xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision e521049d0fdbd76727f036cb77c99c98d7ee1125)
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) 2013, 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 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
696 
697 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
698 		origin->ds_phys->ds_num_children++;
699 
700 		VERIFY0(dsl_dataset_hold_obj(dp,
701 		    origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
702 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
703 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
704 		dsl_dataset_rele(ohds, FTAG);
705 
706 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
707 			if (origin->ds_phys->ds_next_clones_obj == 0) {
708 				origin->ds_phys->ds_next_clones_obj =
709 				    zap_create(mos,
710 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
711 			}
712 			VERIFY0(zap_add_int(mos,
713 			    origin->ds_phys->ds_next_clones_obj, dsobj, tx));
714 		}
715 
716 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
717 		dd->dd_phys->dd_origin_obj = origin->ds_object;
718 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
719 			if (origin->ds_dir->dd_phys->dd_clones == 0) {
720 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
721 				origin->ds_dir->dd_phys->dd_clones =
722 				    zap_create(mos,
723 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
724 			}
725 			VERIFY0(zap_add_int(mos,
726 			    origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
727 		}
728 	}
729 
730 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
731 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
732 
733 	dmu_buf_rele(dbuf, FTAG);
734 
735 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
736 	dd->dd_phys->dd_head_dataset_obj = dsobj;
737 
738 	return (dsobj);
739 }
740 
741 static void
742 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
743 {
744 	objset_t *os;
745 
746 	VERIFY0(dmu_objset_from_ds(ds, &os));
747 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
748 	dsl_dataset_dirty(ds, tx);
749 }
750 
751 uint64_t
752 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
753     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
754 {
755 	dsl_pool_t *dp = pdd->dd_pool;
756 	uint64_t dsobj, ddobj;
757 	dsl_dir_t *dd;
758 
759 	ASSERT(dmu_tx_is_syncing(tx));
760 	ASSERT(lastname[0] != '@');
761 
762 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
763 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
764 
765 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
766 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
767 
768 	dsl_deleg_set_create_perms(dd, tx, cr);
769 
770 	/*
771 	 * Since we're creating a new node we know it's a leaf, so we can
772 	 * initialize the counts if the limit feature is active.
773 	 */
774 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
775 		uint64_t cnt = 0;
776 		objset_t *os = dd->dd_pool->dp_meta_objset;
777 
778 		dsl_dir_zapify(dd, tx);
779 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
780 		    sizeof (cnt), 1, &cnt, tx));
781 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
782 		    sizeof (cnt), 1, &cnt, tx));
783 	}
784 
785 	dsl_dir_rele(dd, FTAG);
786 
787 	/*
788 	 * If we are creating a clone, make sure we zero out any stale
789 	 * data from the origin snapshots zil header.
790 	 */
791 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
792 		dsl_dataset_t *ds;
793 
794 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
795 		dsl_dataset_zero_zil(ds, tx);
796 		dsl_dataset_rele(ds, FTAG);
797 	}
798 
799 	return (dsobj);
800 }
801 
802 /*
803  * The unique space in the head dataset can be calculated by subtracting
804  * the space used in the most recent snapshot, that is still being used
805  * in this file system, from the space currently in use.  To figure out
806  * the space in the most recent snapshot still in use, we need to take
807  * the total space used in the snapshot and subtract out the space that
808  * has been freed up since the snapshot was taken.
809  */
810 void
811 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
812 {
813 	uint64_t mrs_used;
814 	uint64_t dlused, dlcomp, dluncomp;
815 
816 	ASSERT(!dsl_dataset_is_snapshot(ds));
817 
818 	if (ds->ds_phys->ds_prev_snap_obj != 0)
819 		mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
820 	else
821 		mrs_used = 0;
822 
823 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
824 
825 	ASSERT3U(dlused, <=, mrs_used);
826 	ds->ds_phys->ds_unique_bytes =
827 	    ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
828 
829 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
830 	    SPA_VERSION_UNIQUE_ACCURATE)
831 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
832 }
833 
834 void
835 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
836     dmu_tx_t *tx)
837 {
838 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
839 	uint64_t count;
840 	int err;
841 
842 	ASSERT(ds->ds_phys->ds_num_children >= 2);
843 	err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
844 	/*
845 	 * The err should not be ENOENT, but a bug in a previous version
846 	 * of the code could cause upgrade_clones_cb() to not set
847 	 * ds_next_snap_obj when it should, leading to a missing entry.
848 	 * If we knew that the pool was created after
849 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
850 	 * ENOENT.  However, at least we can check that we don't have
851 	 * too many entries in the next_clones_obj even after failing to
852 	 * remove this one.
853 	 */
854 	if (err != ENOENT)
855 		VERIFY0(err);
856 	ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
857 	    &count));
858 	ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
859 }
860 
861 
862 blkptr_t *
863 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
864 {
865 	return (&ds->ds_phys->ds_bp);
866 }
867 
868 void
869 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
870 {
871 	ASSERT(dmu_tx_is_syncing(tx));
872 	/* If it's the meta-objset, set dp_meta_rootbp */
873 	if (ds == NULL) {
874 		tx->tx_pool->dp_meta_rootbp = *bp;
875 	} else {
876 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
877 		ds->ds_phys->ds_bp = *bp;
878 	}
879 }
880 
881 spa_t *
882 dsl_dataset_get_spa(dsl_dataset_t *ds)
883 {
884 	return (ds->ds_dir->dd_pool->dp_spa);
885 }
886 
887 void
888 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
889 {
890 	dsl_pool_t *dp;
891 
892 	if (ds == NULL) /* this is the meta-objset */
893 		return;
894 
895 	ASSERT(ds->ds_objset != NULL);
896 
897 	if (ds->ds_phys->ds_next_snap_obj != 0)
898 		panic("dirtying snapshot!");
899 
900 	dp = ds->ds_dir->dd_pool;
901 
902 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
903 		/* up the hold count until we can be written out */
904 		dmu_buf_add_ref(ds->ds_dbuf, ds);
905 	}
906 }
907 
908 boolean_t
909 dsl_dataset_is_dirty(dsl_dataset_t *ds)
910 {
911 	for (int t = 0; t < TXG_SIZE; t++) {
912 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
913 		    ds, t))
914 			return (B_TRUE);
915 	}
916 	return (B_FALSE);
917 }
918 
919 static int
920 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
921 {
922 	uint64_t asize;
923 
924 	if (!dmu_tx_is_syncing(tx))
925 		return (0);
926 
927 	/*
928 	 * If there's an fs-only reservation, any blocks that might become
929 	 * owned by the snapshot dataset must be accommodated by space
930 	 * outside of the reservation.
931 	 */
932 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
933 	asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
934 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
935 		return (SET_ERROR(ENOSPC));
936 
937 	/*
938 	 * Propagate any reserved space for this snapshot to other
939 	 * snapshot checks in this sync group.
940 	 */
941 	if (asize > 0)
942 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
943 
944 	return (0);
945 }
946 
947 typedef struct dsl_dataset_snapshot_arg {
948 	nvlist_t *ddsa_snaps;
949 	nvlist_t *ddsa_props;
950 	nvlist_t *ddsa_errors;
951 	cred_t *ddsa_cr;
952 } dsl_dataset_snapshot_arg_t;
953 
954 int
955 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
956     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
957 {
958 	int error;
959 	uint64_t value;
960 
961 	ds->ds_trysnap_txg = tx->tx_txg;
962 
963 	if (!dmu_tx_is_syncing(tx))
964 		return (0);
965 
966 	/*
967 	 * We don't allow multiple snapshots of the same txg.  If there
968 	 * is already one, try again.
969 	 */
970 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
971 		return (SET_ERROR(EAGAIN));
972 
973 	/*
974 	 * Check for conflicting snapshot name.
975 	 */
976 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
977 	if (error == 0)
978 		return (SET_ERROR(EEXIST));
979 	if (error != ENOENT)
980 		return (error);
981 
982 	/*
983 	 * We don't allow taking snapshots of inconsistent datasets, such as
984 	 * those into which we are currently receiving.  However, if we are
985 	 * creating this snapshot as part of a receive, this check will be
986 	 * executed atomically with respect to the completion of the receive
987 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
988 	 * case we ignore this, knowing it will be fixed up for us shortly in
989 	 * dmu_recv_end_sync().
990 	 */
991 	if (!recv && DS_IS_INCONSISTENT(ds))
992 		return (SET_ERROR(EBUSY));
993 
994 	/*
995 	 * Skip the check for temporary snapshots or if we have already checked
996 	 * the counts in dsl_dataset_snapshot_check. This means we really only
997 	 * check the count here when we're receiving a stream.
998 	 */
999 	if (cnt != 0 && cr != NULL) {
1000 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1001 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1002 		if (error != 0)
1003 			return (error);
1004 	}
1005 
1006 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1007 	if (error != 0)
1008 		return (error);
1009 
1010 	return (0);
1011 }
1012 
1013 static int
1014 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1015 {
1016 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1017 	dsl_pool_t *dp = dmu_tx_pool(tx);
1018 	nvpair_t *pair;
1019 	int rv = 0;
1020 
1021 	/*
1022 	 * Pre-compute how many total new snapshots will be created for each
1023 	 * level in the tree and below. This is needed for validating the
1024 	 * snapshot limit when either taking a recursive snapshot or when
1025 	 * taking multiple snapshots.
1026 	 *
1027 	 * The problem is that the counts are not actually adjusted when
1028 	 * we are checking, only when we finally sync. For a single snapshot,
1029 	 * this is easy, the count will increase by 1 at each node up the tree,
1030 	 * but its more complicated for the recursive/multiple snapshot case.
