xref: /titanic_50/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 2321aa36382ca9bc1d3f0437d553acc4e342c81b)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/dmu_objset.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_prop.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/arc.h>
36 #include <sys/zio.h>
37 #include <sys/zap.h>
38 #include <sys/unique.h>
39 #include <sys/zfs_context.h>
40 #include <sys/zfs_ioctl.h>
41 #include <sys/spa.h>
42 #include <sys/sunddi.h>
43 
44 static char *dsl_reaper = "the grim reaper";
45 
46 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
47 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
48 static dsl_checkfunc_t dsl_dataset_rollback_check;
49 static dsl_syncfunc_t dsl_dataset_rollback_sync;
50 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
51 
52 #define	DS_REF_MAX	(1ULL << 62)
53 
54 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
55 
56 #define	DSL_DATASET_IS_DESTROYED(ds)	((ds)->ds_owner == dsl_reaper)
57 
58 static void dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag);
59 
60 /*
61  * Figure out how much of this delta should be propogated to the dsl_dir
62  * layer.  If there's a refreservation, that space has already been
63  * partially accounted for in our ancestors.
64  */
65 static int64_t
66 parent_delta(dsl_dataset_t *ds, int64_t delta)
67 {
68 	uint64_t old_bytes, new_bytes;
69 
70 	if (ds->ds_reserved == 0)
71 		return (delta);
72 
73 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
74 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
75 
76 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
77 	return (new_bytes - old_bytes);
78 }
79 
80 void
81 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
82 {
83 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
84 	int compressed = BP_GET_PSIZE(bp);
85 	int uncompressed = BP_GET_UCSIZE(bp);
86 	int64_t delta;
87 
88 	dprintf_bp(bp, "born, ds=%p\n", ds);
89 
90 	ASSERT(dmu_tx_is_syncing(tx));
91 	/* It could have been compressed away to nothing */
92 	if (BP_IS_HOLE(bp))
93 		return;
94 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
95 	ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
96 	if (ds == NULL) {
97 		/*
98 		 * Account for the meta-objset space in its placeholder
99 		 * dsl_dir.
100 		 */
101 		ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
102 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
103 		    used, compressed, uncompressed, tx);
104 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
105 		return;
106 	}
107 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
108 	mutex_enter(&ds->ds_lock);
109 	delta = parent_delta(ds, used);
110 	ds->ds_phys->ds_used_bytes += used;
111 	ds->ds_phys->ds_compressed_bytes += compressed;
112 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
113 	ds->ds_phys->ds_unique_bytes += used;
114 	mutex_exit(&ds->ds_lock);
115 	dsl_dir_diduse_space(ds->ds_dir, delta, compressed, uncompressed, tx);
116 }
117 
118 int
119 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
120     dmu_tx_t *tx)
121 {
122 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
123 	int compressed = BP_GET_PSIZE(bp);
124 	int uncompressed = BP_GET_UCSIZE(bp);
125 
126 	ASSERT(dmu_tx_is_syncing(tx));
127 	/* No block pointer => nothing to free */
128 	if (BP_IS_HOLE(bp))
129 		return (0);
130 
131 	ASSERT(used > 0);
132 	if (ds == NULL) {
133 		int err;
134 		/*
135 		 * Account for the meta-objset space in its placeholder
136 		 * dataset.
137 		 */
138 		err = arc_free(pio, tx->tx_pool->dp_spa,
139 		    tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
140 		ASSERT(err == 0);
141 
142 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
143 		    -used, -compressed, -uncompressed, tx);
144 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
145 		return (used);
146 	}
147 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
148 
149 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
150 
151 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
152 		int err;
153 		int64_t delta;
154 
155 		dprintf_bp(bp, "freeing: %s", "");
156 		err = arc_free(pio, tx->tx_pool->dp_spa,
157 		    tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
158 		ASSERT(err == 0);
159 
160 		mutex_enter(&ds->ds_lock);
161 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
162 		    !DS_UNIQUE_IS_ACCURATE(ds));
163 		delta = parent_delta(ds, -used);
164 		ds->ds_phys->ds_unique_bytes -= used;
165 		mutex_exit(&ds->ds_lock);
166 		dsl_dir_diduse_space(ds->ds_dir,
167 		    delta, -compressed, -uncompressed, tx);
168 	} else {
169 		dprintf_bp(bp, "putting on dead list: %s", "");
170 		VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
171 		ASSERT3U(ds->ds_prev->ds_object, ==,
172 		    ds->ds_phys->ds_prev_snap_obj);
173 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
174 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
175 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
176 		    ds->ds_object && bp->blk_birth >
177 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
178 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
179 			mutex_enter(&ds->ds_prev->ds_lock);
180 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
181 			mutex_exit(&ds->ds_prev->ds_lock);
182 		}
183 	}
184 	mutex_enter(&ds->ds_lock);
185 	ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
186 	ds->ds_phys->ds_used_bytes -= used;
187 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
188 	ds->ds_phys->ds_compressed_bytes -= compressed;
189 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
190 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
191 	mutex_exit(&ds->ds_lock);
192 
193 	return (used);
194 }
195 
196 uint64_t
197 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
198 {
199 	uint64_t trysnap = 0;
200 
201 	if (ds == NULL)
202 		return (0);
203 	/*
204 	 * The snapshot creation could fail, but that would cause an
205 	 * incorrect FALSE return, which would only result in an
206 	 * overestimation of the amount of space that an operation would
207 	 * consume, which is OK.
