xref: /illumos-gate/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 447b1e1fca22e4de5e04623965fbb1460857930c)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
25  * Copyright (c) 2014 RackTop Systems.
26  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27  * Copyright (c) 2014 Integros [integros.com]
28  * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
29  */
30 
31 #include <sys/dmu_objset.h>
32 #include <sys/dsl_dataset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_prop.h>
35 #include <sys/dsl_synctask.h>
36 #include <sys/dmu_traverse.h>
37 #include <sys/dmu_impl.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/arc.h>
40 #include <sys/zio.h>
41 #include <sys/zap.h>
42 #include <sys/zfeature.h>
43 #include <sys/unique.h>
44 #include <sys/zfs_context.h>
45 #include <sys/zfs_ioctl.h>
46 #include <sys/spa.h>
47 #include <sys/zfs_znode.h>
48 #include <sys/zfs_onexit.h>
49 #include <sys/zvol.h>
50 #include <sys/dsl_scan.h>
51 #include <sys/dsl_deadlist.h>
52 #include <sys/dsl_destroy.h>
53 #include <sys/dsl_userhold.h>
54 #include <sys/dsl_bookmark.h>
55 #include <sys/dmu_send.h>
56 #include <sys/zio_checksum.h>
57 #include <sys/zio_compress.h>
58 #include <zfs_fletcher.h>
59 
60 /*
61  * The SPA supports block sizes up to 16MB.  However, very large blocks
62  * can have an impact on i/o latency (e.g. tying up a spinning disk for
63  * ~300ms), and also potentially on the memory allocator.  Therefore,
64  * we do not allow the recordsize to be set larger than zfs_max_recordsize
65  * (default 1MB).  Larger blocks can be created by changing this tunable,
66  * and pools with larger blocks can always be imported and used, regardless
67  * of this setting.
68  */
69 int zfs_max_recordsize = 1 * 1024 * 1024;
70 
71 #define	SWITCH64(x, y) \
72 	{ \
73 		uint64_t __tmp = (x); \
74 		(x) = (y); \
75 		(y) = __tmp; \
76 	}
77 
78 #define	DS_REF_MAX	(1ULL << 62)
79 
80 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
81 
82 extern int spa_asize_inflation;
83 
84 /*
85  * Figure out how much of this delta should be propogated to the dsl_dir
86  * layer.  If there's a refreservation, that space has already been
87  * partially accounted for in our ancestors.
88  */
89 static int64_t
90 parent_delta(dsl_dataset_t *ds, int64_t delta)
91 {
92 	dsl_dataset_phys_t *ds_phys;
93 	uint64_t old_bytes, new_bytes;
94 
95 	if (ds->ds_reserved == 0)
96 		return (delta);
97 
98 	ds_phys = dsl_dataset_phys(ds);
99 	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
100 	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
101 
102 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
103 	return (new_bytes - old_bytes);
104 }
105 
106 void
107 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
108 {
109 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
110 	int compressed = BP_GET_PSIZE(bp);
111 	int uncompressed = BP_GET_UCSIZE(bp);
112 	int64_t delta;
113 
114 	dprintf_bp(bp, "ds=%p", ds);
115 
116 	ASSERT(dmu_tx_is_syncing(tx));
117 	/* It could have been compressed away to nothing */
118 	if (BP_IS_HOLE(bp))
119 		return;
120 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
121 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
122 	if (ds == NULL) {
123 		dsl_pool_mos_diduse_space(tx->tx_pool,
124 		    used, compressed, uncompressed);
125 		return;
126 	}
127 
128 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
129 	mutex_enter(&ds->ds_lock);
130 	delta = parent_delta(ds, used);
131 	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
132 	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
133 	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
134 	dsl_dataset_phys(ds)->ds_unique_bytes += used;
135 
136 	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
137 		ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
138 		    B_TRUE;
139 	}
140 
141 	spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
142 	if (f != SPA_FEATURE_NONE)
143 		ds->ds_feature_activation_needed[f] = B_TRUE;
144 
145 	mutex_exit(&ds->ds_lock);
146 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
147 	    compressed, uncompressed, tx);
148 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
149 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
150 }
151 
152 int
153 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
154     boolean_t async)
155 {
156 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
157 	int compressed = BP_GET_PSIZE(bp);
158 	int uncompressed = BP_GET_UCSIZE(bp);
159 
160 	if (BP_IS_HOLE(bp))
161 		return (0);
162 
163 	ASSERT(dmu_tx_is_syncing(tx));
164 	ASSERT(bp->blk_birth <= tx->tx_txg);
165 
166 	if (ds == NULL) {
167 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
168 		dsl_pool_mos_diduse_space(tx->tx_pool,
169 		    -used, -compressed, -uncompressed);
170 		return (used);
171 	}
172 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
173 
174 	ASSERT(!ds->ds_is_snapshot);
175 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
176 
177 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
178 		int64_t delta;
179 
180 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
181 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
182 
183 		mutex_enter(&ds->ds_lock);
184 		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
185 		    !DS_UNIQUE_IS_ACCURATE(ds));
186 		delta = parent_delta(ds, -used);
187 		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
188 		mutex_exit(&ds->ds_lock);
189 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
190 		    delta, -compressed, -uncompressed, tx);
191 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
192 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
193 	} else {
194 		dprintf_bp(bp, "putting on dead list: %s", "");
195 		if (async) {
196 			/*
197 			 * We are here as part of zio's write done callback,
198 			 * which means we're a zio interrupt thread.  We can't
199 			 * call dsl_deadlist_insert() now because it may block
200 			 * waiting for I/O.  Instead, put bp on the deferred
201 			 * queue and let dsl_pool_sync() finish the job.
202 			 */
203 			bplist_append(&ds->ds_pending_deadlist, bp);
204 		} else {
205 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
206 		}
207 		ASSERT3U(ds->ds_prev->ds_object, ==,
208 		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
209 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
210 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
211 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
212 		    ds->ds_object && bp->blk_birth >
213 		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
214 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
215 			mutex_enter(&ds->ds_prev->ds_lock);
216 			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
217 			mutex_exit(&ds->ds_prev->ds_lock);
218 		}
219 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
220 			dsl_dir_transfer_space(ds->ds_dir, used,
221 			    DD_USED_HEAD, DD_USED_SNAP, tx);
222 		}
223 	}
224 	mutex_enter(&ds->ds_lock);
225 	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
226 	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
227 	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
228 	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
229 	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
230 	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
231 	mutex_exit(&ds->ds_lock);
232 
233 	return (used);
234 }
235 
236 uint64_t
237 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
238 {
239 	uint64_t trysnap = 0;
240 
241 	if (ds == NULL)
242 		return (0);
243 	/*
244 	 * The snapshot creation could fail, but that would cause an
245 	 * incorrect FALSE return, which would only result in an
246 	 * overestimation of the amount of space that an operation would
247 	 * consume, which is OK.
248 	 *
249 	 * There's also a small window where we could miss a pending
250 	 * snapshot, because we could set the sync task in the quiescing
251 	 * phase.  So this should only be used as a guess.
252 	 */
253 	if (ds->ds_trysnap_txg >
254 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
255 		trysnap = ds->ds_trysnap_txg;
256 	return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
257 }
258 
259 boolean_t
260 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
261     uint64_t blk_birth)
262 {
263 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
264 	    (bp != NULL && BP_IS_HOLE(bp)))
265 		return (B_FALSE);
266 
267 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
268 
269 	return (B_TRUE);
270 }
271 
272 static void
273 dsl_dataset_evict(void *dbu)
274 {
275 	dsl_dataset_t *ds = dbu;
276 
277 	ASSERT(ds->ds_owner == NULL);
278 
279 	ds->ds_dbuf = NULL;
280 
281 	unique_remove(ds->ds_fsid_guid);
282 
283 	if (ds->ds_objset != NULL)
284 		dmu_objset_evict(ds->ds_objset);
285 
286 	if (ds->ds_prev) {
287 		dsl_dataset_rele(ds->ds_prev, ds);
288 		ds->ds_prev = NULL;
289 	}
290 
291 	bplist_destroy(&ds->ds_pending_deadlist);
292 	if (ds->ds_deadlist.dl_os != NULL)
293 		dsl_deadlist_close(&ds->ds_deadlist);
294 	if (ds->ds_dir)
295 		dsl_dir_async_rele(ds->ds_dir, ds);
296 
297 	ASSERT(!list_link_active(&ds->ds_synced_link));
298 
299 	list_destroy(&ds->ds_prop_cbs);
300 	mutex_destroy(&ds->ds_lock);
301 	mutex_destroy(&ds->ds_opening_lock);
302 	mutex_destroy(&ds->ds_sendstream_lock);
303 	refcount_destroy(&ds->ds_longholds);
304 
305 	kmem_free(ds, sizeof (dsl_dataset_t));
306 }
307 
308 int
309 dsl_dataset_get_snapname(dsl_dataset_t *ds)
310 {
311 	dsl_dataset_phys_t *headphys;
312 	int err;
313 	dmu_buf_t *headdbuf;
314 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
315 	objset_t *mos = dp->dp_meta_objset;
316 
317 	if (ds->ds_snapname[0])
318 		return (0);
319 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
320 		return (0);
321 
322 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
323 	    FTAG, &headdbuf);
324 	if (err != 0)
325 		return (err);
326 	headphys = headdbuf->db_data;
327 	err = zap_value_search(dp->dp_meta_objset,
328 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
329 	dmu_buf_rele(headdbuf, FTAG);
330 	return (err);
331 }
332 
333 int
334 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
335 {
336 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
337 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
338 	matchtype_t mt;
339 	int err;
340 
341 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
342 		mt = MT_FIRST;
343 	else
344 		mt = MT_EXACT;
345 
346 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
347 	    value, mt, NULL, 0, NULL);
348 	if (err == ENOTSUP && mt == MT_FIRST)
349 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
350 	return (err);
351 }
352 
353 int
354 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
355     boolean_t adj_cnt)
356 {
357 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
358 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
359 	matchtype_t mt;
360 	int err;
361 
362 	dsl_dir_snap_cmtime_update(ds->ds_dir);
363 
364 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
365 		mt = MT_FIRST;
366 	else
367 		mt = MT_EXACT;
368 
369 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
370 	if (err == ENOTSUP && mt == MT_FIRST)
371 		err = zap_remove(mos, snapobj, name, tx);
372 
373 	if (err == 0 && adj_cnt)
374 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
375 		    DD_FIELD_SNAPSHOT_COUNT, tx);
376 
377 	return (err);
378 }
379 
380 boolean_t
381 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
382 {
383 	dmu_buf_t *dbuf = ds->ds_dbuf;
384 	boolean_t result = B_FALSE;
385 
386 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
387 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
388 
389 		if (ds == dmu_buf_get_user(dbuf))
390 			result = B_TRUE;
391 		else
392 			dmu_buf_rele(dbuf, tag);
393 	}
394 
395 	return (result);
396 }
397 
398 int
399 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
400     dsl_dataset_t **dsp)
401 {
402 	objset_t *mos = dp->dp_meta_objset;
403 	dmu_buf_t *dbuf;
404 	dsl_dataset_t *ds;
405 	int err;
406 	dmu_object_info_t doi;
407 
408 	ASSERT(dsl_pool_config_held(dp));
409 
410 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
411 	if (err != 0)
412 		return (err);
413 
414 	/* Make sure dsobj has the correct object type. */
415 	dmu_object_info_from_db(dbuf, &doi);
416 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
417 		dmu_buf_rele(dbuf, tag);
418 		return (SET_ERROR(EINVAL));
419 	}
420 
421 	ds = dmu_buf_get_user(dbuf);
422 	if (ds == NULL) {
423 		dsl_dataset_t *winner = NULL;
424 
425 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
426 		ds->ds_dbuf = dbuf;
427 		ds->ds_object = dsobj;
428 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
429 
430 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
431 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
432 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
433 		refcount_create(&ds->ds_longholds);
434 
435 		bplist_create(&ds->ds_pending_deadlist);
436 		dsl_deadlist_open(&ds->ds_deadlist,
437 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
438 
439 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
440 		    offsetof(dmu_sendarg_t, dsa_link));
441 
442 		list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
443 		    offsetof(dsl_prop_cb_record_t, cbr_ds_node));
444 
445 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
446 			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
447 				if (!(spa_feature_table[f].fi_flags &
448 				    ZFEATURE_FLAG_PER_DATASET))
449 					continue;
450 				err = zap_contains(mos, dsobj,
451 				    spa_feature_table[f].fi_guid);
452 				if (err == 0) {
453 					ds->ds_feature_inuse[f] = B_TRUE;
454 				} else {
455 					ASSERT3U(err, ==, ENOENT);
456 					err = 0;
457 				}
458 			}
459 		}
460 
461 		err = dsl_dir_hold_obj(dp,
462 		    dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
463 		if (err != 0) {
464 			mutex_destroy(&ds->ds_lock);