1031 	 *
1032 	 * The dsl_fs_ss_limit_check function does recursively check the count
1033 	 * at each level up the tree but since it is validating each snapshot
1034 	 * independently we need to be sure that we are validating the complete
1035 	 * count for the entire set of snapshots. We do this by rolling up the
1036 	 * counts for each component of the name into an nvlist and then
1037 	 * checking each of those cases with the aggregated count.
1038 	 *
1039 	 * This approach properly handles not only the recursive snapshot
1040 	 * case (where we get all of those on the ddsa_snaps list) but also
1041 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1042 	 * validate the limit on 'a' using a count of 2).
1043 	 *
1044 	 * We validate the snapshot names in the third loop and only report
1045 	 * name errors once.
1046 	 */
1047 	if (dmu_tx_is_syncing(tx)) {
1048 		nvlist_t *cnt_track = NULL;
1049 		cnt_track = fnvlist_alloc();
1050 
1051 		/* Rollup aggregated counts into the cnt_track list */
1052 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1053 		    pair != NULL;
1054 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1055 			char *pdelim;
1056 			uint64_t val;
1057 			char nm[MAXPATHLEN];
1058 
1059 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1060 			pdelim = strchr(nm, '@');
1061 			if (pdelim == NULL)
1062 				continue;
1063 			*pdelim = '\0';
1064 
1065 			do {
1066 				if (nvlist_lookup_uint64(cnt_track, nm,
1067 				    &val) == 0) {
1068 					/* update existing entry */
1069 					fnvlist_add_uint64(cnt_track, nm,
1070 					    val + 1);
1071 				} else {
1072 					/* add to list */
1073 					fnvlist_add_uint64(cnt_track, nm, 1);
1074 				}
1075 
1076 				pdelim = strrchr(nm, '/');
1077 				if (pdelim != NULL)
1078 					*pdelim = '\0';
1079 			} while (pdelim != NULL);
1080 		}
1081 
1082 		/* Check aggregated counts at each level */
1083 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1084 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1085 			int error = 0;
1086 			char *name;
1087 			uint64_t cnt = 0;
1088 			dsl_dataset_t *ds;
1089 
1090 			name = nvpair_name(pair);
1091 			cnt = fnvpair_value_uint64(pair);
1092 			ASSERT(cnt > 0);
1093 
1094 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1095 			if (error == 0) {
1096 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1097 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1098 				    ddsa->ddsa_cr);
1099 				dsl_dataset_rele(ds, FTAG);
1100 			}
1101 
1102 			if (error != 0) {
1103 				if (ddsa->ddsa_errors != NULL)
1104 					fnvlist_add_int32(ddsa->ddsa_errors,
1105 					    name, error);
1106 				rv = error;
1107 				/* only report one error for this check */
1108 				break;
1109 			}
1110 		}
1111 		nvlist_free(cnt_track);
1112 	}
1113 
1114 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1115 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1116 		int error = 0;
1117 		dsl_dataset_t *ds;
1118 		char *name, *atp;
1119 		char dsname[MAXNAMELEN];
1120 
1121 		name = nvpair_name(pair);
1122 		if (strlen(name) >= MAXNAMELEN)
1123 			error = SET_ERROR(ENAMETOOLONG);
1124 		if (error == 0) {
1125 			atp = strchr(name, '@');
1126 			if (atp == NULL)
1127 				error = SET_ERROR(EINVAL);
1128 			if (error == 0)
1129 				(void) strlcpy(dsname, name, atp - name + 1);
1130 		}
1131 		if (error == 0)
1132 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1133 		if (error == 0) {
1134 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1135 			error = dsl_dataset_snapshot_check_impl(ds,
1136 			    atp + 1, tx, B_FALSE, 0, NULL);
1137 			dsl_dataset_rele(ds, FTAG);
1138 		}
1139 
1140 		if (error != 0) {
1141 			if (ddsa->ddsa_errors != NULL) {
1142 				fnvlist_add_int32(ddsa->ddsa_errors,
1143 				    name, error);
1144 			}
1145 			rv = error;
1146 		}
1147 	}
1148 
1149 	return (rv);
1150 }
1151 
1152 void
1153 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1154     dmu_tx_t *tx)
1155 {
1156 	static zil_header_t zero_zil;
1157 
1158 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1159 	dmu_buf_t *dbuf;
1160 	dsl_dataset_phys_t *dsphys;
1161 	uint64_t dsobj, crtxg;
1162 	objset_t *mos = dp->dp_meta_objset;
1163 	objset_t *os;
1164 
1165 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1166 
1167 	/*
1168 	 * If we are on an old pool, the zil must not be active, in which
1169 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1170 	 */
1171 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1172 	    dmu_objset_from_ds(ds, &os) != 0 ||
1173 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1174 	    sizeof (zero_zil)) == 0);
1175 
1176 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1177 
1178 	/*
1179 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1180 	 */
1181 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1182 		crtxg = 1;
1183 	else
1184 		crtxg = tx->tx_txg;
1185 
1186 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1187 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1188 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1189 	dmu_buf_will_dirty(dbuf, tx);
1190 	dsphys = dbuf->db_data;
1191 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1192 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1193 	dsphys->ds_fsid_guid = unique_create();
1194 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1195 	    sizeof (dsphys->ds_guid));
1196 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1197 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1198 	dsphys->ds_next_snap_obj = ds->ds_object;
1199 	dsphys->ds_num_children = 1;
1200 	dsphys->ds_creation_time = gethrestime_sec();
1201 	dsphys->ds_creation_txg = crtxg;
1202 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1203 	dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
1204 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1205 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1206 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1207 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1208 	dmu_buf_rele(dbuf, FTAG);
1209 
1210 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1211 	if (ds->ds_prev) {
1212 		uint64_t next_clones_obj =
1213 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1214 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1215 		    ds->ds_object ||
1216 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1217 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1218 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1219 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1220 			    ds->ds_prev->ds_phys->ds_creation_txg);
1221 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1222 		} else if (next_clones_obj != 0) {
1223 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1224 			    dsphys->ds_next_snap_obj, tx);
1225 			VERIFY0(zap_add_int(mos,
1226 			    next_clones_obj, dsobj, tx));
1227 		}
1228 	}
1229 
1230 	/*
1231 	 * If we have a reference-reservation on this dataset, we will
1232 	 * need to increase the amount of refreservation being charged
1233 	 * since our unique space is going to zero.
1234 	 */
1235 	if (ds->ds_reserved) {
1236 		int64_t delta;
1237 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1238 		delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1239 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1240 		    delta, 0, 0, tx);
1241 	}
1242 
1243 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1244 	ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
1245 	    UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
1246 	dsl_deadlist_close(&ds->ds_deadlist);
1247 	dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1248 	dsl_deadlist_add_key(&ds->ds_deadlist,
1249 	    ds->ds_phys->ds_prev_snap_txg, tx);
1250 
1251 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1252 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1253 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1254 	ds->ds_phys->ds_unique_bytes = 0;
1255 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1256 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1257 
1258 	VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1259 	    snapname, 8, 1, &dsobj, tx));
1260 
1261 	if (ds->ds_prev)
1262 		dsl_dataset_rele(ds->ds_prev, ds);
1263 	VERIFY0(dsl_dataset_hold_obj(dp,
1264 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1265 
1266 	dsl_scan_ds_snapshotted(ds, tx);
1267 
1268 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1269 
1270 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1271 }
1272 
1273 static void
1274 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1275 {
1276 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1277 	dsl_pool_t *dp = dmu_tx_pool(tx);
1278 	nvpair_t *pair;
1279 
1280 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1281 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1282 		dsl_dataset_t *ds;
1283 		char *name, *atp;
1284 		char dsname[MAXNAMELEN];
1285 
1286 		name = nvpair_name(pair);
1287 		atp = strchr(name, '@');
1288 		(void) strlcpy(dsname, name, atp - name + 1);
1289 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1290 
1291 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1292 		if (ddsa->ddsa_props != NULL) {
1293 			dsl_props_set_sync_impl(ds->ds_prev,
1294 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1295 		}
1296 		dsl_dataset_rele(ds, FTAG);
1297 	}
1298 }
1299 
1300 /*
1301  * The snapshots must all be in the same pool.
1302  * All-or-nothing: if there are any failures, nothing will be modified.