208 	 *
209 	 * There's also a small window where we could miss a pending
210 	 * snapshot, because we could set the sync task in the quiescing
211 	 * phase.  So this should only be used as a guess.
212 	 */
213 	if (ds->ds_trysnap_txg >
214 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
215 		trysnap = ds->ds_trysnap_txg;
216 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
217 }
218 
219 int
220 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
221 {
222 	return (blk_birth > dsl_dataset_prev_snap_txg(ds));
223 }
224 
225 /* ARGSUSED */
226 static void
227 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
228 {
229 	dsl_dataset_t *ds = dsv;
230 
231 	ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
232 
233 	dprintf_ds(ds, "evicting %s\n", "");
234 
235 	unique_remove(ds->ds_fsid_guid);
236 
237 	if (ds->ds_user_ptr != NULL)
238 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
239 
240 	if (ds->ds_prev) {
241 		dsl_dataset_drop_ref(ds->ds_prev, ds);
242 		ds->ds_prev = NULL;
243 	}
244 
245 	bplist_close(&ds->ds_deadlist);
246 	if (ds->ds_dir)
247 		dsl_dir_close(ds->ds_dir, ds);
248 
249 	ASSERT(!list_link_active(&ds->ds_synced_link));
250 
251 	mutex_destroy(&ds->ds_lock);
252 	mutex_destroy(&ds->ds_opening_lock);
253 	mutex_destroy(&ds->ds_deadlist.bpl_lock);
254 	rw_destroy(&ds->ds_rwlock);
255 	cv_destroy(&ds->ds_exclusive_cv);
256 
257 	kmem_free(ds, sizeof (dsl_dataset_t));
258 }
259 
260 static int
261 dsl_dataset_get_snapname(dsl_dataset_t *ds)
262 {
263 	dsl_dataset_phys_t *headphys;
264 	int err;
265 	dmu_buf_t *headdbuf;
266 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
267 	objset_t *mos = dp->dp_meta_objset;
268 
269 	if (ds->ds_snapname[0])
270 		return (0);
271 	if (ds->ds_phys->ds_next_snap_obj == 0)
272 		return (0);
273 
274 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
275 	    FTAG, &headdbuf);
276 	if (err)
277 		return (err);
278 	headphys = headdbuf->db_data;
279 	err = zap_value_search(dp->dp_meta_objset,
280 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
281 	dmu_buf_rele(headdbuf, FTAG);
282 	return (err);
283 }
284 
285 static int
286 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
287 {
288 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
289 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
290 	matchtype_t mt;
291 	int err;
292 
293 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
294 		mt = MT_FIRST;
295 	else
296 		mt = MT_EXACT;
297 
298 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
299 	    value, mt, NULL, 0, NULL);
300 	if (err == ENOTSUP && mt == MT_FIRST)
301 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
302 	return (err);
303 }
304 
305 static int
306 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
307 {
308 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
309 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
310 	matchtype_t mt;
311 	int err;
312 
313 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
314 		mt = MT_FIRST;
315 	else
316 		mt = MT_EXACT;
317 
318 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
319 	if (err == ENOTSUP && mt == MT_FIRST)
320 		err = zap_remove(mos, snapobj, name, tx);
321 	return (err);
322 }
323 
324 static int
325 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
326     dsl_dataset_t **dsp)
327 {
328 	objset_t *mos = dp->dp_meta_objset;
329 	dmu_buf_t *dbuf;
330 	dsl_dataset_t *ds;
331 	int err;
332 
333 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
334 	    dsl_pool_sync_context(dp));
335 
336 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
337 	if (err)
338 		return (err);
339 	ds = dmu_buf_get_user(dbuf);
340 	if (ds == NULL) {
341 		dsl_dataset_t *winner;
342 
343 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
344 		ds->ds_dbuf = dbuf;
345 		ds->ds_object = dsobj;
346 		ds->ds_phys = dbuf->db_data;
347 
348 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
349 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
350 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
351 		    NULL);
352 		rw_init(&ds->ds_rwlock, 0, 0, 0);
353 		cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
354 
355 		err = bplist_open(&ds->ds_deadlist,
356 		    mos, ds->ds_phys->ds_deadlist_obj);
357 		if (err == 0) {
358 			err = dsl_dir_open_obj(dp,
359 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
360 		}
361 		if (err) {
362 			/*
363 			 * we don't really need to close the blist if we
364 			 * just opened it.
