xref: /titanic_50/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision cffcfaee1e6b29ef9ceb7d80e4e053ffd029906b)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
24  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
25  * Copyright (c) 2014 RackTop Systems.
26  */
27 
28 #include <sys/dmu_objset.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_prop.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dmu_impl.h>
35 #include <sys/dmu_tx.h>
36 #include <sys/arc.h>
37 #include <sys/zio.h>
38 #include <sys/zap.h>
39 #include <sys/zfeature.h>
40 #include <sys/unique.h>
41 #include <sys/zfs_context.h>
42 #include <sys/zfs_ioctl.h>
43 #include <sys/spa.h>
44 #include <sys/zfs_znode.h>
45 #include <sys/zfs_onexit.h>
46 #include <sys/zvol.h>
47 #include <sys/dsl_scan.h>
48 #include <sys/dsl_deadlist.h>
49 #include <sys/dsl_destroy.h>
50 #include <sys/dsl_userhold.h>
51 #include <sys/dsl_bookmark.h>
52 
53 #define	SWITCH64(x, y) \
54 	{ \
55 		uint64_t __tmp = (x); \
56 		(x) = (y); \
57 		(y) = __tmp; \
58 	}
59 
60 #define	DS_REF_MAX	(1ULL << 62)
61 
62 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
63 
64 /*
65  * Figure out how much of this delta should be propogated to the dsl_dir
66  * layer.  If there's a refreservation, that space has already been
67  * partially accounted for in our ancestors.
68  */
69 static int64_t
70 parent_delta(dsl_dataset_t *ds, int64_t delta)
71 {
72 	uint64_t old_bytes, new_bytes;
73 
74 	if (ds->ds_reserved == 0)
75 		return (delta);
76 
77 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
78 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
79 
80 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
81 	return (new_bytes - old_bytes);
82 }
83 
84 void
85 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
86 {
87 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
88 	int compressed = BP_GET_PSIZE(bp);
89 	int uncompressed = BP_GET_UCSIZE(bp);
90 	int64_t delta;
91 
92 	dprintf_bp(bp, "ds=%p", ds);
93 
94 	ASSERT(dmu_tx_is_syncing(tx));
95 	/* It could have been compressed away to nothing */
96 	if (BP_IS_HOLE(bp))
97 		return;
98 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
99 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
100 	if (ds == NULL) {
101 		dsl_pool_mos_diduse_space(tx->tx_pool,
102 		    used, compressed, uncompressed);
103 		return;
104 	}
105 
106 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
107 	mutex_enter(&ds->ds_lock);
108 	delta = parent_delta(ds, used);
109 	ds->ds_phys->ds_referenced_bytes += used;
110 	ds->ds_phys->ds_compressed_bytes += compressed;
111 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
112 	ds->ds_phys->ds_unique_bytes += used;
113 	mutex_exit(&ds->ds_lock);
114 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
115 	    compressed, uncompressed, tx);
116 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
117 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
118 }
119 
120 int
121 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
122     boolean_t async)
123 {
124 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
125 	int compressed = BP_GET_PSIZE(bp);
126 	int uncompressed = BP_GET_UCSIZE(bp);
127 
128 	if (BP_IS_HOLE(bp))
129 		return (0);
130 
131 	ASSERT(dmu_tx_is_syncing(tx));
132 	ASSERT(bp->blk_birth <= tx->tx_txg);
133 
134 	if (ds == NULL) {
135 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
136 		dsl_pool_mos_diduse_space(tx->tx_pool,
137 		    -used, -compressed, -uncompressed);
138 		return (used);
139 	}
140 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
141 
142 	ASSERT(!dsl_dataset_is_snapshot(ds));
143 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
144 
145 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
146 		int64_t delta;
147 
148 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
149 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
150 
151 		mutex_enter(&ds->ds_lock);
152 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
153 		    !DS_UNIQUE_IS_ACCURATE(ds));
154 		delta = parent_delta(ds, -used);
155 		ds->ds_phys->ds_unique_bytes -= used;
156 		mutex_exit(&ds->ds_lock);
157 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
158 		    delta, -compressed, -uncompressed, tx);
159 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
160 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
161 	} else {
162 		dprintf_bp(bp, "putting on dead list: %s", "");
163 		if (async) {
164 			/*
165 			 * We are here as part of zio's write done callback,
166 			 * which means we're a zio interrupt thread.  We can't
167 			 * call dsl_deadlist_insert() now because it may block
168 			 * waiting for I/O.  Instead, put bp on the deferred
169 			 * queue and let dsl_pool_sync() finish the job.
170 			 */
171 			bplist_append(&ds->ds_pending_deadlist, bp);
172 		} else {
173 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
174 		}
175 		ASSERT3U(ds->ds_prev->ds_object, ==,
176 		    ds->ds_phys->ds_prev_snap_obj);
177 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
178 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
179 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
180 		    ds->ds_object && bp->blk_birth >
181 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
182 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
183 			mutex_enter(&ds->ds_prev->ds_lock);
184 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
185 			mutex_exit(&ds->ds_prev->ds_lock);
186 		}
187 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
188 			dsl_dir_transfer_space(ds->ds_dir, used,
189 			    DD_USED_HEAD, DD_USED_SNAP, tx);
190 		}
191 	}
192 	mutex_enter(&ds->ds_lock);
193 	ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
194 	ds->ds_phys->ds_referenced_bytes -= used;
195 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
196 	ds->ds_phys->ds_compressed_bytes -= compressed;
197 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
198 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
199 	mutex_exit(&ds->ds_lock);
200 
201 	return (used);
202 }
203 
204 uint64_t
205 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
206 {
207 	uint64_t trysnap = 0;
208 
209 	if (ds == NULL)
210 		return (0);
211 	/*
212 	 * The snapshot creation could fail, but that would cause an
213 	 * incorrect FALSE return, which would only result in an
214 	 * overestimation of the amount of space that an operation would
215 	 * consume, which is OK.
216 	 *
217 	 * There's also a small window where we could miss a pending
218 	 * snapshot, because we could set the sync task in the quiescing
219 	 * phase.  So this should only be used as a guess.
220 	 */
221 	if (ds->ds_trysnap_txg >
222 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
223 		trysnap = ds->ds_trysnap_txg;
224 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
225 }
226 
227 boolean_t
228 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
229     uint64_t blk_birth)
230 {
231 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
232 	    (bp != NULL && BP_IS_HOLE(bp)))
233 		return (B_FALSE);
234 
235 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
236 
237 	return (B_TRUE);
238 }
239 
240 /* ARGSUSED */
241 static void
242 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
243 {
244 	dsl_dataset_t *ds = dsv;
245 
246 	ASSERT(ds->ds_owner == NULL);
247 
248 	unique_remove(ds->ds_fsid_guid);
249 
250 	if (ds->ds_objset != NULL)
251 		dmu_objset_evict(ds->ds_objset);
252 
253 	if (ds->ds_prev) {
254 		dsl_dataset_rele(ds->ds_prev, ds);
255 		ds->ds_prev = NULL;
256 	}
257 
258 	bplist_destroy(&ds->ds_pending_deadlist);
259 	if (ds->ds_phys->ds_deadlist_obj != 0)
260 		dsl_deadlist_close(&ds->ds_deadlist);
261 	if (ds->ds_dir)
262 		dsl_dir_rele(ds->ds_dir, ds);
263 
264 	ASSERT(!list_link_active(&ds->ds_synced_link));
265 
266 	mutex_destroy(&ds->ds_lock);
267 	mutex_destroy(&ds->ds_opening_lock);
268 	mutex_destroy(&ds->ds_sendstream_lock);
269 	refcount_destroy(&ds->ds_longholds);
270 
271 	kmem_free(ds, sizeof (dsl_dataset_t));
272 }
273 
274 int
275 dsl_dataset_get_snapname(dsl_dataset_t *ds)
276 {
277 	dsl_dataset_phys_t *headphys;
278 	int err;
279 	dmu_buf_t *headdbuf;
280 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
281 	objset_t *mos = dp->dp_meta_objset;
282 
283 	if (ds->ds_snapname[0])
284 		return (0);
285 	if (ds->ds_phys->ds_next_snap_obj == 0)
286 		return (0);
287 
288 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
289 	    FTAG, &headdbuf);
290 	if (err != 0)
291 		return (err);
292 	headphys = headdbuf->db_data;
293 	err = zap_value_search(dp->dp_meta_objset,
294 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
295 	dmu_buf_rele(headdbuf, FTAG);
296 	return (err);
297 }
298 
299 int
300 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
301 {
302 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
303 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
304 	matchtype_t mt;
305 	int err;
306 
307 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
308 		mt = MT_FIRST;
309 	else
310 		mt = MT_EXACT;
311 
312 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
313 	    value, mt, NULL, 0, NULL);
314 	if (err == ENOTSUP && mt == MT_FIRST)
315 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
316 	return (err);
317 }
318 
319 int
320 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
321     boolean_t adj_cnt)
322 {
323 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
324 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
325 	matchtype_t mt;
326 	int err;
327 
328 	dsl_dir_snap_cmtime_update(ds->ds_dir);
329 
330 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
331 		mt = MT_FIRST;
332 	else
333 		mt = MT_EXACT;
334 
335 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
336 	if (err == ENOTSUP && mt == MT_FIRST)
337 		err = zap_remove(mos, snapobj, name, tx);
338 
339 	if (err == 0 && adj_cnt)
340 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
341 		    DD_FIELD_SNAPSHOT_COUNT, tx);
342 
343 	return (err);
344 }
345 
346 int
347 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
348     dsl_dataset_t **dsp)
349 {
350 	objset_t *mos = dp->dp_meta_objset;
351 	dmu_buf_t *dbuf;
352 	dsl_dataset_t *ds;
353 	int err;
354 	dmu_object_info_t doi;
355 
356 	ASSERT(dsl_pool_config_held(dp));
357 
358 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
359 	if (err != 0)
360 		return (err);
361 
362 	/* Make sure dsobj has the correct object type. */
363 	dmu_object_info_from_db(dbuf, &doi);
364 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
365 		dmu_buf_rele(dbuf, tag);
366 		return (SET_ERROR(EINVAL));
367 	}
368 
369 	ds = dmu_buf_get_user(dbuf);
370 	if (ds == NULL) {
371 		dsl_dataset_t *winner = NULL;
372 
373 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
374 		ds->ds_dbuf = dbuf;
375 		ds->ds_object = dsobj;
376 		ds->ds_phys = dbuf->db_data;
377 
378 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
379 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
380 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
381 		refcount_create(&ds->ds_longholds);
382 
383 		bplist_create(&ds->ds_pending_deadlist);
384 		dsl_deadlist_open(&ds->ds_deadlist,
385 		    mos, ds->ds_phys->ds_deadlist_obj);
386 
387 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
388 		    offsetof(dmu_sendarg_t, dsa_link));
389 
390 		if (err == 0) {
391 			err = dsl_dir_hold_obj(dp,
392 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
393 		}
394 		if (err != 0) {
395 			mutex_destroy(&ds->ds_lock);
396 			mutex_destroy(&ds->ds_opening_lock);
397 			mutex_destroy(&ds->ds_sendstream_lock);
398 			refcount_destroy(&ds->ds_longholds);
399 			bplist_destroy(&ds->ds_pending_deadlist);
400 			dsl_deadlist_close(&ds->ds_deadlist);
401 			kmem_free(ds, sizeof (dsl_dataset_t));
402 			dmu_buf_rele(dbuf, tag);
403 			return (err);
404 		}
405 
406 		if (!dsl_dataset_is_snapshot(ds)) {
407 			ds->ds_snapname[0] = '\0';
408 			if (ds->ds_phys->ds_prev_snap_obj != 0) {
409 				err = dsl_dataset_hold_obj(dp,
410 				    ds->ds_phys->ds_prev_snap_obj,
411 				    ds, &ds->ds_prev);
412 			}
413 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
414 				int zaperr = zap_lookup(mos, ds->ds_object,
415 				    DS_FIELD_BOOKMARK_NAMES,
416 				    sizeof (ds->ds_bookmarks), 1,
417 				    &ds->ds_bookmarks);
