144 	 * independently we need to be sure that we are validating the complete
1145 	 * count for the entire set of snapshots. We do this by rolling up the
1146 	 * counts for each component of the name into an nvlist and then
1147 	 * checking each of those cases with the aggregated count.
1148 	 *
1149 	 * This approach properly handles not only the recursive snapshot
1150 	 * case (where we get all of those on the ddsa_snaps list) but also
1151 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1152 	 * validate the limit on 'a' using a count of 2).
1153 	 *
1154 	 * We validate the snapshot names in the third loop and only report
1155 	 * name errors once.
1156 	 */
1157 	if (dmu_tx_is_syncing(tx)) {
1158 		nvlist_t *cnt_track = NULL;
1159 		cnt_track = fnvlist_alloc();
1160 
1161 		/* Rollup aggregated counts into the cnt_track list */
1162 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1163 		    pair != NULL;
1164 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1165 			char *pdelim;
1166 			uint64_t val;
1167 			char nm[MAXPATHLEN];
1168 
1169 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1170 			pdelim = strchr(nm, '@');
1171 			if (pdelim == NULL)
1172 				continue;
1173 			*pdelim = '\0';
1174 
1175 			do {
1176 				if (nvlist_lookup_uint64(cnt_track, nm,
1177 				    &val) == 0) {
1178 					/* update existing entry */
1179 					fnvlist_add_uint64(cnt_track, nm,
1180 					    val + 1);
1181 				} else {
1182 					/* add to list */
1183 					fnvlist_add_uint64(cnt_track, nm, 1);
1184 				}
1185 
1186 				pdelim = strrchr(nm, '/');
1187 				if (pdelim != NULL)
1188 					*pdelim = '\0';
1189 			} while (pdelim != NULL);
1190 		}
1191 
1192 		/* Check aggregated counts at each level */
1193 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1194 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1195 			int error = 0;
1196 			char *name;
1197 			uint64_t cnt = 0;
1198 			dsl_dataset_t *ds;
1199 
1200 			name = nvpair_name(pair);
1201 			cnt = fnvpair_value_uint64(pair);
1202 			ASSERT(cnt > 0);
1203 
1204 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1205 			if (error == 0) {
1206 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1207 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1208 				    ddsa->ddsa_cr);
1209 				dsl_dataset_rele(ds, FTAG);
1210 			}
1211 
1212 			if (error != 0) {
1213 				if (ddsa->ddsa_errors != NULL)
1214 					fnvlist_add_int32(ddsa->ddsa_errors,
1215 					    name, error);
1216 				rv = error;
1217 				/* only report one error for this check */
1218 				break;
1219 			}
1220 		}
1221 		nvlist_free(cnt_track);
1222 	}
1223 
1224 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1225 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1226 		int error = 0;
1227 		dsl_dataset_t *ds;
1228 		char *name, *atp;
1229 		char dsname[MAXNAMELEN];
1230 
1231 		name = nvpair_name(pair);
1232 		if (strlen(name) >= MAXNAMELEN)
1233 			error = SET_ERROR(ENAMETOOLONG);
1234 		if (error == 0) {
1235 			atp = strchr(name, '@');
1236 			if (atp == NULL)
1237 				error = SET_ERROR(EINVAL);
1238 			if (error == 0)
1239 				(void) strlcpy(dsname, name, atp - name + 1);
1240 		}
1241 		if (error == 0)
1242 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1243 		if (error == 0) {
1244 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1245 			error = dsl_dataset_snapshot_check_impl(ds,
1246 			    atp + 1, tx, B_FALSE, 0, NULL);
1247 			dsl_dataset_rele(ds, FTAG);
1248 		}
1249 
1250 		if (error != 0) {
1251 			if (ddsa->ddsa_errors != NULL) {
1252 				fnvlist_add_int32(ddsa->ddsa_errors,
1253 				    name, error);
1254 			}
1255 			rv = error;
1256 		}
1257 	}
1258 
1259 	return (rv);
1260 }
1261 
1262 void
1263 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1264     dmu_tx_t *tx)
1265 {
1266 	static zil_header_t zero_zil;
1267 
1268 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1269 	dmu_buf_t *dbuf;
1270 	dsl_dataset_phys_t *dsphys;
1271 	uint64_t dsobj, crtxg;
1272 	objset_t *mos = dp->dp_meta_objset;
1273 	objset_t *os;
1274 
1275 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1276 
1277 	/*
1278 	 * If we are on an old pool, the zil must not be active, in which
1279 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1280 	 */
1281 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1282 	    dmu_objset_from_ds(ds, &os) != 0 ||
1283 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1284 	    sizeof (zero_zil)) == 0);
1285 
1286 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1287 
1288 	/*
1289 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1290 	 */
1291 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1292 		crtxg = 1;
1293 	else
1294 		crtxg = tx->tx_txg;
1295 
1296 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1297 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1298 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1299 	dmu_buf_will_dirty(dbuf, tx);
1300 	dsphys = dbuf->db_data;
1301 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1302 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1303 	dsphys->ds_fsid_guid = unique_create();
1304 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1305 	    sizeof (dsphys->ds_guid));
1306 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1307 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1308 	dsphys->ds_next_snap_obj = ds->ds_object;
1309 	dsphys->ds_num_children = 1;
1310 	dsphys->ds_creation_time = gethrestime_sec();
1311 	dsphys->ds_creation_txg = crtxg;
1312 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1313 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1314 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1315 	dsphys->ds_uncompressed_bytes =
1316 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1317 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1318 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1319 	dmu_buf_rele(dbuf, FTAG);
1320 
1321 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1322 		if (ds->ds_feature_inuse[f])
1323 			dsl_dataset_activate_feature(dsobj, f, tx);
1324 	}
1325 
1326 	ASSERT3U(ds->ds_prev != 0, ==,
1327 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1328 	if (ds->ds_prev) {
1329 		uint64_t next_clones_obj =
1330 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1331 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1332 		    ds->ds_object ||
1333 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1334 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1335 		    ds->ds_object) {
1336 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1337 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1338 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1339 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1340 		} else if (next_clones_obj != 0) {
1341 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1342 			    dsphys->ds_next_snap_obj, tx);
1343 			VERIFY0(zap_add_int(mos,
1344 			    next_clones_obj, dsobj, tx));
1345 		}
1346 	}
1347 
1348 	/*
1349 	 * If we have a reference-reservation on this dataset, we will
1350 	 * need to increase the amount of refreservation being charged
1351 	 * since our unique space is going to zero.
1352 	 */
1353 	if (ds->ds_reserved) {
1354 		int64_t delta;
1355 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1356 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1357 		    ds->ds_reserved);
1358 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1359 		    delta, 0, 0, tx);
1360 	}
1361 
1362 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1363 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1364 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1365 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1366 	dsl_deadlist_close(&ds->ds_deadlist);
1367 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1368 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1369 	dsl_deadlist_add_key(&ds->ds_deadlist,
1370 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1371 
1372 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1373 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1374 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1375 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1376 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1377 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1378 
1379 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1380 	    snapname, 8, 1, &dsobj, tx));
1381 
1382 	if (ds->ds_prev)
1383 		dsl_dataset_rele(ds->ds_prev, ds);
1384 	VERIFY0(dsl_dataset_hold_obj(dp,
1385 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1386 
1387 	dsl_scan_ds_snapshotted(ds, tx);
1388 
1389 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1390 
1391 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1392 }
1393 
1394 static void
1395 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1396 {
1397 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1398 	dsl_pool_t *dp = dmu_tx_pool(tx);
1399 	nvpair_t *pair;
1400 
1401 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1402 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1403 		dsl_dataset_t *ds;
1404 		char *name, *atp;
1405 		char dsname[MAXNAMELEN];
1406 
1407 		name = nvpair_name(pair);
1408 		atp = strchr(name, '@');
1409 		(void) strlcpy(dsname, name, atp - name + 1);
1410 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1411 
1412 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1413 		if (ddsa->ddsa_props != NULL) {
1414 			dsl_props_set_sync_impl(ds->ds_prev,
1415 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1416 		}
1417 		dsl_dataset_rele(ds, FTAG);
1418 	}
1419 }
1420 
1421 /*
1422  * The snapshots must all be in the same pool.