465 			mutex_destroy(&ds->ds_opening_lock);
466 			mutex_destroy(&ds->ds_sendstream_lock);
467 			refcount_destroy(&ds->ds_longholds);
468 			bplist_destroy(&ds->ds_pending_deadlist);
469 			dsl_deadlist_close(&ds->ds_deadlist);
470 			kmem_free(ds, sizeof (dsl_dataset_t));
471 			dmu_buf_rele(dbuf, tag);
472 			return (err);
473 		}
474 
475 		if (!ds->ds_is_snapshot) {
476 			ds->ds_snapname[0] = '\0';
477 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
478 				err = dsl_dataset_hold_obj(dp,
479 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
480 				    ds, &ds->ds_prev);
481 			}
482 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
483 				int zaperr = zap_lookup(mos, ds->ds_object,
484 				    DS_FIELD_BOOKMARK_NAMES,
485 				    sizeof (ds->ds_bookmarks), 1,
486 				    &ds->ds_bookmarks);
487 				if (zaperr != ENOENT)
488 					VERIFY0(zaperr);
489 			}
490 		} else {
491 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
492 				err = dsl_dataset_get_snapname(ds);
493 			if (err == 0 &&
494 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
495 				err = zap_count(
496 				    ds->ds_dir->dd_pool->dp_meta_objset,
497 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
498 				    &ds->ds_userrefs);
499 			}
500 		}
501 
502 		if (err == 0 && !ds->ds_is_snapshot) {
503 			err = dsl_prop_get_int_ds(ds,
504 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
505 			    &ds->ds_reserved);
506 			if (err == 0) {
507 				err = dsl_prop_get_int_ds(ds,
508 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
509 				    &ds->ds_quota);
510 			}
511 		} else {
512 			ds->ds_reserved = ds->ds_quota = 0;
513 		}
514 
515 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
516 		if (err == 0)
517 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
518 
519 		if (err != 0 || winner != NULL) {
520 			bplist_destroy(&ds->ds_pending_deadlist);
521 			dsl_deadlist_close(&ds->ds_deadlist);
522 			if (ds->ds_prev)
523 				dsl_dataset_rele(ds->ds_prev, ds);
524 			dsl_dir_rele(ds->ds_dir, ds);
525 			mutex_destroy(&ds->ds_lock);
526 			mutex_destroy(&ds->ds_opening_lock);
527 			mutex_destroy(&ds->ds_sendstream_lock);
528 			refcount_destroy(&ds->ds_longholds);
529 			kmem_free(ds, sizeof (dsl_dataset_t));
530 			if (err != 0) {
531 				dmu_buf_rele(dbuf, tag);
532 				return (err);
533 			}
534 			ds = winner;
535 		} else {
536 			ds->ds_fsid_guid =
537 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
538 		}
539 	}
540 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
541 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
542 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
543 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
544 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
545 	*dsp = ds;
546 	return (0);
547 }
548 
549 int
550 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
551     void *tag, dsl_dataset_t **dsp)
552 {
553 	dsl_dir_t *dd;
554 	const char *snapname;
555 	uint64_t obj;
556 	int err = 0;
557 	dsl_dataset_t *ds;
558 
559 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
560 	if (err != 0)
561 		return (err);
562 
563 	ASSERT(dsl_pool_config_held(dp));
564 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
565 	if (obj != 0)
566 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
567 	else
568 		err = SET_ERROR(ENOENT);
569 
570 	/* we may be looking for a snapshot */
571 	if (err == 0 && snapname != NULL) {
572 		dsl_dataset_t *snap_ds;
573 
574 		if (*snapname++ != '@') {
575 			dsl_dataset_rele(ds, tag);
576 			dsl_dir_rele(dd, FTAG);
577 			return (SET_ERROR(ENOENT));
578 		}
579 
580 		dprintf("looking for snapshot '%s'\n", snapname);
581 		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
582 		if (err == 0)
583 			err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
584 		dsl_dataset_rele(ds, tag);
585 
586 		if (err == 0) {
587 			mutex_enter(&snap_ds->ds_lock);
588 			if (snap_ds->ds_snapname[0] == 0)
589 				(void) strlcpy(snap_ds->ds_snapname, snapname,
590 				    sizeof (snap_ds->ds_snapname));
591 			mutex_exit(&snap_ds->ds_lock);
592 			ds = snap_ds;
593 		}
594 	}
595 	if (err == 0)
596 		*dsp = ds;
597 	dsl_dir_rele(dd, FTAG);
598 	return (err);
599 }
600 
601 int
602 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
603     void *tag, dsl_dataset_t **dsp)
604 {
605 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
606 	if (err != 0)
607 		return (err);
608 	if (!dsl_dataset_tryown(*dsp, tag)) {
609 		dsl_dataset_rele(*dsp, tag);
610 		*dsp = NULL;
611 		return (SET_ERROR(EBUSY));
612 	}
613 	return (0);
614 }
615 
616 int
617 dsl_dataset_own(dsl_pool_t *dp, const char *name,
618     void *tag, dsl_dataset_t **dsp)
619 {
620 	int err = dsl_dataset_hold(dp, name, tag, dsp);
621 	if (err != 0)
622 		return (err);
623 	if (!dsl_dataset_tryown(*dsp, tag)) {
624 		dsl_dataset_rele(*dsp, tag);
625 		return (SET_ERROR(EBUSY));
626 	}
627 	return (0);
628 }
629 
630 /*
631  * See the comment above dsl_pool_hold() for details.  In summary, a long
632  * hold is used to prevent destruction of a dataset while the pool hold
633  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
634  *
635  * The dataset and pool must be held when this function is called.  After it
636  * is called, the pool hold may be released while the dataset is still held
637  * and accessed.
638  */
639 void
640 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
641 {
642 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
643 	(void) refcount_add(&ds->ds_longholds, tag);
644 }
645 
646 void
647 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
648 {
649 	(void) refcount_remove(&ds->ds_longholds, tag);
650 }
651 
652 /* Return B_TRUE if there are any long holds on this dataset. */
653 boolean_t
654 dsl_dataset_long_held(dsl_dataset_t *ds)
655 {
656 	return (!refcount_is_zero(&ds->ds_longholds));
657 }
658 
659 void
660 dsl_dataset_name(dsl_dataset_t *ds, char *name)
661 {
662 	if (ds == NULL) {
663 		(void) strcpy(name, "mos");
664 	} else {
665 		dsl_dir_name(ds->ds_dir, name);
666 		VERIFY0(dsl_dataset_get_snapname(ds));
667 		if (ds->ds_snapname[0]) {
668 			VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
669 			    <, ZFS_MAX_DATASET_NAME_LEN);
670 			/*
671 			 * We use a "recursive" mutex so that we
672 			 * can call dprintf_ds() with ds_lock held.
673 			 */
674 			if (!MUTEX_HELD(&ds->ds_lock)) {
675 				mutex_enter(&ds->ds_lock);
676 				VERIFY3U(strlcat(name, ds->ds_snapname,
677 				    ZFS_MAX_DATASET_NAME_LEN), <,
678 				    ZFS_MAX_DATASET_NAME_LEN);
679 				mutex_exit(&ds->ds_lock);
680 			} else {
681 				VERIFY3U(strlcat(name, ds->ds_snapname,
682 				    ZFS_MAX_DATASET_NAME_LEN), <,
683 				    ZFS_MAX_DATASET_NAME_LEN);
684 			}
685 		}
686 	}
687 }
688 
689 int
690 dsl_dataset_namelen(dsl_dataset_t *ds)
691 {
692 	VERIFY0(dsl_dataset_get_snapname(ds));
693 	mutex_enter(&ds->ds_lock);
694 	int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
695 	mutex_exit(&ds->ds_lock);
696 	return (len);
697 }
698 
699 void
700 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
701 {
702 	dmu_buf_rele(ds->ds_dbuf, tag);
703 }
704 
705 void
706 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
707 {
708 	ASSERT3P(ds->ds_owner, ==, tag);
709 	ASSERT(ds->ds_dbuf != NULL);
710 
711 	mutex_enter(&ds->ds_lock);
712 	ds->ds_owner = NULL;
713 	mutex_exit(&ds->ds_lock);
714 	dsl_dataset_long_rele(ds, tag);
715 	dsl_dataset_rele(ds, tag);
716 }
717 
718 boolean_t
719 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
720 {
721 	boolean_t gotit = FALSE;
722 
723 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
724 	mutex_enter(&ds->ds_lock);
725 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
726 		ds->ds_owner = tag;
727 		dsl_dataset_long_hold(ds, tag);
728 		gotit = TRUE;
729 	}
730 	mutex_exit(&ds->ds_lock);
731 	return (gotit);
732 }
733 
734 boolean_t
735 dsl_dataset_has_owner(dsl_dataset_t *ds)
736 {
737 	boolean_t rv;
738 	mutex_enter(&ds->ds_lock);
739 	rv = (ds->ds_owner != NULL);
740 	mutex_exit(&ds->ds_lock);
741 	return (rv);
742 }
743 
744 static void
745 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
746 {
747 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
748 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
749 	uint64_t zero = 0;
750 
751 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
752 
753 	spa_feature_incr(spa, f, tx);
754 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
755 
756 	VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
757 	    sizeof (zero), 1, &zero, tx));
758 }
759 
760 void
761 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
762 {
763 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
764 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
765 
766 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
767 
768 	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
769 	spa_feature_decr(spa, f, tx);
770 }
771 
772 uint64_t
773 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
774     uint64_t flags, dmu_tx_t *tx)
775 {
776 	dsl_pool_t *dp = dd->dd_pool;
777 	dmu_buf_t *dbuf;
778 	dsl_dataset_phys_t *dsphys;
779 	uint64_t dsobj;
780 	objset_t *mos = dp->dp_meta_objset;
781 
782 	if (origin == NULL)
783 		origin = dp->dp_origin_snap;
784 
785 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
786 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
787 	ASSERT(dmu_tx_is_syncing(tx));
788 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
789 
790 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
791 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
792 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
793 	dmu_buf_will_dirty(dbuf, tx);
794 	dsphys = dbuf->db_data;
795 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
796 	dsphys->ds_dir_obj = dd->dd_object;
797 	dsphys->ds_flags = flags;
798 	dsphys->ds_fsid_guid = unique_create();
799 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
800 	    sizeof (dsphys->ds_guid));
801 	dsphys->ds_snapnames_zapobj =
802 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
803 	    DMU_OT_NONE, 0, tx);
804 	dsphys->ds_creation_time = gethrestime_sec();
805 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
806 
807 	if (origin == NULL) {
808 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
809 	} else {
810 		dsl_dataset_t *ohds; /* head of the origin snapshot */
811 
812 		dsphys->ds_prev_snap_obj = origin->ds_object;
813 		dsphys->ds_prev_snap_txg =
814 		    dsl_dataset_phys(origin)->ds_creation_txg;
815 		dsphys->ds_referenced_bytes =
816 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
817 		dsphys->ds_compressed_bytes =
818 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
819 		dsphys->ds_uncompressed_bytes =
820 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
821 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
822 
823 		/*
824 		 * Inherit flags that describe the dataset's contents
825 		 * (INCONSISTENT) or properties (Case Insensitive).
826 		 */
827 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
828 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
829 
830 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
831 			if (origin->ds_feature_inuse[f])
832 				dsl_dataset_activate_feature(dsobj, f, tx);
833 		}
834 
835 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
836 		dsl_dataset_phys(origin)->ds_num_children++;
837 
838 		VERIFY0(dsl_dataset_hold_obj(dp,
839 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
840 		    FTAG, &ohds));
841 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
842 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
843 		dsl_dataset_rele(ohds, FTAG);
844 
845 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
846 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
847 				dsl_dataset_phys(origin)->ds_next_clones_obj =
848 				    zap_create(mos,
849 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
850 			}
851 			VERIFY0(zap_add_int(mos,
852 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
853 			    dsobj, tx));
854 		}
855 
856 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
857 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
858 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
859 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
860 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
861 				dsl_dir_phys(origin->ds_dir)->dd_clones =
862 				    zap_create(mos,
863 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
864 			}
865 			VERIFY0(zap_add_int(mos,
866 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
867 			    dsobj, tx));
868 		}
869 	}
870 
871 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
872 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
873 
874 	dmu_buf_rele(dbuf, FTAG);
875 
876 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
877 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
878 
879 	return (dsobj);
880 }
881 
882 static void
883 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
884 {
885 	objset_t *os;
886 
887 	VERIFY0(dmu_objset_from_ds(ds, &os));
888 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
889 	dsl_dataset_dirty(ds, tx);
890 }
891 
892 uint64_t
893 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
894     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
895 {
896 	dsl_pool_t *dp = pdd->dd_pool;
897 	uint64_t dsobj, ddobj;
898 	dsl_dir_t *dd;
899 
900 	ASSERT(dmu_tx_is_syncing(tx));
901 	ASSERT(lastname[0] != '@');
902 
903 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
904 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
905 
906 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
907 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
908 
909 	dsl_deleg_set_create_perms(dd, tx, cr);
910 
911 	/*
912 	 * Since we're creating a new node we know it's a leaf, so we can
913 	 * initialize the counts if the limit feature is active.