418 				if (zaperr != ENOENT)
419 					VERIFY0(zaperr);
420 			}
421 		} else {
422 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
423 				err = dsl_dataset_get_snapname(ds);
424 			if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
425 				err = zap_count(
426 				    ds->ds_dir->dd_pool->dp_meta_objset,
427 				    ds->ds_phys->ds_userrefs_obj,
428 				    &ds->ds_userrefs);
429 			}
430 		}
431 
432 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
433 			err = dsl_prop_get_int_ds(ds,
434 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
435 			    &ds->ds_reserved);
436 			if (err == 0) {
437 				err = dsl_prop_get_int_ds(ds,
438 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
439 				    &ds->ds_quota);
440 			}
441 		} else {
442 			ds->ds_reserved = ds->ds_quota = 0;
443 		}
444 
445 		if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
446 		    &ds->ds_phys, dsl_dataset_evict)) != NULL) {
447 			bplist_destroy(&ds->ds_pending_deadlist);
448 			dsl_deadlist_close(&ds->ds_deadlist);
449 			if (ds->ds_prev)
450 				dsl_dataset_rele(ds->ds_prev, ds);
451 			dsl_dir_rele(ds->ds_dir, ds);
452 			mutex_destroy(&ds->ds_lock);
453 			mutex_destroy(&ds->ds_opening_lock);
454 			mutex_destroy(&ds->ds_sendstream_lock);
455 			refcount_destroy(&ds->ds_longholds);
456 			kmem_free(ds, sizeof (dsl_dataset_t));
457 			if (err != 0) {
458 				dmu_buf_rele(dbuf, tag);
459 				return (err);
460 			}
461 			ds = winner;
462 		} else {
463 			ds->ds_fsid_guid =
464 			    unique_insert(ds->ds_phys->ds_fsid_guid);
465 		}
466 	}
467 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
468 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
469 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
470 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
471 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
472 	*dsp = ds;
473 	return (0);
474 }
475 
476 int
477 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
478     void *tag, dsl_dataset_t **dsp)
479 {
480 	dsl_dir_t *dd;
481 	const char *snapname;
482 	uint64_t obj;
483 	int err = 0;
484 
485 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
486 	if (err != 0)
487 		return (err);
488 
489 	ASSERT(dsl_pool_config_held(dp));
490 	obj = dd->dd_phys->dd_head_dataset_obj;
491 	if (obj != 0)
492 		err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
493 	else
494 		err = SET_ERROR(ENOENT);
495 
496 	/* we may be looking for a snapshot */
497 	if (err == 0 && snapname != NULL) {
498 		dsl_dataset_t *ds;
499 
500 		if (*snapname++ != '@') {
501 			dsl_dataset_rele(*dsp, tag);
502 			dsl_dir_rele(dd, FTAG);
503 			return (SET_ERROR(ENOENT));
504 		}
505 
506 		dprintf("looking for snapshot '%s'\n", snapname);
507 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
508 		if (err == 0)
509 			err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
510 		dsl_dataset_rele(*dsp, tag);
511 
512 		if (err == 0) {
513 			mutex_enter(&ds->ds_lock);
514 			if (ds->ds_snapname[0] == 0)
515 				(void) strlcpy(ds->ds_snapname, snapname,
516 				    sizeof (ds->ds_snapname));
517 			mutex_exit(&ds->ds_lock);
518 			*dsp = ds;
519 		}
520 	}
521 
522 	dsl_dir_rele(dd, FTAG);
523 	return (err);
524 }
525 
526 int
527 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
528     void *tag, dsl_dataset_t **dsp)
529 {
530 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
531 	if (err != 0)
532 		return (err);
533 	if (!dsl_dataset_tryown(*dsp, tag)) {
534 		dsl_dataset_rele(*dsp, tag);
535 		*dsp = NULL;
536 		return (SET_ERROR(EBUSY));
537 	}
538 	return (0);
539 }
540 
541 int
542 dsl_dataset_own(dsl_pool_t *dp, const char *name,
543     void *tag, dsl_dataset_t **dsp)
544 {
545 	int err = dsl_dataset_hold(dp, name, tag, dsp);
546 	if (err != 0)
547 		return (err);
548 	if (!dsl_dataset_tryown(*dsp, tag)) {
549 		dsl_dataset_rele(*dsp, tag);
550 		return (SET_ERROR(EBUSY));
551 	}
552 	return (0);
553 }
554 
555 /*
556  * See the comment above dsl_pool_hold() for details.  In summary, a long
557  * hold is used to prevent destruction of a dataset while the pool hold
558  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
559  *
560  * The dataset and pool must be held when this function is called.  After it
561  * is called, the pool hold may be released while the dataset is still held
562  * and accessed.
563  */
564 void
565 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
566 {
567 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
568 	(void) refcount_add(&ds->ds_longholds, tag);
569 }
570 
571 void
572 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
573 {
574 	(void) refcount_remove(&ds->ds_longholds, tag);
575 }
576 
577 /* Return B_TRUE if there are any long holds on this dataset. */
578 boolean_t
579 dsl_dataset_long_held(dsl_dataset_t *ds)
580 {
581 	return (!refcount_is_zero(&ds->ds_longholds));
582 }
583 
584 void
585 dsl_dataset_name(dsl_dataset_t *ds, char *name)
586 {
587 	if (ds == NULL) {
588 		(void) strcpy(name, "mos");
589 	} else {
590 		dsl_dir_name(ds->ds_dir, name);
591 		VERIFY0(dsl_dataset_get_snapname(ds));
592 		if (ds->ds_snapname[0]) {
593 			(void) strcat(name, "@");
594 			/*
595 			 * We use a "recursive" mutex so that we
596 			 * can call dprintf_ds() with ds_lock held.
597 			 */
598 			if (!MUTEX_HELD(&ds->ds_lock)) {
599 				mutex_enter(&ds->ds_lock);
600 				(void) strcat(name, ds->ds_snapname);
601 				mutex_exit(&ds->ds_lock);
602 			} else {
603 				(void) strcat(name, ds->ds_snapname);
604 			}
605 		}
606 	}
607 }
608 
609 void
610 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
611 {
612 	dmu_buf_rele(ds->ds_dbuf, tag);
613 }
614 
615 void
616 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
617 {
618 	ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL);
619 
620 	mutex_enter(&ds->ds_lock);
621 	ds->ds_owner = NULL;
622 	mutex_exit(&ds->ds_lock);
623 	dsl_dataset_long_rele(ds, tag);
624 	if (ds->ds_dbuf != NULL)
625 		dsl_dataset_rele(ds, tag);
626 	else
627 		dsl_dataset_evict(NULL, ds);
628 }
629 
630 boolean_t
631 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
632 {
633 	boolean_t gotit = FALSE;
634 
635 	mutex_enter(&ds->ds_lock);
636 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
637 		ds->ds_owner = tag;
638 		dsl_dataset_long_hold(ds, tag);
639 		gotit = TRUE;
640 	}
641 	mutex_exit(&ds->ds_lock);
642 	return (gotit);
643 }
644 
645 uint64_t
646 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
647     uint64_t flags, dmu_tx_t *tx)
648 {
649 	dsl_pool_t *dp = dd->dd_pool;
650 	dmu_buf_t *dbuf;
651 	dsl_dataset_phys_t *dsphys;
652 	uint64_t dsobj;
653 	objset_t *mos = dp->dp_meta_objset;
654 
655 	if (origin == NULL)
656 		origin = dp->dp_origin_snap;
657 
658 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
659 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
660 	ASSERT(dmu_tx_is_syncing(tx));
661 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
662 
663 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
664 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
665 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
666 	dmu_buf_will_dirty(dbuf, tx);
667 	dsphys = dbuf->db_data;
668 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
669 	dsphys->ds_dir_obj = dd->dd_object;
670 	dsphys->ds_flags = flags;
671 	dsphys->ds_fsid_guid = unique_create();
672 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
673 	    sizeof (dsphys->ds_guid));
674 	dsphys->ds_snapnames_zapobj =
675 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
676 	    DMU_OT_NONE, 0, tx);
677 	dsphys->ds_creation_time = gethrestime_sec();
678 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
679 
680 	if (origin == NULL) {
681 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
682 	} else {
683 		dsl_dataset_t *ohds; /* head of the origin snapshot */
684 
685 		dsphys->ds_prev_snap_obj = origin->ds_object;
686 		dsphys->ds_prev_snap_txg =
687 		    origin->ds_phys->ds_creation_txg;
688 		dsphys->ds_referenced_bytes =
689 		    origin->ds_phys->ds_referenced_bytes;
690 		dsphys->ds_compressed_bytes =
691 		    origin->ds_phys->ds_compressed_bytes;
692 		dsphys->ds_uncompressed_bytes =
693 		    origin->ds_phys->ds_uncompressed_bytes;
694 		dsphys->ds_bp = origin->ds_phys->ds_bp;
695 
696 		/*
697 		 * Inherit flags that describe the dataset's contents
698 		 * (INCONSISTENT) or properties (Case Insensitive).
699 		 */
700 		dsphys->ds_flags |= origin->ds_phys->ds_flags &
701 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
702 
703 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
704 		origin->ds_phys->ds_num_children++;
705 
706 		VERIFY0(dsl_dataset_hold_obj(dp,
707 		    origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
708 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
709 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
710 		dsl_dataset_rele(ohds, FTAG);
711 
712 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
713 			if (origin->ds_phys->ds_next_clones_obj == 0) {
714 				origin->ds_phys->ds_next_clones_obj =
715 				    zap_create(mos,
716 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
717 			}
718 			VERIFY0(zap_add_int(mos,
719 			    origin->ds_phys->ds_next_clones_obj, dsobj, tx));
720 		}
721 
722 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
723 		dd->dd_phys->dd_origin_obj = origin->ds_object;
724 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
725 			if (origin->ds_dir->dd_phys->dd_clones == 0) {
726 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
727 				origin->ds_dir->dd_phys->dd_clones =
728 				    zap_create(mos,
729 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
730 			}
731 			VERIFY0(zap_add_int(mos,
732 			    origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
733 		}
734 	}
735 
736 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
737 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
738 
739 	dmu_buf_rele(dbuf, FTAG);
740 
741 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
742 	dd->dd_phys->dd_head_dataset_obj = dsobj;
743 
744 	return (dsobj);
745 }
746 
747 static void
748 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
749 {
750 	objset_t *os;
751 
752 	VERIFY0(dmu_objset_from_ds(ds, &os));
753 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
754 	dsl_dataset_dirty(ds, tx);
755 }
756 
757 uint64_t
758 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
759     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
760 {
761 	dsl_pool_t *dp = pdd->dd_pool;
762 	uint64_t dsobj, ddobj;
763 	dsl_dir_t *dd;
764 
765 	ASSERT(dmu_tx_is_syncing(tx));
766 	ASSERT(lastname[0] != '@');
767 
768 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
769 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
770 
771 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
772 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
773 
774 	dsl_deleg_set_create_perms(dd, tx, cr);
775 
776 	/*
777 	 * Since we're creating a new node we know it's a leaf, so we can
778 	 * initialize the counts if the limit feature is active.
779 	 */
780 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
781 		uint64_t cnt = 0;
782 		objset_t *os = dd->dd_pool->dp_meta_objset;
783 
784 		dsl_dir_zapify(dd, tx);
785 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
786 		    sizeof (cnt), 1, &cnt, tx));
787 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
788 		    sizeof (cnt), 1, &cnt, tx));
789 	}
790 
791 	dsl_dir_rele(dd, FTAG);
792 
793 	/*
794 	 * If we are creating a clone, make sure we zero out any stale
795 	 * data from the origin snapshots zil header.
796 	 */
797 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
798 		dsl_dataset_t *ds;
799 
800 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
801 		dsl_dataset_zero_zil(ds, tx);
802 		dsl_dataset_rele(ds, FTAG);
803 	}
804 
805 	return (dsobj);
806 }
807 
808 /*
809  * The unique space in the head dataset can be calculated by subtracting
810  * the space used in the most recent snapshot, that is still being used
811  * in this file system, from the space currently in use.  To figure out
812  * the space in the most recent snapshot still in use, we need to take
813  * the total space used in the snapshot and subtract out the space that
814  * has been freed up since the snapshot was taken.