1303  */
1304 int
1305 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1306 {
1307 	dsl_dataset_snapshot_arg_t ddsa;
1308 	nvpair_t *pair;
1309 	boolean_t needsuspend;
1310 	int error;
1311 	spa_t *spa;
1312 	char *firstname;
1313 	nvlist_t *suspended = NULL;
1314 
1315 	pair = nvlist_next_nvpair(snaps, NULL);
1316 	if (pair == NULL)
1317 		return (0);
1318 	firstname = nvpair_name(pair);
1319 
1320 	error = spa_open(firstname, &spa, FTAG);
1321 	if (error != 0)
1322 		return (error);
1323 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1324 	spa_close(spa, FTAG);
1325 
1326 	if (needsuspend) {
1327 		suspended = fnvlist_alloc();
1328 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1329 		    pair = nvlist_next_nvpair(snaps, pair)) {
1330 			char fsname[MAXNAMELEN];
1331 			char *snapname = nvpair_name(pair);
1332 			char *atp;
1333 			void *cookie;
1334 
1335 			atp = strchr(snapname, '@');
1336 			if (atp == NULL) {
1337 				error = SET_ERROR(EINVAL);
1338 				break;
1339 			}
1340 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1341 
1342 			error = zil_suspend(fsname, &cookie);
1343 			if (error != 0)
1344 				break;
1345 			fnvlist_add_uint64(suspended, fsname,
1346 			    (uintptr_t)cookie);
1347 		}
1348 	}
1349 
1350 	ddsa.ddsa_snaps = snaps;
1351 	ddsa.ddsa_props = props;
1352 	ddsa.ddsa_errors = errors;
1353 	ddsa.ddsa_cr = CRED();
1354 
1355 	if (error == 0) {
1356 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1357 		    dsl_dataset_snapshot_sync, &ddsa,
1358 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1359 	}
1360 
1361 	if (suspended != NULL) {
1362 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1363 		    pair = nvlist_next_nvpair(suspended, pair)) {
1364 			zil_resume((void *)(uintptr_t)
1365 			    fnvpair_value_uint64(pair));
1366 		}
1367 		fnvlist_free(suspended);
1368 	}
1369 
1370 	return (error);
1371 }
1372 
1373 typedef struct dsl_dataset_snapshot_tmp_arg {
1374 	const char *ddsta_fsname;
1375 	const char *ddsta_snapname;
1376 	minor_t ddsta_cleanup_minor;
1377 	const char *ddsta_htag;
1378 } dsl_dataset_snapshot_tmp_arg_t;
1379 
1380 static int
1381 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1382 {
1383 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1384 	dsl_pool_t *dp = dmu_tx_pool(tx);
1385 	dsl_dataset_t *ds;
1386 	int error;
1387 
1388 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1389 	if (error != 0)
1390 		return (error);
1391 
1392 	/* NULL cred means no limit check for tmp snapshot */
1393 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1394 	    tx, B_FALSE, 0, NULL);
1395 	if (error != 0) {
1396 		dsl_dataset_rele(ds, FTAG);
1397 		return (error);
1398 	}
1399 
1400 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1401 		dsl_dataset_rele(ds, FTAG);
1402 		return (SET_ERROR(ENOTSUP));
1403 	}
1404 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1405 	    B_TRUE, tx);
1406 	if (error != 0) {
1407 		dsl_dataset_rele(ds, FTAG);
1408 		return (error);
1409 	}
1410 
1411 	dsl_dataset_rele(ds, FTAG);
1412 	return (0);
1413 }
1414 
1415 static void
1416 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1417 {
1418 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1419 	dsl_pool_t *dp = dmu_tx_pool(tx);
1420 	dsl_dataset_t *ds;
1421 
1422 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1423 
1424 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1425 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1426 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1427 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1428 
1429 	dsl_dataset_rele(ds, FTAG);
1430 }
1431 
1432 int
1433 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1434     minor_t cleanup_minor, const char *htag)
1435 {
1436 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1437 	int error;
1438 	spa_t *spa;
1439 	boolean_t needsuspend;
1440 	void *cookie;
1441 
1442 	ddsta.ddsta_fsname = fsname;
1443 	ddsta.ddsta_snapname = snapname;
1444 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1445 	ddsta.ddsta_htag = htag;
1446 
1447 	error = spa_open(fsname, &spa, FTAG);
1448 	if (error != 0)
1449 		return (error);
1450 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1451 	spa_close(spa, FTAG);
1452 
1453 	if (needsuspend) {
1454 		error = zil_suspend(fsname, &cookie);
1455 		if (error != 0)
1456 			return (error);
1457 	}
1458 
1459 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1460 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1461 
1462 	if (needsuspend)
1463 		zil_resume(cookie);
1464 	return (error);
1465 }
1466 
1467 
1468 void
1469 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1470 {
1471 	ASSERT(dmu_tx_is_syncing(tx));
1472 	ASSERT(ds->ds_objset != NULL);
1473 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1474 
1475 	/*
1476 	 * in case we had to change ds_fsid_guid when we opened it,
1477 	 * sync it out now.
1478 	 */
1479 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1480 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1481 
1482 	dmu_objset_sync(ds->ds_objset, zio, tx);
1483 }
1484 
1485 static void
1486 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1487 {
1488 	uint64_t count = 0;
1489 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1490 	zap_cursor_t zc;
1491 	zap_attribute_t za;
1492 	nvlist_t *propval = fnvlist_alloc();
1493 	nvlist_t *val = fnvlist_alloc();
1494 
1495 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1496 
1497 	/*
1498 	 * There may be missing entries in ds_next_clones_obj
1499 	 * due to a bug in a previous version of the code.
1500 	 * Only trust it if it has the right number of entries.
1501 	 */
1502 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1503 		VERIFY0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1504 		    &count));
1505 	}
1506 	if (count != ds->ds_phys->ds_num_children - 1)
1507 		goto fail;
1508 	for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
1509 	    zap_cursor_retrieve(&zc, &za) == 0;
1510 	    zap_cursor_advance(&zc)) {
1511 		dsl_dataset_t *clone;
1512 		char buf[ZFS_MAXNAMELEN];
1513 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1514 		    za.za_first_integer, FTAG, &clone));
1515 		dsl_dir_name(clone->ds_dir, buf);
1516 		fnvlist_add_boolean(val, buf);
1517 		dsl_dataset_rele(clone, FTAG);
1518 	}
1519 	zap_cursor_fini(&zc);
1520 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1521 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1522 fail:
1523 	nvlist_free(val);
1524 	nvlist_free(propval);
1525 }
1526 
1527 void
1528 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1529 {
1530 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1531 	uint64_t refd, avail, uobjs, aobjs, ratio;
1532 
1533 	ASSERT(dsl_pool_config_held(dp));
1534 
1535 	ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1536 	    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1537 	    ds->ds_phys->ds_compressed_bytes);
1538 
1539 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1540 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1541 	    ds->ds_phys->ds_uncompressed_bytes);
1542 
1543 	if (dsl_dataset_is_snapshot(ds)) {
1544 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1545 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1546 		    ds->ds_phys->ds_unique_bytes);
1547 		get_clones_stat(ds, nv);
1548 	} else {
1549 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1550 			char buf[MAXNAMELEN];
1551 			dsl_dataset_name(ds->ds_prev, buf);
1552 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1553 		}
1554 
1555 		dsl_dir_stats(ds->ds_dir, nv);
1556 	}
1557 
1558 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1559 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1560 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1561 
1562 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1563 	    ds->ds_phys->ds_creation_time);
1564 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1565 	    ds->ds_phys->ds_creation_txg);
1566 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1567 	    ds->ds_quota);
1568 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1569 	    ds->ds_reserved);
1570 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1571 	    ds->ds_phys->ds_guid);
1572 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1573 	    ds->ds_phys->ds_unique_bytes);
1574 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1575 	    ds->ds_object);
1576 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1577 	    ds->ds_userrefs);
1578 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1579 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1580 
1581 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1582 		uint64_t written, comp, uncomp;
1583 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1584 		dsl_dataset_t *prev;
1585 
1586 		int err = dsl_dataset_hold_obj(dp,
1587 		    ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
1588 		if (err == 0) {
1589 			err = dsl_dataset_space_written(prev, ds, &written,
1590 			    &comp, &uncomp);
1591 			dsl_dataset_rele(prev, FTAG);
1592 			if (err == 0) {
1593 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1594 				    written);
1595 			}
1596 		}
1597 	}
1598 }
1599 
1600 void
1601 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1602 {
1603 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1604 	ASSERT(dsl_pool_config_held(dp));
1605 
1606 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1607 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1608 	stat->dds_guid = ds->ds_phys->ds_guid;
1609 	stat->dds_origin[0] = '\0';
1610 	if (dsl_dataset_is_snapshot(ds)) {
1611 		stat->dds_is_snapshot = B_TRUE;
1612 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1613 	} else {
1614 		stat->dds_is_snapshot = B_FALSE;
1615 		stat->dds_num_clones = 0;
1616 
1617 		if (dsl_dir_is_clone(ds->ds_dir)) {
1618 			dsl_dataset_t *ods;
1619 
1620 			VERIFY0(dsl_dataset_hold_obj(dp,
1621 			    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1622 			dsl_dataset_name(ods, stat->dds_origin);
1623 			dsl_dataset_rele(ods, FTAG);
1624 		}
1625 	}
1626 }
1627 
1628 uint64_t
1629 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1630 {
1631 	return (ds->ds_fsid_guid);
1632 }
1633 
1634 void
1635 dsl_dataset_space(dsl_dataset_t *ds,
1636     uint64_t *refdbytesp, uint64_t *availbytesp,
1637     uint64_t *usedobjsp, uint64_t *availobjsp)
1638 {
1639 	*refdbytesp = ds->ds_phys->ds_referenced_bytes;
1640 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1641 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1642 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1643 	if (ds->ds_quota != 0) {
1644 		/*
1645 		 * Adjust available bytes according to refquota
1646 		 */
1647 		if (*refdbytesp < ds->ds_quota)
1648 			*availbytesp = MIN(*availbytesp,
1649 			    ds->ds_quota - *refdbytesp);
1650 		else
1651 			*availbytesp = 0;
1652 	}
1653 	*usedobjsp = BP_GET_FILL(&ds->ds_phys->ds_bp);
1654 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1655 }
1656 
1657 boolean_t
1658 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1659 {
1660 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1661 
1662 	ASSERT(dsl_pool_config_held(dp));
1663 	if (snap == NULL)
1664 		return (B_FALSE);
1665 	if (ds->ds_phys->ds_bp.blk_birth >
1666 	    snap->ds_phys->ds_creation_txg) {
1667 		objset_t *os, *os_snap;
1668 		/*
1669 		 * It may be that only the ZIL differs, because it was
1670 		 * reset in the head.  Don't count that as being
1671 		 * modified.