914 	 */
915 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
916 		uint64_t cnt = 0;
917 		objset_t *os = dd->dd_pool->dp_meta_objset;
918 
919 		dsl_dir_zapify(dd, tx);
920 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
921 		    sizeof (cnt), 1, &cnt, tx));
922 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
923 		    sizeof (cnt), 1, &cnt, tx));
924 	}
925 
926 	dsl_dir_rele(dd, FTAG);
927 
928 	/*
929 	 * If we are creating a clone, make sure we zero out any stale
930 	 * data from the origin snapshots zil header.
931 	 */
932 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
933 		dsl_dataset_t *ds;
934 
935 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
936 		dsl_dataset_zero_zil(ds, tx);
937 		dsl_dataset_rele(ds, FTAG);
938 	}
939 
940 	return (dsobj);
941 }
942 
943 /*
944  * The unique space in the head dataset can be calculated by subtracting
945  * the space used in the most recent snapshot, that is still being used
946  * in this file system, from the space currently in use.  To figure out
947  * the space in the most recent snapshot still in use, we need to take
948  * the total space used in the snapshot and subtract out the space that
949  * has been freed up since the snapshot was taken.
950  */
951 void
952 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
953 {
954 	uint64_t mrs_used;
955 	uint64_t dlused, dlcomp, dluncomp;
956 
957 	ASSERT(!ds->ds_is_snapshot);
958 
959 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
960 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
961 	else
962 		mrs_used = 0;
963 
964 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
965 
966 	ASSERT3U(dlused, <=, mrs_used);
967 	dsl_dataset_phys(ds)->ds_unique_bytes =
968 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
969 
970 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
971 	    SPA_VERSION_UNIQUE_ACCURATE)
972 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
973 }
974 
975 void
976 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
977     dmu_tx_t *tx)
978 {
979 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
980 	uint64_t count;
981 	int err;
982 
983 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
984 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
985 	    obj, tx);
986 	/*
987 	 * The err should not be ENOENT, but a bug in a previous version
988 	 * of the code could cause upgrade_clones_cb() to not set
989 	 * ds_next_snap_obj when it should, leading to a missing entry.
990 	 * If we knew that the pool was created after
991 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
992 	 * ENOENT.  However, at least we can check that we don't have
993 	 * too many entries in the next_clones_obj even after failing to
994 	 * remove this one.
995 	 */
996 	if (err != ENOENT)
997 		VERIFY0(err);
998 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
999 	    &count));
1000 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1001 }
1002 
1003 
1004 blkptr_t *
1005 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1006 {
1007 	return (&dsl_dataset_phys(ds)->ds_bp);
1008 }
1009 
1010 spa_t *
1011 dsl_dataset_get_spa(dsl_dataset_t *ds)
1012 {
1013 	return (ds->ds_dir->dd_pool->dp_spa);
1014 }
1015 
1016 void
1017 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1018 {
1019 	dsl_pool_t *dp;
1020 
1021 	if (ds == NULL) /* this is the meta-objset */
1022 		return;
1023 
1024 	ASSERT(ds->ds_objset != NULL);
1025 
1026 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1027 		panic("dirtying snapshot!");
1028 
1029 	dp = ds->ds_dir->dd_pool;
1030 
1031 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1032 		/* up the hold count until we can be written out */
1033 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1034 	}
1035 }
1036 
1037 boolean_t
1038 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1039 {
1040 	for (int t = 0; t < TXG_SIZE; t++) {
1041 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1042 		    ds, t))
1043 			return (B_TRUE);
1044 	}
1045 	return (B_FALSE);
1046 }
1047 
1048 static int
1049 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1050 {
1051 	uint64_t asize;
1052 
1053 	if (!dmu_tx_is_syncing(tx))
1054 		return (0);
1055 
1056 	/*
1057 	 * If there's an fs-only reservation, any blocks that might become
1058 	 * owned by the snapshot dataset must be accommodated by space
1059 	 * outside of the reservation.
1060 	 */
1061 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1062 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1063 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1064 		return (SET_ERROR(ENOSPC));
1065 
1066 	/*
1067 	 * Propagate any reserved space for this snapshot to other
1068 	 * snapshot checks in this sync group.
1069 	 */
1070 	if (asize > 0)
1071 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1072 
1073 	return (0);
1074 }
1075 
1076 typedef struct dsl_dataset_snapshot_arg {
1077 	nvlist_t *ddsa_snaps;
1078 	nvlist_t *ddsa_props;
1079 	nvlist_t *ddsa_errors;
1080 	cred_t *ddsa_cr;
1081 } dsl_dataset_snapshot_arg_t;
1082 
1083 int
1084 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1085     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1086 {
1087 	int error;
1088 	uint64_t value;
1089 
1090 	ds->ds_trysnap_txg = tx->tx_txg;
1091 
1092 	if (!dmu_tx_is_syncing(tx))
1093 		return (0);
1094 
1095 	/*
1096 	 * We don't allow multiple snapshots of the same txg.  If there
1097 	 * is already one, try again.
1098 	 */
1099 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1100 		return (SET_ERROR(EAGAIN));
1101 
1102 	/*
1103 	 * Check for conflicting snapshot name.
1104 	 */
1105 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1106 	if (error == 0)
1107 		return (SET_ERROR(EEXIST));
1108 	if (error != ENOENT)
1109 		return (error);
1110 
1111 	/*
1112 	 * We don't allow taking snapshots of inconsistent datasets, such as
1113 	 * those into which we are currently receiving.  However, if we are
1114 	 * creating this snapshot as part of a receive, this check will be
1115 	 * executed atomically with respect to the completion of the receive
1116 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1117 	 * case we ignore this, knowing it will be fixed up for us shortly in
1118 	 * dmu_recv_end_sync().
1119 	 */
1120 	if (!recv && DS_IS_INCONSISTENT(ds))
1121 		return (SET_ERROR(EBUSY));
1122 
1123 	/*
1124 	 * Skip the check for temporary snapshots or if we have already checked
1125 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1126 	 * check the count here when we're receiving a stream.
1127 	 */
1128 	if (cnt != 0 && cr != NULL) {
1129 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1130 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1131 		if (error != 0)
1132 			return (error);
1133 	}
1134 
1135 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1136 	if (error != 0)
1137 		return (error);
1138 
1139 	return (0);
1140 }
1141 
1142 static int
1143 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1144 {
1145 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1146 	dsl_pool_t *dp = dmu_tx_pool(tx);
1147 	nvpair_t *pair;
1148 	int rv = 0;
1149 
1150 	/*
1151 	 * Pre-compute how many total new snapshots will be created for each
1152 	 * level in the tree and below. This is needed for validating the
1153 	 * snapshot limit when either taking a recursive snapshot or when
1154 	 * taking multiple snapshots.
1155 	 *
1156 	 * The problem is that the counts are not actually adjusted when
1157 	 * we are checking, only when we finally sync. For a single snapshot,
1158 	 * this is easy, the count will increase by 1 at each node up the tree,
1159 	 * but its more complicated for the recursive/multiple snapshot case.
1160 	 *
1161 	 * The dsl_fs_ss_limit_check function does recursively check the count
1162 	 * at each level up the tree but since it is validating each snapshot
1163 	 * independently we need to be sure that we are validating the complete
1164 	 * count for the entire set of snapshots. We do this by rolling up the
1165 	 * counts for each component of the name into an nvlist and then
1166 	 * checking each of those cases with the aggregated count.
1167 	 *
1168 	 * This approach properly handles not only the recursive snapshot
1169 	 * case (where we get all of those on the ddsa_snaps list) but also
1170 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1171 	 * validate the limit on 'a' using a count of 2).
1172 	 *
1173 	 * We validate the snapshot names in the third loop and only report
1174 	 * name errors once.
1175 	 */
1176 	if (dmu_tx_is_syncing(tx)) {
1177 		nvlist_t *cnt_track = NULL;
1178 		cnt_track = fnvlist_alloc();
1179 
1180 		/* Rollup aggregated counts into the cnt_track list */
1181 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1182 		    pair != NULL;
1183 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1184 			char *pdelim;
1185 			uint64_t val;
1186 			char nm[MAXPATHLEN];
1187 
1188 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1189 			pdelim = strchr(nm, '@');
1190 			if (pdelim == NULL)
1191 				continue;
1192 			*pdelim = '\0';
1193 
1194 			do {
1195 				if (nvlist_lookup_uint64(cnt_track, nm,
1196 				    &val) == 0) {
1197 					/* update existing entry */
1198 					fnvlist_add_uint64(cnt_track, nm,
1199 					    val + 1);
1200 				} else {
1201 					/* add to list */
1202 					fnvlist_add_uint64(cnt_track, nm, 1);
1203 				}
1204 
1205 				pdelim = strrchr(nm, '/');
1206 				if (pdelim != NULL)
1207 					*pdelim = '\0';
1208 			} while (pdelim != NULL);
1209 		}
1210 
1211 		/* Check aggregated counts at each level */
1212 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1213 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1214 			int error = 0;
1215 			char *name;
1216 			uint64_t cnt = 0;
1217 			dsl_dataset_t *ds;
1218 
1219 			name = nvpair_name(pair);
1220 			cnt = fnvpair_value_uint64(pair);
1221 			ASSERT(cnt > 0);
1222 
1223 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1224 			if (error == 0) {
1225 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1226 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1227 				    ddsa->ddsa_cr);
1228 				dsl_dataset_rele(ds, FTAG);
1229 			}
1230 
1231 			if (error != 0) {
1232 				if (ddsa->ddsa_errors != NULL)
1233 					fnvlist_add_int32(ddsa->ddsa_errors,
1234 					    name, error);
1235 				rv = error;
1236 				/* only report one error for this check */
1237 				break;
1238 			}
1239 		}
1240 		nvlist_free(cnt_track);
1241 	}
1242 
1243 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1244 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1245 		int error = 0;
1246 		dsl_dataset_t *ds;
1247 		char *name, *atp;
1248 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1249 
1250 		name = nvpair_name(pair);
1251 		if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1252 			error = SET_ERROR(ENAMETOOLONG);
1253 		if (error == 0) {
1254 			atp = strchr(name, '@');
1255 			if (atp == NULL)
1256 				error = SET_ERROR(EINVAL);
1257 			if (error == 0)
1258 				(void) strlcpy(dsname, name, atp - name + 1);
1259 		}
1260 		if (error == 0)
1261 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1262 		if (error == 0) {
1263 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1264 			error = dsl_dataset_snapshot_check_impl(ds,
1265 			    atp + 1, tx, B_FALSE, 0, NULL);
1266 			dsl_dataset_rele(ds, FTAG);
1267 		}
1268 
1269 		if (error != 0) {
1270 			if (ddsa->ddsa_errors != NULL) {
1271 				fnvlist_add_int32(ddsa->ddsa_errors,
1272 				    name, error);
1273 			}
1274 			rv = error;
1275 		}
1276 	}
1277 
1278 	return (rv);
1279 }
1280 
1281 void
1282 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1283     dmu_tx_t *tx)
1284 {
1285 	static zil_header_t zero_zil;
1286 
1287 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1288 	dmu_buf_t *dbuf;
1289 	dsl_dataset_phys_t *dsphys;
1290 	uint64_t dsobj, crtxg;
1291 	objset_t *mos = dp->dp_meta_objset;
1292 	objset_t *os;
1293 
1294 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1295 
1296 	/*
1297 	 * If we are on an old pool, the zil must not be active, in which
1298 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1299 	 */
1300 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1301 	    dmu_objset_from_ds(ds, &os) != 0 ||
1302 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1303 	    sizeof (zero_zil)) == 0);
1304 
1305 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1306 
1307 	/*
1308 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1309 	 */
1310 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1311 		crtxg = 1;
1312 	else
1313 		crtxg = tx->tx_txg;
1314 
1315 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1316 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1317 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1318 	dmu_buf_will_dirty(dbuf, tx);
1319 	dsphys = dbuf->db_data;
1320 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1321 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1322 	dsphys->ds_fsid_guid = unique_create();
1323 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1324 	    sizeof (dsphys->ds_guid));
1325 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1326 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1327 	dsphys->ds_next_snap_obj = ds->ds_object;
1328 	dsphys->ds_num_children = 1;
1329 	dsphys->ds_creation_time = gethrestime_sec();
1330 	dsphys->ds_creation_txg = crtxg;
1331 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1332 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1333 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1334 	dsphys->ds_uncompressed_bytes =
1335 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1336 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1337 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1338 	dmu_buf_rele(dbuf, FTAG);
1339 
1340 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1341 		if (ds->ds_feature_inuse[f])
1342 			dsl_dataset_activate_feature(dsobj, f, tx);
1343 	}
1344 
1345 	ASSERT3U(ds->ds_prev != 0, ==,
1346 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1347 	if (ds->ds_prev) {
1348 		uint64_t next_clones_obj =
1349 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1350 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1351 		    ds->ds_object ||
1352 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1353 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1354 		    ds->ds_object) {
1355 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1356 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1357 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1358 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1359 		} else if (next_clones_obj != 0) {
1360 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1361 			    dsphys->ds_next_snap_obj, tx);
1362 			VERIFY0(zap_add_int(mos,
1363 			    next_clones_obj, dsobj, tx));
1364 		}
1365 	}
1366 
1367 	/*
1368 	 * If we have a reference-reservation on this dataset, we will
1369 	 * need to increase the amount of refreservation being charged
1370 	 * since our unique space is going to zero.