815  */
816 void
817 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
818 {
819 	uint64_t mrs_used;
820 	uint64_t dlused, dlcomp, dluncomp;
821 
822 	ASSERT(!dsl_dataset_is_snapshot(ds));
823 
824 	if (ds->ds_phys->ds_prev_snap_obj != 0)
825 		mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
826 	else
827 		mrs_used = 0;
828 
829 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
830 
831 	ASSERT3U(dlused, <=, mrs_used);
832 	ds->ds_phys->ds_unique_bytes =
833 	    ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
834 
835 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
836 	    SPA_VERSION_UNIQUE_ACCURATE)
837 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
838 }
839 
840 void
841 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
842     dmu_tx_t *tx)
843 {
844 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
845 	uint64_t count;
846 	int err;
847 
848 	ASSERT(ds->ds_phys->ds_num_children >= 2);
849 	err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
850 	/*
851 	 * The err should not be ENOENT, but a bug in a previous version
852 	 * of the code could cause upgrade_clones_cb() to not set
853 	 * ds_next_snap_obj when it should, leading to a missing entry.
854 	 * If we knew that the pool was created after
855 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
856 	 * ENOENT.  However, at least we can check that we don't have
857 	 * too many entries in the next_clones_obj even after failing to
858 	 * remove this one.
859 	 */
860 	if (err != ENOENT)
861 		VERIFY0(err);
862 	ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
863 	    &count));
864 	ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
865 }
866 
867 
868 blkptr_t *
869 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
870 {
871 	return (&ds->ds_phys->ds_bp);
872 }
873 
874 void
875 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
876 {
877 	ASSERT(dmu_tx_is_syncing(tx));
878 	/* If it's the meta-objset, set dp_meta_rootbp */
879 	if (ds == NULL) {
880 		tx->tx_pool->dp_meta_rootbp = *bp;
881 	} else {
882 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
883 		ds->ds_phys->ds_bp = *bp;
884 	}
885 }
886 
887 spa_t *
888 dsl_dataset_get_spa(dsl_dataset_t *ds)
889 {
890 	return (ds->ds_dir->dd_pool->dp_spa);
891 }
892 
893 void
894 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
895 {
896 	dsl_pool_t *dp;
897 
898 	if (ds == NULL) /* this is the meta-objset */
899 		return;
900 
901 	ASSERT(ds->ds_objset != NULL);
902 
903 	if (ds->ds_phys->ds_next_snap_obj != 0)
904 		panic("dirtying snapshot!");
905 
906 	dp = ds->ds_dir->dd_pool;
907 
908 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
909 		/* up the hold count until we can be written out */
910 		dmu_buf_add_ref(ds->ds_dbuf, ds);
911 	}
912 }
913 
914 boolean_t
915 dsl_dataset_is_dirty(dsl_dataset_t *ds)
916 {
917 	for (int t = 0; t < TXG_SIZE; t++) {
918 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
919 		    ds, t))
920 			return (B_TRUE);
921 	}
922 	return (B_FALSE);
923 }
924 
925 static int
926 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
927 {
928 	uint64_t asize;
929 
930 	if (!dmu_tx_is_syncing(tx))
931 		return (0);
932 
933 	/*
934 	 * If there's an fs-only reservation, any blocks that might become
935 	 * owned by the snapshot dataset must be accommodated by space
936 	 * outside of the reservation.
937 	 */
938 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
939 	asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
940 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
941 		return (SET_ERROR(ENOSPC));
942 
943 	/*
944 	 * Propagate any reserved space for this snapshot to other
945 	 * snapshot checks in this sync group.
946 	 */
947 	if (asize > 0)
948 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
949 
950 	return (0);
951 }
952 
953 typedef struct dsl_dataset_snapshot_arg {
954 	nvlist_t *ddsa_snaps;
955 	nvlist_t *ddsa_props;
956 	nvlist_t *ddsa_errors;
957 	cred_t *ddsa_cr;
958 } dsl_dataset_snapshot_arg_t;
959 
960 int
961 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
962     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
963 {
964 	int error;
965 	uint64_t value;
966 
967 	ds->ds_trysnap_txg = tx->tx_txg;
968 
969 	if (!dmu_tx_is_syncing(tx))
970 		return (0);
971 
972 	/*
973 	 * We don't allow multiple snapshots of the same txg.  If there
974 	 * is already one, try again.
975 	 */
976 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
977 		return (SET_ERROR(EAGAIN));
978 
979 	/*
980 	 * Check for conflicting snapshot name.
981 	 */
982 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
983 	if (error == 0)
984 		return (SET_ERROR(EEXIST));
985 	if (error != ENOENT)
986 		return (error);
987 
988 	/*
989 	 * We don't allow taking snapshots of inconsistent datasets, such as
990 	 * those into which we are currently receiving.  However, if we are
991 	 * creating this snapshot as part of a receive, this check will be
992 	 * executed atomically with respect to the completion of the receive
993 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
994 	 * case we ignore this, knowing it will be fixed up for us shortly in
995 	 * dmu_recv_end_sync().
996 	 */
997 	if (!recv && DS_IS_INCONSISTENT(ds))
998 		return (SET_ERROR(EBUSY));
999 
1000 	/*
1001 	 * Skip the check for temporary snapshots or if we have already checked
1002 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1003 	 * check the count here when we're receiving a stream.
1004 	 */
1005 	if (cnt != 0 && cr != NULL) {
1006 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1007 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1008 		if (error != 0)
1009 			return (error);
1010 	}
1011 
1012 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1013 	if (error != 0)
1014 		return (error);
1015 
1016 	return (0);
1017 }
1018 
1019 static int
1020 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1021 {
1022 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1023 	dsl_pool_t *dp = dmu_tx_pool(tx);
1024 	nvpair_t *pair;
1025 	int rv = 0;
1026 
1027 	/*
1028 	 * Pre-compute how many total new snapshots will be created for each
1029 	 * level in the tree and below. This is needed for validating the
1030 	 * snapshot limit when either taking a recursive snapshot or when
1031 	 * taking multiple snapshots.
1032 	 *
1033 	 * The problem is that the counts are not actually adjusted when
1034 	 * we are checking, only when we finally sync. For a single snapshot,
1035 	 * this is easy, the count will increase by 1 at each node up the tree,
1036 	 * but its more complicated for the recursive/multiple snapshot case.
1037 	 *
1038 	 * The dsl_fs_ss_limit_check function does recursively check the count
1039 	 * at each level up the tree but since it is validating each snapshot
1040 	 * independently we need to be sure that we are validating the complete
1041 	 * count for the entire set of snapshots. We do this by rolling up the
1042 	 * counts for each component of the name into an nvlist and then
1043 	 * checking each of those cases with the aggregated count.
1044 	 *
1045 	 * This approach properly handles not only the recursive snapshot
1046 	 * case (where we get all of those on the ddsa_snaps list) but also
1047 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1048 	 * validate the limit on 'a' using a count of 2).
1049 	 *
1050 	 * We validate the snapshot names in the third loop and only report
1051 	 * name errors once.
1052 	 */
1053 	if (dmu_tx_is_syncing(tx)) {
1054 		nvlist_t *cnt_track = NULL;
1055 		cnt_track = fnvlist_alloc();
1056 
1057 		/* Rollup aggregated counts into the cnt_track list */
1058 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1059 		    pair != NULL;
1060 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1061 			char *pdelim;
1062 			uint64_t val;
1063 			char nm[MAXPATHLEN];
1064 
1065 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1066 			pdelim = strchr(nm, '@');
1067 			if (pdelim == NULL)
1068 				continue;
1069 			*pdelim = '\0';
1070 
1071 			do {
1072 				if (nvlist_lookup_uint64(cnt_track, nm,
1073 				    &val) == 0) {
1074 					/* update existing entry */
1075 					fnvlist_add_uint64(cnt_track, nm,
1076 					    val + 1);
1077 				} else {
1078 					/* add to list */
1079 					fnvlist_add_uint64(cnt_track, nm, 1);
1080 				}
1081 
1082 				pdelim = strrchr(nm, '/');
1083 				if (pdelim != NULL)
1084 					*pdelim = '\0';
1085 			} while (pdelim != NULL);
1086 		}
1087 
1088 		/* Check aggregated counts at each level */
1089 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1090 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1091 			int error = 0;
1092 			char *name;
1093 			uint64_t cnt = 0;
1094 			dsl_dataset_t *ds;
1095 
1096 			name = nvpair_name(pair);
1097 			cnt = fnvpair_value_uint64(pair);
1098 			ASSERT(cnt > 0);
1099 
1100 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1101 			if (error == 0) {
1102 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1103 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1104 				    ddsa->ddsa_cr);
1105 				dsl_dataset_rele(ds, FTAG);
1106 			}
1107 
1108 			if (error != 0) {
1109 				if (ddsa->ddsa_errors != NULL)
1110 					fnvlist_add_int32(ddsa->ddsa_errors,
1111 					    name, error);
1112 				rv = error;
1113 				/* only report one error for this check */
1114 				break;
1115 			}
1116 		}
1117 		nvlist_free(cnt_track);
1118 	}
1119 
1120 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1121 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1122 		int error = 0;
1123 		dsl_dataset_t *ds;
1124 		char *name, *atp;
1125 		char dsname[MAXNAMELEN];
1126 
1127 		name = nvpair_name(pair);
1128 		if (strlen(name) >= MAXNAMELEN)
1129 			error = SET_ERROR(ENAMETOOLONG);
1130 		if (error == 0) {
1131 			atp = strchr(name, '@');
1132 			if (atp == NULL)
1133 				error = SET_ERROR(EINVAL);
1134 			if (error == 0)
1135 				(void) strlcpy(dsname, name, atp - name + 1);
1136 		}
1137 		if (error == 0)
1138 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1139 		if (error == 0) {
1140 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1141 			error = dsl_dataset_snapshot_check_impl(ds,
1142 			    atp + 1, tx, B_FALSE, 0, NULL);
1143 			dsl_dataset_rele(ds, FTAG);
1144 		}
1145 
1146 		if (error != 0) {
1147 			if (ddsa->ddsa_errors != NULL) {
1148 				fnvlist_add_int32(ddsa->ddsa_errors,
1149 				    name, error);
1150 			}
1151 			rv = error;
1152 		}
1153 	}
1154 
1155 	return (rv);
1156 }
1157 
1158 void
1159 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1160     dmu_tx_t *tx)
1161 {
1162 	static zil_header_t zero_zil;
1163 
1164 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1165 	dmu_buf_t *dbuf;
1166 	dsl_dataset_phys_t *dsphys;
1167 	uint64_t dsobj, crtxg;
1168 	objset_t *mos = dp->dp_meta_objset;
1169 	objset_t *os;
1170 
1171 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1172 
1173 	/*
1174 	 * If we are on an old pool, the zil must not be active, in which
1175 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1176 	 */
1177 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1178 	    dmu_objset_from_ds(ds, &os) != 0 ||
1179 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1180 	    sizeof (zero_zil)) == 0);
1181 
1182 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1183 
1184 	/*
1185 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1186 	 */
1187 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1188 		crtxg = 1;
1189 	else
1190 		crtxg = tx->tx_txg;
1191 
1192 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1193 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1194 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1195 	dmu_buf_will_dirty(dbuf, tx);
1196 	dsphys = dbuf->db_data;
1197 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1198 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1199 	dsphys->ds_fsid_guid = unique_create();
1200 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1201 	    sizeof (dsphys->ds_guid));
1202 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1203 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1204 	dsphys->ds_next_snap_obj = ds->ds_object;
1205 	dsphys->ds_num_children = 1;
1206 	dsphys->ds_creation_time = gethrestime_sec();
1207 	dsphys->ds_creation_txg = crtxg;
1208 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1209 	dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
1210 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1211 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1212 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1213 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1214 	dmu_buf_rele(dbuf, FTAG);
1215 
1216 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1217 	if (ds->ds_prev) {
1218 		uint64_t next_clones_obj =
1219 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1220 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1221 		    ds->ds_object ||
1222 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1223 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1224 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1225 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1226 			    ds->ds_prev->ds_phys->ds_creation_txg);
1227 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1228 		} else if (next_clones_obj != 0) {
1229 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1230 			    dsphys->ds_next_snap_obj, tx);
1231 			VERIFY0(zap_add_int(mos,
1232 			    next_clones_obj, dsobj, tx));
1233 		}
1234 	}
1235 
1236 	/*
1237 	 * If we have a reference-reservation on this dataset, we will
1238 	 * need to increase the amount of refreservation being charged
1239 	 * since our unique space is going to zero.