1423  * All-or-nothing: if there are any failures, nothing will be modified.
1424  */
1425 int
1426 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1427 {
1428 	dsl_dataset_snapshot_arg_t ddsa;
1429 	nvpair_t *pair;
1430 	boolean_t needsuspend;
1431 	int error;
1432 	spa_t *spa;
1433 	char *firstname;
1434 	nvlist_t *suspended = NULL;
1435 
1436 	pair = nvlist_next_nvpair(snaps, NULL);
1437 	if (pair == NULL)
1438 		return (0);
1439 	firstname = nvpair_name(pair);
1440 
1441 	error = spa_open(firstname, &spa, FTAG);
1442 	if (error != 0)
1443 		return (error);
1444 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1445 	spa_close(spa, FTAG);
1446 
1447 	if (needsuspend) {
1448 		suspended = fnvlist_alloc();
1449 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1450 		    pair = nvlist_next_nvpair(snaps, pair)) {
1451 			char fsname[MAXNAMELEN];
1452 			char *snapname = nvpair_name(pair);
1453 			char *atp;
1454 			void *cookie;
1455 
1456 			atp = strchr(snapname, '@');
1457 			if (atp == NULL) {
1458 				error = SET_ERROR(EINVAL);
1459 				break;
1460 			}
1461 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1462 
1463 			error = zil_suspend(fsname, &cookie);
1464 			if (error != 0)
1465 				break;
1466 			fnvlist_add_uint64(suspended, fsname,
1467 			    (uintptr_t)cookie);
1468 		}
1469 	}
1470 
1471 	ddsa.ddsa_snaps = snaps;
1472 	ddsa.ddsa_props = props;
1473 	ddsa.ddsa_errors = errors;
1474 	ddsa.ddsa_cr = CRED();
1475 
1476 	if (error == 0) {
1477 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1478 		    dsl_dataset_snapshot_sync, &ddsa,
1479 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1480 	}
1481 
1482 	if (suspended != NULL) {
1483 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1484 		    pair = nvlist_next_nvpair(suspended, pair)) {
1485 			zil_resume((void *)(uintptr_t)
1486 			    fnvpair_value_uint64(pair));
1487 		}
1488 		fnvlist_free(suspended);
1489 	}
1490 
1491 	return (error);
1492 }
1493 
1494 typedef struct dsl_dataset_snapshot_tmp_arg {
1495 	const char *ddsta_fsname;
1496 	const char *ddsta_snapname;
1497 	minor_t ddsta_cleanup_minor;
1498 	const char *ddsta_htag;
1499 } dsl_dataset_snapshot_tmp_arg_t;
1500 
1501 static int
1502 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1503 {
1504 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1505 	dsl_pool_t *dp = dmu_tx_pool(tx);
1506 	dsl_dataset_t *ds;
1507 	int error;
1508 
1509 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1510 	if (error != 0)
1511 		return (error);
1512 
1513 	/* NULL cred means no limit check for tmp snapshot */
1514 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1515 	    tx, B_FALSE, 0, NULL);
1516 	if (error != 0) {
1517 		dsl_dataset_rele(ds, FTAG);
1518 		return (error);
1519 	}
1520 
1521 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1522 		dsl_dataset_rele(ds, FTAG);
1523 		return (SET_ERROR(ENOTSUP));
1524 	}
1525 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1526 	    B_TRUE, tx);
1527 	if (error != 0) {
1528 		dsl_dataset_rele(ds, FTAG);
1529 		return (error);
1530 	}
1531 
1532 	dsl_dataset_rele(ds, FTAG);
1533 	return (0);
1534 }
1535 
1536 static void
1537 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1538 {
1539 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1540 	dsl_pool_t *dp = dmu_tx_pool(tx);
1541 	dsl_dataset_t *ds;
1542 
1543 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1544 
1545 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1546 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1547 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1548 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1549 
1550 	dsl_dataset_rele(ds, FTAG);
1551 }
1552 
1553 int
1554 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1555     minor_t cleanup_minor, const char *htag)
1556 {
1557 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1558 	int error;
1559 	spa_t *spa;
1560 	boolean_t needsuspend;
1561 	void *cookie;
1562 
1563 	ddsta.ddsta_fsname = fsname;
1564 	ddsta.ddsta_snapname = snapname;
1565 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1566 	ddsta.ddsta_htag = htag;
1567 
1568 	error = spa_open(fsname, &spa, FTAG);
1569 	if (error != 0)
1570 		return (error);
1571 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1572 	spa_close(spa, FTAG);
1573 
1574 	if (needsuspend) {
1575 		error = zil_suspend(fsname, &cookie);
1576 		if (error != 0)
1577 			return (error);
1578 	}
1579 
1580 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1581 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1582 
1583 	if (needsuspend)
1584 		zil_resume(cookie);
1585 	return (error);
1586 }
1587 
1588 
1589 void
1590 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1591 {
1592 	ASSERT(dmu_tx_is_syncing(tx));
1593 	ASSERT(ds->ds_objset != NULL);
1594 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1595 
1596 	/*
1597 	 * in case we had to change ds_fsid_guid when we opened it,
1598 	 * sync it out now.
1599 	 */
1600 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1601 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1602 
1603 	dmu_objset_sync(ds->ds_objset, zio, tx);
1604 
1605 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1606 		if (ds->ds_feature_activation_needed[f]) {
1607 			if (ds->ds_feature_inuse[f])
1608 				continue;
1609 			dsl_dataset_activate_feature(ds->ds_object, f, tx);
1610 			ds->ds_feature_inuse[f] = B_TRUE;
1611 		}
1612 	}
1613 }
1614 
1615 static void
1616 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1617 {
1618 	uint64_t count = 0;
1619 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1620 	zap_cursor_t zc;
1621 	zap_attribute_t za;
1622 	nvlist_t *propval = fnvlist_alloc();
1623 	nvlist_t *val = fnvlist_alloc();
1624 
1625 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1626 
1627 	/*
1628 	 * There may be missing entries in ds_next_clones_obj
1629 	 * due to a bug in a previous version of the code.
1630 	 * Only trust it if it has the right number of entries.