1371 	 */
1372 	if (ds->ds_reserved) {
1373 		int64_t delta;
1374 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1375 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1376 		    ds->ds_reserved);
1377 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1378 		    delta, 0, 0, tx);
1379 	}
1380 
1381 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1382 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1383 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1384 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1385 	dsl_deadlist_close(&ds->ds_deadlist);
1386 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1387 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1388 	dsl_deadlist_add_key(&ds->ds_deadlist,
1389 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1390 
1391 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1392 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1393 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1394 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1395 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1396 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1397 
1398 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1399 	    snapname, 8, 1, &dsobj, tx));
1400 
1401 	if (ds->ds_prev)
1402 		dsl_dataset_rele(ds->ds_prev, ds);
1403 	VERIFY0(dsl_dataset_hold_obj(dp,
1404 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1405 
1406 	dsl_scan_ds_snapshotted(ds, tx);
1407 
1408 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1409 
1410 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1411 }
1412 
1413 static void
1414 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1415 {
1416 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1417 	dsl_pool_t *dp = dmu_tx_pool(tx);
1418 	nvpair_t *pair;
1419 
1420 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1421 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1422 		dsl_dataset_t *ds;
1423 		char *name, *atp;
1424 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1425 
1426 		name = nvpair_name(pair);
1427 		atp = strchr(name, '@');
1428 		(void) strlcpy(dsname, name, atp - name + 1);
1429 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1430 
1431 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1432 		if (ddsa->ddsa_props != NULL) {
1433 			dsl_props_set_sync_impl(ds->ds_prev,
1434 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1435 		}
1436 		dsl_dataset_rele(ds, FTAG);
1437 	}
1438 }
1439 
1440 /*
1441  * The snapshots must all be in the same pool.
1442  * All-or-nothing: if there are any failures, nothing will be modified.
1443  */
1444 int
1445 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1446 {
1447 	dsl_dataset_snapshot_arg_t ddsa;
1448 	nvpair_t *pair;
1449 	boolean_t needsuspend;
1450 	int error;
1451 	spa_t *spa;
1452 	char *firstname;
1453 	nvlist_t *suspended = NULL;
1454 
1455 	pair = nvlist_next_nvpair(snaps, NULL);
1456 	if (pair == NULL)
1457 		return (0);
1458 	firstname = nvpair_name(pair);
1459 
1460 	error = spa_open(firstname, &spa, FTAG);
1461 	if (error != 0)
1462 		return (error);
1463 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1464 	spa_close(spa, FTAG);
1465 
1466 	if (needsuspend) {
1467 		suspended = fnvlist_alloc();
1468 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1469 		    pair = nvlist_next_nvpair(snaps, pair)) {
1470 			char fsname[ZFS_MAX_DATASET_NAME_LEN];
1471 			char *snapname = nvpair_name(pair);
1472 			char *atp;
1473 			void *cookie;
1474 
1475 			atp = strchr(snapname, '@');
1476 			if (atp == NULL) {
1477 				error = SET_ERROR(EINVAL);
1478 				break;
1479 			}
1480 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1481 
1482 			error = zil_suspend(fsname, &cookie);
1483 			if (error != 0)
1484 				break;
1485 			fnvlist_add_uint64(suspended, fsname,
1486 			    (uintptr_t)cookie);
1487 		}
1488 	}
1489 
1490 	ddsa.ddsa_snaps = snaps;
1491 	ddsa.ddsa_props = props;
1492 	ddsa.ddsa_errors = errors;
1493 	ddsa.ddsa_cr = CRED();
1494 
1495 	if (error == 0) {
1496 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1497 		    dsl_dataset_snapshot_sync, &ddsa,
1498 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1499 	}
1500 
1501 	if (suspended != NULL) {
1502 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1503 		    pair = nvlist_next_nvpair(suspended, pair)) {
1504 			zil_resume((void *)(uintptr_t)
1505 			    fnvpair_value_uint64(pair));
1506 		}
1507 		fnvlist_free(suspended);
1508 	}
1509 
1510 	return (error);
1511 }
1512 
1513 typedef struct dsl_dataset_snapshot_tmp_arg {
1514 	const char *ddsta_fsname;
1515 	const char *ddsta_snapname;
1516 	minor_t ddsta_cleanup_minor;
1517 	const char *ddsta_htag;
1518 } dsl_dataset_snapshot_tmp_arg_t;
1519 
1520 static int
1521 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1522 {
1523 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1524 	dsl_pool_t *dp = dmu_tx_pool(tx);
1525 	dsl_dataset_t *ds;
1526 	int error;
1527 
1528 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1529 	if (error != 0)
1530 		return (error);
1531 
1532 	/* NULL cred means no limit check for tmp snapshot */
1533 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1534 	    tx, B_FALSE, 0, NULL);
1535 	if (error != 0) {
1536 		dsl_dataset_rele(ds, FTAG);
1537 		return (error);
1538 	}
1539 
1540 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1541 		dsl_dataset_rele(ds, FTAG);
1542 		return (SET_ERROR(ENOTSUP));
1543 	}
1544 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1545 	    B_TRUE, tx);
1546 	if (error != 0) {
1547 		dsl_dataset_rele(ds, FTAG);
1548 		return (error);
1549 	}
1550 
1551 	dsl_dataset_rele(ds, FTAG);
1552 	return (0);
1553 }
1554 
1555 static void
1556 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1557 {
1558 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1559 	dsl_pool_t *dp = dmu_tx_pool(tx);
1560 	dsl_dataset_t *ds;
1561 
1562 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1563 
1564 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1565 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1566 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1567 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1568 
1569 	dsl_dataset_rele(ds, FTAG);
1570 }
1571 
1572 int
1573 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1574     minor_t cleanup_minor, const char *htag)
1575 {
1576 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1577 	int error;
1578 	spa_t *spa;
1579 	boolean_t needsuspend;
1580 	void *cookie;
1581 
1582 	ddsta.ddsta_fsname = fsname;
1583 	ddsta.ddsta_snapname = snapname;
1584 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1585 	ddsta.ddsta_htag = htag;
1586 
1587 	error = spa_open(fsname, &spa, FTAG);
1588 	if (error != 0)
1589 		return (error);
1590 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1591 	spa_close(spa, FTAG);
1592 
1593 	if (needsuspend) {
1594 		error = zil_suspend(fsname, &cookie);
1595 		if (error != 0)
1596 			return (error);
1597 	}
1598 
1599 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1600 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1601 
1602 	if (needsuspend)
1603 		zil_resume(cookie);
1604 	return (error);
1605 }
1606 
1607 
1608 void
1609 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1610 {
1611 	ASSERT(dmu_tx_is_syncing(tx));
1612 	ASSERT(ds->ds_objset != NULL);
1613 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1614 
1615 	/*
1616 	 * in case we had to change ds_fsid_guid when we opened it,
1617 	 * sync it out now.
1618 	 */
1619 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1620 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1621 
1622 	if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1623 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1624 		    ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1625 		    &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1626 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1627 		    ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1628 		    &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1629 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1630 		    ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1631 		    &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1632 		ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1633 		ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1634 		ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1635 	}
1636 
1637 	dmu_objset_sync(ds->ds_objset, zio, tx);
1638 
1639 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1640 		if (ds->ds_feature_activation_needed[f]) {
1641 			if (ds->ds_feature_inuse[f])
1642 				continue;
1643 			dsl_dataset_activate_feature(ds->ds_object, f, tx);
1644 			ds->ds_feature_inuse[f] = B_TRUE;
1645 		}
1646 	}
1647 }
1648 
1649 static void
1650 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1651 {
1652 	uint64_t count = 0;
1653 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1654 	zap_cursor_t zc;
1655 	zap_attribute_t za;
1656 	nvlist_t *propval = fnvlist_alloc();
1657 	nvlist_t *val = fnvlist_alloc();
1658 
1659 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1660 
1661 	/*
1662 	 * There may be missing entries in ds_next_clones_obj
1663 	 * due to a bug in a previous version of the code.
1664 	 * Only trust it if it has the right number of entries.
1665 	 */
1666 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1667 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1668 		    &count));
1669 	}
1670 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1671 		goto fail;
1672 	for (zap_cursor_init(&zc, mos,
1673 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1674 	    zap_cursor_retrieve(&zc, &za) == 0;
1675 	    zap_cursor_advance(&zc)) {
1676 		dsl_dataset_t *clone;
1677 		char buf[ZFS_MAX_DATASET_NAME_LEN];
1678 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1679 		    za.za_first_integer, FTAG, &clone));
1680 		dsl_dir_name(clone->ds_dir, buf);
1681 		fnvlist_add_boolean(val, buf);
1682 		dsl_dataset_rele(clone, FTAG);
1683 	}
1684 	zap_cursor_fini(&zc);
1685 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1686 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1687 fail:
1688 	nvlist_free(val);
1689 	nvlist_free(propval);
1690 }
1691 
1692 static void
1693 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1694 {
1695 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1696 
1697 	if (dsl_dataset_has_resume_receive_state(ds)) {
1698 		char *str;
1699 		void *packed;
1700 		uint8_t *compressed;
1701 		uint64_t val;
1702 		nvlist_t *token_nv = fnvlist_alloc();
1703 		size_t packed_size, compressed_size;
1704 
1705 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1706 		    DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1707 			fnvlist_add_uint64(token_nv, "fromguid", val);
1708 		}
1709 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1710 		    DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1711 			fnvlist_add_uint64(token_nv, "object", val);
1712 		}
1713 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1714 		    DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1715 			fnvlist_add_uint64(token_nv, "offset", val);
1716 		}
1717 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1718 		    DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1719 			fnvlist_add_uint64(token_nv, "bytes", val);
1720 		}
1721 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1722 		    DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1723 			fnvlist_add_uint64(token_nv, "toguid", val);
1724 		}
1725 		char buf[256];
1726 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1727 		    DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1728 			fnvlist_add_string(token_nv, "toname", buf);
1729 		}
1730 		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1731 		    DS_FIELD_RESUME_EMBEDOK) == 0) {
1732 			fnvlist_add_boolean(token_nv, "embedok");
1733 		}
1734 		packed = fnvlist_pack(token_nv, &packed_size);
1735 		fnvlist_free(token_nv);
1736 		compressed = kmem_alloc(packed_size, KM_SLEEP);
1737 
1738 		compressed_size = gzip_compress(packed, compressed,
1739 		    packed_size, packed_size, 6);
1740 
1741 		zio_cksum_t cksum;
1742 		fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1743 
1744 		str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1745 		for (int i = 0; i < compressed_size; i++) {
1746 			(void) sprintf(str + i * 2, "%02x", compressed[i]);
1747 		}
1748 		str[compressed_size * 2] = '\0';
1749 		char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1750 		    ZFS_SEND_RESUME_TOKEN_VERSION,
1751 		    (longlong_t)cksum.zc_word[0],
1752 		    (longlong_t)packed_size, str);
1753 		dsl_prop_nvlist_add_string(nv,
1754 		    ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1755 		kmem_free(packed, packed_size);
1756 		kmem_free(str, compressed_size * 2 + 1);
1757 		kmem_free(compressed, packed_size);
1758 		strfree(propval);
1759 	}
1760 }
1761 
1762 void
1763 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1764 {
1765 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1766 	uint64_t refd, avail, uobjs, aobjs, ratio;
1767 
1768 	ASSERT(dsl_pool_config_held(dp));
1769 
1770 	ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1771 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1772 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1773 
1774 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1775 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1776 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1777 
1778 	if (ds->ds_is_snapshot) {
1779 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1780 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1781 		    dsl_dataset_phys(ds)->ds_unique_bytes);
1782 		get_clones_stat(ds, nv);
1783 	} else {
1784 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1785 			char buf[ZFS_MAX_DATASET_NAME_LEN];
1786 			dsl_dataset_name(ds->ds_prev, buf);
1787 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1788 		}
1789 
1790 		dsl_dir_stats(ds->ds_dir, nv);
1791 	}
1792 
1793 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1794 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1795 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1796 
1797 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1798 	    dsl_dataset_phys(ds)->ds_creation_time);
1799 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1800 	    dsl_dataset_phys(ds)->ds_creation_txg);
1801 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1802 	    ds->ds_quota);
1803 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1804 	    ds->ds_reserved);
1805 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1806 	    dsl_dataset_phys(ds)->ds_guid);
1807 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1808 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1809 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1810 	    ds->ds_object);
1811 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1812 	    ds->ds_userrefs);
1813 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1814 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1815 
1816 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1817 		uint64_t written, comp, uncomp;
1818 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1819 		dsl_dataset_t *prev;
1820 
1821 		int err = dsl_dataset_hold_obj(dp,
1822 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1823 		if (err == 0) {
1824 			err = dsl_dataset_space_written(prev, ds, &written,
1825 			    &comp, &uncomp);
1826 			dsl_dataset_rele(prev, FTAG);
1827 			if (err == 0) {
1828 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1829 				    written);
1830 			}
1831 		}
1832 	}
1833 
1834 	if (!dsl_dataset_is_snapshot(ds)) {
1835 		/*
1836 		 * A failed "newfs" (e.g. full) resumable receive leaves
1837 		 * the stats set on this dataset.  Check here for the prop.