1672 		 */
1673 		if (dmu_objset_from_ds(ds, &os) != 0)
1674 			return (B_TRUE);
1675 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1676 			return (B_TRUE);
1677 		return (bcmp(&os->os_phys->os_meta_dnode,
1678 		    &os_snap->os_phys->os_meta_dnode,
1679 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1680 	}
1681 	return (B_FALSE);
1682 }
1683 
1684 typedef struct dsl_dataset_rename_snapshot_arg {
1685 	const char *ddrsa_fsname;
1686 	const char *ddrsa_oldsnapname;
1687 	const char *ddrsa_newsnapname;
1688 	boolean_t ddrsa_recursive;
1689 	dmu_tx_t *ddrsa_tx;
1690 } dsl_dataset_rename_snapshot_arg_t;
1691 
1692 /* ARGSUSED */
1693 static int
1694 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1695     dsl_dataset_t *hds, void *arg)
1696 {
1697 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1698 	int error;
1699 	uint64_t val;
1700 
1701 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1702 	if (error != 0) {
1703 		/* ignore nonexistent snapshots */
1704 		return (error == ENOENT ? 0 : error);
1705 	}
1706 
1707 	/* new name should not exist */
1708 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1709 	if (error == 0)
1710 		error = SET_ERROR(EEXIST);
1711 	else if (error == ENOENT)
1712 		error = 0;
1713 
1714 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1715 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1716 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1717 		error = SET_ERROR(ENAMETOOLONG);
1718 
1719 	return (error);
1720 }
1721 
1722 static int
1723 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1724 {
1725 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1726 	dsl_pool_t *dp = dmu_tx_pool(tx);
1727 	dsl_dataset_t *hds;
1728 	int error;
1729 
1730 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1731 	if (error != 0)
1732 		return (error);
1733 
1734 	if (ddrsa->ddrsa_recursive) {
1735 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1736 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1737 		    DS_FIND_CHILDREN);
1738 	} else {
1739 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1740 	}
1741 	dsl_dataset_rele(hds, FTAG);
1742 	return (error);
1743 }
1744 
1745 static int
1746 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1747     dsl_dataset_t *hds, void *arg)
1748 {
1749 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1750 	dsl_dataset_t *ds;
1751 	uint64_t val;
1752 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1753 	int error;
1754 
1755 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1756 	ASSERT(error == 0 || error == ENOENT);
1757 	if (error == ENOENT) {
1758 		/* ignore nonexistent snapshots */
1759 		return (0);
1760 	}
1761 
1762 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1763 
1764 	/* log before we change the name */
1765 	spa_history_log_internal_ds(ds, "rename", tx,
1766 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1767 
1768 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1769 	    B_FALSE));
1770 	mutex_enter(&ds->ds_lock);
1771 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1772 	mutex_exit(&ds->ds_lock);
1773 	VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj,
1774 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1775 
1776 	dsl_dataset_rele(ds, FTAG);
1777 	return (0);
1778 }
1779 
1780 static void
1781 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1782 {
1783 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1784 	dsl_pool_t *dp = dmu_tx_pool(tx);
1785 	dsl_dataset_t *hds;
1786 
1787 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1788 	ddrsa->ddrsa_tx = tx;
1789 	if (ddrsa->ddrsa_recursive) {
1790 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1791 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1792 		    DS_FIND_CHILDREN));
1793 	} else {
1794 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1795 	}
1796 	dsl_dataset_rele(hds, FTAG);
1797 }
1798 
1799 int
1800 dsl_dataset_rename_snapshot(const char *fsname,
1801     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1802 {
1803 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1804 
1805 	ddrsa.ddrsa_fsname = fsname;
1806 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1807 	ddrsa.ddrsa_newsnapname = newsnapname;
1808 	ddrsa.ddrsa_recursive = recursive;
1809 
1810 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1811 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
1812 	    1, ZFS_SPACE_CHECK_RESERVED));
1813 }
1814 
1815 /*
1816  * If we're doing an ownership handoff, we need to make sure that there is
1817  * only one long hold on the dataset.  We're not allowed to change anything here
1818  * so we don't permanently release the long hold or regular hold here.  We want
1819  * to do this only when syncing to avoid the dataset unexpectedly going away
1820  * when we release the long hold.
1821  */
1822 static int
1823 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1824 {
1825 	boolean_t held;
1826 
1827 	if (!dmu_tx_is_syncing(tx))
1828 		return (0);
1829 
1830 	if (owner != NULL) {
1831 		VERIFY3P(ds->ds_owner, ==, owner);
1832 		dsl_dataset_long_rele(ds, owner);
1833 	}
1834 
1835 	held = dsl_dataset_long_held(ds);
1836 
1837 	if (owner != NULL)
1838 		dsl_dataset_long_hold(ds, owner);
1839 
1840 	if (held)
1841 		return (SET_ERROR(EBUSY));
1842 
1843 	return (0);
1844 }
1845 
1846 typedef struct dsl_dataset_rollback_arg {
1847 	const char *ddra_fsname;
1848 	void *ddra_owner;
1849 	nvlist_t *ddra_result;
1850 } dsl_dataset_rollback_arg_t;
1851 
1852 static int
1853 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1854 {
1855 	dsl_dataset_rollback_arg_t *ddra = arg;
1856 	dsl_pool_t *dp = dmu_tx_pool(tx);
1857 	dsl_dataset_t *ds;
1858 	int64_t unused_refres_delta;
1859 	int error;
1860 
1861 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1862 	if (error != 0)
1863 		return (error);
1864 
1865 	/* must not be a snapshot */
1866 	if (dsl_dataset_is_snapshot(ds)) {
1867 		dsl_dataset_rele(ds, FTAG);
1868 		return (SET_ERROR(EINVAL));
1869 	}
1870 
1871 	/* must have a most recent snapshot */
1872 	if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) {
1873 		dsl_dataset_rele(ds, FTAG);
1874 		return (SET_ERROR(EINVAL));
1875 	}
1876 
1877 	/* must not have any bookmarks after the most recent snapshot */
1878 	nvlist_t *proprequest = fnvlist_alloc();
1879 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
1880 	nvlist_t *bookmarks = fnvlist_alloc();
1881 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
1882 	fnvlist_free(proprequest);
1883 	if (error != 0)
1884 		return (error);
1885 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
1886 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
1887 		nvlist_t *valuenv =
1888 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
1889 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
1890 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
1891 		if (createtxg > ds->ds_phys->ds_prev_snap_txg) {
1892 			fnvlist_free(bookmarks);
1893 			dsl_dataset_rele(ds, FTAG);
1894 			return (SET_ERROR(EEXIST));
1895 		}
1896 	}
1897 	fnvlist_free(bookmarks);
1898 
1899 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
1900 	if (error != 0) {
1901 		dsl_dataset_rele(ds, FTAG);
1902 		return (error);
1903 	}
1904 
1905 	/*
1906 	 * Check if the snap we are rolling back to uses more than
1907 	 * the refquota.
1908 	 */
1909 	if (ds->ds_quota != 0 &&
1910 	    ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) {
1911 		dsl_dataset_rele(ds, FTAG);
1912 		return (SET_ERROR(EDQUOT));
1913 	}
1914 
1915 	/*
1916 	 * When we do the clone swap, we will temporarily use more space
1917 	 * due to the refreservation (the head will no longer have any
1918 	 * unique space, so the entire amount of the refreservation will need
1919 	 * to be free).  We will immediately destroy the clone, freeing
1920 	 * this space, but the freeing happens over many txg's.
1921 	 */
1922 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1923 	    ds->ds_phys->ds_unique_bytes);
1924 
1925 	if (unused_refres_delta > 0 &&
1926 	    unused_refres_delta >
1927 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1928 		dsl_dataset_rele(ds, FTAG);
1929 		return (SET_ERROR(ENOSPC));
1930 	}
1931 
1932 	dsl_dataset_rele(ds, FTAG);
1933 	return (0);
1934 }
1935 
1936 static void
1937 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1938 {
1939 	dsl_dataset_rollback_arg_t *ddra = arg;
1940 	dsl_pool_t *dp = dmu_tx_pool(tx);
1941 	dsl_dataset_t *ds, *clone;
1942 	uint64_t cloneobj;
1943 	char namebuf[ZFS_MAXNAMELEN];
1944 
1945 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
1946 
1947 	dsl_dataset_name(ds->ds_prev, namebuf);
1948 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
1949 
1950 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
1951 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
1952 
1953 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
1954 
1955 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
1956 	dsl_dataset_zero_zil(ds, tx);
1957 
1958 	dsl_destroy_head_sync_impl(clone, tx);
1959 
1960 	dsl_dataset_rele(clone, FTAG);
1961 	dsl_dataset_rele(ds, FTAG);
1962 }
1963 
1964 /*
1965  * Rolls back the given filesystem or volume to the most recent snapshot.
1966  * The name of the most recent snapshot will be returned under key "target"
1967  * in the result nvlist.
1968  *
1969  * If owner != NULL:
1970  * - The existing dataset MUST be owned by the specified owner at entry
1971  * - Upon return, dataset will still be held by the same owner, whether we
1972  *   succeed or not.
1973  *
1974  * This mode is required any time the existing filesystem is mounted.  See
1975  * notes above zfs_suspend_fs() for further details.