1631 	 */
1632 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1633 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1634 		    &count));
1635 	}
1636 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1637 		goto fail;
1638 	for (zap_cursor_init(&zc, mos,
1639 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1640 	    zap_cursor_retrieve(&zc, &za) == 0;
1641 	    zap_cursor_advance(&zc)) {
1642 		dsl_dataset_t *clone;
1643 		char buf[ZFS_MAXNAMELEN];
1644 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1645 		    za.za_first_integer, FTAG, &clone));
1646 		dsl_dir_name(clone->ds_dir, buf);
1647 		fnvlist_add_boolean(val, buf);
1648 		dsl_dataset_rele(clone, FTAG);
1649 	}
1650 	zap_cursor_fini(&zc);
1651 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1652 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1653 fail:
1654 	nvlist_free(val);
1655 	nvlist_free(propval);
1656 }
1657 
1658 void
1659 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1660 {
1661 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1662 	uint64_t refd, avail, uobjs, aobjs, ratio;
1663 
1664 	ASSERT(dsl_pool_config_held(dp));
1665 
1666 	ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1667 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1668 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1669 
1670 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1671 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1672 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1673 
1674 	if (ds->ds_is_snapshot) {
1675 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1676 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1677 		    dsl_dataset_phys(ds)->ds_unique_bytes);
1678 		get_clones_stat(ds, nv);
1679 	} else {
1680 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1681 			char buf[MAXNAMELEN];
1682 			dsl_dataset_name(ds->ds_prev, buf);
1683 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1684 		}
1685 
1686 		dsl_dir_stats(ds->ds_dir, nv);
1687 	}
1688 
1689 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1690 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1691 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1692 
1693 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1694 	    dsl_dataset_phys(ds)->ds_creation_time);
1695 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1696 	    dsl_dataset_phys(ds)->ds_creation_txg);
1697 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1698 	    ds->ds_quota);
1699 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1700 	    ds->ds_reserved);
1701 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1702 	    dsl_dataset_phys(ds)->ds_guid);
1703 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1704 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1705 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1706 	    ds->ds_object);
1707 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1708 	    ds->ds_userrefs);
1709 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1710 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1711 
1712 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1713 		uint64_t written, comp, uncomp;
1714 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1715 		dsl_dataset_t *prev;
1716 
1717 		int err = dsl_dataset_hold_obj(dp,
1718 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1719 		if (err == 0) {
1720 			err = dsl_dataset_space_written(prev, ds, &written,
1721 			    &comp, &uncomp);
1722 			dsl_dataset_rele(prev, FTAG);
1723 			if (err == 0) {
1724 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1725 				    written);
1726 			}
1727 		}
1728 	}
1729 }
1730 
1731 void
1732 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1733 {
1734 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1735 	ASSERT(dsl_pool_config_held(dp));
1736 
1737 	stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1738 	stat->dds_inconsistent =
1739 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1740 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1741 	stat->dds_origin[0] = '\0';
1742 	if (ds->ds_is_snapshot) {
1743 		stat->dds_is_snapshot = B_TRUE;
1744 		stat->dds_num_clones =
1745 		    dsl_dataset_phys(ds)->ds_num_children - 1;
1746 	} else {
1747 		stat->dds_is_snapshot = B_FALSE;
1748 		stat->dds_num_clones = 0;
1749 
1750 		if (dsl_dir_is_clone(ds->ds_dir)) {
1751 			dsl_dataset_t *ods;
1752 
1753 			VERIFY0(dsl_dataset_hold_obj(dp,
1754 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1755 			    FTAG, &ods));
1756 			dsl_dataset_name(ods, stat->dds_origin);
1757 			dsl_dataset_rele(ods, FTAG);
1758 		}
1759 	}
1760 }
1761 
1762 uint64_t
1763 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1764 {
1765 	return (ds->ds_fsid_guid);
1766 }
1767 
1768 void
1769 dsl_dataset_space(dsl_dataset_t *ds,
1770     uint64_t *refdbytesp, uint64_t *availbytesp,
1771     uint64_t *usedobjsp, uint64_t *availobjsp)
1772 {
1773 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1774 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1775 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1776 		*availbytesp +=
1777 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1778 	if (ds->ds_quota != 0) {
1779 		/*
1780 		 * Adjust available bytes according to refquota
1781 		 */
1782 		if (*refdbytesp < ds->ds_quota)
1783 			*availbytesp = MIN(*availbytesp,
1784 			    ds->ds_quota - *refdbytesp);
1785 		else
1786 			*availbytesp = 0;
1787 	}
1788 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1789 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1790 }
1791 
1792 boolean_t
1793 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1794 {
1795 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1796 
1797 	ASSERT(dsl_pool_config_held(dp));
1798 	if (snap == NULL)
1799 		return (B_FALSE);
1800 	if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1801 	    dsl_dataset_phys(snap)->ds_creation_txg) {
1802 		objset_t *os, *os_snap;
1803 		/*
1804 		 * It may be that only the ZIL differs, because it was
1805 		 * reset in the head.  Don't count that as being
1806 		 * modified.
1807 		 */
1808 		if (dmu_objset_from_ds(ds, &os) != 0)
1809 			return (B_TRUE);
1810 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1811 			return (B_TRUE);
1812 		return (bcmp(&os->os_phys->os_meta_dnode,
1813 		    &os_snap->os_phys->os_meta_dnode,
1814 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1815 	}
1816 	return (B_FALSE);
1817 }
1818 
1819 typedef struct dsl_dataset_rename_snapshot_arg {
1820 	const char *ddrsa_fsname;
1821 	const char *ddrsa_oldsnapname;
1822 	const char *ddrsa_newsnapname;
1823 	boolean_t ddrsa_recursive;
1824 	dmu_tx_t *ddrsa_tx;
1825 } dsl_dataset_rename_snapshot_arg_t;
1826 
1827 /* ARGSUSED */
1828 static int
1829 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1830     dsl_dataset_t *hds, void *arg)
1831 {
1832 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1833 	int error;
1834 	uint64_t val;
1835 
1836 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1837 	if (error != 0) {
1838 		/* ignore nonexistent snapshots */
1839 		return (error == ENOENT ? 0 : error);
1840 	}
1841 
1842 	/* new name should not exist */
1843 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1844 	if (error == 0)
1845 		error = SET_ERROR(EEXIST);
1846 	else if (error == ENOENT)
1847 		error = 0;
1848 
1849 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1850 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1851 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1852 		error = SET_ERROR(ENAMETOOLONG);
1853 
1854 	return (error);
1855 }
1856 
1857 static int
1858 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1859 {
1860 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1861 	dsl_pool_t *dp = dmu_tx_pool(tx);
1862 	dsl_dataset_t *hds;
1863 	int error;
1864 
1865 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1866 	if (error != 0)
1867 		return (error);
1868 
1869 	if (ddrsa->ddrsa_recursive) {
1870 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1871 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1872 		    DS_FIND_CHILDREN);
1873 	} else {
1874 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1875 	}
1876 	dsl_dataset_rele(hds, FTAG);
1877 	return (error);
1878 }
1879 
1880 static int
1881 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1882     dsl_dataset_t *hds, void *arg)
1883 {
1884 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1885 	dsl_dataset_t *ds;
1886 	uint64_t val;
1887 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1888 	int error;
1889 
1890 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1891 	ASSERT(error == 0 || error == ENOENT);
1892 	if (error == ENOENT) {
1893 		/* ignore nonexistent snapshots */
1894 		return (0);
1895 	}
1896 
1897 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1898 
1899 	/* log before we change the name */
1900 	spa_history_log_internal_ds(ds, "rename", tx,
1901 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1902 
1903 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1904 	    B_FALSE));
1905 	mutex_enter(&ds->ds_lock);
1906 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1907 	mutex_exit(&ds->ds_lock);
1908 	VERIFY0(zap_add(dp->dp_meta_objset,
1909 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
1910 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1911 
1912 	dsl_dataset_rele(ds, FTAG);
1913 	return (0);
1914 }
1915 
1916 static void
1917 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1918 {
1919 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1920 	dsl_pool_t *dp = dmu_tx_pool(tx);
1921 	dsl_dataset_t *hds;
1922 
1923 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1924 	ddrsa->ddrsa_tx = tx;
1925 	if (ddrsa->ddrsa_recursive) {
1926 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1927 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1928 		    DS_FIND_CHILDREN));
1929 	} else {
1930 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1931 	}
1932 	dsl_dataset_rele(hds, FTAG);
1933 }
1934 
1935 int
1936 dsl_dataset_rename_snapshot(const char *fsname,
1937     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1938 {
1939 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1940 
1941 	ddrsa.ddrsa_fsname = fsname;
1942 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1943 	ddrsa.ddrsa_newsnapname = newsnapname;
1944 	ddrsa.ddrsa_recursive = recursive;
1945 
1946 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1947 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
1948 	    1, ZFS_SPACE_CHECK_RESERVED));
1949 }
1950 
1951 /*
1952  * If we're doing an ownership handoff, we need to make sure that there is
1953  * only one long hold on the dataset.  We're not allowed to change anything here
1954  * so we don't permanently release the long hold or regular hold here.  We want
1955  * to do this only when syncing to avoid the dataset unexpectedly going away
1956  * when we release the long hold.