1838 		 */
1839 		get_receive_resume_stats(ds, nv);
1840 
1841 		/*
1842 		 * A failed incremental resumable receive leaves the
1843 		 * stats set on our child named "%recv".  Check the child
1844 		 * for the prop.
1845 		 */
1846 		/* 6 extra bytes for /%recv */
1847 		char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1848 		dsl_dataset_t *recv_ds;
1849 		dsl_dataset_name(ds, recvname);
1850 		if (strlcat(recvname, "/", sizeof (recvname)) <
1851 		    sizeof (recvname) &&
1852 		    strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1853 		    sizeof (recvname) &&
1854 		    dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1855 			get_receive_resume_stats(recv_ds, nv);
1856 			dsl_dataset_rele(recv_ds, FTAG);
1857 		}
1858 	}
1859 }
1860 
1861 void
1862 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1863 {
1864 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1865 	ASSERT(dsl_pool_config_held(dp));
1866 
1867 	stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1868 	stat->dds_inconsistent =
1869 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1870 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1871 	stat->dds_origin[0] = '\0';
1872 	if (ds->ds_is_snapshot) {
1873 		stat->dds_is_snapshot = B_TRUE;
1874 		stat->dds_num_clones =
1875 		    dsl_dataset_phys(ds)->ds_num_children - 1;
1876 	} else {
1877 		stat->dds_is_snapshot = B_FALSE;
1878 		stat->dds_num_clones = 0;
1879 
1880 		if (dsl_dir_is_clone(ds->ds_dir)) {
1881 			dsl_dataset_t *ods;
1882 
1883 			VERIFY0(dsl_dataset_hold_obj(dp,
1884 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1885 			    FTAG, &ods));
1886 			dsl_dataset_name(ods, stat->dds_origin);
1887 			dsl_dataset_rele(ods, FTAG);
1888 		}
1889 	}
1890 }
1891 
1892 uint64_t
1893 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1894 {
1895 	return (ds->ds_fsid_guid);
1896 }
1897 
1898 void
1899 dsl_dataset_space(dsl_dataset_t *ds,
1900     uint64_t *refdbytesp, uint64_t *availbytesp,
1901     uint64_t *usedobjsp, uint64_t *availobjsp)
1902 {
1903 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1904 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1905 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1906 		*availbytesp +=
1907 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1908 	if (ds->ds_quota != 0) {
1909 		/*
1910 		 * Adjust available bytes according to refquota
1911 		 */
1912 		if (*refdbytesp < ds->ds_quota)
1913 			*availbytesp = MIN(*availbytesp,
1914 			    ds->ds_quota - *refdbytesp);
1915 		else
1916 			*availbytesp = 0;
1917 	}
1918 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1919 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1920 }
1921 
1922 boolean_t
1923 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1924 {
1925 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1926 
1927 	ASSERT(dsl_pool_config_held(dp));
1928 	if (snap == NULL)
1929 		return (B_FALSE);
1930 	if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1931 	    dsl_dataset_phys(snap)->ds_creation_txg) {
1932 		objset_t *os, *os_snap;
1933 		/*
1934 		 * It may be that only the ZIL differs, because it was
1935 		 * reset in the head.  Don't count that as being
1936 		 * modified.
1937 		 */
1938 		if (dmu_objset_from_ds(ds, &os) != 0)
1939 			return (B_TRUE);
1940 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1941 			return (B_TRUE);
1942 		return (bcmp(&os->os_phys->os_meta_dnode,
1943 		    &os_snap->os_phys->os_meta_dnode,
1944 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1945 	}
1946 	return (B_FALSE);
1947 }
1948 
1949 typedef struct dsl_dataset_rename_snapshot_arg {
1950 	const char *ddrsa_fsname;
1951 	const char *ddrsa_oldsnapname;
1952 	const char *ddrsa_newsnapname;
1953 	boolean_t ddrsa_recursive;
1954 	dmu_tx_t *ddrsa_tx;
1955 } dsl_dataset_rename_snapshot_arg_t;
1956 
1957 /* ARGSUSED */
1958 static int
1959 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1960     dsl_dataset_t *hds, void *arg)
1961 {
1962 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1963 	int error;
1964 	uint64_t val;
1965 
1966 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1967 	if (error != 0) {
1968 		/* ignore nonexistent snapshots */
1969 		return (error == ENOENT ? 0 : error);
1970 	}
1971 
1972 	/* new name should not exist */
1973 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1974 	if (error == 0)
1975 		error = SET_ERROR(EEXIST);
1976 	else if (error == ENOENT)
1977 		error = 0;
1978 
1979 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1980 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1981 	    strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
1982 		error = SET_ERROR(ENAMETOOLONG);
1983 
1984 	return (error);
1985 }
1986 
1987 static int
1988 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1989 {
1990 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1991 	dsl_pool_t *dp = dmu_tx_pool(tx);
1992 	dsl_dataset_t *hds;
1993 	int error;
1994 
1995 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1996 	if (error != 0)
1997 		return (error);
1998 
1999 	if (ddrsa->ddrsa_recursive) {
2000 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2001 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
2002 		    DS_FIND_CHILDREN);
2003 	} else {
2004 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2005 	}
2006 	dsl_dataset_rele(hds, FTAG);
2007 	return (error);
2008 }
2009 
2010 static int
2011 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2012     dsl_dataset_t *hds, void *arg)
2013 {
2014 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2015 	dsl_dataset_t *ds;
2016 	uint64_t val;
2017 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
2018 	int error;
2019 
2020 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2021 	ASSERT(error == 0 || error == ENOENT);
2022 	if (error == ENOENT) {
2023 		/* ignore nonexistent snapshots */
2024 		return (0);
2025 	}
2026 
2027 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2028 
2029 	/* log before we change the name */
2030 	spa_history_log_internal_ds(ds, "rename", tx,
2031 	    "-> @%s", ddrsa->ddrsa_newsnapname);
2032 
2033 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2034 	    B_FALSE));
2035 	mutex_enter(&ds->ds_lock);
2036 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2037 	mutex_exit(&ds->ds_lock);
2038 	VERIFY0(zap_add(dp->dp_meta_objset,
2039 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2040 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2041 
2042 	dsl_dataset_rele(ds, FTAG);
2043 	return (0);
2044 }
2045 
2046 static void
2047 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2048 {
2049 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2050 	dsl_pool_t *dp = dmu_tx_pool(tx);
2051 	dsl_dataset_t *hds;
2052 
2053 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2054 	ddrsa->ddrsa_tx = tx;
2055 	if (ddrsa->ddrsa_recursive) {
2056 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2057 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2058 		    DS_FIND_CHILDREN));
2059 	} else {
2060 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2061 	}
2062 	dsl_dataset_rele(hds, FTAG);
2063 }
2064 
2065 int
2066 dsl_dataset_rename_snapshot(const char *fsname,
2067     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2068 {
2069 	dsl_dataset_rename_snapshot_arg_t ddrsa;
2070 
2071 	ddrsa.ddrsa_fsname = fsname;
2072 	ddrsa.ddrsa_oldsnapname = oldsnapname;
2073 	ddrsa.ddrsa_newsnapname = newsnapname;
2074 	ddrsa.ddrsa_recursive = recursive;
2075 
2076 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2077 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
2078 	    1, ZFS_SPACE_CHECK_RESERVED));
2079 }
2080 
2081 /*
2082  * If we're doing an ownership handoff, we need to make sure that there is
2083  * only one long hold on the dataset.  We're not allowed to change anything here
2084  * so we don't permanently release the long hold or regular hold here.  We want
2085  * to do this only when syncing to avoid the dataset unexpectedly going away
2086  * when we release the long hold.
2087  */
2088 static int
2089 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2090 {
2091 	boolean_t held;
2092 
2093 	if (!dmu_tx_is_syncing(tx))
2094 		return (0);
2095 
2096 	if (owner != NULL) {
2097 		VERIFY3P(ds->ds_owner, ==, owner);
2098 		dsl_dataset_long_rele(ds, owner);
2099 	}
2100 
2101 	held = dsl_dataset_long_held(ds);
2102 
2103 	if (owner != NULL)
2104 		dsl_dataset_long_hold(ds, owner);
2105 
2106 	if (held)
2107 		return (SET_ERROR(EBUSY));
2108 
2109 	return (0);
2110 }
2111 
2112 typedef struct dsl_dataset_rollback_arg {
2113 	const char *ddra_fsname;
2114 	void *ddra_owner;
2115 	nvlist_t *ddra_result;
2116 } dsl_dataset_rollback_arg_t;
2117 
2118 static int
2119 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2120 {
2121 	dsl_dataset_rollback_arg_t *ddra = arg;
2122 	dsl_pool_t *dp = dmu_tx_pool(tx);
2123 	dsl_dataset_t *ds;
2124 	int64_t unused_refres_delta;
2125 	int error;
2126 
2127 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2128 	if (error != 0)
2129 		return (error);
2130 
2131 	/* must not be a snapshot */
2132 	if (ds->ds_is_snapshot) {
2133 		dsl_dataset_rele(ds, FTAG);
2134 		return (SET_ERROR(EINVAL));
2135 	}
2136 
2137 	/* must have a most recent snapshot */
2138 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2139 		dsl_dataset_rele(ds, FTAG);
2140 		return (SET_ERROR(EINVAL));
2141 	}
2142 
2143 	/* must not have any bookmarks after the most recent snapshot */
2144 	nvlist_t *proprequest = fnvlist_alloc();
2145 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2146 	nvlist_t *bookmarks = fnvlist_alloc();
2147 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2148 	fnvlist_free(proprequest);
2149 	if (error != 0)
2150 		return (error);
2151 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2152 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2153 		nvlist_t *valuenv =
2154 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2155 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
2156 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2157 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2158 			fnvlist_free(bookmarks);
2159 			dsl_dataset_rele(ds, FTAG);
2160 			return (SET_ERROR(EEXIST));
2161 		}
2162 	}
2163 	fnvlist_free(bookmarks);
2164 
2165 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2166 	if (error != 0) {
2167 		dsl_dataset_rele(ds, FTAG);
2168 		return (error);
2169 	}
2170 
2171 	/*
2172 	 * Check if the snap we are rolling back to uses more than
2173 	 * the refquota.
2174 	 */
2175 	if (ds->ds_quota != 0 &&
2176 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2177 		dsl_dataset_rele(ds, FTAG);
2178 		return (SET_ERROR(EDQUOT));
2179 	}
2180 
2181 	/*
2182 	 * When we do the clone swap, we will temporarily use more space
2183 	 * due to the refreservation (the head will no longer have any
2184 	 * unique space, so the entire amount of the refreservation will need
2185 	 * to be free).  We will immediately destroy the clone, freeing
2186 	 * this space, but the freeing happens over many txg's.
2187 	 */
2188 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2189 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2190 
2191 	if (unused_refres_delta > 0 &&
2192 	    unused_refres_delta >
2193 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2194 		dsl_dataset_rele(ds, FTAG);
2195 		return (SET_ERROR(ENOSPC));
2196 	}
2197 
2198 	dsl_dataset_rele(ds, FTAG);
2199 	return (0);
2200 }
2201 
2202 static void
2203 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2204 {
2205 	dsl_dataset_rollback_arg_t *ddra = arg;
2206 	dsl_pool_t *dp = dmu_tx_pool(tx);
2207 	dsl_dataset_t *ds, *clone;
2208 	uint64_t cloneobj;
2209 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2210 
2211 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2212 
2213 	dsl_dataset_name(ds->ds_prev, namebuf);
2214 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2215 
2216 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2217 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2218 
2219 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2220 
2221 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2222 	dsl_dataset_zero_zil(ds, tx);
2223 
2224 	dsl_destroy_head_sync_impl(clone, tx);
2225 
2226 	dsl_dataset_rele(clone, FTAG);
2227 	dsl_dataset_rele(ds, FTAG);
2228 }
2229 
2230 /*
2231  * Rolls back the given filesystem or volume to the most recent snapshot.
2232  * The name of the most recent snapshot will be returned under key "target"
2233  * in the result nvlist.