1976  */
1977 int
1978 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
1979 {
1980 	dsl_dataset_rollback_arg_t ddra;
1981 
1982 	ddra.ddra_fsname = fsname;
1983 	ddra.ddra_owner = owner;
1984 	ddra.ddra_result = result;
1985 
1986 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
1987 	    dsl_dataset_rollback_sync, &ddra,
1988 	    1, ZFS_SPACE_CHECK_RESERVED));
1989 }
1990 
1991 struct promotenode {
1992 	list_node_t link;
1993 	dsl_dataset_t *ds;
1994 };
1995 
1996 typedef struct dsl_dataset_promote_arg {
1997 	const char *ddpa_clonename;
1998 	dsl_dataset_t *ddpa_clone;
1999 	list_t shared_snaps, origin_snaps, clone_snaps;
2000 	dsl_dataset_t *origin_origin; /* origin of the origin */
2001 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2002 	char *err_ds;
2003 	cred_t *cr;
2004 } dsl_dataset_promote_arg_t;
2005 
2006 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2007 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2008     void *tag);
2009 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2010 
2011 static int
2012 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2013 {
2014 	dsl_dataset_promote_arg_t *ddpa = arg;
2015 	dsl_pool_t *dp = dmu_tx_pool(tx);
2016 	dsl_dataset_t *hds;
2017 	struct promotenode *snap;
2018 	dsl_dataset_t *origin_ds;
2019 	int err;
2020 	uint64_t unused;
2021 	uint64_t ss_mv_cnt;
2022 
2023 	err = promote_hold(ddpa, dp, FTAG);
2024 	if (err != 0)
2025 		return (err);
2026 
2027 	hds = ddpa->ddpa_clone;
2028 
2029 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
2030 		promote_rele(ddpa, FTAG);
2031 		return (SET_ERROR(EXDEV));
2032 	}
2033 
2034 	/*
2035 	 * Compute and check the amount of space to transfer.  Since this is
2036 	 * so expensive, don't do the preliminary check.
2037 	 */
2038 	if (!dmu_tx_is_syncing(tx)) {
2039 		promote_rele(ddpa, FTAG);
2040 		return (0);
2041 	}
2042 
2043 	snap = list_head(&ddpa->shared_snaps);
2044 	origin_ds = snap->ds;
2045 
2046 	/* compute origin's new unique space */
2047 	snap = list_tail(&ddpa->clone_snaps);
2048 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2049 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2050 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2051 	    &ddpa->unique, &unused, &unused);
2052 
2053 	/*
2054 	 * Walk the snapshots that we are moving
2055 	 *
2056 	 * Compute space to transfer.  Consider the incremental changes
2057 	 * to used by each snapshot:
2058 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2059 	 * So each snapshot gave birth to:
2060 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2061 	 * So a sequence would look like:
2062 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2063 	 * Which simplifies to:
2064 	 * uN + kN + kN-1 + ... + k1 + k0
2065 	 * Note however, if we stop before we reach the ORIGIN we get:
2066 	 * uN + kN + kN-1 + ... + kM - uM-1
2067 	 */
2068 	ss_mv_cnt = 0;
2069 	ddpa->used = origin_ds->ds_phys->ds_referenced_bytes;
2070 	ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2071 	ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2072 	for (snap = list_head(&ddpa->shared_snaps); snap;
2073 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2074 		uint64_t val, dlused, dlcomp, dluncomp;
2075 		dsl_dataset_t *ds = snap->ds;
2076 
2077 		ss_mv_cnt++;
2078 
2079 		/*
2080 		 * If there are long holds, we won't be able to evict
2081 		 * the objset.
2082 		 */
2083 		if (dsl_dataset_long_held(ds)) {
2084 			err = SET_ERROR(EBUSY);
2085 			goto out;
2086 		}
2087 
2088 		/* Check that the snapshot name does not conflict */
2089 		VERIFY0(dsl_dataset_get_snapname(ds));
2090 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2091 		if (err == 0) {
2092 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2093 			err = SET_ERROR(EEXIST);
2094 			goto out;
2095 		}
2096 		if (err != ENOENT)
2097 			goto out;
2098 
2099 		/* The very first snapshot does not have a deadlist */
2100 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2101 			continue;
2102 
2103 		dsl_deadlist_space(&ds->ds_deadlist,
2104 		    &dlused, &dlcomp, &dluncomp);
2105 		ddpa->used += dlused;
2106 		ddpa->comp += dlcomp;
2107 		ddpa->uncomp += dluncomp;
2108 	}
2109 
2110 	/*
2111 	 * If we are a clone of a clone then we never reached ORIGIN,
2112 	 * so we need to subtract out the clone origin's used space.
2113 	 */
2114 	if (ddpa->origin_origin) {
2115 		ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes;
2116 		ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes;
2117 		ddpa->uncomp -=
2118 		    ddpa->origin_origin->ds_phys->ds_uncompressed_bytes;
2119 	}
2120 
2121 	/* Check that there is enough space and limit headroom here */
2122 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2123 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2124 	if (err != 0)
2125 		goto out;
2126 
2127 	/*
2128 	 * Compute the amounts of space that will be used by snapshots
2129 	 * after the promotion (for both origin and clone).  For each,
2130 	 * it is the amount of space that will be on all of their
2131 	 * deadlists (that was not born before their new origin).
2132 	 */
2133 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2134 		uint64_t space;
2135 
2136 		/*
2137 		 * Note, typically this will not be a clone of a clone,
2138 		 * so dd_origin_txg will be < TXG_INITIAL, so
2139 		 * these snaplist_space() -> dsl_deadlist_space_range()
2140 		 * calls will be fast because they do not have to
2141 		 * iterate over all bps.
2142 		 */
2143 		snap = list_head(&ddpa->origin_snaps);
2144 		err = snaplist_space(&ddpa->shared_snaps,
2145 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2146 		if (err != 0)
2147 			goto out;
2148 
2149 		err = snaplist_space(&ddpa->clone_snaps,
2150 		    snap->ds->ds_dir->dd_origin_txg, &space);
2151 		if (err != 0)
2152 			goto out;
2153 		ddpa->cloneusedsnap += space;
2154 	}
2155 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2156 		err = snaplist_space(&ddpa->origin_snaps,
2157 		    origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap);
2158 		if (err != 0)
2159 			goto out;
2160 	}
2161 
2162 out:
2163 	promote_rele(ddpa, FTAG);
2164 	return (err);
2165 }
2166 
2167 static void
2168 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2169 {
2170 	dsl_dataset_promote_arg_t *ddpa = arg;
2171 	dsl_pool_t *dp = dmu_tx_pool(tx);
2172 	dsl_dataset_t *hds;
2173 	struct promotenode *snap;
2174 	dsl_dataset_t *origin_ds;
2175 	dsl_dataset_t *origin_head;
2176 	dsl_dir_t *dd;
2177 	dsl_dir_t *odd = NULL;
2178 	uint64_t oldnext_obj;
2179 	int64_t delta;
2180 
2181 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2182 	hds = ddpa->ddpa_clone;
2183 
2184 	ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE);
2185 
2186 	snap = list_head(&ddpa->shared_snaps);
2187 	origin_ds = snap->ds;
2188 	dd = hds->ds_dir;
2189 
2190 	snap = list_head(&ddpa->origin_snaps);
2191 	origin_head = snap->ds;
2192 
2193 	/*
2194 	 * We need to explicitly open odd, since origin_ds's dd will be
2195 	 * changing.
2196 	 */
2197 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2198 	    NULL, FTAG, &odd));
2199 
2200 	/* change origin's next snap */
2201 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2202 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2203 	snap = list_tail(&ddpa->clone_snaps);
2204 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2205 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2206 
2207 	/* change the origin's next clone */
2208 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2209 		dsl_dataset_remove_from_next_clones(origin_ds,
2210 		    snap->ds->ds_object, tx);
2211 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2212 		    origin_ds->ds_phys->ds_next_clones_obj,
2213 		    oldnext_obj, tx));
2214 	}
2215 
2216 	/* change origin */
2217 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2218 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2219 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2220 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2221 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2222 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2223 	origin_head->ds_dir->dd_origin_txg =
2224 	    origin_ds->ds_phys->ds_creation_txg;
2225 
2226 	/* change dd_clone entries */
2227 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2228 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2229 		    odd->dd_phys->dd_clones, hds->ds_object, tx));
2230 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2231 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2232 		    hds->ds_object, tx));
2233 
2234 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2235 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2236 		    origin_head->ds_object, tx));
2237 		if (dd->dd_phys->dd_clones == 0) {
2238 			dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2239 			    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2240 		}
2241 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2242 		    dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2243 	}
2244 
2245 	/* move snapshots to this dir */
2246 	for (snap = list_head(&ddpa->shared_snaps); snap;
2247 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2248 		dsl_dataset_t *ds = snap->ds;
2249 
2250 		/*
2251 		 * Property callbacks are registered to a particular
2252 		 * dsl_dir.  Since ours is changing, evict the objset
2253 		 * so that they will be unregistered from the old dsl_dir.
2254 		 */
2255 		if (ds->ds_objset) {
2256 			dmu_objset_evict(ds->ds_objset);
2257 			ds->ds_objset = NULL;
2258 		}
2259 
2260 		/* move snap name entry */
2261 		VERIFY0(dsl_dataset_get_snapname(ds));
2262 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2263 		    ds->ds_snapname, tx, B_TRUE));
2264 		VERIFY0(zap_add(dp->dp_meta_objset,
2265 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2266 		    8, 1, &ds->ds_object, tx));
2267 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2268 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2269 
2270 		/* change containing dsl_dir */
2271 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2272 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2273 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2274 		ASSERT3P(ds->ds_dir, ==, odd);
2275 		dsl_dir_rele(ds->ds_dir, ds);
2276 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2277 		    NULL, ds, &ds->ds_dir));
2278 
2279 		/* move any clone references */
2280 		if (ds->ds_phys->ds_next_clones_obj &&
2281 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2282 			zap_cursor_t zc;
2283 			zap_attribute_t za;
2284 
2285 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2286 			    ds->ds_phys->ds_next_clones_obj);
2287 			    zap_cursor_retrieve(&zc, &za) == 0;
2288 			    zap_cursor_advance(&zc)) {
2289 				dsl_dataset_t *cnds;
2290 				uint64_t o;
2291 
2292 				if (za.za_first_integer == oldnext_obj) {
2293 					/*
2294 					 * We've already moved the
2295 					 * origin's reference.