1957  */
1958 static int
1959 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1960 {
1961 	boolean_t held;
1962 
1963 	if (!dmu_tx_is_syncing(tx))
1964 		return (0);
1965 
1966 	if (owner != NULL) {
1967 		VERIFY3P(ds->ds_owner, ==, owner);
1968 		dsl_dataset_long_rele(ds, owner);
1969 	}
1970 
1971 	held = dsl_dataset_long_held(ds);
1972 
1973 	if (owner != NULL)
1974 		dsl_dataset_long_hold(ds, owner);
1975 
1976 	if (held)
1977 		return (SET_ERROR(EBUSY));
1978 
1979 	return (0);
1980 }
1981 
1982 typedef struct dsl_dataset_rollback_arg {
1983 	const char *ddra_fsname;
1984 	void *ddra_owner;
1985 	nvlist_t *ddra_result;
1986 } dsl_dataset_rollback_arg_t;
1987 
1988 static int
1989 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1990 {
1991 	dsl_dataset_rollback_arg_t *ddra = arg;
1992 	dsl_pool_t *dp = dmu_tx_pool(tx);
1993 	dsl_dataset_t *ds;
1994 	int64_t unused_refres_delta;
1995 	int error;
1996 
1997 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1998 	if (error != 0)
1999 		return (error);
2000 
2001 	/* must not be a snapshot */
2002 	if (ds->ds_is_snapshot) {
2003 		dsl_dataset_rele(ds, FTAG);
2004 		return (SET_ERROR(EINVAL));
2005 	}
2006 
2007 	/* must have a most recent snapshot */
2008 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2009 		dsl_dataset_rele(ds, FTAG);
2010 		return (SET_ERROR(EINVAL));
2011 	}
2012 
2013 	/* must not have any bookmarks after the most recent snapshot */
2014 	nvlist_t *proprequest = fnvlist_alloc();
2015 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2016 	nvlist_t *bookmarks = fnvlist_alloc();
2017 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2018 	fnvlist_free(proprequest);
2019 	if (error != 0)
2020 		return (error);
2021 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2022 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2023 		nvlist_t *valuenv =
2024 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2025 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
2026 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2027 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2028 			fnvlist_free(bookmarks);
2029 			dsl_dataset_rele(ds, FTAG);
2030 			return (SET_ERROR(EEXIST));
2031 		}
2032 	}
2033 	fnvlist_free(bookmarks);
2034 
2035 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2036 	if (error != 0) {
2037 		dsl_dataset_rele(ds, FTAG);
2038 		return (error);
2039 	}
2040 
2041 	/*
2042 	 * Check if the snap we are rolling back to uses more than
2043 	 * the refquota.
2044 	 */
2045 	if (ds->ds_quota != 0 &&
2046 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2047 		dsl_dataset_rele(ds, FTAG);
2048 		return (SET_ERROR(EDQUOT));
2049 	}
2050 
2051 	/*
2052 	 * When we do the clone swap, we will temporarily use more space
2053 	 * due to the refreservation (the head will no longer have any
2054 	 * unique space, so the entire amount of the refreservation will need
2055 	 * to be free).  We will immediately destroy the clone, freeing
2056 	 * this space, but the freeing happens over many txg's.
2057 	 */
2058 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2059 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2060 
2061 	if (unused_refres_delta > 0 &&
2062 	    unused_refres_delta >
2063 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2064 		dsl_dataset_rele(ds, FTAG);
2065 		return (SET_ERROR(ENOSPC));
2066 	}
2067 
2068 	dsl_dataset_rele(ds, FTAG);
2069 	return (0);
2070 }
2071 
2072 static void
2073 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2074 {
2075 	dsl_dataset_rollback_arg_t *ddra = arg;
2076 	dsl_pool_t *dp = dmu_tx_pool(tx);
2077 	dsl_dataset_t *ds, *clone;
2078 	uint64_t cloneobj;
2079 	char namebuf[ZFS_MAXNAMELEN];
2080 
2081 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2082 
2083 	dsl_dataset_name(ds->ds_prev, namebuf);
2084 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2085 
2086 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2087 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2088 
2089 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2090 
2091 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2092 	dsl_dataset_zero_zil(ds, tx);
2093 
2094 	dsl_destroy_head_sync_impl(clone, tx);
2095 
2096 	dsl_dataset_rele(clone, FTAG);
2097 	dsl_dataset_rele(ds, FTAG);
2098 }
2099 
2100 /*
2101  * Rolls back the given filesystem or volume to the most recent snapshot.
2102  * The name of the most recent snapshot will be returned under key "target"
2103  * in the result nvlist.
2104  *
2105  * If owner != NULL:
2106  * - The existing dataset MUST be owned by the specified owner at entry
2107  * - Upon return, dataset will still be held by the same owner, whether we
2108  *   succeed or not.
2109  *
2110  * This mode is required any time the existing filesystem is mounted.  See
2111  * notes above zfs_suspend_fs() for further details.
2112  */
2113 int
2114 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2115 {
2116 	dsl_dataset_rollback_arg_t ddra;
2117 
2118 	ddra.ddra_fsname = fsname;
2119 	ddra.ddra_owner = owner;
2120 	ddra.ddra_result = result;
2121 
2122 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2123 	    dsl_dataset_rollback_sync, &ddra,
2124 	    1, ZFS_SPACE_CHECK_RESERVED));
2125 }
2126 
2127 struct promotenode {
2128 	list_node_t link;
2129 	dsl_dataset_t *ds;
2130 };
2131 
2132 typedef struct dsl_dataset_promote_arg {
2133 	const char *ddpa_clonename;
2134 	dsl_dataset_t *ddpa_clone;
2135 	list_t shared_snaps, origin_snaps, clone_snaps;
2136 	dsl_dataset_t *origin_origin; /* origin of the origin */
2137 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2138 	char *err_ds;
2139 	cred_t *cr;
2140 } dsl_dataset_promote_arg_t;
2141 
2142 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2143 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2144     void *tag);
2145 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2146 
2147 static int
2148 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2149 {
2150 	dsl_dataset_promote_arg_t *ddpa = arg;
2151 	dsl_pool_t *dp = dmu_tx_pool(tx);
2152 	dsl_dataset_t *hds;
2153 	struct promotenode *snap;
2154 	dsl_dataset_t *origin_ds;
2155 	int err;
2156 	uint64_t unused;
2157 	uint64_t ss_mv_cnt;
2158 	size_t max_snap_len;
2159 
2160 	err = promote_hold(ddpa, dp, FTAG);
2161 	if (err != 0)
2162 		return (err);
2163 
2164 	hds = ddpa->ddpa_clone;
2165 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2166 
2167 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2168 		promote_rele(ddpa, FTAG);
2169 		return (SET_ERROR(EXDEV));
2170 	}
2171 
2172 	/*
2173 	 * Compute and check the amount of space to transfer.  Since this is
2174 	 * so expensive, don't do the preliminary check.
2175 	 */
2176 	if (!dmu_tx_is_syncing(tx)) {
2177 		promote_rele(ddpa, FTAG);
2178 		return (0);
2179 	}
2180 
2181 	snap = list_head(&ddpa->shared_snaps);
2182 	origin_ds = snap->ds;
2183 
2184 	/* compute origin's new unique space */
2185 	snap = list_tail(&ddpa->clone_snaps);
2186 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2187 	    origin_ds->ds_object);
2188 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2189 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2190 	    &ddpa->unique, &unused, &unused);
2191 
2192 	/*
2193 	 * Walk the snapshots that we are moving
2194 	 *
2195 	 * Compute space to transfer.  Consider the incremental changes
2196 	 * to used by each snapshot:
2197 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2198 	 * So each snapshot gave birth to:
2199 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2200 	 * So a sequence would look like:
2201 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2202 	 * Which simplifies to:
2203 	 * uN + kN + kN-1 + ... + k1 + k0
2204 	 * Note however, if we stop before we reach the ORIGIN we get:
2205 	 * uN + kN + kN-1 + ... + kM - uM-1
2206 	 */
2207 	ss_mv_cnt = 0;
2208 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2209 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2210 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2211 	for (snap = list_head(&ddpa->shared_snaps); snap;
2212 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2213 		uint64_t val, dlused, dlcomp, dluncomp;
2214 		dsl_dataset_t *ds = snap->ds;
2215 
2216 		ss_mv_cnt++;
2217 
2218 		/*
2219 		 * If there are long holds, we won't be able to evict
2220 		 * the objset.
2221 		 */
2222 		if (dsl_dataset_long_held(ds)) {
2223 			err = SET_ERROR(EBUSY);
2224 			goto out;
2225 		}
2226 
2227 		/* Check that the snapshot name does not conflict */
2228 		VERIFY0(dsl_dataset_get_snapname(ds));
2229 		if (strlen(ds->ds_snapname) >= max_snap_len) {
2230 			err = SET_ERROR(ENAMETOOLONG);
2231 			goto out;
2232 		}
2233 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2234 		if (err == 0) {
2235 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2236 			err = SET_ERROR(EEXIST);
2237 			goto out;
2238 		}
2239 		if (err != ENOENT)
2240 			goto out;
2241 
2242 		/* The very first snapshot does not have a deadlist */
2243 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2244 			continue;
2245 
2246 		dsl_deadlist_space(&ds->ds_deadlist,
2247 		    &dlused, &dlcomp, &dluncomp);
2248 		ddpa->used += dlused;
2249 		ddpa->comp += dlcomp;
2250 		ddpa->uncomp += dluncomp;
2251 	}
2252 
2253 	/*
2254 	 * If we are a clone of a clone then we never reached ORIGIN,
2255 	 * so we need to subtract out the clone origin's used space.