2234  *
2235  * If owner != NULL:
2236  * - The existing dataset MUST be owned by the specified owner at entry
2237  * - Upon return, dataset will still be held by the same owner, whether we
2238  *   succeed or not.
2239  *
2240  * This mode is required any time the existing filesystem is mounted.  See
2241  * notes above zfs_suspend_fs() for further details.
2242  */
2243 int
2244 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2245 {
2246 	dsl_dataset_rollback_arg_t ddra;
2247 
2248 	ddra.ddra_fsname = fsname;
2249 	ddra.ddra_owner = owner;
2250 	ddra.ddra_result = result;
2251 
2252 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2253 	    dsl_dataset_rollback_sync, &ddra,
2254 	    1, ZFS_SPACE_CHECK_RESERVED));
2255 }
2256 
2257 struct promotenode {
2258 	list_node_t link;
2259 	dsl_dataset_t *ds;
2260 };
2261 
2262 typedef struct dsl_dataset_promote_arg {
2263 	const char *ddpa_clonename;
2264 	dsl_dataset_t *ddpa_clone;
2265 	list_t shared_snaps, origin_snaps, clone_snaps;
2266 	dsl_dataset_t *origin_origin; /* origin of the origin */
2267 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2268 	char *err_ds;
2269 	cred_t *cr;
2270 } dsl_dataset_promote_arg_t;
2271 
2272 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2273 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2274     void *tag);
2275 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2276 
2277 static int
2278 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2279 {
2280 	dsl_dataset_promote_arg_t *ddpa = arg;
2281 	dsl_pool_t *dp = dmu_tx_pool(tx);
2282 	dsl_dataset_t *hds;
2283 	struct promotenode *snap;
2284 	dsl_dataset_t *origin_ds;
2285 	int err;
2286 	uint64_t unused;
2287 	uint64_t ss_mv_cnt;
2288 	size_t max_snap_len;
2289 
2290 	err = promote_hold(ddpa, dp, FTAG);
2291 	if (err != 0)
2292 		return (err);
2293 
2294 	hds = ddpa->ddpa_clone;
2295 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2296 
2297 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2298 		promote_rele(ddpa, FTAG);
2299 		return (SET_ERROR(EXDEV));
2300 	}
2301 
2302 	/*
2303 	 * Compute and check the amount of space to transfer.  Since this is
2304 	 * so expensive, don't do the preliminary check.
2305 	 */
2306 	if (!dmu_tx_is_syncing(tx)) {
2307 		promote_rele(ddpa, FTAG);
2308 		return (0);
2309 	}
2310 
2311 	snap = list_head(&ddpa->shared_snaps);
2312 	origin_ds = snap->ds;
2313 
2314 	/* compute origin's new unique space */
2315 	snap = list_tail(&ddpa->clone_snaps);
2316 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2317 	    origin_ds->ds_object);
2318 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2319 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2320 	    &ddpa->unique, &unused, &unused);
2321 
2322 	/*
2323 	 * Walk the snapshots that we are moving
2324 	 *
2325 	 * Compute space to transfer.  Consider the incremental changes
2326 	 * to used by each snapshot:
2327 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2328 	 * So each snapshot gave birth to:
2329 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2330 	 * So a sequence would look like:
2331 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2332 	 * Which simplifies to:
2333 	 * uN + kN + kN-1 + ... + k1 + k0
2334 	 * Note however, if we stop before we reach the ORIGIN we get:
2335 	 * uN + kN + kN-1 + ... + kM - uM-1
2336 	 */
2337 	ss_mv_cnt = 0;
2338 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2339 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2340 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2341 	for (snap = list_head(&ddpa->shared_snaps); snap;
2342 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2343 		uint64_t val, dlused, dlcomp, dluncomp;
2344 		dsl_dataset_t *ds = snap->ds;
2345 
2346 		ss_mv_cnt++;
2347 
2348 		/*
2349 		 * If there are long holds, we won't be able to evict
2350 		 * the objset.
2351 		 */
2352 		if (dsl_dataset_long_held(ds)) {
2353 			err = SET_ERROR(EBUSY);
2354 			goto out;
2355 		}
2356 
2357 		/* Check that the snapshot name does not conflict */
2358 		VERIFY0(dsl_dataset_get_snapname(ds));
2359 		if (strlen(ds->ds_snapname) >= max_snap_len) {
2360 			err = SET_ERROR(ENAMETOOLONG);
2361 			goto out;
2362 		}
2363 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2364 		if (err == 0) {
2365 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2366 			err = SET_ERROR(EEXIST);
2367 			goto out;
2368 		}
2369 		if (err != ENOENT)
2370 			goto out;
2371 
2372 		/* The very first snapshot does not have a deadlist */
2373 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2374 			continue;
2375 
2376 		dsl_deadlist_space(&ds->ds_deadlist,
2377 		    &dlused, &dlcomp, &dluncomp);
2378 		ddpa->used += dlused;
2379 		ddpa->comp += dlcomp;
2380 		ddpa->uncomp += dluncomp;
2381 	}
2382 
2383 	/*
2384 	 * If we are a clone of a clone then we never reached ORIGIN,
2385 	 * so we need to subtract out the clone origin's used space.
2386 	 */
2387 	if (ddpa->origin_origin) {
2388 		ddpa->used -=
2389 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2390 		ddpa->comp -=
2391 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2392 		ddpa->uncomp -=
2393 		    dsl_dataset_phys(ddpa->origin_origin)->
2394 		    ds_uncompressed_bytes;
2395 	}
2396 
2397 	/* Check that there is enough space and limit headroom here */
2398 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2399 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2400 	if (err != 0)
2401 		goto out;
2402 
2403 	/*
2404 	 * Compute the amounts of space that will be used by snapshots
2405 	 * after the promotion (for both origin and clone).  For each,
2406 	 * it is the amount of space that will be on all of their
2407 	 * deadlists (that was not born before their new origin).
2408 	 */
2409 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2410 		uint64_t space;
2411 
2412 		/*
2413 		 * Note, typically this will not be a clone of a clone,
2414 		 * so dd_origin_txg will be < TXG_INITIAL, so
2415 		 * these snaplist_space() -> dsl_deadlist_space_range()
2416 		 * calls will be fast because they do not have to
2417 		 * iterate over all bps.
2418 		 */
2419 		snap = list_head(&ddpa->origin_snaps);
2420 		err = snaplist_space(&ddpa->shared_snaps,
2421 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2422 		if (err != 0)
2423 			goto out;
2424 
2425 		err = snaplist_space(&ddpa->clone_snaps,
2426 		    snap->ds->ds_dir->dd_origin_txg, &space);
2427 		if (err != 0)
2428 			goto out;
2429 		ddpa->cloneusedsnap += space;
2430 	}
2431 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2432 	    DD_FLAG_USED_BREAKDOWN) {
2433 		err = snaplist_space(&ddpa->origin_snaps,
2434 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2435 		    &ddpa->originusedsnap);
2436 		if (err != 0)
2437 			goto out;
2438 	}
2439 
2440 out:
2441 	promote_rele(ddpa, FTAG);
2442 	return (err);
2443 }
2444 
2445 static void
2446 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2447 {
2448 	dsl_dataset_promote_arg_t *ddpa = arg;
2449 	dsl_pool_t *dp = dmu_tx_pool(tx);
2450 	dsl_dataset_t *hds;
2451 	struct promotenode *snap;
2452 	dsl_dataset_t *origin_ds;
2453 	dsl_dataset_t *origin_head;
2454 	dsl_dir_t *dd;
2455 	dsl_dir_t *odd = NULL;
2456 	uint64_t oldnext_obj;
2457 	int64_t delta;
2458 
2459 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2460 	hds = ddpa->ddpa_clone;
2461 
2462 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2463 
2464 	snap = list_head(&ddpa->shared_snaps);
2465 	origin_ds = snap->ds;
2466 	dd = hds->ds_dir;
2467 
2468 	snap = list_head(&ddpa->origin_snaps);
2469 	origin_head = snap->ds;
2470 
2471 	/*
2472 	 * We need to explicitly open odd, since origin_ds's dd will be
2473 	 * changing.
2474 	 */
2475 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2476 	    NULL, FTAG, &odd));
2477 
2478 	/* change origin's next snap */
2479 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2480 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2481 	snap = list_tail(&ddpa->clone_snaps);
2482 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2483 	    origin_ds->ds_object);
2484 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2485 
2486 	/* change the origin's next clone */
2487 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2488 		dsl_dataset_remove_from_next_clones(origin_ds,
2489 		    snap->ds->ds_object, tx);
2490 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2491 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2492 		    oldnext_obj, tx));
2493 	}
2494 
2495 	/* change origin */
2496 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2497 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2498 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2499 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2500 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2501 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2502 	origin_head->ds_dir->dd_origin_txg =
2503 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2504 
2505 	/* change dd_clone entries */
2506 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2507 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2508 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2509 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2510 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2511 		    hds->ds_object, tx));
2512 
2513 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2514 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2515 		    origin_head->ds_object, tx));
2516 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2517 			dsl_dir_phys(dd)->dd_clones =
2518 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2519 			    DMU_OT_NONE, 0, tx);
2520 		}
2521 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2522 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2523 	}
2524 
2525 	/* move snapshots to this dir */
2526 	for (snap = list_head(&ddpa->shared_snaps); snap;
2527 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2528 		dsl_dataset_t *ds = snap->ds;
2529 
2530 		/*
2531 		 * Property callbacks are registered to a particular
2532 		 * dsl_dir.  Since ours is changing, evict the objset
2533 		 * so that they will be unregistered from the old dsl_dir.
2534 		 */
2535 		if (ds->ds_objset) {
2536 			dmu_objset_evict(ds->ds_objset);
2537 			ds->ds_objset = NULL;
2538 		}
2539 
2540 		/* move snap name entry */
2541 		VERIFY0(dsl_dataset_get_snapname(ds));
2542 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2543 		    ds->ds_snapname, tx, B_TRUE));
2544 		VERIFY0(zap_add(dp->dp_meta_objset,
2545 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2546 		    8, 1, &ds->ds_object, tx));
2547 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2548 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2549 
2550 		/* change containing dsl_dir */
2551 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2552 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2553 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2554 		ASSERT3P(ds->ds_dir, ==, odd);
2555 		dsl_dir_rele(ds->ds_dir, ds);
2556 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2557 		    NULL, ds, &ds->ds_dir));
2558 
2559 		/* move any clone references */
2560 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2561 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2562 			zap_cursor_t zc;
2563 			zap_attribute_t za;
2564 
2565 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2566 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
2567 			    zap_cursor_retrieve(&zc, &za) == 0;
2568 			    zap_cursor_advance(&zc)) {
2569 				dsl_dataset_t *cnds;
2570 				uint64_t o;
2571 
2572 				if (za.za_first_integer == oldnext_obj) {
2573 					/*
2574 					 * We've already moved the
2575 					 * origin's reference.
2576 					 */
2577 					continue;
2578 				}
2579 
2580 				VERIFY0(dsl_dataset_hold_obj(dp,
2581 				    za.za_first_integer, FTAG, &cnds));
2582 				o = dsl_dir_phys(cnds->ds_dir)->
2583 				    dd_head_dataset_obj;
2584 
2585 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2586 				    dsl_dir_phys(odd)->dd_clones, o, tx));
2587 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2588 				    dsl_dir_phys(dd)->dd_clones, o, tx));
2589 				dsl_dataset_rele(cnds, FTAG);
2590 			}
2591 			zap_cursor_fini(&zc);
2592 		}
2593 
2594 		ASSERT(!dsl_prop_hascb(ds));
2595 	}
2596 
2597 	/*
2598 	 * Change space accounting.
2599 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2600 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2601 	 * is true for each of {clone,origin} independently.
2602 	 */
2603 
2604 	delta = ddpa->cloneusedsnap -
2605 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2606 	ASSERT3S(delta, >=, 0);
2607 	ASSERT3U(ddpa->used, >=, delta);
2608 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2609 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2610 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2611 
2612 	delta = ddpa->originusedsnap -
2613 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2614 	ASSERT3S(delta, <=, 0);
2615 	ASSERT3U(ddpa->used, >=, -delta);
2616 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2617 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2618 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2619 
2620 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2621 
2622 	/* log history record */
2623 	spa_history_log_internal_ds(hds, "promote", tx, "");
2624 
2625 	dsl_dir_rele(odd, FTAG);
2626 	promote_rele(ddpa, FTAG);
2627 }
2628 
2629 /*
2630  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2631  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2632  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2633  * snapshots back to this dataset's origin.