2296 					 */
2297 					continue;
2298 				}
2299 
2300 				VERIFY0(dsl_dataset_hold_obj(dp,
2301 				    za.za_first_integer, FTAG, &cnds));
2302 				o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2303 
2304 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2305 				    odd->dd_phys->dd_clones, o, tx));
2306 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2307 				    dd->dd_phys->dd_clones, o, tx));
2308 				dsl_dataset_rele(cnds, FTAG);
2309 			}
2310 			zap_cursor_fini(&zc);
2311 		}
2312 
2313 		ASSERT(!dsl_prop_hascb(ds));
2314 	}
2315 
2316 	/*
2317 	 * Change space accounting.
2318 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2319 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2320 	 * is true for each of {clone,origin} independently.
2321 	 */
2322 
2323 	delta = ddpa->cloneusedsnap -
2324 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2325 	ASSERT3S(delta, >=, 0);
2326 	ASSERT3U(ddpa->used, >=, delta);
2327 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2328 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2329 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2330 
2331 	delta = ddpa->originusedsnap -
2332 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2333 	ASSERT3S(delta, <=, 0);
2334 	ASSERT3U(ddpa->used, >=, -delta);
2335 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2336 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2337 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2338 
2339 	origin_ds->ds_phys->ds_unique_bytes = ddpa->unique;
2340 
2341 	/* log history record */
2342 	spa_history_log_internal_ds(hds, "promote", tx, "");
2343 
2344 	dsl_dir_rele(odd, FTAG);
2345 	promote_rele(ddpa, FTAG);
2346 }
2347 
2348 /*
2349  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2350  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2351  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2352  * snapshots back to this dataset's origin.
2353  */
2354 static int
2355 snaplist_make(dsl_pool_t *dp,
2356     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2357 {
2358 	uint64_t obj = last_obj;
2359 
2360 	list_create(l, sizeof (struct promotenode),
2361 	    offsetof(struct promotenode, link));
2362 
2363 	while (obj != first_obj) {
2364 		dsl_dataset_t *ds;
2365 		struct promotenode *snap;
2366 		int err;
2367 
2368 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2369 		ASSERT(err != ENOENT);
2370 		if (err != 0)
2371 			return (err);
2372 
2373 		if (first_obj == 0)
2374 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2375 
2376 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2377 		snap->ds = ds;
2378 		list_insert_tail(l, snap);
2379 		obj = ds->ds_phys->ds_prev_snap_obj;
2380 	}
2381 
2382 	return (0);
2383 }
2384 
2385 static int
2386 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2387 {
2388 	struct promotenode *snap;
2389 
2390 	*spacep = 0;
2391 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2392 		uint64_t used, comp, uncomp;
2393 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2394 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2395 		*spacep += used;
2396 	}
2397 	return (0);
2398 }
2399 
2400 static void
2401 snaplist_destroy(list_t *l, void *tag)
2402 {
2403 	struct promotenode *snap;
2404 
2405 	if (l == NULL || !list_link_active(&l->list_head))
2406 		return;
2407 
2408 	while ((snap = list_tail(l)) != NULL) {
2409 		list_remove(l, snap);
2410 		dsl_dataset_rele(snap->ds, tag);
2411 		kmem_free(snap, sizeof (*snap));
2412 	}
2413 	list_destroy(l);
2414 }
2415 
2416 static int
2417 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2418 {
2419 	int error;
2420 	dsl_dir_t *dd;
2421 	struct promotenode *snap;
2422 
2423 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2424 	    &ddpa->ddpa_clone);
2425 	if (error != 0)
2426 		return (error);
2427 	dd = ddpa->ddpa_clone->ds_dir;
2428 
2429 	if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2430 	    !dsl_dir_is_clone(dd)) {
2431 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2432 		return (SET_ERROR(EINVAL));
2433 	}
2434 
2435 	error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj,
2436 	    &ddpa->shared_snaps, tag);
2437 	if (error != 0)
2438 		goto out;
2439 
2440 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2441 	    &ddpa->clone_snaps, tag);
2442 	if (error != 0)
2443 		goto out;
2444 
2445 	snap = list_head(&ddpa->shared_snaps);
2446 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2447 	error = snaplist_make(dp, dd->dd_phys->dd_origin_obj,
2448 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj,
2449 	    &ddpa->origin_snaps, tag);
2450 	if (error != 0)
2451 		goto out;
2452 
2453 	if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2454 		error = dsl_dataset_hold_obj(dp,
2455 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2456 		    tag, &ddpa->origin_origin);
2457 		if (error != 0)
2458 			goto out;
2459 	}
2460 out:
2461 	if (error != 0)
2462 		promote_rele(ddpa, tag);
2463 	return (error);
2464 }
2465 
2466 static void
2467 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2468 {
2469 	snaplist_destroy(&ddpa->shared_snaps, tag);
2470 	snaplist_destroy(&ddpa->clone_snaps, tag);
2471 	snaplist_destroy(&ddpa->origin_snaps, tag);
2472 	if (ddpa->origin_origin != NULL)
2473 		dsl_dataset_rele(ddpa->origin_origin, tag);
2474 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2475 }
2476 
2477 /*
2478  * Promote a clone.
2479  *
2480  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2481  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2482  */
2483 int
2484 dsl_dataset_promote(const char *name, char *conflsnap)
2485 {
2486 	dsl_dataset_promote_arg_t ddpa = { 0 };
2487 	uint64_t numsnaps;
2488 	int error;
2489 	objset_t *os;
2490 
2491 	/*
2492 	 * We will modify space proportional to the number of
2493 	 * snapshots.  Compute numsnaps.
2494 	 */
2495 	error = dmu_objset_hold(name, FTAG, &os);
2496 	if (error != 0)
2497 		return (error);
2498 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2499 	    dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps);
2500 	dmu_objset_rele(os, FTAG);
2501 	if (error != 0)
2502 		return (error);
2503 
2504 	ddpa.ddpa_clonename = name;
2505 	ddpa.err_ds = conflsnap;
2506 	ddpa.cr = CRED();
2507 
2508 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2509 	    dsl_dataset_promote_sync, &ddpa,
2510 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2511 }
2512 
2513 int
2514 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2515     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2516 {
2517 	int64_t unused_refres_delta;
2518 
2519 	/* they should both be heads */
2520 	if (dsl_dataset_is_snapshot(clone) ||
2521 	    dsl_dataset_is_snapshot(origin_head))
2522 		return (SET_ERROR(EINVAL));
2523 
2524 	/* if we are not forcing, the branch point should be just before them */
2525 	if (!force && clone->ds_prev != origin_head->ds_prev)
2526 		return (SET_ERROR(EINVAL));
2527 
2528 	/* clone should be the clone (unless they are unrelated) */
2529 	if (clone->ds_prev != NULL &&
2530 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2531 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2532 		return (SET_ERROR(EINVAL));
2533 
2534 	/* the clone should be a child of the origin */
2535 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2536 		return (SET_ERROR(EINVAL));
2537 
2538 	/* origin_head shouldn't be modified unless 'force' */
2539 	if (!force &&
2540 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2541 		return (SET_ERROR(ETXTBSY));
2542 
2543 	/* origin_head should have no long holds (e.g. is not mounted) */
2544 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2545 		return (SET_ERROR(EBUSY));
2546 
2547 	/* check amount of any unconsumed refreservation */
2548 	unused_refres_delta =
2549 	    (int64_t)MIN(origin_head->ds_reserved,
2550 	    origin_head->ds_phys->ds_unique_bytes) -
2551 	    (int64_t)MIN(origin_head->ds_reserved,
2552 	    clone->ds_phys->ds_unique_bytes);
2553 
2554 	if (unused_refres_delta > 0 &&
2555 	    unused_refres_delta >
2556 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2557 		return (SET_ERROR(ENOSPC));
2558 
2559 	/* clone can't be over the head's refquota */
2560 	if (origin_head->ds_quota != 0 &&
2561 	    clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota)
2562 		return (SET_ERROR(EDQUOT));
2563 
2564 	return (0);
2565 }
2566 
2567 void
2568 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2569     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2570 {
2571 	dsl_pool_t *dp = dmu_tx_pool(tx);
2572 	int64_t unused_refres_delta;
2573 
2574 	ASSERT(clone->ds_reserved == 0);
2575 	ASSERT(origin_head->ds_quota == 0 ||
2576 	    clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota);
2577 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2578 
2579 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2580 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2581 
2582 	if (clone->ds_objset != NULL) {
2583 		dmu_objset_evict(clone->ds_objset);
2584 		clone->ds_objset = NULL;
2585 	}
2586 
2587 	if (origin_head->ds_objset != NULL) {
2588 		dmu_objset_evict(origin_head->ds_objset);
2589 		origin_head->ds_objset = NULL;
2590 	}
2591 
2592 	unused_refres_delta =
2593 	    (int64_t)MIN(origin_head->ds_reserved,
2594 	    origin_head->ds_phys->ds_unique_bytes) -
2595 	    (int64_t)MIN(origin_head->ds_reserved,
2596 	    clone->ds_phys->ds_unique_bytes);
2597 
2598 	/*
2599 	 * Reset origin's unique bytes, if it exists.