2256 	 */
2257 	if (ddpa->origin_origin) {
2258 		ddpa->used -=
2259 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2260 		ddpa->comp -=
2261 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2262 		ddpa->uncomp -=
2263 		    dsl_dataset_phys(ddpa->origin_origin)->
2264 		    ds_uncompressed_bytes;
2265 	}
2266 
2267 	/* Check that there is enough space and limit headroom here */
2268 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2269 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2270 	if (err != 0)
2271 		goto out;
2272 
2273 	/*
2274 	 * Compute the amounts of space that will be used by snapshots
2275 	 * after the promotion (for both origin and clone).  For each,
2276 	 * it is the amount of space that will be on all of their
2277 	 * deadlists (that was not born before their new origin).
2278 	 */
2279 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2280 		uint64_t space;
2281 
2282 		/*
2283 		 * Note, typically this will not be a clone of a clone,
2284 		 * so dd_origin_txg will be < TXG_INITIAL, so
2285 		 * these snaplist_space() -> dsl_deadlist_space_range()
2286 		 * calls will be fast because they do not have to
2287 		 * iterate over all bps.
2288 		 */
2289 		snap = list_head(&ddpa->origin_snaps);
2290 		err = snaplist_space(&ddpa->shared_snaps,
2291 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2292 		if (err != 0)
2293 			goto out;
2294 
2295 		err = snaplist_space(&ddpa->clone_snaps,
2296 		    snap->ds->ds_dir->dd_origin_txg, &space);
2297 		if (err != 0)
2298 			goto out;
2299 		ddpa->cloneusedsnap += space;
2300 	}
2301 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2302 	    DD_FLAG_USED_BREAKDOWN) {
2303 		err = snaplist_space(&ddpa->origin_snaps,
2304 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2305 		    &ddpa->originusedsnap);
2306 		if (err != 0)
2307 			goto out;
2308 	}
2309 
2310 out:
2311 	promote_rele(ddpa, FTAG);
2312 	return (err);
2313 }
2314 
2315 static void
2316 dsl_dataset_promote_sync(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 	dsl_dataset_t *origin_head;
2324 	dsl_dir_t *dd;
2325 	dsl_dir_t *odd = NULL;
2326 	uint64_t oldnext_obj;
2327 	int64_t delta;
2328 
2329 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2330 	hds = ddpa->ddpa_clone;
2331 
2332 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2333 
2334 	snap = list_head(&ddpa->shared_snaps);
2335 	origin_ds = snap->ds;
2336 	dd = hds->ds_dir;
2337 
2338 	snap = list_head(&ddpa->origin_snaps);
2339 	origin_head = snap->ds;
2340 
2341 	/*
2342 	 * We need to explicitly open odd, since origin_ds's dd will be
2343 	 * changing.
2344 	 */
2345 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2346 	    NULL, FTAG, &odd));
2347 
2348 	/* change origin's next snap */
2349 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2350 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2351 	snap = list_tail(&ddpa->clone_snaps);
2352 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2353 	    origin_ds->ds_object);
2354 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2355 
2356 	/* change the origin's next clone */
2357 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2358 		dsl_dataset_remove_from_next_clones(origin_ds,
2359 		    snap->ds->ds_object, tx);
2360 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2361 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2362 		    oldnext_obj, tx));
2363 	}
2364 
2365 	/* change origin */
2366 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2367 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2368 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2369 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2370 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2371 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2372 	origin_head->ds_dir->dd_origin_txg =
2373 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2374 
2375 	/* change dd_clone entries */
2376 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2377 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2378 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2379 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2380 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2381 		    hds->ds_object, tx));
2382 
2383 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2384 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2385 		    origin_head->ds_object, tx));
2386 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2387 			dsl_dir_phys(dd)->dd_clones =
2388 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2389 			    DMU_OT_NONE, 0, tx);
2390 		}
2391 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2392 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2393 	}
2394 
2395 	/* move snapshots to this dir */
2396 	for (snap = list_head(&ddpa->shared_snaps); snap;
2397 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2398 		dsl_dataset_t *ds = snap->ds;
2399 
2400 		/*
2401 		 * Property callbacks are registered to a particular
2402 		 * dsl_dir.  Since ours is changing, evict the objset
2403 		 * so that they will be unregistered from the old dsl_dir.
2404 		 */
2405 		if (ds->ds_objset) {
2406 			dmu_objset_evict(ds->ds_objset);
2407 			ds->ds_objset = NULL;
2408 		}
2409 
2410 		/* move snap name entry */
2411 		VERIFY0(dsl_dataset_get_snapname(ds));
2412 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2413 		    ds->ds_snapname, tx, B_TRUE));
2414 		VERIFY0(zap_add(dp->dp_meta_objset,
2415 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2416 		    8, 1, &ds->ds_object, tx));
2417 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2418 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2419 
2420 		/* change containing dsl_dir */
2421 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2422 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2423 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2424 		ASSERT3P(ds->ds_dir, ==, odd);
2425 		dsl_dir_rele(ds->ds_dir, ds);
2426 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2427 		    NULL, ds, &ds->ds_dir));
2428 
2429 		/* move any clone references */
2430 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2431 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2432 			zap_cursor_t zc;
2433 			zap_attribute_t za;
2434 
2435 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2436 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
2437 			    zap_cursor_retrieve(&zc, &za) == 0;
2438 			    zap_cursor_advance(&zc)) {
2439 				dsl_dataset_t *cnds;
2440 				uint64_t o;
2441 
2442 				if (za.za_first_integer == oldnext_obj) {
2443 					/*
2444 					 * We've already moved the
2445 					 * origin's reference.
2446 					 */
2447 					continue;
2448 				}
2449 
2450 				VERIFY0(dsl_dataset_hold_obj(dp,
2451 				    za.za_first_integer, FTAG, &cnds));
2452 				o = dsl_dir_phys(cnds->ds_dir)->
2453 				    dd_head_dataset_obj;
2454 
2455 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2456 				    dsl_dir_phys(odd)->dd_clones, o, tx));
2457 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2458 				    dsl_dir_phys(dd)->dd_clones, o, tx));
2459 				dsl_dataset_rele(cnds, FTAG);
2460 			}
2461 			zap_cursor_fini(&zc);
2462 		}
2463 
2464 		ASSERT(!dsl_prop_hascb(ds));
2465 	}
2466 
2467 	/*
2468 	 * Change space accounting.
2469 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2470 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2471 	 * is true for each of {clone,origin} independently.
2472 	 */
2473 
2474 	delta = ddpa->cloneusedsnap -
2475 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2476 	ASSERT3S(delta, >=, 0);
2477 	ASSERT3U(ddpa->used, >=, delta);
2478 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2479 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2480 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2481 
2482 	delta = ddpa->originusedsnap -
2483 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2484 	ASSERT3S(delta, <=, 0);
2485 	ASSERT3U(ddpa->used, >=, -delta);
2486 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2487 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2488 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2489 
2490 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2491 
2492 	/* log history record */
2493 	spa_history_log_internal_ds(hds, "promote", tx, "");
2494 
2495 	dsl_dir_rele(odd, FTAG);
2496 	promote_rele(ddpa, FTAG);
2497 }
2498 
2499 /*
2500  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2501  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2502  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2503  * snapshots back to this dataset's origin.