2634  */
2635 static int
2636 snaplist_make(dsl_pool_t *dp,
2637     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2638 {
2639 	uint64_t obj = last_obj;
2640 
2641 	list_create(l, sizeof (struct promotenode),
2642 	    offsetof(struct promotenode, link));
2643 
2644 	while (obj != first_obj) {
2645 		dsl_dataset_t *ds;
2646 		struct promotenode *snap;
2647 		int err;
2648 
2649 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2650 		ASSERT(err != ENOENT);
2651 		if (err != 0)
2652 			return (err);
2653 
2654 		if (first_obj == 0)
2655 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2656 
2657 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2658 		snap->ds = ds;
2659 		list_insert_tail(l, snap);
2660 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2661 	}
2662 
2663 	return (0);
2664 }
2665 
2666 static int
2667 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2668 {
2669 	struct promotenode *snap;
2670 
2671 	*spacep = 0;
2672 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2673 		uint64_t used, comp, uncomp;
2674 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2675 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2676 		*spacep += used;
2677 	}
2678 	return (0);
2679 }
2680 
2681 static void
2682 snaplist_destroy(list_t *l, void *tag)
2683 {
2684 	struct promotenode *snap;
2685 
2686 	if (l == NULL || !list_link_active(&l->list_head))
2687 		return;
2688 
2689 	while ((snap = list_tail(l)) != NULL) {
2690 		list_remove(l, snap);
2691 		dsl_dataset_rele(snap->ds, tag);
2692 		kmem_free(snap, sizeof (*snap));
2693 	}
2694 	list_destroy(l);
2695 }
2696 
2697 static int
2698 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2699 {
2700 	int error;
2701 	dsl_dir_t *dd;
2702 	struct promotenode *snap;
2703 
2704 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2705 	    &ddpa->ddpa_clone);
2706 	if (error != 0)
2707 		return (error);
2708 	dd = ddpa->ddpa_clone->ds_dir;
2709 
2710 	if (ddpa->ddpa_clone->ds_is_snapshot ||
2711 	    !dsl_dir_is_clone(dd)) {
2712 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2713 		return (SET_ERROR(EINVAL));
2714 	}
2715 
2716 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2717 	    &ddpa->shared_snaps, tag);
2718 	if (error != 0)
2719 		goto out;
2720 
2721 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2722 	    &ddpa->clone_snaps, tag);
2723 	if (error != 0)
2724 		goto out;
2725 
2726 	snap = list_head(&ddpa->shared_snaps);
2727 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2728 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2729 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2730 	    &ddpa->origin_snaps, tag);
2731 	if (error != 0)
2732 		goto out;
2733 
2734 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2735 		error = dsl_dataset_hold_obj(dp,
2736 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2737 		    tag, &ddpa->origin_origin);
2738 		if (error != 0)
2739 			goto out;
2740 	}
2741 out:
2742 	if (error != 0)
2743 		promote_rele(ddpa, tag);
2744 	return (error);
2745 }
2746 
2747 static void
2748 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2749 {
2750 	snaplist_destroy(&ddpa->shared_snaps, tag);
2751 	snaplist_destroy(&ddpa->clone_snaps, tag);
2752 	snaplist_destroy(&ddpa->origin_snaps, tag);
2753 	if (ddpa->origin_origin != NULL)
2754 		dsl_dataset_rele(ddpa->origin_origin, tag);
2755 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2756 }
2757 
2758 /*
2759  * Promote a clone.
2760  *
2761  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2762  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2763  */
2764 int
2765 dsl_dataset_promote(const char *name, char *conflsnap)
2766 {
2767 	dsl_dataset_promote_arg_t ddpa = { 0 };
2768 	uint64_t numsnaps;
2769 	int error;
2770 	objset_t *os;
2771 
2772 	/*
2773 	 * We will modify space proportional to the number of
2774 	 * snapshots.  Compute numsnaps.
2775 	 */
2776 	error = dmu_objset_hold(name, FTAG, &os);
2777 	if (error != 0)
2778 		return (error);
2779 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2780 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2781 	    &numsnaps);
2782 	dmu_objset_rele(os, FTAG);
2783 	if (error != 0)
2784 		return (error);
2785 
2786 	ddpa.ddpa_clonename = name;
2787 	ddpa.err_ds = conflsnap;
2788 	ddpa.cr = CRED();
2789 
2790 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2791 	    dsl_dataset_promote_sync, &ddpa,
2792 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2793 }
2794 
2795 int
2796 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2797     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2798 {
2799 	/*
2800 	 * "slack" factor for received datasets with refquota set on them.
2801 	 * See the bottom of this function for details on its use.
2802 	 */
2803 	uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
2804 	int64_t unused_refres_delta;
2805 
2806 	/* they should both be heads */
2807 	if (clone->ds_is_snapshot ||
2808 	    origin_head->ds_is_snapshot)
2809 		return (SET_ERROR(EINVAL));
2810 
2811 	/* if we are not forcing, the branch point should be just before them */
2812 	if (!force && clone->ds_prev != origin_head->ds_prev)
2813 		return (SET_ERROR(EINVAL));
2814 
2815 	/* clone should be the clone (unless they are unrelated) */
2816 	if (clone->ds_prev != NULL &&
2817 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2818 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2819 		return (SET_ERROR(EINVAL));
2820 
2821 	/* the clone should be a child of the origin */
2822 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2823 		return (SET_ERROR(EINVAL));
2824 
2825 	/* origin_head shouldn't be modified unless 'force' */
2826 	if (!force &&
2827 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2828 		return (SET_ERROR(ETXTBSY));
2829 
2830 	/* origin_head should have no long holds (e.g. is not mounted) */
2831 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2832 		return (SET_ERROR(EBUSY));
2833 
2834 	/* check amount of any unconsumed refreservation */
2835 	unused_refres_delta =
2836 	    (int64_t)MIN(origin_head->ds_reserved,
2837 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2838 	    (int64_t)MIN(origin_head->ds_reserved,
2839 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2840 
2841 	if (unused_refres_delta > 0 &&
2842 	    unused_refres_delta >
2843 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2844 		return (SET_ERROR(ENOSPC));
2845 
2846 	/*
2847 	 * The clone can't be too much over the head's refquota.
2848 	 *
2849 	 * To ensure that the entire refquota can be used, we allow one
2850 	 * transaction to exceed the the refquota.  Therefore, this check
2851 	 * needs to also allow for the space referenced to be more than the
2852 	 * refquota.  The maximum amount of space that one transaction can use
2853 	 * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
2854 	 * overage ensures that we are able to receive a filesystem that
2855 	 * exceeds the refquota on the source system.
2856 	 *
2857 	 * So that overage is the refquota_slack we use below.
2858 	 */
2859 	if (origin_head->ds_quota != 0 &&
2860 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
2861 	    origin_head->ds_quota + refquota_slack)
2862 		return (SET_ERROR(EDQUOT));
2863 
2864 	return (0);
2865 }
2866 
2867 void
2868 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2869     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2870 {
2871 	dsl_pool_t *dp = dmu_tx_pool(tx);
2872 	int64_t unused_refres_delta;
2873 
2874 	ASSERT(clone->ds_reserved == 0);
2875 	/*
2876 	 * NOTE: On DEBUG kernels there could be a race between this and
2877 	 * the check function if spa_asize_inflation is adjusted...
2878 	 */
2879 	ASSERT(origin_head->ds_quota == 0 ||
2880 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
2881 	    DMU_MAX_ACCESS * spa_asize_inflation);
2882 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2883 
2884 	/*
2885 	 * Swap per-dataset feature flags.
2886 	 */
2887 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2888 		if (!(spa_feature_table[f].fi_flags &
2889 		    ZFEATURE_FLAG_PER_DATASET)) {
2890 			ASSERT(!clone->ds_feature_inuse[f]);
2891 			ASSERT(!origin_head->ds_feature_inuse[f]);
2892 			continue;
2893 		}
2894 
2895 		boolean_t clone_inuse = clone->ds_feature_inuse[f];
2896 		boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2897 
2898 		if (clone_inuse) {
2899 			dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2900 			clone->ds_feature_inuse[f] = B_FALSE;
2901 		}
2902 		if (origin_head_inuse) {
2903 			dsl_dataset_deactivate_feature(origin_head->ds_object,
2904 			    f, tx);
2905 			origin_head->ds_feature_inuse[f] = B_FALSE;
2906 		}
2907 		if (clone_inuse) {
2908 			dsl_dataset_activate_feature(origin_head->ds_object,
2909 			    f, tx);
2910 			origin_head->ds_feature_inuse[f] = B_TRUE;
2911 		}
2912 		if (origin_head_inuse) {
2913 			dsl_dataset_activate_feature(clone->ds_object, f, tx);
2914 			clone->ds_feature_inuse[f] = B_TRUE;
2915 		}
2916 	}
2917 
2918 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2919 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2920 
2921 	if (clone->ds_objset != NULL) {
2922 		dmu_objset_evict(clone->ds_objset);
2923 		clone->ds_objset = NULL;
2924 	}
2925 
2926 	if (origin_head->ds_objset != NULL) {
2927 		dmu_objset_evict(origin_head->ds_objset);
2928 		origin_head->ds_objset = NULL;
2929 	}
2930 
2931 	unused_refres_delta =
2932 	    (int64_t)MIN(origin_head->ds_reserved,
2933 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2934 	    (int64_t)MIN(origin_head->ds_reserved,
2935 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2936 
2937 	/*
2938 	 * Reset origin's unique bytes, if it exists.
2939 	 */
2940 	if (clone->ds_prev) {
2941 		dsl_dataset_t *origin = clone->ds_prev;
2942 		uint64_t comp, uncomp;
2943 
2944 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2945 		dsl_deadlist_space_range(&clone->ds_deadlist,
2946 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2947 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2948 	}
2949 
2950 	/* swap blkptrs */
2951 	{
2952 		blkptr_t tmp;
2953 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
2954 		dsl_dataset_phys(origin_head)->ds_bp =
2955 		    dsl_dataset_phys(clone)->ds_bp;
2956 		dsl_dataset_phys(clone)->ds_bp = tmp;
2957 	}
2958 
2959 	/* set dd_*_bytes */
2960 	{
2961 		int64_t dused, dcomp, duncomp;
2962 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2963 		uint64_t odl_used, odl_comp, odl_uncomp;
2964 
2965 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2966 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2967 
2968 		dsl_deadlist_space(&clone->ds_deadlist,
2969 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2970 		dsl_deadlist_space(&origin_head->ds_deadlist,
2971 		    &odl_used, &odl_comp, &odl_uncomp);
2972 
2973 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2974 		    cdl_used -
2975 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2976 		    odl_used);
2977 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2978 		    cdl_comp -
2979 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2980 		    odl_comp);
2981 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2982 		    cdl_uncomp -
2983 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2984 		    odl_uncomp);
2985 
2986 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2987 		    dused, dcomp, duncomp, tx);
2988 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2989 		    -dused, -dcomp, -duncomp, tx);
2990 
2991 		/*
2992 		 * The difference in the space used by snapshots is the
2993 		 * difference in snapshot space due to the head's
2994 		 * deadlist (since that's the only thing that's
2995 		 * changing that affects the snapused).
2996 		 */
2997 		dsl_deadlist_space_range(&clone->ds_deadlist,
2998 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2999 		    &cdl_used, &cdl_comp, &cdl_uncomp);
3000 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
3001 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3002 		    &odl_used, &odl_comp, &odl_uncomp);
3003 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3004 		    DD_USED_HEAD, DD_USED_SNAP, tx);
3005 	}
3006 
3007 	/* swap ds_*_bytes */
3008 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3009 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
3010 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3011 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
3012 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3013 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3014 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3015 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3016 
3017 	/* apply any parent delta for change in unconsumed refreservation */
3018 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3019 	    unused_refres_delta, 0, 0, tx);
3020 
3021 	/*
3022 	 * Swap deadlists.
3023 	 */
3024 	dsl_deadlist_close(&clone->ds_deadlist);
3025 	dsl_deadlist_close(&origin_head->ds_deadlist);
3026 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3027 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3028 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3029 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3030 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3031 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3032 
3033 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3034 
3035 	spa_history_log_internal_ds(clone, "clone swap", tx,
3036 	    "parent=%s", origin_head->ds_dir->dd_myname);
3037 }
3038 
3039 /*
3040  * Given a pool name and a dataset object number in that pool,
3041  * return the name of that dataset.
3042  */
3043 int
3044 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3045 {
3046 	dsl_pool_t *dp;
3047 	dsl_dataset_t *ds;
3048 	int error;
3049 
3050 	error = dsl_pool_hold(pname, FTAG, &dp);
3051 	if (error != 0)
3052 		return (error);
3053 
3054 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3055 	if (error == 0) {
3056 		dsl_dataset_name(ds, buf);
3057 		dsl_dataset_rele(ds, FTAG);
3058 	}
3059 	dsl_pool_rele(dp, FTAG);
3060 
3061 	return (error);
3062 }
3063 
3064 int
3065 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3066     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3067 {
3068 	int error = 0;
3069 
3070 	ASSERT3S(asize, >, 0);
3071 
3072 	/*
3073 	 * *ref_rsrv is the portion of asize that will come from any
3074 	 * unconsumed refreservation space.
3075 	 */
3076 	*ref_rsrv = 0;
3077 
3078 	mutex_enter(&ds->ds_lock);
3079 	/*
3080 	 * Make a space adjustment for reserved bytes.