1240 	 */
1241 	if (ds->ds_reserved) {
1242 		int64_t delta;
1243 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1244 		delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1245 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1246 		    delta, 0, 0, tx);
1247 	}
1248 
1249 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1250 	ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
1251 	    UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
1252 	dsl_deadlist_close(&ds->ds_deadlist);
1253 	dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1254 	dsl_deadlist_add_key(&ds->ds_deadlist,
1255 	    ds->ds_phys->ds_prev_snap_txg, tx);
1256 
1257 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1258 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1259 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1260 	ds->ds_phys->ds_unique_bytes = 0;
1261 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1262 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1263 
1264 	VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1265 	    snapname, 8, 1, &dsobj, tx));
1266 
1267 	if (ds->ds_prev)
1268 		dsl_dataset_rele(ds->ds_prev, ds);
1269 	VERIFY0(dsl_dataset_hold_obj(dp,
1270 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1271 
1272 	dsl_scan_ds_snapshotted(ds, tx);
1273 
1274 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1275 
1276 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1277 }
1278 
1279 static void
1280 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1281 {
1282 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1283 	dsl_pool_t *dp = dmu_tx_pool(tx);
1284 	nvpair_t *pair;
1285 
1286 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1287 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1288 		dsl_dataset_t *ds;
1289 		char *name, *atp;
1290 		char dsname[MAXNAMELEN];
1291 
1292 		name = nvpair_name(pair);
1293 		atp = strchr(name, '@');
1294 		(void) strlcpy(dsname, name, atp - name + 1);
1295 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1296 
1297 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1298 		if (ddsa->ddsa_props != NULL) {
1299 			dsl_props_set_sync_impl(ds->ds_prev,
1300 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1301 		}
1302 		dsl_dataset_rele(ds, FTAG);
1303 	}
1304 }
1305 
1306 /*
1307  * The snapshots must all be in the same pool.
1308  * All-or-nothing: if there are any failures, nothing will be modified.
1309  */
1310 int
1311 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1312 {
1313 	dsl_dataset_snapshot_arg_t ddsa;
1314 	nvpair_t *pair;
1315 	boolean_t needsuspend;
1316 	int error;
1317 	spa_t *spa;
1318 	char *firstname;
1319 	nvlist_t *suspended = NULL;
1320 
1321 	pair = nvlist_next_nvpair(snaps, NULL);
1322 	if (pair == NULL)
1323 		return (0);
1324 	firstname = nvpair_name(pair);
1325 
1326 	error = spa_open(firstname, &spa, FTAG);
1327 	if (error != 0)
1328 		return (error);
1329 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1330 	spa_close(spa, FTAG);
1331 
1332 	if (needsuspend) {
1333 		suspended = fnvlist_alloc();
1334 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1335 		    pair = nvlist_next_nvpair(snaps, pair)) {
1336 			char fsname[MAXNAMELEN];
1337 			char *snapname = nvpair_name(pair);
1338 			char *atp;
1339 			void *cookie;
1340 
1341 			atp = strchr(snapname, '@');
1342 			if (atp == NULL) {
1343 				error = SET_ERROR(EINVAL);
1344 				break;
1345 			}
1346 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1347 
1348 			error = zil_suspend(fsname, &cookie);
1349 			if (error != 0)
1350 				break;
1351 			fnvlist_add_uint64(suspended, fsname,
1352 			    (uintptr_t)cookie);
1353 		}
1354 	}
1355 
1356 	ddsa.ddsa_snaps = snaps;
1357 	ddsa.ddsa_props = props;
1358 	ddsa.ddsa_errors = errors;
1359 	ddsa.ddsa_cr = CRED();
1360 
1361 	if (error == 0) {
1362 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1363 		    dsl_dataset_snapshot_sync, &ddsa,
1364 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1365 	}
1366 
1367 	if (suspended != NULL) {
1368 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1369 		    pair = nvlist_next_nvpair(suspended, pair)) {
1370 			zil_resume((void *)(uintptr_t)
1371 			    fnvpair_value_uint64(pair));
1372 		}
1373 		fnvlist_free(suspended);
1374 	}
1375 
1376 	return (error);
1377 }
1378 
1379 typedef struct dsl_dataset_snapshot_tmp_arg {
1380 	const char *ddsta_fsname;
1381 	const char *ddsta_snapname;
1382 	minor_t ddsta_cleanup_minor;
1383 	const char *ddsta_htag;
1384 } dsl_dataset_snapshot_tmp_arg_t;
1385 
1386 static int
1387 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1388 {
1389 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1390 	dsl_pool_t *dp = dmu_tx_pool(tx);
1391 	dsl_dataset_t *ds;
1392 	int error;
1393 
1394 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1395 	if (error != 0)
1396 		return (error);
1397 
1398 	/* NULL cred means no limit check for tmp snapshot */
1399 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1400 	    tx, B_FALSE, 0, NULL);
1401 	if (error != 0) {
1402 		dsl_dataset_rele(ds, FTAG);
1403 		return (error);
1404 	}
1405 
1406 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1407 		dsl_dataset_rele(ds, FTAG);
1408 		return (SET_ERROR(ENOTSUP));
1409 	}
1410 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1411 	    B_TRUE, tx);
1412 	if (error != 0) {
1413 		dsl_dataset_rele(ds, FTAG);
1414 		return (error);
1415 	}
1416 
1417 	dsl_dataset_rele(ds, FTAG);
1418 	return (0);
1419 }
1420 
1421 static void
1422 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1423 {
1424 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1425 	dsl_pool_t *dp = dmu_tx_pool(tx);
1426 	dsl_dataset_t *ds;
1427 
1428 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1429 
1430 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1431 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1432 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1433 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1434 
1435 	dsl_dataset_rele(ds, FTAG);
1436 }
1437 
1438 int
1439 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1440     minor_t cleanup_minor, const char *htag)
1441 {
1442 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1443 	int error;
1444 	spa_t *spa;
1445 	boolean_t needsuspend;
1446 	void *cookie;
1447 
1448 	ddsta.ddsta_fsname = fsname;
1449 	ddsta.ddsta_snapname = snapname;
1450 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1451 	ddsta.ddsta_htag = htag;
1452 
1453 	error = spa_open(fsname, &spa, FTAG);
1454 	if (error != 0)
1455 		return (error);
1456 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1457 	spa_close(spa, FTAG);
1458 
1459 	if (needsuspend) {
1460 		error = zil_suspend(fsname, &cookie);
1461 		if (error != 0)
1462 			return (error);
1463 	}
1464 
1465 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1466 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1467 
1468 	if (needsuspend)
1469 		zil_resume(cookie);
1470 	return (error);
1471 }
1472 
1473 
1474 void
1475 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1476 {
1477 	ASSERT(dmu_tx_is_syncing(tx));
1478 	ASSERT(ds->ds_objset != NULL);
1479 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1480 
1481 	/*
1482 	 * in case we had to change ds_fsid_guid when we opened it,
1483 	 * sync it out now.
1484 	 */
1485 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1486 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1487 
1488 	dmu_objset_sync(ds->ds_objset, zio, tx);
1489 }
1490 
1491 static void
1492 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1493 {
1494 	uint64_t count = 0;
1495 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1496 	zap_cursor_t zc;
1497 	zap_attribute_t za;
1498 	nvlist_t *propval = fnvlist_alloc();
1499 	nvlist_t *val = fnvlist_alloc();
1500 
1501 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1502 
1503 	/*
1504 	 * There may be missing entries in ds_next_clones_obj
1505 	 * due to a bug in a previous version of the code.
1506 	 * Only trust it if it has the right number of entries.
1507 	 */
1508 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1509 		VERIFY0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1510 		    &count));
1511 	}
1512 	if (count != ds->ds_phys->ds_num_children - 1)
1513 		goto fail;
1514 	for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
1515 	    zap_cursor_retrieve(&zc, &za) == 0;
1516 	    zap_cursor_advance(&zc)) {
1517 		dsl_dataset_t *clone;
1518 		char buf[ZFS_MAXNAMELEN];
1519 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1520 		    za.za_first_integer, FTAG, &clone));
1521 		dsl_dir_name(clone->ds_dir, buf);
1522 		fnvlist_add_boolean(val, buf);
1523 		dsl_dataset_rele(clone, FTAG);
1524 	}
1525 	zap_cursor_fini(&zc);
1526 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1527 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1528 fail:
1529 	nvlist_free(val);
1530 	nvlist_free(propval);
1531 }
1532 
1533 void
1534 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1535 {
1536 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1537 	uint64_t refd, avail, uobjs, aobjs, ratio;
1538 
1539 	ASSERT(dsl_pool_config_held(dp));
1540 
1541 	ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1542 	    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1543 	    ds->ds_phys->ds_compressed_bytes);
1544 
1545 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1546 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1547 	    ds->ds_phys->ds_uncompressed_bytes);
1548 
1549 	if (dsl_dataset_is_snapshot(ds)) {
1550 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1551 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1552 		    ds->ds_phys->ds_unique_bytes);
1553 		get_clones_stat(ds, nv);
1554 	} else {
1555 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1556 			char buf[MAXNAMELEN];
1557 			dsl_dataset_name(ds->ds_prev, buf);
1558 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1559 		}
1560 
1561 		dsl_dir_stats(ds->ds_dir, nv);
1562 	}
1563 
1564 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1565 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1566 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1567 
1568 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1569 	    ds->ds_phys->ds_creation_time);
1570 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1571 	    ds->ds_phys->ds_creation_txg);
1572 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1573 	    ds->ds_quota);
1574 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1575 	    ds->ds_reserved);
1576 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1577 	    ds->ds_phys->ds_guid);
1578 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1579 	    ds->ds_phys->ds_unique_bytes);
1580 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1581 	    ds->ds_object);
1582 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1583 	    ds->ds_userrefs);
1584 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1585 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1586 
1587 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1588 		uint64_t written, comp, uncomp;
1589 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1590 		dsl_dataset_t *prev;
1591 
1592 		int err = dsl_dataset_hold_obj(dp,
1593 		    ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
1594 		if (err == 0) {
1595 			err = dsl_dataset_space_written(prev, ds, &written,
1596 			    &comp, &uncomp);
1597 			dsl_dataset_rele(prev, FTAG);
1598 			if (err == 0) {
1599 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1600 				    written);
1601 			}
1602 		}
1603 	}
1604 }
1605 
1606 void
1607 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1608 {
1609 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1610 	ASSERT(dsl_pool_config_held(dp));
1611 
1612 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1613 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1614 	stat->dds_guid = ds->ds_phys->ds_guid;
1615 	stat->dds_origin[0] = '\0';
1616 	if (dsl_dataset_is_snapshot(ds)) {
1617 		stat->dds_is_snapshot = B_TRUE;
1618 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1619 	} else {
1620 		stat->dds_is_snapshot = B_FALSE;
1621 		stat->dds_num_clones = 0;
1622 
1623 		if (dsl_dir_is_clone(ds->ds_dir)) {
1624 			dsl_dataset_t *ods;
1625 
1626 			VERIFY0(dsl_dataset_hold_obj(dp,
1627 			    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1628 			dsl_dataset_name(ods, stat->dds_origin);
1629 			dsl_dataset_rele(ods, FTAG);
1630 		}
1631 	}
1632 }
1633 
1634 uint64_t
1635 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1636 {
1637 	return (ds->ds_fsid_guid);
1638 }
1639 
1640 void
1641 dsl_dataset_space(dsl_dataset_t *ds,
1642     uint64_t *refdbytesp, uint64_t *availbytesp,
1643     uint64_t *usedobjsp, uint64_t *availobjsp)
1644 {
1645 	*refdbytesp = ds->ds_phys->ds_referenced_bytes;
1646 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1647 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1648 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1649 	if (ds->ds_quota != 0) {
1650 		/*
1651 		 * Adjust available bytes according to refquota
1652 		 */
1653 		if (*refdbytesp < ds->ds_quota)
1654 			*availbytesp = MIN(*availbytesp,
1655 			    ds->ds_quota - *refdbytesp);
1656 		else
1657 			*availbytesp = 0;
1658 	}
1659 	*usedobjsp = BP_GET_FILL(&ds->ds_phys->ds_bp);
1660 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1661 }
1662 
1663 boolean_t
1664 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1665 {
1666 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1667 
1668 	ASSERT(dsl_pool_config_held(dp));
1669 	if (snap == NULL)
1670 		return (B_FALSE);
1671 	if (ds->ds_phys->ds_bp.blk_birth >
1672 	    snap->ds_phys->ds_creation_txg) {
1673 		objset_t *os, *os_snap;
1674 		/*
1675 		 * It may be that only the ZIL differs, because it was
1676 		 * reset in the head.  Don't count that as being
1677 		 * modified.