2600 	 */
2601 	if (clone->ds_prev) {
2602 		dsl_dataset_t *origin = clone->ds_prev;
2603 		uint64_t comp, uncomp;
2604 
2605 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2606 		dsl_deadlist_space_range(&clone->ds_deadlist,
2607 		    origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2608 		    &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
2609 	}
2610 
2611 	/* swap blkptrs */
2612 	{
2613 		blkptr_t tmp;
2614 		tmp = origin_head->ds_phys->ds_bp;
2615 		origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp;
2616 		clone->ds_phys->ds_bp = tmp;
2617 	}
2618 
2619 	/* set dd_*_bytes */
2620 	{
2621 		int64_t dused, dcomp, duncomp;
2622 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2623 		uint64_t odl_used, odl_comp, odl_uncomp;
2624 
2625 		ASSERT3U(clone->ds_dir->dd_phys->
2626 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2627 
2628 		dsl_deadlist_space(&clone->ds_deadlist,
2629 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2630 		dsl_deadlist_space(&origin_head->ds_deadlist,
2631 		    &odl_used, &odl_comp, &odl_uncomp);
2632 
2633 		dused = clone->ds_phys->ds_referenced_bytes + cdl_used -
2634 		    (origin_head->ds_phys->ds_referenced_bytes + odl_used);
2635 		dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp -
2636 		    (origin_head->ds_phys->ds_compressed_bytes + odl_comp);
2637 		duncomp = clone->ds_phys->ds_uncompressed_bytes +
2638 		    cdl_uncomp -
2639 		    (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2640 
2641 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2642 		    dused, dcomp, duncomp, tx);
2643 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2644 		    -dused, -dcomp, -duncomp, tx);
2645 
2646 		/*
2647 		 * The difference in the space used by snapshots is the
2648 		 * difference in snapshot space due to the head's
2649 		 * deadlist (since that's the only thing that's
2650 		 * changing that affects the snapused).
2651 		 */
2652 		dsl_deadlist_space_range(&clone->ds_deadlist,
2653 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2654 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2655 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2656 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2657 		    &odl_used, &odl_comp, &odl_uncomp);
2658 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2659 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2660 	}
2661 
2662 	/* swap ds_*_bytes */
2663 	SWITCH64(origin_head->ds_phys->ds_referenced_bytes,
2664 	    clone->ds_phys->ds_referenced_bytes);
2665 	SWITCH64(origin_head->ds_phys->ds_compressed_bytes,
2666 	    clone->ds_phys->ds_compressed_bytes);
2667 	SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes,
2668 	    clone->ds_phys->ds_uncompressed_bytes);
2669 	SWITCH64(origin_head->ds_phys->ds_unique_bytes,
2670 	    clone->ds_phys->ds_unique_bytes);
2671 
2672 	/* apply any parent delta for change in unconsumed refreservation */
2673 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2674 	    unused_refres_delta, 0, 0, tx);
2675 
2676 	/*
2677 	 * Swap deadlists.
2678 	 */
2679 	dsl_deadlist_close(&clone->ds_deadlist);
2680 	dsl_deadlist_close(&origin_head->ds_deadlist);
2681 	SWITCH64(origin_head->ds_phys->ds_deadlist_obj,
2682 	    clone->ds_phys->ds_deadlist_obj);
2683 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2684 	    clone->ds_phys->ds_deadlist_obj);
2685 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2686 	    origin_head->ds_phys->ds_deadlist_obj);
2687 
2688 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2689 
2690 	spa_history_log_internal_ds(clone, "clone swap", tx,
2691 	    "parent=%s", origin_head->ds_dir->dd_myname);
2692 }
2693 
2694 /*
2695  * Given a pool name and a dataset object number in that pool,
2696  * return the name of that dataset.
2697  */
2698 int
2699 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2700 {
2701 	dsl_pool_t *dp;
2702 	dsl_dataset_t *ds;
2703 	int error;
2704 
2705 	error = dsl_pool_hold(pname, FTAG, &dp);
2706 	if (error != 0)
2707 		return (error);
2708 
2709 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2710 	if (error == 0) {
2711 		dsl_dataset_name(ds, buf);
2712 		dsl_dataset_rele(ds, FTAG);
2713 	}
2714 	dsl_pool_rele(dp, FTAG);
2715 
2716 	return (error);
2717 }
2718 
2719 int
2720 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2721     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2722 {
2723 	int error = 0;
2724 
2725 	ASSERT3S(asize, >, 0);
2726 
2727 	/*
2728 	 * *ref_rsrv is the portion of asize that will come from any
2729 	 * unconsumed refreservation space.
2730 	 */
2731 	*ref_rsrv = 0;
2732 
2733 	mutex_enter(&ds->ds_lock);
2734 	/*
2735 	 * Make a space adjustment for reserved bytes.
2736 	 */
2737 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2738 		ASSERT3U(*used, >=,
2739 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2740 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2741 		*ref_rsrv =
2742 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2743 	}
2744 
2745 	if (!check_quota || ds->ds_quota == 0) {
2746 		mutex_exit(&ds->ds_lock);
2747 		return (0);
2748 	}
2749 	/*
2750 	 * If they are requesting more space, and our current estimate
2751 	 * is over quota, they get to try again unless the actual
2752 	 * on-disk is over quota and there are no pending changes (which
2753 	 * may free up space for us).
2754 	 */
2755 	if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
2756 		if (inflight > 0 ||
2757 		    ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
2758 			error = SET_ERROR(ERESTART);
2759 		else
2760 			error = SET_ERROR(EDQUOT);
2761 	}
2762 	mutex_exit(&ds->ds_lock);
2763 
2764 	return (error);
2765 }
2766 
2767 typedef struct dsl_dataset_set_qr_arg {
2768 	const char *ddsqra_name;
2769 	zprop_source_t ddsqra_source;
2770 	uint64_t ddsqra_value;
2771 } dsl_dataset_set_qr_arg_t;
2772 
2773 
2774 /* ARGSUSED */
2775 static int
2776 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2777 {
2778 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2779 	dsl_pool_t *dp = dmu_tx_pool(tx);
2780 	dsl_dataset_t *ds;
2781 	int error;
2782 	uint64_t newval;
2783 
2784 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2785 		return (SET_ERROR(ENOTSUP));
2786 
2787 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2788 	if (error != 0)
2789 		return (error);
2790 
2791 	if (dsl_dataset_is_snapshot(ds)) {
2792 		dsl_dataset_rele(ds, FTAG);
2793 		return (SET_ERROR(EINVAL));
2794 	}
2795 
2796 	error = dsl_prop_predict(ds->ds_dir,
2797 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2798 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2799 	if (error != 0) {
2800 		dsl_dataset_rele(ds, FTAG);
2801 		return (error);
2802 	}
2803 
2804 	if (newval == 0) {
2805 		dsl_dataset_rele(ds, FTAG);
2806 		return (0);
2807 	}
2808 
2809 	if (newval < ds->ds_phys->ds_referenced_bytes ||
2810 	    newval < ds->ds_reserved) {
2811 		dsl_dataset_rele(ds, FTAG);
2812 		return (SET_ERROR(ENOSPC));
2813 	}
2814 
2815 	dsl_dataset_rele(ds, FTAG);
2816 	return (0);
2817 }
2818 
2819 static void
2820 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2821 {
2822 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2823 	dsl_pool_t *dp = dmu_tx_pool(tx);
2824 	dsl_dataset_t *ds;
2825 	uint64_t newval;
2826 
2827 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2828 
2829 	dsl_prop_set_sync_impl(ds,
2830 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2831 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2832 	    &ddsqra->ddsqra_value, tx);
2833 
2834 	VERIFY0(dsl_prop_get_int_ds(ds,
2835 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2836 
2837 	if (ds->ds_quota != newval) {
2838 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2839 		ds->ds_quota = newval;
2840 	}
2841 	dsl_dataset_rele(ds, FTAG);
2842 }
2843 
2844 int
2845 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2846     uint64_t refquota)
2847 {
2848 	dsl_dataset_set_qr_arg_t ddsqra;
2849 
2850 	ddsqra.ddsqra_name = dsname;
2851 	ddsqra.ddsqra_source = source;
2852 	ddsqra.ddsqra_value = refquota;
2853 
2854 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2855 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
2856 }
2857 
2858 static int
2859 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2860 {
2861 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2862 	dsl_pool_t *dp = dmu_tx_pool(tx);
2863 	dsl_dataset_t *ds;
2864 	int error;
2865 	uint64_t newval, unique;
2866 
2867 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2868 		return (SET_ERROR(ENOTSUP));
2869 
2870 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2871 	if (error != 0)
2872 		return (error);
2873 
2874 	if (dsl_dataset_is_snapshot(ds)) {
2875 		dsl_dataset_rele(ds, FTAG);
2876 		return (SET_ERROR(EINVAL));
2877 	}
2878 
2879 	error = dsl_prop_predict(ds->ds_dir,
2880 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2881 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2882 	if (error != 0) {
2883 		dsl_dataset_rele(ds, FTAG);
2884 		return (error);
2885 	}
2886 
2887 	/*
2888 	 * If we are doing the preliminary check in open context, the
2889 	 * space estimates may be inaccurate.