3081 	 */
3082 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3083 		ASSERT3U(*used, >=,
3084 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3085 		*used -=
3086 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3087 		*ref_rsrv =
3088 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3089 	}
3090 
3091 	if (!check_quota || ds->ds_quota == 0) {
3092 		mutex_exit(&ds->ds_lock);
3093 		return (0);
3094 	}
3095 	/*
3096 	 * If they are requesting more space, and our current estimate
3097 	 * is over quota, they get to try again unless the actual
3098 	 * on-disk is over quota and there are no pending changes (which
3099 	 * may free up space for us).
3100 	 */
3101 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3102 	    ds->ds_quota) {
3103 		if (inflight > 0 ||
3104 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3105 			error = SET_ERROR(ERESTART);
3106 		else
3107 			error = SET_ERROR(EDQUOT);
3108 	}
3109 	mutex_exit(&ds->ds_lock);
3110 
3111 	return (error);
3112 }
3113 
3114 typedef struct dsl_dataset_set_qr_arg {
3115 	const char *ddsqra_name;
3116 	zprop_source_t ddsqra_source;
3117 	uint64_t ddsqra_value;
3118 } dsl_dataset_set_qr_arg_t;
3119 
3120 
3121 /* ARGSUSED */
3122 static int
3123 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3124 {
3125 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3126 	dsl_pool_t *dp = dmu_tx_pool(tx);
3127 	dsl_dataset_t *ds;
3128 	int error;
3129 	uint64_t newval;
3130 
3131 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3132 		return (SET_ERROR(ENOTSUP));
3133 
3134 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3135 	if (error != 0)
3136 		return (error);
3137 
3138 	if (ds->ds_is_snapshot) {
3139 		dsl_dataset_rele(ds, FTAG);
3140 		return (SET_ERROR(EINVAL));
3141 	}
3142 
3143 	error = dsl_prop_predict(ds->ds_dir,
3144 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3145 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3146 	if (error != 0) {
3147 		dsl_dataset_rele(ds, FTAG);
3148 		return (error);
3149 	}
3150 
3151 	if (newval == 0) {
3152 		dsl_dataset_rele(ds, FTAG);
3153 		return (0);
3154 	}
3155 
3156 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3157 	    newval < ds->ds_reserved) {
3158 		dsl_dataset_rele(ds, FTAG);
3159 		return (SET_ERROR(ENOSPC));
3160 	}
3161 
3162 	dsl_dataset_rele(ds, FTAG);
3163 	return (0);
3164 }
3165 
3166 static void
3167 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3168 {
3169 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3170 	dsl_pool_t *dp = dmu_tx_pool(tx);
3171 	dsl_dataset_t *ds;
3172 	uint64_t newval;
3173 
3174 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3175 
3176 	dsl_prop_set_sync_impl(ds,
3177 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3178 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3179 	    &ddsqra->ddsqra_value, tx);
3180 
3181 	VERIFY0(dsl_prop_get_int_ds(ds,
3182 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3183 
3184 	if (ds->ds_quota != newval) {
3185 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3186 		ds->ds_quota = newval;
3187 	}
3188 	dsl_dataset_rele(ds, FTAG);
3189 }
3190 
3191 int
3192 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3193     uint64_t refquota)
3194 {
3195 	dsl_dataset_set_qr_arg_t ddsqra;
3196 
3197 	ddsqra.ddsqra_name = dsname;
3198 	ddsqra.ddsqra_source = source;
3199 	ddsqra.ddsqra_value = refquota;
3200 
3201 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3202 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3203 }
3204 
3205 static int
3206 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3207 {
3208 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3209 	dsl_pool_t *dp = dmu_tx_pool(tx);
3210 	dsl_dataset_t *ds;
3211 	int error;
3212 	uint64_t newval, unique;
3213 
3214 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3215 		return (SET_ERROR(ENOTSUP));
3216 
3217 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3218 	if (error != 0)
3219 		return (error);
3220 
3221 	if (ds->ds_is_snapshot) {
3222 		dsl_dataset_rele(ds, FTAG);
3223 		return (SET_ERROR(EINVAL));
3224 	}
3225 
3226 	error = dsl_prop_predict(ds->ds_dir,
3227 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3228 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3229 	if (error != 0) {
3230 		dsl_dataset_rele(ds, FTAG);
3231 		return (error);
3232 	}
3233 
3234 	/*
3235 	 * If we are doing the preliminary check in open context, the
3236 	 * space estimates may be inaccurate.
3237 	 */
3238 	if (!dmu_tx_is_syncing(tx)) {
3239 		dsl_dataset_rele(ds, FTAG);
3240 		return (0);
3241 	}
3242 
3243 	mutex_enter(&ds->ds_lock);
3244 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3245 		dsl_dataset_recalc_head_uniq(ds);
3246 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3247 	mutex_exit(&ds->ds_lock);
3248 
3249 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3250 		uint64_t delta = MAX(unique, newval) -
3251 		    MAX(unique, ds->ds_reserved);
3252 
3253 		if (delta >
3254 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3255 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3256 			dsl_dataset_rele(ds, FTAG);
3257 			return (SET_ERROR(ENOSPC));
3258 		}
3259 	}
3260 
3261 	dsl_dataset_rele(ds, FTAG);
3262 	return (0);
3263 }
3264 
3265 void
3266 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3267     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3268 {
3269 	uint64_t newval;
3270 	uint64_t unique;
3271 	int64_t delta;
3272 
3273 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3274 	    source, sizeof (value), 1, &value, tx);
3275 
3276 	VERIFY0(dsl_prop_get_int_ds(ds,
3277 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3278 
3279 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3280 	mutex_enter(&ds->ds_dir->dd_lock);
3281 	mutex_enter(&ds->ds_lock);
3282 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3283 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3284 	delta = MAX(0, (int64_t)(newval - unique)) -
3285 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3286 	ds->ds_reserved = newval;
3287 	mutex_exit(&ds->ds_lock);
3288 
3289 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3290 	mutex_exit(&ds->ds_dir->dd_lock);
3291 }
3292 
3293 static void
3294 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3295 {
3296 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3297 	dsl_pool_t *dp = dmu_tx_pool(tx);
3298 	dsl_dataset_t *ds;
3299 
3300 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3301 	dsl_dataset_set_refreservation_sync_impl(ds,
3302 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3303 	dsl_dataset_rele(ds, FTAG);
3304 }
3305 
3306 int
3307 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3308     uint64_t refreservation)
3309 {
3310 	dsl_dataset_set_qr_arg_t ddsqra;
3311 
3312 	ddsqra.ddsqra_name = dsname;
3313 	ddsqra.ddsqra_source = source;
3314 	ddsqra.ddsqra_value = refreservation;
3315 
3316 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3317 	    dsl_dataset_set_refreservation_sync, &ddsqra,
3318 	    0, ZFS_SPACE_CHECK_NONE));
3319 }
3320 
3321 /*
3322  * Return (in *usedp) the amount of space written in new that is not
3323  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3324  * a snapshot before new, in new's filesystem (or its origin).  If not then
3325  * fail and return EINVAL.
3326  *
3327  * The written space is calculated by considering two components:  First, we
3328  * ignore any freed space, and calculate the written as new's used space
3329  * minus old's used space.  Next, we add in the amount of space that was freed
3330  * between the two snapshots, thus reducing new's used space relative to old's.
3331  * Specifically, this is the space that was born before old->ds_creation_txg,
3332  * and freed before new (ie. on new's deadlist or a previous deadlist).
3333  *
3334  * space freed                         [---------------------]
3335  * snapshots                       ---O-------O--------O-------O------
3336  *                                         oldsnap            new
3337  */
3338 int
3339 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3340     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3341 {
3342 	int err = 0;
3343 	uint64_t snapobj;
3344 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3345 
3346 	ASSERT(dsl_pool_config_held(dp));
3347 
3348 	*usedp = 0;
3349 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3350 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3351 
3352 	*compp = 0;
3353 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3354 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3355 
3356 	*uncompp = 0;
3357 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3358 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3359 
3360 	snapobj = new->ds_object;
3361 	while (snapobj != oldsnap->ds_object) {
3362 		dsl_dataset_t *snap;
3363 		uint64_t used, comp, uncomp;
3364 
3365 		if (snapobj == new->ds_object) {
3366 			snap = new;
3367 		} else {
3368 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3369 			if (err != 0)
3370 				break;
3371 		}
3372 
3373 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3374 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3375 			/*
3376 			 * The blocks in the deadlist can not be born after
3377 			 * ds_prev_snap_txg, so get the whole deadlist space,
3378 			 * which is more efficient (especially for old-format
3379 			 * deadlists).  Unfortunately the deadlist code
3380 			 * doesn't have enough information to make this
3381 			 * optimization itself.
3382 			 */
3383 			dsl_deadlist_space(&snap->ds_deadlist,
3384 			    &used, &comp, &uncomp);
3385 		} else {
3386 			dsl_deadlist_space_range(&snap->ds_deadlist,
3387 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3388 			    &used, &comp, &uncomp);
3389 		}
3390 		*usedp += used;
3391 		*compp += comp;
3392 		*uncompp += uncomp;
3393 
3394 		/*
3395 		 * If we get to the beginning of the chain of snapshots
3396 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3397 		 * was not a snapshot of/before new.
3398 		 */
3399 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3400 		if (snap != new)
3401 			dsl_dataset_rele(snap, FTAG);
3402 		if (snapobj == 0) {
3403 			err = SET_ERROR(EINVAL);
3404 			break;
3405 		}
3406 
3407 	}
3408 	return (err);
3409 }
3410 
3411 /*
3412  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3413  * lastsnap, and all snapshots in between are deleted.
3414  *
3415  * blocks that would be freed            [---------------------------]
3416  * snapshots                       ---O-------O--------O-------O--------O
3417  *                                        firstsnap        lastsnap
3418  *
3419  * This is the set of blocks that were born after the snap before firstsnap,
3420  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3421  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3422  * We calculate this by iterating over the relevant deadlists (from the snap
3423  * after lastsnap, backward to the snap after firstsnap), summing up the
3424  * space on the deadlist that was born after the snap before firstsnap.
3425  */
3426 int
3427 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3428     dsl_dataset_t *lastsnap,
3429     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3430 {
3431 	int err = 0;
3432 	uint64_t snapobj;
3433 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3434 
3435 	ASSERT(firstsnap->ds_is_snapshot);
3436 	ASSERT(lastsnap->ds_is_snapshot);
3437 
3438 	/*
3439 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3440 	 * is before lastsnap.
3441 	 */
3442 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3443 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3444 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3445 		return (SET_ERROR(EINVAL));
3446 
3447 	*usedp = *compp = *uncompp = 0;
3448 
3449 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3450 	while (snapobj != firstsnap->ds_object) {
3451 		dsl_dataset_t *ds;
3452 		uint64_t used, comp, uncomp;
3453 
3454 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3455 		if (err != 0)
3456 			break;
3457 
3458 		dsl_deadlist_space_range(&ds->ds_deadlist,
3459 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3460 		    &used, &comp, &uncomp);
3461 		*usedp += used;
3462 		*compp += comp;
3463 		*uncompp += uncomp;
3464 
3465 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3466 		ASSERT3U(snapobj, !=, 0);
3467 		dsl_dataset_rele(ds, FTAG);
3468 	}
3469 	return (err);
3470 }
3471 
3472 /*
3473  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3474  * For example, they could both be snapshots of the same filesystem, and
3475  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3476  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3477  * filesystem.  Or 'earlier' could be the origin's origin.
3478  *
3479  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3480  */
3481 boolean_t
3482 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3483     uint64_t earlier_txg)
3484 {
3485 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3486 	int error;
3487 	boolean_t ret;
3488 
3489 	ASSERT(dsl_pool_config_held(dp));
3490 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3491 
3492 	if (earlier_txg == 0)
3493 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3494 
3495 	if (later->ds_is_snapshot &&
3496 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3497 		return (B_FALSE);
3498 
3499 	if (later->ds_dir == earlier->ds_dir)
3500 		return (B_TRUE);
3501 	if (!dsl_dir_is_clone(later->ds_dir))
3502 		return (B_FALSE);
3503 
3504 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3505 		return (B_TRUE);
3506 	dsl_dataset_t *origin;
3507 	error = dsl_dataset_hold_obj(dp,
3508 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3509 	if (error != 0)
3510 		return (B_FALSE);
3511 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3512 	dsl_dataset_rele(origin, FTAG);
3513 	return (ret);
3514 }
3515 
3516 void
3517 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3518 {
3519 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3520 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3521 }
3522 
3523 boolean_t
3524 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3525 {
3526 	dmu_object_info_t doi;
3527 
3528 	dmu_object_info_from_db(ds->ds_dbuf, &doi);
3529 	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3530 }
3531 
3532 boolean_t
3533 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3534 {
3535 	return (dsl_dataset_is_zapified(ds) &&
3536 	    zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3537 	    ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3538 }
3539