1678 		 */
1679 		if (dmu_objset_from_ds(ds, &os) != 0)
1680 			return (B_TRUE);
1681 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1682 			return (B_TRUE);
1683 		return (bcmp(&os->os_phys->os_meta_dnode,
1684 		    &os_snap->os_phys->os_meta_dnode,
1685 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1686 	}
1687 	return (B_FALSE);
1688 }
1689 
1690 typedef struct dsl_dataset_rename_snapshot_arg {
1691 	const char *ddrsa_fsname;
1692 	const char *ddrsa_oldsnapname;
1693 	const char *ddrsa_newsnapname;
1694 	boolean_t ddrsa_recursive;
1695 	dmu_tx_t *ddrsa_tx;
1696 } dsl_dataset_rename_snapshot_arg_t;
1697 
1698 /* ARGSUSED */
1699 static int
1700 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1701     dsl_dataset_t *hds, void *arg)
1702 {
1703 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1704 	int error;
1705 	uint64_t val;
1706 
1707 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1708 	if (error != 0) {
1709 		/* ignore nonexistent snapshots */
1710 		return (error == ENOENT ? 0 : error);
1711 	}
1712 
1713 	/* new name should not exist */
1714 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1715 	if (error == 0)
1716 		error = SET_ERROR(EEXIST);
1717 	else if (error == ENOENT)
1718 		error = 0;
1719 
1720 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1721 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1722 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1723 		error = SET_ERROR(ENAMETOOLONG);
1724 
1725 	return (error);
1726 }
1727 
1728 static int
1729 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1730 {
1731 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1732 	dsl_pool_t *dp = dmu_tx_pool(tx);
1733 	dsl_dataset_t *hds;
1734 	int error;
1735 
1736 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1737 	if (error != 0)
1738 		return (error);
1739 
1740 	if (ddrsa->ddrsa_recursive) {
1741 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1742 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1743 		    DS_FIND_CHILDREN);
1744 	} else {
1745 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1746 	}
1747 	dsl_dataset_rele(hds, FTAG);
1748 	return (error);
1749 }
1750 
1751 static int
1752 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1753     dsl_dataset_t *hds, void *arg)
1754 {
1755 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1756 	dsl_dataset_t *ds;
1757 	uint64_t val;
1758 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1759 	int error;
1760 
1761 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1762 	ASSERT(error == 0 || error == ENOENT);
1763 	if (error == ENOENT) {
1764 		/* ignore nonexistent snapshots */
1765 		return (0);
1766 	}
1767 
1768 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1769 
1770 	/* log before we change the name */
1771 	spa_history_log_internal_ds(ds, "rename", tx,
1772 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1773 
1774 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1775 	    B_FALSE));
1776 	mutex_enter(&ds->ds_lock);
1777 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1778 	mutex_exit(&ds->ds_lock);
1779 	VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj,
1780 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1781 
1782 	dsl_dataset_rele(ds, FTAG);
1783 	return (0);
1784 }
1785 
1786 static void
1787 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1788 {
1789 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1790 	dsl_pool_t *dp = dmu_tx_pool(tx);
1791 	dsl_dataset_t *hds;
1792 
1793 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1794 	ddrsa->ddrsa_tx = tx;
1795 	if (ddrsa->ddrsa_recursive) {
1796 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1797 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1798 		    DS_FIND_CHILDREN));
1799 	} else {
1800 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1801 	}
1802 	dsl_dataset_rele(hds, FTAG);
1803 }
1804 
1805 int
1806 dsl_dataset_rename_snapshot(const char *fsname,
1807     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1808 {
1809 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1810 
1811 	ddrsa.ddrsa_fsname = fsname;
1812 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1813 	ddrsa.ddrsa_newsnapname = newsnapname;
1814 	ddrsa.ddrsa_recursive = recursive;
1815 
1816 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1817 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
1818 	    1, ZFS_SPACE_CHECK_RESERVED));
1819 }
1820 
1821 /*
1822  * If we're doing an ownership handoff, we need to make sure that there is
1823  * only one long hold on the dataset.  We're not allowed to change anything here
1824  * so we don't permanently release the long hold or regular hold here.  We want
1825  * to do this only when syncing to avoid the dataset unexpectedly going away
1826  * when we release the long hold.
1827  */
1828 static int
1829 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1830 {
1831 	boolean_t held;
1832 
1833 	if (!dmu_tx_is_syncing(tx))
1834 		return (0);
1835 
1836 	if (owner != NULL) {
1837 		VERIFY3P(ds->ds_owner, ==, owner);
1838 		dsl_dataset_long_rele(ds, owner);
1839 	}
1840 
1841 	held = dsl_dataset_long_held(ds);
1842 
1843 	if (owner != NULL)
1844 		dsl_dataset_long_hold(ds, owner);
1845 
1846 	if (held)
1847 		return (SET_ERROR(EBUSY));
1848 
1849 	return (0);
1850 }
1851 
1852 typedef struct dsl_dataset_rollback_arg {
1853 	const char *ddra_fsname;
1854 	void *ddra_owner;
1855 	nvlist_t *ddra_result;
1856 } dsl_dataset_rollback_arg_t;
1857 
1858 static int
1859 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1860 {
1861 	dsl_dataset_rollback_arg_t *ddra = arg;
1862 	dsl_pool_t *dp = dmu_tx_pool(tx);
1863 	dsl_dataset_t *ds;
1864 	int64_t unused_refres_delta;
1865 	int error;
1866 
1867 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1868 	if (error != 0)
1869 		return (error);
1870 
1871 	/* must not be a snapshot */
1872 	if (dsl_dataset_is_snapshot(ds)) {
1873 		dsl_dataset_rele(ds, FTAG);
1874 		return (SET_ERROR(EINVAL));
1875 	}
1876 
1877 	/* must have a most recent snapshot */
1878 	if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) {
1879 		dsl_dataset_rele(ds, FTAG);
1880 		return (SET_ERROR(EINVAL));
1881 	}
1882 
1883 	/* must not have any bookmarks after the most recent snapshot */
1884 	nvlist_t *proprequest = fnvlist_alloc();
1885 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
1886 	nvlist_t *bookmarks = fnvlist_alloc();
1887 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
1888 	fnvlist_free(proprequest);
1889 	if (error != 0)
1890 		return (error);
1891 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
1892 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
1893 		nvlist_t *valuenv =
1894 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
1895 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
1896 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
1897 		if (createtxg > ds->ds_phys->ds_prev_snap_txg) {
1898 			fnvlist_free(bookmarks);
1899 			dsl_dataset_rele(ds, FTAG);
1900 			return (SET_ERROR(EEXIST));
1901 		}
1902 	}
1903 	fnvlist_free(bookmarks);
1904 
1905 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
1906 	if (error != 0) {
1907 		dsl_dataset_rele(ds, FTAG);
1908 		return (error);
1909 	}
1910 
1911 	/*
1912 	 * Check if the snap we are rolling back to uses more than
1913 	 * the refquota.
1914 	 */
1915 	if (ds->ds_quota != 0 &&
1916 	    ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) {
1917 		dsl_dataset_rele(ds, FTAG);
1918 		return (SET_ERROR(EDQUOT));
1919 	}
1920 
1921 	/*
1922 	 * When we do the clone swap, we will temporarily use more space
1923 	 * due to the refreservation (the head will no longer have any
1924 	 * unique space, so the entire amount of the refreservation will need
1925 	 * to be free).  We will immediately destroy the clone, freeing
1926 	 * this space, but the freeing happens over many txg's.
1927 	 */
1928 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1929 	    ds->ds_phys->ds_unique_bytes);
1930 
1931 	if (unused_refres_delta > 0 &&
1932 	    unused_refres_delta >
1933 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1934 		dsl_dataset_rele(ds, FTAG);
1935 		return (SET_ERROR(ENOSPC));
1936 	}
1937 
1938 	dsl_dataset_rele(ds, FTAG);
1939 	return (0);
1940 }
1941 
1942 static void
1943 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1944 {
1945 	dsl_dataset_rollback_arg_t *ddra = arg;
1946 	dsl_pool_t *dp = dmu_tx_pool(tx);
1947 	dsl_dataset_t *ds, *clone;
1948 	uint64_t cloneobj;
1949 	char namebuf[ZFS_MAXNAMELEN];
1950 
1951 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
1952 
1953 	dsl_dataset_name(ds->ds_prev, namebuf);
1954 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
1955 
1956 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
1957 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
1958 
1959 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
1960 
1961 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
1962 	dsl_dataset_zero_zil(ds, tx);
1963 
1964 	dsl_destroy_head_sync_impl(clone, tx);
1965 
1966 	dsl_dataset_rele(clone, FTAG);
1967 	dsl_dataset_rele(ds, FTAG);
1968 }
1969 
1970 /*
1971  * Rolls back the given filesystem or volume to the most recent snapshot.
1972  * The name of the most recent snapshot will be returned under key "target"
1973  * in the result nvlist.
1974  *
1975  * If owner != NULL:
1976  * - The existing dataset MUST be owned by the specified owner at entry
1977  * - Upon return, dataset will still be held by the same owner, whether we
1978  *   succeed or not.
1979  *
1980  * This mode is required any time the existing filesystem is mounted.  See
1981  * notes above zfs_suspend_fs() for further details.
1982  */
1983 int
1984 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
1985 {
1986 	dsl_dataset_rollback_arg_t ddra;
1987 
1988 	ddra.ddra_fsname = fsname;
1989 	ddra.ddra_owner = owner;
1990 	ddra.ddra_result = result;
1991 
1992 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
1993 	    dsl_dataset_rollback_sync, &ddra,
1994 	    1, ZFS_SPACE_CHECK_RESERVED));
1995 }
1996 
1997 struct promotenode {
1998 	list_node_t link;
1999 	dsl_dataset_t *ds;
2000 };
2001 
2002 typedef struct dsl_dataset_promote_arg {
2003 	const char *ddpa_clonename;
2004 	dsl_dataset_t *ddpa_clone;
2005 	list_t shared_snaps, origin_snaps, clone_snaps;
2006 	dsl_dataset_t *origin_origin; /* origin of the origin */
2007 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2008 	char *err_ds;
2009 	cred_t *cr;
2010 } dsl_dataset_promote_arg_t;
2011 
2012 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2013 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2014     void *tag);
2015 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2016 
2017 static int
2018 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2019 {
2020 	dsl_dataset_promote_arg_t *ddpa = arg;
2021 	dsl_pool_t *dp = dmu_tx_pool(tx);
2022 	dsl_dataset_t *hds;
2023 	struct promotenode *snap;
2024 	dsl_dataset_t *origin_ds;
2025 	int err;
2026 	uint64_t unused;
2027 	uint64_t ss_mv_cnt;
2028 
2029 	err = promote_hold(ddpa, dp, FTAG);
2030 	if (err != 0)
2031 		return (err);
2032 
2033 	hds = ddpa->ddpa_clone;
2034 
2035 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
2036 		promote_rele(ddpa, FTAG);
2037 		return (SET_ERROR(EXDEV));
2038 	}
2039 
2040 	/*
2041 	 * Compute and check the amount of space to transfer.  Since this is
2042 	 * so expensive, don't do the preliminary check.