2504  */
2505 static int
2506 snaplist_make(dsl_pool_t *dp,
2507     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2508 {
2509 	uint64_t obj = last_obj;
2510 
2511 	list_create(l, sizeof (struct promotenode),
2512 	    offsetof(struct promotenode, link));
2513 
2514 	while (obj != first_obj) {
2515 		dsl_dataset_t *ds;
2516 		struct promotenode *snap;
2517 		int err;
2518 
2519 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2520 		ASSERT(err != ENOENT);
2521 		if (err != 0)
2522 			return (err);
2523 
2524 		if (first_obj == 0)
2525 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2526 
2527 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2528 		snap->ds = ds;
2529 		list_insert_tail(l, snap);
2530 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2531 	}
2532 
2533 	return (0);
2534 }
2535 
2536 static int
2537 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2538 {
2539 	struct promotenode *snap;
2540 
2541 	*spacep = 0;
2542 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2543 		uint64_t used, comp, uncomp;
2544 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2545 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2546 		*spacep += used;
2547 	}
2548 	return (0);
2549 }
2550 
2551 static void
2552 snaplist_destroy(list_t *l, void *tag)
2553 {
2554 	struct promotenode *snap;
2555 
2556 	if (l == NULL || !list_link_active(&l->list_head))
2557 		return;
2558 
2559 	while ((snap = list_tail(l)) != NULL) {
2560 		list_remove(l, snap);
2561 		dsl_dataset_rele(snap->ds, tag);
2562 		kmem_free(snap, sizeof (*snap));
2563 	}
2564 	list_destroy(l);
2565 }
2566 
2567 static int
2568 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2569 {
2570 	int error;
2571 	dsl_dir_t *dd;
2572 	struct promotenode *snap;
2573 
2574 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2575 	    &ddpa->ddpa_clone);
2576 	if (error != 0)
2577 		return (error);
2578 	dd = ddpa->ddpa_clone->ds_dir;
2579 
2580 	if (ddpa->ddpa_clone->ds_is_snapshot ||
2581 	    !dsl_dir_is_clone(dd)) {
2582 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2583 		return (SET_ERROR(EINVAL));
2584 	}
2585 
2586 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2587 	    &ddpa->shared_snaps, tag);
2588 	if (error != 0)
2589 		goto out;
2590 
2591 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2592 	    &ddpa->clone_snaps, tag);
2593 	if (error != 0)
2594 		goto out;
2595 
2596 	snap = list_head(&ddpa->shared_snaps);
2597 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2598 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2599 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2600 	    &ddpa->origin_snaps, tag);
2601 	if (error != 0)
2602 		goto out;
2603 
2604 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2605 		error = dsl_dataset_hold_obj(dp,
2606 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2607 		    tag, &ddpa->origin_origin);
2608 		if (error != 0)
2609 			goto out;
2610 	}
2611 out:
2612 	if (error != 0)
2613 		promote_rele(ddpa, tag);
2614 	return (error);
2615 }
2616 
2617 static void
2618 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2619 {
2620 	snaplist_destroy(&ddpa->shared_snaps, tag);
2621 	snaplist_destroy(&ddpa->clone_snaps, tag);
2622 	snaplist_destroy(&ddpa->origin_snaps, tag);
2623 	if (ddpa->origin_origin != NULL)
2624 		dsl_dataset_rele(ddpa->origin_origin, tag);
2625 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2626 }
2627 
2628 /*
2629  * Promote a clone.
2630  *
2631  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2632  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2633  */
2634 int
2635 dsl_dataset_promote(const char *name, char *conflsnap)
2636 {
2637 	dsl_dataset_promote_arg_t ddpa = { 0 };
2638 	uint64_t numsnaps;
2639 	int error;
2640 	objset_t *os;
2641 
2642 	/*
2643 	 * We will modify space proportional to the number of
2644 	 * snapshots.  Compute numsnaps.
2645 	 */
2646 	error = dmu_objset_hold(name, FTAG, &os);
2647 	if (error != 0)
2648 		return (error);
2649 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2650 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2651 	    &numsnaps);
2652 	dmu_objset_rele(os, FTAG);
2653 	if (error != 0)
2654 		return (error);
2655 
2656 	ddpa.ddpa_clonename = name;
2657 	ddpa.err_ds = conflsnap;
2658 	ddpa.cr = CRED();
2659 
2660 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2661 	    dsl_dataset_promote_sync, &ddpa,
2662 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2663 }
2664 
2665 int
2666 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2667     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2668 {
2669 	int64_t unused_refres_delta;
2670 
2671 	/* they should both be heads */
2672 	if (clone->ds_is_snapshot ||
2673 	    origin_head->ds_is_snapshot)
2674 		return (SET_ERROR(EINVAL));
2675 
2676 	/* if we are not forcing, the branch point should be just before them */
2677 	if (!force && clone->ds_prev != origin_head->ds_prev)
2678 		return (SET_ERROR(EINVAL));
2679 
2680 	/* clone should be the clone (unless they are unrelated) */
2681 	if (clone->ds_prev != NULL &&
2682 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2683 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2684 		return (SET_ERROR(EINVAL));
2685 
2686 	/* the clone should be a child of the origin */
2687 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2688 		return (SET_ERROR(EINVAL));
2689 
2690 	/* origin_head shouldn't be modified unless 'force' */
2691 	if (!force &&
2692 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2693 		return (SET_ERROR(ETXTBSY));
2694 
2695 	/* origin_head should have no long holds (e.g. is not mounted) */
2696 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2697 		return (SET_ERROR(EBUSY));
2698 
2699 	/* check amount of any unconsumed refreservation */
2700 	unused_refres_delta =
2701 	    (int64_t)MIN(origin_head->ds_reserved,
2702 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2703 	    (int64_t)MIN(origin_head->ds_reserved,
2704 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2705 
2706 	if (unused_refres_delta > 0 &&
2707 	    unused_refres_delta >
2708 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2709 		return (SET_ERROR(ENOSPC));
2710 
2711 	/* clone can't be over the head's refquota */
2712 	if (origin_head->ds_quota != 0 &&
2713 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
2714 	    origin_head->ds_quota)
2715 		return (SET_ERROR(EDQUOT));
2716 
2717 	return (0);
2718 }
2719 
2720 void
2721 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2722     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2723 {
2724 	dsl_pool_t *dp = dmu_tx_pool(tx);
2725 	int64_t unused_refres_delta;
2726 
2727 	ASSERT(clone->ds_reserved == 0);
2728 	ASSERT(origin_head->ds_quota == 0 ||
2729 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2730 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2731 
2732 	/*
2733 	 * Swap per-dataset feature flags.
2734 	 */
2735 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2736 		if (!(spa_feature_table[f].fi_flags &
2737 		    ZFEATURE_FLAG_PER_DATASET)) {
2738 			ASSERT(!clone->ds_feature_inuse[f]);
2739 			ASSERT(!origin_head->ds_feature_inuse[f]);
2740 			continue;
2741 		}
2742 
2743 		boolean_t clone_inuse = clone->ds_feature_inuse[f];
2744 		boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2745 
2746 		if (clone_inuse) {
2747 			dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2748 			clone->ds_feature_inuse[f] = B_FALSE;
2749 		}
2750 		if (origin_head_inuse) {
2751 			dsl_dataset_deactivate_feature(origin_head->ds_object,
2752 			    f, tx);
2753 			origin_head->ds_feature_inuse[f] = B_FALSE;
2754 		}
2755 		if (clone_inuse) {
2756 			dsl_dataset_activate_feature(origin_head->ds_object,
2757 			    f, tx);
2758 			origin_head->ds_feature_inuse[f] = B_TRUE;
2759 		}
2760 		if (origin_head_inuse) {
2761 			dsl_dataset_activate_feature(clone->ds_object, f, tx);
2762 			clone->ds_feature_inuse[f] = B_TRUE;
2763 		}
2764 	}
2765 
2766 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2767 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2768 
2769 	if (clone->ds_objset != NULL) {
2770 		dmu_objset_evict(clone->ds_objset);
2771 		clone->ds_objset = NULL;
2772 	}
2773 
2774 	if (origin_head->ds_objset != NULL) {
2775 		dmu_objset_evict(origin_head->ds_objset);
2776 		origin_head->ds_objset = NULL;
2777 	}
2778 
2779 	unused_refres_delta =
2780 	    (int64_t)MIN(origin_head->ds_reserved,
2781 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2782 	    (int64_t)MIN(origin_head->ds_reserved,
2783 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2784 
2785 	/*
2786 	 * Reset origin's unique bytes, if it exists.