2890 	 */
2891 	if (!dmu_tx_is_syncing(tx)) {
2892 		dsl_dataset_rele(ds, FTAG);
2893 		return (0);
2894 	}
2895 
2896 	mutex_enter(&ds->ds_lock);
2897 	if (!DS_UNIQUE_IS_ACCURATE(ds))
2898 		dsl_dataset_recalc_head_uniq(ds);
2899 	unique = ds->ds_phys->ds_unique_bytes;
2900 	mutex_exit(&ds->ds_lock);
2901 
2902 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2903 		uint64_t delta = MAX(unique, newval) -
2904 		    MAX(unique, ds->ds_reserved);
2905 
2906 		if (delta >
2907 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2908 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2909 			dsl_dataset_rele(ds, FTAG);
2910 			return (SET_ERROR(ENOSPC));
2911 		}
2912 	}
2913 
2914 	dsl_dataset_rele(ds, FTAG);
2915 	return (0);
2916 }
2917 
2918 void
2919 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2920     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2921 {
2922 	uint64_t newval;
2923 	uint64_t unique;
2924 	int64_t delta;
2925 
2926 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2927 	    source, sizeof (value), 1, &value, tx);
2928 
2929 	VERIFY0(dsl_prop_get_int_ds(ds,
2930 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
2931 
2932 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2933 	mutex_enter(&ds->ds_dir->dd_lock);
2934 	mutex_enter(&ds->ds_lock);
2935 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2936 	unique = ds->ds_phys->ds_unique_bytes;
2937 	delta = MAX(0, (int64_t)(newval - unique)) -
2938 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
2939 	ds->ds_reserved = newval;
2940 	mutex_exit(&ds->ds_lock);
2941 
2942 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
2943 	mutex_exit(&ds->ds_dir->dd_lock);
2944 }
2945 
2946 static void
2947 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
2948 {
2949 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2950 	dsl_pool_t *dp = dmu_tx_pool(tx);
2951 	dsl_dataset_t *ds;
2952 
2953 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2954 	dsl_dataset_set_refreservation_sync_impl(ds,
2955 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
2956 	dsl_dataset_rele(ds, FTAG);
2957 }
2958 
2959 int
2960 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
2961     uint64_t refreservation)
2962 {
2963 	dsl_dataset_set_qr_arg_t ddsqra;
2964 
2965 	ddsqra.ddsqra_name = dsname;
2966 	ddsqra.ddsqra_source = source;
2967 	ddsqra.ddsqra_value = refreservation;
2968 
2969 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
2970 	    dsl_dataset_set_refreservation_sync, &ddsqra,
2971 	    0, ZFS_SPACE_CHECK_NONE));
2972 }
2973 
2974 /*
2975  * Return (in *usedp) the amount of space written in new that is not
2976  * present in oldsnap.  New may be a snapshot or the head.  Old must be
2977  * a snapshot before new, in new's filesystem (or its origin).  If not then
2978  * fail and return EINVAL.
2979  *
2980  * The written space is calculated by considering two components:  First, we
2981  * ignore any freed space, and calculate the written as new's used space
2982  * minus old's used space.  Next, we add in the amount of space that was freed
2983  * between the two snapshots, thus reducing new's used space relative to old's.
2984  * Specifically, this is the space that was born before old->ds_creation_txg,
2985  * and freed before new (ie. on new's deadlist or a previous deadlist).
2986  *
2987  * space freed                         [---------------------]
2988  * snapshots                       ---O-------O--------O-------O------
2989  *                                         oldsnap            new
2990  */
2991 int
2992 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
2993     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2994 {
2995 	int err = 0;
2996 	uint64_t snapobj;
2997 	dsl_pool_t *dp = new->ds_dir->dd_pool;
2998 
2999 	ASSERT(dsl_pool_config_held(dp));
3000 
3001 	*usedp = 0;
3002 	*usedp += new->ds_phys->ds_referenced_bytes;
3003 	*usedp -= oldsnap->ds_phys->ds_referenced_bytes;
3004 
3005 	*compp = 0;
3006 	*compp += new->ds_phys->ds_compressed_bytes;
3007 	*compp -= oldsnap->ds_phys->ds_compressed_bytes;
3008 
3009 	*uncompp = 0;
3010 	*uncompp += new->ds_phys->ds_uncompressed_bytes;
3011 	*uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
3012 
3013 	snapobj = new->ds_object;
3014 	while (snapobj != oldsnap->ds_object) {
3015 		dsl_dataset_t *snap;
3016 		uint64_t used, comp, uncomp;
3017 
3018 		if (snapobj == new->ds_object) {
3019 			snap = new;
3020 		} else {
3021 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3022 			if (err != 0)
3023 				break;
3024 		}
3025 
3026 		if (snap->ds_phys->ds_prev_snap_txg ==
3027 		    oldsnap->ds_phys->ds_creation_txg) {
3028 			/*
3029 			 * The blocks in the deadlist can not be born after
3030 			 * ds_prev_snap_txg, so get the whole deadlist space,
3031 			 * which is more efficient (especially for old-format
3032 			 * deadlists).  Unfortunately the deadlist code
3033 			 * doesn't have enough information to make this
3034 			 * optimization itself.
3035 			 */
3036 			dsl_deadlist_space(&snap->ds_deadlist,
3037 			    &used, &comp, &uncomp);
3038 		} else {
3039 			dsl_deadlist_space_range(&snap->ds_deadlist,
3040 			    0, oldsnap->ds_phys->ds_creation_txg,
3041 			    &used, &comp, &uncomp);
3042 		}
3043 		*usedp += used;
3044 		*compp += comp;
3045 		*uncompp += uncomp;
3046 
3047 		/*
3048 		 * If we get to the beginning of the chain of snapshots
3049 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3050 		 * was not a snapshot of/before new.
3051 		 */
3052 		snapobj = snap->ds_phys->ds_prev_snap_obj;
3053 		if (snap != new)
3054 			dsl_dataset_rele(snap, FTAG);
3055 		if (snapobj == 0) {
3056 			err = SET_ERROR(EINVAL);
3057 			break;
3058 		}
3059 
3060 	}
3061 	return (err);
3062 }
3063 
3064 /*
3065  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3066  * lastsnap, and all snapshots in between are deleted.
3067  *
3068  * blocks that would be freed            [---------------------------]
3069  * snapshots                       ---O-------O--------O-------O--------O
3070  *                                        firstsnap        lastsnap
3071  *
3072  * This is the set of blocks that were born after the snap before firstsnap,
3073  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3074  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3075  * We calculate this by iterating over the relevant deadlists (from the snap
3076  * after lastsnap, backward to the snap after firstsnap), summing up the
3077  * space on the deadlist that was born after the snap before firstsnap.
3078  */
3079 int
3080 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3081     dsl_dataset_t *lastsnap,
3082     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3083 {
3084 	int err = 0;
3085 	uint64_t snapobj;
3086 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3087 
3088 	ASSERT(dsl_dataset_is_snapshot(firstsnap));
3089 	ASSERT(dsl_dataset_is_snapshot(lastsnap));
3090 
3091 	/*
3092 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3093 	 * is before lastsnap.
3094 	 */
3095 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3096 	    firstsnap->ds_phys->ds_creation_txg >
3097 	    lastsnap->ds_phys->ds_creation_txg)
3098 		return (SET_ERROR(EINVAL));
3099 
3100 	*usedp = *compp = *uncompp = 0;
3101 
3102 	snapobj = lastsnap->ds_phys->ds_next_snap_obj;
3103 	while (snapobj != firstsnap->ds_object) {
3104 		dsl_dataset_t *ds;
3105 		uint64_t used, comp, uncomp;
3106 
3107 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3108 		if (err != 0)
3109 			break;
3110 
3111 		dsl_deadlist_space_range(&ds->ds_deadlist,
3112 		    firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3113 		    &used, &comp, &uncomp);
3114 		*usedp += used;
3115 		*compp += comp;
3116 		*uncompp += uncomp;
3117 
3118 		snapobj = ds->ds_phys->ds_prev_snap_obj;
3119 		ASSERT3U(snapobj, !=, 0);
3120 		dsl_dataset_rele(ds, FTAG);
3121 	}
3122 	return (err);
3123 }
3124 
3125 /*
3126  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3127  * For example, they could both be snapshots of the same filesystem, and
3128  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3129  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3130  * filesystem.  Or 'earlier' could be the origin's origin.
3131  *
3132  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3133  */
3134 boolean_t
3135 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3136 	uint64_t earlier_txg)
3137 {
3138 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3139 	int error;
3140 	boolean_t ret;
3141 
3142 	ASSERT(dsl_pool_config_held(dp));
3143 	ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
3144 
3145 	if (earlier_txg == 0)
3146 		earlier_txg = earlier->ds_phys->ds_creation_txg;
3147 
3148 	if (dsl_dataset_is_snapshot(later) &&
3149 	    earlier_txg >= later->ds_phys->ds_creation_txg)
3150 		return (B_FALSE);
3151 
3152 	if (later->ds_dir == earlier->ds_dir)
3153 		return (B_TRUE);
3154 	if (!dsl_dir_is_clone(later->ds_dir))
3155 		return (B_FALSE);
3156 
3157 	if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object)
3158 		return (B_TRUE);
3159 	dsl_dataset_t *origin;
3160 	error = dsl_dataset_hold_obj(dp,
3161 	    later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
3162 	if (error != 0)
3163 		return (B_FALSE);
3164 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3165 	dsl_dataset_rele(origin, FTAG);
3166 	return (ret);
3167 }
3168 
3169 
3170 void
3171 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3172 {
3173 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3174 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3175 }
3176