2043 	 */
2044 	if (!dmu_tx_is_syncing(tx)) {
2045 		promote_rele(ddpa, FTAG);
2046 		return (0);
2047 	}
2048 
2049 	snap = list_head(&ddpa->shared_snaps);
2050 	origin_ds = snap->ds;
2051 
2052 	/* compute origin's new unique space */
2053 	snap = list_tail(&ddpa->clone_snaps);
2054 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2055 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2056 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2057 	    &ddpa->unique, &unused, &unused);
2058 
2059 	/*
2060 	 * Walk the snapshots that we are moving
2061 	 *
2062 	 * Compute space to transfer.  Consider the incremental changes
2063 	 * to used by each snapshot:
2064 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2065 	 * So each snapshot gave birth to:
2066 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2067 	 * So a sequence would look like:
2068 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2069 	 * Which simplifies to:
2070 	 * uN + kN + kN-1 + ... + k1 + k0
2071 	 * Note however, if we stop before we reach the ORIGIN we get:
2072 	 * uN + kN + kN-1 + ... + kM - uM-1
2073 	 */
2074 	ss_mv_cnt = 0;
2075 	ddpa->used = origin_ds->ds_phys->ds_referenced_bytes;
2076 	ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2077 	ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2078 	for (snap = list_head(&ddpa->shared_snaps); snap;
2079 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2080 		uint64_t val, dlused, dlcomp, dluncomp;
2081 		dsl_dataset_t *ds = snap->ds;
2082 
2083 		ss_mv_cnt++;
2084 
2085 		/*
2086 		 * If there are long holds, we won't be able to evict
2087 		 * the objset.
2088 		 */
2089 		if (dsl_dataset_long_held(ds)) {
2090 			err = SET_ERROR(EBUSY);
2091 			goto out;
2092 		}
2093 
2094 		/* Check that the snapshot name does not conflict */
2095 		VERIFY0(dsl_dataset_get_snapname(ds));
2096 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2097 		if (err == 0) {
2098 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2099 			err = SET_ERROR(EEXIST);
2100 			goto out;
2101 		}
2102 		if (err != ENOENT)
2103 			goto out;
2104 
2105 		/* The very first snapshot does not have a deadlist */
2106 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2107 			continue;
2108 
2109 		dsl_deadlist_space(&ds->ds_deadlist,
2110 		    &dlused, &dlcomp, &dluncomp);
2111 		ddpa->used += dlused;
2112 		ddpa->comp += dlcomp;
2113 		ddpa->uncomp += dluncomp;
2114 	}
2115 
2116 	/*
2117 	 * If we are a clone of a clone then we never reached ORIGIN,
2118 	 * so we need to subtract out the clone origin's used space.
2119 	 */
2120 	if (ddpa->origin_origin) {
2121 		ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes;
2122 		ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes;
2123 		ddpa->uncomp -=
2124 		    ddpa->origin_origin->ds_phys->ds_uncompressed_bytes;
2125 	}
2126 
2127 	/* Check that there is enough space and limit headroom here */
2128 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2129 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2130 	if (err != 0)
2131 		goto out;
2132 
2133 	/*
2134 	 * Compute the amounts of space that will be used by snapshots
2135 	 * after the promotion (for both origin and clone).  For each,
2136 	 * it is the amount of space that will be on all of their
2137 	 * deadlists (that was not born before their new origin).
2138 	 */
2139 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2140 		uint64_t space;
2141 
2142 		/*
2143 		 * Note, typically this will not be a clone of a clone,
2144 		 * so dd_origin_txg will be < TXG_INITIAL, so
2145 		 * these snaplist_space() -> dsl_deadlist_space_range()
2146 		 * calls will be fast because they do not have to
2147 		 * iterate over all bps.
2148 		 */
2149 		snap = list_head(&ddpa->origin_snaps);
2150 		err = snaplist_space(&ddpa->shared_snaps,
2151 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2152 		if (err != 0)
2153 			goto out;
2154 
2155 		err = snaplist_space(&ddpa->clone_snaps,
2156 		    snap->ds->ds_dir->dd_origin_txg, &space);
2157 		if (err != 0)
2158 			goto out;
2159 		ddpa->cloneusedsnap += space;
2160 	}
2161 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2162 		err = snaplist_space(&ddpa->origin_snaps,
2163 		    origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap);
2164 		if (err != 0)
2165 			goto out;
2166 	}
2167 
2168 out:
2169 	promote_rele(ddpa, FTAG);
2170 	return (err);
2171 }
2172 
2173 static void
2174 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2175 {
2176 	dsl_dataset_promote_arg_t *ddpa = arg;
2177 	dsl_pool_t *dp = dmu_tx_pool(tx);
2178 	dsl_dataset_t *hds;
2179 	struct promotenode *snap;
2180 	dsl_dataset_t *origin_ds;
2181 	dsl_dataset_t *origin_head;
2182 	dsl_dir_t *dd;
2183 	dsl_dir_t *odd = NULL;
2184 	uint64_t oldnext_obj;
2185 	int64_t delta;
2186 
2187 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2188 	hds = ddpa->ddpa_clone;
2189 
2190 	ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE);
2191 
2192 	snap = list_head(&ddpa->shared_snaps);
2193 	origin_ds = snap->ds;
2194 	dd = hds->ds_dir;
2195 
2196 	snap = list_head(&ddpa->origin_snaps);
2197 	origin_head = snap->ds;
2198 
2199 	/*
2200 	 * We need to explicitly open odd, since origin_ds's dd will be
2201 	 * changing.
2202 	 */
2203 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2204 	    NULL, FTAG, &odd));
2205 
2206 	/* change origin's next snap */
2207 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2208 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2209 	snap = list_tail(&ddpa->clone_snaps);
2210 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2211 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2212 
2213 	/* change the origin's next clone */
2214 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2215 		dsl_dataset_remove_from_next_clones(origin_ds,
2216 		    snap->ds->ds_object, tx);
2217 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2218 		    origin_ds->ds_phys->ds_next_clones_obj,
2219 		    oldnext_obj, tx));
2220 	}
2221 
2222 	/* change origin */
2223 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2224 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2225 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2226 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2227 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2228 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2229 	origin_head->ds_dir->dd_origin_txg =
2230 	    origin_ds->ds_phys->ds_creation_txg;
2231 
2232 	/* change dd_clone entries */
2233 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2234 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2235 		    odd->dd_phys->dd_clones, hds->ds_object, tx));
2236 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2237 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2238 		    hds->ds_object, tx));
2239 
2240 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2241 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2242 		    origin_head->ds_object, tx));
2243 		if (dd->dd_phys->dd_clones == 0) {
2244 			dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2245 			    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2246 		}
2247 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2248 		    dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2249 	}
2250 
2251 	/* move snapshots to this dir */
2252 	for (snap = list_head(&ddpa->shared_snaps); snap;
2253 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2254 		dsl_dataset_t *ds = snap->ds;
2255 
2256 		/*
2257 		 * Property callbacks are registered to a particular
2258 		 * dsl_dir.  Since ours is changing, evict the objset
2259 		 * so that they will be unregistered from the old dsl_dir.
2260 		 */
2261 		if (ds->ds_objset) {
2262 			dmu_objset_evict(ds->ds_objset);
2263 			ds->ds_objset = NULL;
2264 		}
2265 
2266 		/* move snap name entry */
2267 		VERIFY0(dsl_dataset_get_snapname(ds));
2268 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2269 		    ds->ds_snapname, tx, B_TRUE));
2270 		VERIFY0(zap_add(dp->dp_meta_objset,
2271 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2272 		    8, 1, &ds->ds_object, tx));
2273 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2274 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2275 
2276 		/* change containing dsl_dir */
2277 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2278 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2279 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2280 		ASSERT3P(ds->ds_dir, ==, odd);
2281 		dsl_dir_rele(ds->ds_dir, ds);
2282 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2283 		    NULL, ds, &ds->ds_dir));
2284 
2285 		/* move any clone references */
2286 		if (ds->ds_phys->ds_next_clones_obj &&
2287 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2288 			zap_cursor_t zc;
2289 			zap_attribute_t za;
2290 
2291 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2292 			    ds->ds_phys->ds_next_clones_obj);
2293 			    zap_cursor_retrieve(&zc, &za) == 0;
2294 			    zap_cursor_advance(&zc)) {
2295 				dsl_dataset_t *cnds;
2296 				uint64_t o;
2297 
2298 				if (za.za_first_integer == oldnext_obj) {
2299 					/*
2300 					 * We've already moved the
2301 					 * origin's reference.
2302 					 */
2303 					continue;
2304 				}
2305 
2306 				VERIFY0(dsl_dataset_hold_obj(dp,
2307 				    za.za_first_integer, FTAG, &cnds));
2308 				o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2309 
2310 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2311 				    odd->dd_phys->dd_clones, o, tx));
2312 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2313 				    dd->dd_phys->dd_clones, o, tx));
2314 				dsl_dataset_rele(cnds, FTAG);
2315 			}
2316 			zap_cursor_fini(&zc);
2317 		}
2318 
2319 		ASSERT(!dsl_prop_hascb(ds));
2320 	}
2321 
2322 	/*
2323 	 * Change space accounting.
2324 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2325 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2326 	 * is true for each of {clone,origin} independently.
2327 	 */
2328 
2329 	delta = ddpa->cloneusedsnap -
2330 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2331 	ASSERT3S(delta, >=, 0);
2332 	ASSERT3U(ddpa->used, >=, delta);
2333 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2334 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2335 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2336 
2337 	delta = ddpa->originusedsnap -
2338 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2339 	ASSERT3S(delta, <=, 0);
2340 	ASSERT3U(ddpa->used, >=, -delta);
2341 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2342 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2343 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2344 
2345 	origin_ds->ds_phys->ds_unique_bytes = ddpa->unique;
2346 
2347 	/* log history record */
2348 	spa_history_log_internal_ds(hds, "promote", tx, "");
2349 
2350 	dsl_dir_rele(odd, FTAG);
2351 	promote_rele(ddpa, FTAG);
2352 }
2353 
2354 /*
2355  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2356  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2357  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2358  * snapshots back to this dataset's origin.
2359  */
2360 static int
2361 snaplist_make(dsl_pool_t *dp,
2362     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2363 {
2364 	uint64_t obj = last_obj;
2365 
2366 	list_create(l, sizeof (struct promotenode),
2367 	    offsetof(struct promotenode, link));
2368 
2369 	while (obj != first_obj) {
2370 		dsl_dataset_t *ds;
2371 		struct promotenode *snap;
2372 		int err;
2373 
2374 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2375 		ASSERT(err != ENOENT);
2376 		if (err != 0)
2377 			return (err);
2378 
2379 		if (first_obj == 0)
2380 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2381 
2382 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2383 		snap->ds = ds;
2384 		list_insert_tail(l, snap);
2385 		obj = ds->ds_phys->ds_prev_snap_obj;
2386 	}
2387 
2388 	return (0);
2389 }
2390 
2391 static int
2392 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2393 {
2394 	struct promotenode *snap;
2395 
2396 	*spacep = 0;
2397 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2398 		uint64_t used, comp, uncomp;
2399 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2400 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2401 		*spacep += used;
2402 	}
2403 	return (0);
2404 }
2405 
2406 static void
2407 snaplist_destroy(list_t *l, void *tag)
2408 {
2409 	struct promotenode *snap;
2410 
2411 	if (l == NULL || !list_link_active(&l->list_head))
2412 		return;
2413 
2414 	while ((snap = list_tail(l)) != NULL) {
2415 		list_remove(l, snap);
2416 		dsl_dataset_rele(snap->ds, tag);
2417 		kmem_free(snap, sizeof (*snap));
2418 	}
2419 	list_destroy(l);
2420 }
2421 
2422 static int
2423 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2424 {
2425 	int error;
2426 	dsl_dir_t *dd;
2427 	struct promotenode *snap;
2428 
2429 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2430 	    &ddpa->ddpa_clone);
2431 	if (error != 0)
2432 		return (error);
2433 	dd = ddpa->ddpa_clone->ds_dir;
2434 
2435 	if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2436 	    !dsl_dir_is_clone(dd)) {
2437 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2438 		return (SET_ERROR(EINVAL));
2439 	}
2440 
2441 	error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj,
2442 	    &ddpa->shared_snaps, tag);
2443 	if (error != 0)
2444 		goto out;
2445 
2446 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2447 	    &ddpa->clone_snaps, tag);
2448 	if (error != 0)
2449 		goto out;
2450 
2451 	snap = list_head(&ddpa->shared_snaps);
2452 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2453 	error = snaplist_make(dp, dd->dd_phys->dd_origin_obj,
2454 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj,
2455 	    &ddpa->origin_snaps, tag);
2456 	if (error != 0)
2457 		goto out;
2458 
2459 	if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2460 		error = dsl_dataset_hold_obj(dp,
2461 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2462 		    tag, &ddpa->origin_origin);
2463 		if (error != 0)
2464 			goto out;
2465 	}
2466 out:
2467 	if (error != 0)
2468 		promote_rele(ddpa, tag);
2469 	return (error);
2470 }
2471 
2472 static void
2473 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2474 {
2475 	snaplist_destroy(&ddpa->shared_snaps, tag);
2476 	snaplist_destroy(&ddpa->clone_snaps, tag);
2477 	snaplist_destroy(&ddpa->origin_snaps, tag);
2478 	if (ddpa->origin_origin != NULL)
2479 		dsl_dataset_rele(ddpa->origin_origin, tag);
2480 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2481 }
2482 
2483 /*
2484  * Promote a clone.