2787 	 */
2788 	if (clone->ds_prev) {
2789 		dsl_dataset_t *origin = clone->ds_prev;
2790 		uint64_t comp, uncomp;
2791 
2792 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2793 		dsl_deadlist_space_range(&clone->ds_deadlist,
2794 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2795 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2796 	}
2797 
2798 	/* swap blkptrs */
2799 	{
2800 		blkptr_t tmp;
2801 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
2802 		dsl_dataset_phys(origin_head)->ds_bp =
2803 		    dsl_dataset_phys(clone)->ds_bp;
2804 		dsl_dataset_phys(clone)->ds_bp = tmp;
2805 	}
2806 
2807 	/* set dd_*_bytes */
2808 	{
2809 		int64_t dused, dcomp, duncomp;
2810 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2811 		uint64_t odl_used, odl_comp, odl_uncomp;
2812 
2813 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2814 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2815 
2816 		dsl_deadlist_space(&clone->ds_deadlist,
2817 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2818 		dsl_deadlist_space(&origin_head->ds_deadlist,
2819 		    &odl_used, &odl_comp, &odl_uncomp);
2820 
2821 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2822 		    cdl_used -
2823 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2824 		    odl_used);
2825 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2826 		    cdl_comp -
2827 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2828 		    odl_comp);
2829 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2830 		    cdl_uncomp -
2831 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2832 		    odl_uncomp);
2833 
2834 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2835 		    dused, dcomp, duncomp, tx);
2836 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2837 		    -dused, -dcomp, -duncomp, tx);
2838 
2839 		/*
2840 		 * The difference in the space used by snapshots is the
2841 		 * difference in snapshot space due to the head's
2842 		 * deadlist (since that's the only thing that's
2843 		 * changing that affects the snapused).
2844 		 */
2845 		dsl_deadlist_space_range(&clone->ds_deadlist,
2846 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2847 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2848 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2849 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2850 		    &odl_used, &odl_comp, &odl_uncomp);
2851 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2852 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2853 	}
2854 
2855 	/* swap ds_*_bytes */
2856 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2857 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
2858 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2859 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
2860 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2861 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2862 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2863 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2864 
2865 	/* apply any parent delta for change in unconsumed refreservation */
2866 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2867 	    unused_refres_delta, 0, 0, tx);
2868 
2869 	/*
2870 	 * Swap deadlists.
2871 	 */
2872 	dsl_deadlist_close(&clone->ds_deadlist);
2873 	dsl_deadlist_close(&origin_head->ds_deadlist);
2874 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2875 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2876 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2877 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2878 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2879 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
2880 
2881 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2882 
2883 	spa_history_log_internal_ds(clone, "clone swap", tx,
2884 	    "parent=%s", origin_head->ds_dir->dd_myname);
2885 }
2886 
2887 /*
2888  * Given a pool name and a dataset object number in that pool,
2889  * return the name of that dataset.
2890  */
2891 int
2892 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2893 {
2894 	dsl_pool_t *dp;
2895 	dsl_dataset_t *ds;
2896 	int error;
2897 
2898 	error = dsl_pool_hold(pname, FTAG, &dp);
2899 	if (error != 0)
2900 		return (error);
2901 
2902 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2903 	if (error == 0) {
2904 		dsl_dataset_name(ds, buf);
2905 		dsl_dataset_rele(ds, FTAG);
2906 	}
2907 	dsl_pool_rele(dp, FTAG);
2908 
2909 	return (error);
2910 }
2911 
2912 int
2913 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2914     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2915 {
2916 	int error = 0;
2917 
2918 	ASSERT3S(asize, >, 0);
2919 
2920 	/*
2921 	 * *ref_rsrv is the portion of asize that will come from any
2922 	 * unconsumed refreservation space.
2923 	 */
2924 	*ref_rsrv = 0;
2925 
2926 	mutex_enter(&ds->ds_lock);
2927 	/*
2928 	 * Make a space adjustment for reserved bytes.
2929 	 */
2930 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2931 		ASSERT3U(*used, >=,
2932 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2933 		*used -=
2934 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2935 		*ref_rsrv =
2936 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2937 	}
2938 
2939 	if (!check_quota || ds->ds_quota == 0) {
2940 		mutex_exit(&ds->ds_lock);
2941 		return (0);
2942 	}
2943 	/*
2944 	 * If they are requesting more space, and our current estimate
2945 	 * is over quota, they get to try again unless the actual
2946 	 * on-disk is over quota and there are no pending changes (which
2947 	 * may free up space for us).
2948 	 */
2949 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
2950 	    ds->ds_quota) {
2951 		if (inflight > 0 ||
2952 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
2953 			error = SET_ERROR(ERESTART);
2954 		else
2955 			error = SET_ERROR(EDQUOT);
2956 	}
2957 	mutex_exit(&ds->ds_lock);
2958 
2959 	return (error);
2960 }
2961 
2962 typedef struct dsl_dataset_set_qr_arg {
2963 	const char *ddsqra_name;
2964 	zprop_source_t ddsqra_source;
2965 	uint64_t ddsqra_value;
2966 } dsl_dataset_set_qr_arg_t;
2967 
2968 
2969 /* ARGSUSED */
2970 static int
2971 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2972 {
2973 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2974 	dsl_pool_t *dp = dmu_tx_pool(tx);
2975 	dsl_dataset_t *ds;
2976 	int error;
2977 	uint64_t newval;
2978 
2979 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2980 		return (SET_ERROR(ENOTSUP));
2981 
2982 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2983 	if (error != 0)
2984 		return (error);
2985 
2986 	if (ds->ds_is_snapshot) {
2987 		dsl_dataset_rele(ds, FTAG);
2988 		return (SET_ERROR(EINVAL));
2989 	}
2990 
2991 	error = dsl_prop_predict(ds->ds_dir,
2992 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2993 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2994 	if (error != 0) {
2995 		dsl_dataset_rele(ds, FTAG);
2996 		return (error);
2997 	}
2998 
2999 	if (newval == 0) {
3000 		dsl_dataset_rele(ds, FTAG);
3001 		return (0);
3002 	}
3003 
3004 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3005 	    newval < ds->ds_reserved) {
3006 		dsl_dataset_rele(ds, FTAG);
3007 		return (SET_ERROR(ENOSPC));
3008 	}
3009 
3010 	dsl_dataset_rele(ds, FTAG);
3011 	return (0);
3012 }
3013 
3014 static void
3015 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3016 {
3017 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3018 	dsl_pool_t *dp = dmu_tx_pool(tx);
3019 	dsl_dataset_t *ds;
3020 	uint64_t newval;
3021 
3022 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3023 
3024 	dsl_prop_set_sync_impl(ds,
3025 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3026 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3027 	    &ddsqra->ddsqra_value, tx);
3028 
3029 	VERIFY0(dsl_prop_get_int_ds(ds,
3030 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3031 
3032 	if (ds->ds_quota != newval) {
3033 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3034 		ds->ds_quota = newval;
3035 	}
3036 	dsl_dataset_rele(ds, FTAG);
3037 }
3038 
3039 int
3040 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3041     uint64_t refquota)
3042 {
3043 	dsl_dataset_set_qr_arg_t ddsqra;
3044 
3045 	ddsqra.ddsqra_name = dsname;
3046 	ddsqra.ddsqra_source = source;
3047 	ddsqra.ddsqra_value = refquota;
3048 
3049 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3050 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3051 }
3052 
3053 static int
3054 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3055 {
3056 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3057 	dsl_pool_t *dp = dmu_tx_pool(tx);
3058 	dsl_dataset_t *ds;
3059 	int error;
3060 	uint64_t newval, unique;
3061 
3062 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3063 		return (SET_ERROR(ENOTSUP));
3064 
3065 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3066 	if (error != 0)
3067 		return (error);
3068 
3069 	if (ds->ds_is_snapshot) {
3070 		dsl_dataset_rele(ds, FTAG);
3071 		return (SET_ERROR(EINVAL));
3072 	}
3073 
3074 	error = dsl_prop_predict(ds->ds_dir,
3075 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3076 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3077 	if (error != 0) {
3078 		dsl_dataset_rele(ds, FTAG);
3079 		return (error);
3080 	}
3081 
3082 	/*
3083 	 * If we are doing the preliminary check in open context, the
3084 	 * space estimates may be inaccurate.
3085 	 */
3086 	if (!dmu_tx_is_syncing(tx)) {
3087 		dsl_dataset_rele(ds, FTAG);
3088 		return (0);
3089 	}
3090 
3091 	mutex_enter(&ds->ds_lock);
3092 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3093 		dsl_dataset_recalc_head_uniq(ds);
3094 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3095 	mutex_exit(&ds->ds_lock);
3096 
3097 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3098 		uint64_t delta = MAX(unique, newval) -
3099 		    MAX(unique, ds->ds_reserved);
3100 
3101 		if (delta >
3102 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3103 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3104 			dsl_dataset_rele(ds, FTAG);
3105 			return (SET_ERROR(ENOSPC));
3106 		}
3107 	}
3108 
3109 	dsl_dataset_rele(ds, FTAG);
3110 	return (0);
3111 }
3112 
3113 void
3114 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3115     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3116 {
3117 	uint64_t newval;
3118 	uint64_t unique;
3119 	int64_t delta;
3120 
3121 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3122 	    source, sizeof (value), 1, &value, tx);
3123 
3124 	VERIFY0(dsl_prop_get_int_ds(ds,
3125 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3126 
3127 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3128 	mutex_enter(&ds->ds_dir->dd_lock);
3129 	mutex_enter(&ds->ds_lock);
3130 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3131 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3132 	delta = MAX(0, (int64_t)(newval - unique)) -
3133 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3134 	ds->ds_reserved = newval;
3135 	mutex_exit(&ds->ds_lock);
3136 
3137 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3138 	mutex_exit(&ds->ds_dir->dd_lock);
3139 }
3140 
3141 static void
3142 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3143 {
3144 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3145 	dsl_pool_t *dp = dmu_tx_pool(tx);
3146 	dsl_dataset_t *ds;
3147 
3148 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3149 	dsl_dataset_set_refreservation_sync_impl(ds,
3150 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3151 	dsl_dataset_rele(ds, FTAG);
3152 }
3153 
3154 int
3155 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3156     uint64_t refreservation)
3157 {
3158 	dsl_dataset_set_qr_arg_t ddsqra;
3159 
3160 	ddsqra.ddsqra_name = dsname;
3161 	ddsqra.ddsqra_source = source;
3162 	ddsqra.ddsqra_value = refreservation;
3163 
3164 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3165 	    dsl_dataset_set_refreservation_sync, &ddsqra,
3166 	    0, ZFS_SPACE_CHECK_NONE));
3167 }
3168 
3169 /*
3170  * Return (in *usedp) the amount of space written in new that is not
3171  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3172  * a snapshot before new, in new's filesystem (or its origin).  If not then
3173  * fail and return EINVAL.
3174  *
3175  * The written space is calculated by considering two components:  First, we
3176  * ignore any freed space, and calculate the written as new's used space
3177  * minus old's used space.  Next, we add in the amount of space that was freed
3178  * between the two snapshots, thus reducing new's used space relative to old's.
3179  * Specifically, this is the space that was born before old->ds_creation_txg,
3180  * and freed before new (ie. on new's deadlist or a previous deadlist).
3181  *
3182  * space freed                         [---------------------]
3183  * snapshots                       ---O-------O--------O-------O------
3184  *                                         oldsnap            new
3185  */
3186 int
3187 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3188     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3189 {
3190 	int err = 0;
3191 	uint64_t snapobj;
3192 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3193 
3194 	ASSERT(dsl_pool_config_held(dp));
3195 
3196 	*usedp = 0;
3197 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3198 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3199 
3200 	*compp = 0;
3201 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3202 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3203 
3204 	*uncompp = 0;
3205 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3206 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3207 
3208 	snapobj = new->ds_object;
3209 	while (snapobj != oldsnap->ds_object) {
3210 		dsl_dataset_t *snap;
3211 		uint64_t used, comp, uncomp;
3212 
3213 		if (snapobj == new->ds_object) {
3214 			snap = new;
3215 		} else {
3216 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3217 			if (err != 0)
3218 				break;
3219 		}
3220 
3221 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3222 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3223 			/*
3224 			 * The blocks in the deadlist can not be born after
3225 			 * ds_prev_snap_txg, so get the whole deadlist space,
3226 			 * which is more efficient (especially for old-format
3227 			 * deadlists).  Unfortunately the deadlist code
3228 			 * doesn't have enough information to make this
3229 			 * optimization itself.
3230 			 */
3231 			dsl_deadlist_space(&snap->ds_deadlist,
3232 			    &used, &comp, &uncomp);
3233 		} else {
3234 			dsl_deadlist_space_range(&snap->ds_deadlist,
3235 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3236 			    &used, &comp, &uncomp);
3237 		}
3238 		*usedp += used;
3239 		*compp += comp;
3240 		*uncompp += uncomp;
3241 
3242 		/*
3243 		 * If we get to the beginning of the chain of snapshots
3244 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3245 		 * was not a snapshot of/before new.
3246 		 */
3247 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3248 		if (snap != new)
3249 			dsl_dataset_rele(snap, FTAG);
3250 		if (snapobj == 0) {
3251 			err = SET_ERROR(EINVAL);
3252 			break;
3253 		}
3254 
3255 	}
3256 	return (err);
3257 }
3258 
3259 /*
3260  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3261  * lastsnap, and all snapshots in between are deleted.
3262  *
3263  * blocks that would be freed            [---------------------------]
3264  * snapshots                       ---O-------O--------O-------O--------O
3265  *                                        firstsnap        lastsnap
3266  *
3267  * This is the set of blocks that were born after the snap before firstsnap,
3268  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3269  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3270  * We calculate this by iterating over the relevant deadlists (from the snap
3271  * after lastsnap, backward to the snap after firstsnap), summing up the
3272  * space on the deadlist that was born after the snap before firstsnap.
3273  */
3274 int
3275 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3276     dsl_dataset_t *lastsnap,
3277     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3278 {
3279 	int err = 0;
3280 	uint64_t snapobj;
3281 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3282 
3283 	ASSERT(firstsnap->ds_is_snapshot);
3284 	ASSERT(lastsnap->ds_is_snapshot);
3285 
3286 	/*
3287 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3288 	 * is before lastsnap.
3289 	 */
3290 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3291 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3292 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3293 		return (SET_ERROR(EINVAL));
3294 
3295 	*usedp = *compp = *uncompp = 0;
3296 
3297 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3298 	while (snapobj != firstsnap->ds_object) {
3299 		dsl_dataset_t *ds;
3300 		uint64_t used, comp, uncomp;
3301 
3302 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3303 		if (err != 0)
3304 			break;
3305 
3306 		dsl_deadlist_space_range(&ds->ds_deadlist,
3307 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3308 		    &used, &comp, &uncomp);
3309 		*usedp += used;
3310 		*compp += comp;
3311 		*uncompp += uncomp;
3312 
3313 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3314 		ASSERT3U(snapobj, !=, 0);
3315 		dsl_dataset_rele(ds, FTAG);
3316 	}
3317 	return (err);
3318 }
3319 
3320 /*
3321  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3322  * For example, they could both be snapshots of the same filesystem, and
3323  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3324  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3325  * filesystem.  Or 'earlier' could be the origin's origin.
3326  *
3327  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3328  */
3329 boolean_t
3330 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3331 	uint64_t earlier_txg)
3332 {
3333 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3334 	int error;
3335 	boolean_t ret;
3336 
3337 	ASSERT(dsl_pool_config_held(dp));
3338 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3339 
3340 	if (earlier_txg == 0)
3341 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3342 
3343 	if (later->ds_is_snapshot &&
3344 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3345 		return (B_FALSE);
3346 
3347 	if (later->ds_dir == earlier->ds_dir)
3348 		return (B_TRUE);
3349 	if (!dsl_dir_is_clone(later->ds_dir))
3350 		return (B_FALSE);
3351 
3352 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3353 		return (B_TRUE);
3354 	dsl_dataset_t *origin;
3355 	error = dsl_dataset_hold_obj(dp,
3356 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3357 	if (error != 0)
3358 		return (B_FALSE);
3359 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3360 	dsl_dataset_rele(origin, FTAG);
3361 	return (ret);
3362 }
3363 
3364 void
3365 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3366 {
3367 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3368 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3369 }
3370