2485  *
2486  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2487  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2488  */
2489 int
2490 dsl_dataset_promote(const char *name, char *conflsnap)
2491 {
2492 	dsl_dataset_promote_arg_t ddpa = { 0 };
2493 	uint64_t numsnaps;
2494 	int error;
2495 	objset_t *os;
2496 
2497 	/*
2498 	 * We will modify space proportional to the number of
2499 	 * snapshots.  Compute numsnaps.
2500 	 */
2501 	error = dmu_objset_hold(name, FTAG, &os);
2502 	if (error != 0)
2503 		return (error);
2504 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2505 	    dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps);
2506 	dmu_objset_rele(os, FTAG);
2507 	if (error != 0)
2508 		return (error);
2509 
2510 	ddpa.ddpa_clonename = name;
2511 	ddpa.err_ds = conflsnap;
2512 	ddpa.cr = CRED();
2513 
2514 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2515 	    dsl_dataset_promote_sync, &ddpa,
2516 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2517 }
2518 
2519 int
2520 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2521     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2522 {
2523 	int64_t unused_refres_delta;
2524 
2525 	/* they should both be heads */
2526 	if (dsl_dataset_is_snapshot(clone) ||
2527 	    dsl_dataset_is_snapshot(origin_head))
2528 		return (SET_ERROR(EINVAL));
2529 
2530 	/* if we are not forcing, the branch point should be just before them */
2531 	if (!force && clone->ds_prev != origin_head->ds_prev)
2532 		return (SET_ERROR(EINVAL));
2533 
2534 	/* clone should be the clone (unless they are unrelated) */
2535 	if (clone->ds_prev != NULL &&
2536 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2537 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2538 		return (SET_ERROR(EINVAL));
2539 
2540 	/* the clone should be a child of the origin */
2541 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2542 		return (SET_ERROR(EINVAL));
2543 
2544 	/* origin_head shouldn't be modified unless 'force' */
2545 	if (!force &&
2546 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2547 		return (SET_ERROR(ETXTBSY));
2548 
2549 	/* origin_head should have no long holds (e.g. is not mounted) */
2550 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2551 		return (SET_ERROR(EBUSY));
2552 
2553 	/* check amount of any unconsumed refreservation */
2554 	unused_refres_delta =
2555 	    (int64_t)MIN(origin_head->ds_reserved,
2556 	    origin_head->ds_phys->ds_unique_bytes) -
2557 	    (int64_t)MIN(origin_head->ds_reserved,
2558 	    clone->ds_phys->ds_unique_bytes);
2559 
2560 	if (unused_refres_delta > 0 &&
2561 	    unused_refres_delta >
2562 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2563 		return (SET_ERROR(ENOSPC));
2564 
2565 	/* clone can't be over the head's refquota */
2566 	if (origin_head->ds_quota != 0 &&
2567 	    clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota)
2568 		return (SET_ERROR(EDQUOT));
2569 
2570 	return (0);
2571 }
2572 
2573 void
2574 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2575     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2576 {
2577 	dsl_pool_t *dp = dmu_tx_pool(tx);
2578 	int64_t unused_refres_delta;
2579 
2580 	ASSERT(clone->ds_reserved == 0);
2581 	ASSERT(origin_head->ds_quota == 0 ||
2582 	    clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota);
2583 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2584 
2585 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2586 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2587 
2588 	if (clone->ds_objset != NULL) {
2589 		dmu_objset_evict(clone->ds_objset);
2590 		clone->ds_objset = NULL;
2591 	}
2592 
2593 	if (origin_head->ds_objset != NULL) {
2594 		dmu_objset_evict(origin_head->ds_objset);
2595 		origin_head->ds_objset = NULL;
2596 	}
2597 
2598 	unused_refres_delta =
2599 	    (int64_t)MIN(origin_head->ds_reserved,
2600 	    origin_head->ds_phys->ds_unique_bytes) -
2601 	    (int64_t)MIN(origin_head->ds_reserved,
2602 	    clone->ds_phys->ds_unique_bytes);
2603 
2604 	/*
2605 	 * Reset origin's unique bytes, if it exists.
2606 	 */
2607 	if (clone->ds_prev) {
2608 		dsl_dataset_t *origin = clone->ds_prev;
2609 		uint64_t comp, uncomp;
2610 
2611 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2612 		dsl_deadlist_space_range(&clone->ds_deadlist,
2613 		    origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2614 		    &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
2615 	}
2616 
2617 	/* swap blkptrs */
2618 	{
2619 		blkptr_t tmp;
2620 		tmp = origin_head->ds_phys->ds_bp;
2621 		origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp;
2622 		clone->ds_phys->ds_bp = tmp;
2623 	}
2624 
2625 	/* set dd_*_bytes */
2626 	{
2627 		int64_t dused, dcomp, duncomp;
2628 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2629 		uint64_t odl_used, odl_comp, odl_uncomp;
2630 
2631 		ASSERT3U(clone->ds_dir->dd_phys->
2632 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2633 
2634 		dsl_deadlist_space(&clone->ds_deadlist,
2635 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2636 		dsl_deadlist_space(&origin_head->ds_deadlist,
2637 		    &odl_used, &odl_comp, &odl_uncomp);
2638 
2639 		dused = clone->ds_phys->ds_referenced_bytes + cdl_used -
2640 		    (origin_head->ds_phys->ds_referenced_bytes + odl_used);
2641 		dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp -
2642 		    (origin_head->ds_phys->ds_compressed_bytes + odl_comp);
2643 		duncomp = clone->ds_phys->ds_uncompressed_bytes +
2644 		    cdl_uncomp -
2645 		    (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2646 
2647 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2648 		    dused, dcomp, duncomp, tx);
2649 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2650 		    -dused, -dcomp, -duncomp, tx);
2651 
2652 		/*
2653 		 * The difference in the space used by snapshots is the
2654 		 * difference in snapshot space due to the head's
2655 		 * deadlist (since that's the only thing that's
2656 		 * changing that affects the snapused).
2657 		 */
2658 		dsl_deadlist_space_range(&clone->ds_deadlist,
2659 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2660 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2661 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2662 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2663 		    &odl_used, &odl_comp, &odl_uncomp);
2664 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2665 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2666 	}
2667 
2668 	/* swap ds_*_bytes */
2669 	SWITCH64(origin_head->ds_phys->ds_referenced_bytes,
2670 	    clone->ds_phys->ds_referenced_bytes);
2671 	SWITCH64(origin_head->ds_phys->ds_compressed_bytes,
2672 	    clone->ds_phys->ds_compressed_bytes);
2673 	SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes,
2674 	    clone->ds_phys->ds_uncompressed_bytes);
2675 	SWITCH64(origin_head->ds_phys->ds_unique_bytes,
2676 	    clone->ds_phys->ds_unique_bytes);
2677 
2678 	/* apply any parent delta for change in unconsumed refreservation */
2679 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2680 	    unused_refres_delta, 0, 0, tx);
2681 
2682 	/*
2683 	 * Swap deadlists.
2684 	 */
2685 	dsl_deadlist_close(&clone->ds_deadlist);
2686 	dsl_deadlist_close(&origin_head->ds_deadlist);
2687 	SWITCH64(origin_head->ds_phys->ds_deadlist_obj,
2688 	    clone->ds_phys->ds_deadlist_obj);
2689 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2690 	    clone->ds_phys->ds_deadlist_obj);
2691 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2692 	    origin_head->ds_phys->ds_deadlist_obj);
2693 
2694 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2695 
2696 	spa_history_log_internal_ds(clone, "clone swap", tx,
2697 	    "parent=%s", origin_head->ds_dir->dd_myname);
2698 }
2699 
2700 /*
2701  * Given a pool name and a dataset object number in that pool,
2702  * return the name of that dataset.
2703  */
2704 int
2705 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2706 {
2707 	dsl_pool_t *dp;
2708 	dsl_dataset_t *ds;
2709 	int error;
2710 
2711 	error = dsl_pool_hold(pname, FTAG, &dp);
2712 	if (error != 0)
2713 		return (error);
2714 
2715 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2716 	if (error == 0) {
2717 		dsl_dataset_name(ds, buf);
2718 		dsl_dataset_rele(ds, FTAG);
2719 	}
2720 	dsl_pool_rele(dp, FTAG);
2721 
2722 	return (error);
2723 }
2724 
2725 int
2726 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2727     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2728 {
2729 	int error = 0;
2730 
2731 	ASSERT3S(asize, >, 0);
2732 
2733 	/*
2734 	 * *ref_rsrv is the portion of asize that will come from any
2735 	 * unconsumed refreservation space.
2736 	 */
2737 	*ref_rsrv = 0;
2738 
2739 	mutex_enter(&ds->ds_lock);
2740 	/*
2741 	 * Make a space adjustment for reserved bytes.
2742 	 */
2743 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2744 		ASSERT3U(*used, >=,
2745 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2746 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2747 		*ref_rsrv =
2748 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2749 	}
2750 
2751 	if (!check_quota || ds->ds_quota == 0) {
2752 		mutex_exit(&ds->ds_lock);
2753 		return (0);
2754 	}
2755 	/*
2756 	 * If they are requesting more space, and our current estimate
2757 	 * is over quota, they get to try again unless the actual
2758 	 * on-disk is over quota and there are no pending changes (which
2759 	 * may free up space for us).
2760 	 */
2761 	if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
2762 		if (inflight > 0 ||
2763 		    ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
2764 			error = SET_ERROR(ERESTART);
2765 		else
2766 			error = SET_ERROR(EDQUOT);
2767 	}
2768 	mutex_exit(&ds->ds_lock);
2769 
2770 	return (error);
2771 }
2772 
2773 typedef struct dsl_dataset_set_qr_arg {
2774 	const char *ddsqra_name;
2775 	zprop_source_t ddsqra_source;
2776 	uint64_t ddsqra_value;
2777 } dsl_dataset_set_qr_arg_t;
2778 
2779 
2780 /* ARGSUSED */
2781 static int
2782 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2783 {
2784 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2785 	dsl_pool_t *dp = dmu_tx_pool(tx);
2786 	dsl_dataset_t *ds;
2787 	int error;
2788 	uint64_t newval;
2789 
2790 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2791 		return (SET_ERROR(ENOTSUP));
2792 
2793 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2794 	if (error != 0)
2795 		return (error);
2796 
2797 	if (dsl_dataset_is_snapshot(ds)) {
2798 		dsl_dataset_rele(ds, FTAG);
2799 		return (SET_ERROR(EINVAL));
2800 	}
2801 
2802 	error = dsl_prop_predict(ds->ds_dir,
2803 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2804 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2805 	if (error != 0) {
2806 		dsl_dataset_rele(ds, FTAG);
2807 		return (error);
2808 	}
2809 
2810 	if (newval == 0) {
2811 		dsl_dataset_rele(ds, FTAG);
2812 		return (0);
2813 	}
2814 
2815 	if (newval < ds->ds_phys->ds_referenced_bytes ||
2816 	    newval < ds->ds_reserved) {
2817 		dsl_dataset_rele(ds, FTAG);
2818 		return (SET_ERROR(ENOSPC));
2819 	}
2820 
2821 	dsl_dataset_rele(ds, FTAG);
2822 	return (0);
2823 }
2824 
2825 static void
2826 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2827 {
2828 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2829 	dsl_pool_t *dp = dmu_tx_pool(tx);
2830 	dsl_dataset_t *ds;
2831 	uint64_t newval;
2832 
2833 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2834 
2835 	dsl_prop_set_sync_impl(ds,
2836 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2837 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2838 	    &ddsqra->ddsqra_value, tx);
2839 
2840 	VERIFY0(dsl_prop_get_int_ds(ds,
2841 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2842 
2843 	if (ds->ds_quota != newval) {
2844 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2845 		ds->ds_quota = newval;
2846 	}
2847 	dsl_dataset_rele(ds, FTAG);
2848 }
2849 
2850 int
2851 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2852     uint64_t refquota)
2853 {
2854 	dsl_dataset_set_qr_arg_t ddsqra;
2855 
2856 	ddsqra.ddsqra_name = dsname;
2857 	ddsqra.ddsqra_source = source;
2858 	ddsqra.ddsqra_value = refquota;
2859 
2860 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2861 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
2862 }
2863 
2864 static int
2865 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2866 {
2867 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2868 	dsl_pool_t *dp = dmu_tx_pool(tx);
2869 	dsl_dataset_t *ds;
2870 	int error;
2871 	uint64_t newval, unique;
2872 
2873 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2874 		return (SET_ERROR(ENOTSUP));
2875 
2876 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2877 	if (error != 0)
2878 		return (error);
2879 
2880 	if (dsl_dataset_is_snapshot(ds)) {
2881 		dsl_dataset_rele(ds, FTAG);
2882 		return (SET_ERROR(EINVAL));
2883 	}
2884 
2885 	error = dsl_prop_predict(ds->ds_dir,
2886 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2887 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2888 	if (error != 0) {
2889 		dsl_dataset_rele(ds, FTAG);
2890 		return (error);
2891 	}
2892 
2893 	/*
2894 	 * If we are doing the preliminary check in open context, the
2895 	 * space estimates may be inaccurate.
2896 	 */
2897 	if (!dmu_tx_is_syncing(tx)) {
2898 		dsl_dataset_rele(ds, FTAG);
2899 		return (0);
2900 	}
2901 
2902 	mutex_enter(&ds->ds_lock);
2903 	if (!DS_UNIQUE_IS_ACCURATE(ds))
2904 		dsl_dataset_recalc_head_uniq(ds);
2905 	unique = ds->ds_phys->ds_unique_bytes;
2906 	mutex_exit(&ds->ds_lock);
2907 
2908 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2909 		uint64_t delta = MAX(unique, newval) -
2910 		    MAX(unique, ds->ds_reserved);
2911 
2912 		if (delta >
2913 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2914 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2915 			dsl_dataset_rele(ds, FTAG);
2916 			return (SET_ERROR(ENOSPC));
2917 		}
2918 	}
2919 
2920 	dsl_dataset_rele(ds, FTAG);
2921 	return (0);
2922 }
2923 
2924 void
2925 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2926     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2927 {
2928 	uint64_t newval;
2929 	uint64_t unique;
2930 	int64_t delta;
2931 
2932 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2933 	    source, sizeof (value), 1, &value, tx);
2934 
2935 	VERIFY0(dsl_prop_get_int_ds(ds,
2936 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
2937 
2938 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2939 	mutex_enter(&ds->ds_dir->dd_lock);
2940 	mutex_enter(&ds->ds_lock);
2941 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2942 	unique = ds->ds_phys->ds_unique_bytes;
2943 	delta = MAX(0, (int64_t)(newval - unique)) -
2944 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
2945 	ds->ds_reserved = newval;
2946 	mutex_exit(&ds->ds_lock);
2947 
2948 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
2949 	mutex_exit(&ds->ds_dir->dd_lock);
2950 }
2951 
2952 static void
2953 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
2954 {
2955 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2956 	dsl_pool_t *dp = dmu_tx_pool(tx);
2957 	dsl_dataset_t *ds;
2958 
2959 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2960 	dsl_dataset_set_refreservation_sync_impl(ds,
2961 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
2962 	dsl_dataset_rele(ds, FTAG);
2963 }
2964 
2965 int
2966 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
2967     uint64_t refreservation)
2968 {
2969 	dsl_dataset_set_qr_arg_t ddsqra;
2970 
2971 	ddsqra.ddsqra_name = dsname;
2972 	ddsqra.ddsqra_source = source;
2973 	ddsqra.ddsqra_value = refreservation;
2974 
2975 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
2976 	    dsl_dataset_set_refreservation_sync, &ddsqra,
2977 	    0, ZFS_SPACE_CHECK_NONE));
2978 }
2979 
2980 /*
2981  * Return (in *usedp) the amount of space written in new that is not
2982  * present in oldsnap.  New may be a snapshot or the head.  Old must be
2983  * a snapshot before new, in new's filesystem (or its origin).  If not then
2984  * fail and return EINVAL.
2985  *
2986  * The written space is calculated by considering two components:  First, we
2987  * ignore any freed space, and calculate the written as new's used space
2988  * minus old's used space.  Next, we add in the amount of space that was freed
2989  * between the two snapshots, thus reducing new's used space relative to old's.
2990  * Specifically, this is the space that was born before old->ds_creation_txg,
2991  * and freed before new (ie. on new's deadlist or a previous deadlist).
2992  *
2993  * space freed                         [---------------------]
2994  * snapshots                       ---O-------O--------O-------O------
2995  *                                         oldsnap            new
2996  */
2997 int
2998 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
2999     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3000 {
3001 	int err = 0;
3002 	uint64_t snapobj;
3003 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3004 
3005 	ASSERT(dsl_pool_config_held(dp));
3006 
3007 	*usedp = 0;
3008 	*usedp += new->ds_phys->ds_referenced_bytes;
3009 	*usedp -= oldsnap->ds_phys->ds_referenced_bytes;
3010 
3011 	*compp = 0;
3012 	*compp += new->ds_phys->ds_compressed_bytes;
3013 	*compp -= oldsnap->ds_phys->ds_compressed_bytes;
3014 
3015 	*uncompp = 0;
3016 	*uncompp += new->ds_phys->ds_uncompressed_bytes;
3017 	*uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
3018 
3019 	snapobj = new->ds_object;
3020 	while (snapobj != oldsnap->ds_object) {
3021 		dsl_dataset_t *snap;
3022 		uint64_t used, comp, uncomp;
3023 
3024 		if (snapobj == new->ds_object) {
3025 			snap = new;
3026 		} else {
3027 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3028 			if (err != 0)
3029 				break;
3030 		}
3031 
3032 		if (snap->ds_phys->ds_prev_snap_txg ==
3033 		    oldsnap->ds_phys->ds_creation_txg) {
3034 			/*
3035 			 * The blocks in the deadlist can not be born after
3036 			 * ds_prev_snap_txg, so get the whole deadlist space,
3037 			 * which is more efficient (especially for old-format
3038 			 * deadlists).  Unfortunately the deadlist code
3039 			 * doesn't have enough information to make this
3040 			 * optimization itself.
3041 			 */
3042 			dsl_deadlist_space(&snap->ds_deadlist,
3043 			    &used, &comp, &uncomp);
3044 		} else {
3045 			dsl_deadlist_space_range(&snap->ds_deadlist,
3046 			    0, oldsnap->ds_phys->ds_creation_txg,
3047 			    &used, &comp, &uncomp);
3048 		}
3049 		*usedp += used;
3050 		*compp += comp;
3051 		*uncompp += uncomp;
3052 
3053 		/*
3054 		 * If we get to the beginning of the chain of snapshots
3055 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3056 		 * was not a snapshot of/before new.
3057 		 */
3058 		snapobj = snap->ds_phys->ds_prev_snap_obj;
3059 		if (snap != new)
3060 			dsl_dataset_rele(snap, FTAG);
3061 		if (snapobj == 0) {
3062 			err = SET_ERROR(EINVAL);
3063 			break;
3064 		}
3065 
3066 	}
3067 	return (err);
3068 }
3069 
3070 /*
3071  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3072  * lastsnap, and all snapshots in between are deleted.
3073  *
3074  * blocks that would be freed            [---------------------------]
3075  * snapshots                       ---O-------O--------O-------O--------O
3076  *                                        firstsnap        lastsnap
3077  *
3078  * This is the set of blocks that were born after the snap before firstsnap,
3079  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3080  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3081  * We calculate this by iterating over the relevant deadlists (from the snap
3082  * after lastsnap, backward to the snap after firstsnap), summing up the
3083  * space on the deadlist that was born after the snap before firstsnap.
3084  */
3085 int
3086 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3087     dsl_dataset_t *lastsnap,
3088     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3089 {
3090 	int err = 0;
3091 	uint64_t snapobj;
3092 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3093 
3094 	ASSERT(dsl_dataset_is_snapshot(firstsnap));
3095 	ASSERT(dsl_dataset_is_snapshot(lastsnap));
3096 
3097 	/*
3098 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3099 	 * is before lastsnap.
3100 	 */
3101 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3102 	    firstsnap->ds_phys->ds_creation_txg >
3103 	    lastsnap->ds_phys->ds_creation_txg)
3104 		return (SET_ERROR(EINVAL));
3105 
3106 	*usedp = *compp = *uncompp = 0;
3107 
3108 	snapobj = lastsnap->ds_phys->ds_next_snap_obj;
3109 	while (snapobj != firstsnap->ds_object) {
3110 		dsl_dataset_t *ds;
3111 		uint64_t used, comp, uncomp;
3112 
3113 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3114 		if (err != 0)
3115 			break;
3116 
3117 		dsl_deadlist_space_range(&ds->ds_deadlist,
3118 		    firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3119 		    &used, &comp, &uncomp);
3120 		*usedp += used;
3121 		*compp += comp;
3122 		*uncompp += uncomp;
3123 
3124 		snapobj = ds->ds_phys->ds_prev_snap_obj;
3125 		ASSERT3U(snapobj, !=, 0);
3126 		dsl_dataset_rele(ds, FTAG);
3127 	}
3128 	return (err);
3129 }
3130 
3131 /*
3132  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3133  * For example, they could both be snapshots of the same filesystem, and
3134  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3135  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3136  * filesystem.  Or 'earlier' could be the origin's origin.
3137  *
3138  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3139  */
3140 boolean_t
3141 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3142 	uint64_t earlier_txg)
3143 {
3144 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3145 	int error;
3146 	boolean_t ret;
3147 
3148 	ASSERT(dsl_pool_config_held(dp));
3149 	ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
3150 
3151 	if (earlier_txg == 0)
3152 		earlier_txg = earlier->ds_phys->ds_creation_txg;
3153 
3154 	if (dsl_dataset_is_snapshot(later) &&
3155 	    earlier_txg >= later->ds_phys->ds_creation_txg)
3156 		return (B_FALSE);
3157 
3158 	if (later->ds_dir == earlier->ds_dir)
3159 		return (B_TRUE);
3160 	if (!dsl_dir_is_clone(later->ds_dir))
3161 		return (B_FALSE);
3162 
3163 	if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object)
3164 		return (B_TRUE);
3165 	dsl_dataset_t *origin;
3166 	error = dsl_dataset_hold_obj(dp,
3167 	    later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
3168 	if (error != 0)
3169 		return (B_FALSE);
3170 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3171 	dsl_dataset_rele(origin, FTAG);
3172 	return (ret);
3173 }
3174 
3175 
3176 void
3177 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3178 {
3179 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3180 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3181 }
3182