365 			 */
366 			mutex_destroy(&ds->ds_lock);
367 			mutex_destroy(&ds->ds_opening_lock);
368 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
369 			rw_destroy(&ds->ds_rwlock);
370 			cv_destroy(&ds->ds_exclusive_cv);
371 			kmem_free(ds, sizeof (dsl_dataset_t));
372 			dmu_buf_rele(dbuf, tag);
373 			return (err);
374 		}
375 
376 		if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) {
377 			ds->ds_snapname[0] = '\0';
378 			if (ds->ds_phys->ds_prev_snap_obj) {
379 				err = dsl_dataset_get_ref(dp,
380 				    ds->ds_phys->ds_prev_snap_obj,
381 				    ds, &ds->ds_prev);
382 			}
383 		} else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
384 			err = dsl_dataset_get_snapname(ds);
385 		}
386 
387 		if (!dsl_dataset_is_snapshot(ds)) {
388 			/*
389 			 * In sync context, we're called with either no lock
390 			 * or with the write lock.  If we're not syncing,
391 			 * we're always called with the read lock held.
392 			 */
393 			boolean_t need_lock =
394 			    !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
395 			    dsl_pool_sync_context(dp);
396 
397 			if (need_lock)
398 				rw_enter(&dp->dp_config_rwlock, RW_READER);
399 
400 			err = dsl_prop_get_ds_locked(ds->ds_dir,
401 			    "refreservation", sizeof (uint64_t), 1,
402 			    &ds->ds_reserved, NULL);
403 			if (err == 0) {
404 				err = dsl_prop_get_ds_locked(ds->ds_dir,
405 				    "refquota", sizeof (uint64_t), 1,
406 				    &ds->ds_quota, NULL);
407 			}
408 
409 			if (need_lock)
410 				rw_exit(&dp->dp_config_rwlock);
411 		} else {
412 			ds->ds_reserved = ds->ds_quota = 0;
413 		}
414 
415 		if (err == 0) {
416 			winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
417 			    dsl_dataset_evict);
418 		}
419 		if (err || winner) {
420 			bplist_close(&ds->ds_deadlist);
421 			if (ds->ds_prev)
422 				dsl_dataset_drop_ref(ds->ds_prev, ds);
423 			dsl_dir_close(ds->ds_dir, ds);
424 			mutex_destroy(&ds->ds_lock);
425 			mutex_destroy(&ds->ds_opening_lock);
426 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
427 			rw_destroy(&ds->ds_rwlock);
428 			cv_destroy(&ds->ds_exclusive_cv);
429 			kmem_free(ds, sizeof (dsl_dataset_t));
430 			if (err) {
431 				dmu_buf_rele(dbuf, tag);
432 				return (err);
433 			}
434 			ds = winner;
435 		} else {
436 			ds->ds_fsid_guid =
437 			    unique_insert(ds->ds_phys->ds_fsid_guid);
438 		}
439 	}
440 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
441 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
442 	mutex_enter(&ds->ds_lock);
443 	if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
444 		mutex_exit(&ds->ds_lock);
445 		dmu_buf_rele(ds->ds_dbuf, tag);
446 		return (ENOENT);
447 	}
448 	mutex_exit(&ds->ds_lock);
449 	*dsp = ds;
450 	return (0);
451 }
452 
453 static int
454 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
455 {
456 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
457 
458 	/*
459 	 * In syncing context we don't want the rwlock lock: there
460 	 * may be an existing writer waiting for sync phase to
461 	 * finish.  We don't need to worry about such writers, since
462 	 * sync phase is single-threaded, so the writer can't be
463 	 * doing anything while we are active.
464 	 */
465 	if (dsl_pool_sync_context(dp)) {
466 		ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
467 		return (0);
468 	}
469 
470 	/*
471 	 * Normal users will hold the ds_rwlock as a READER until they
472 	 * are finished (i.e., call dsl_dataset_rele()).  "Owners" will
473 	 * drop their READER lock after they set the ds_owner field.
474 	 *
475 	 * If the dataset is being destroyed, the destroy thread will
476 	 * obtain a WRITER lock for exclusive access after it's done its
477 	 * open-context work and then change the ds_owner to
478 	 * dsl_reaper once destruction is assured.  So threads
479 	 * may block here temporarily, until the "destructability" of
480 	 * the dataset is determined.
481 	 */
482 	ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
483 	mutex_enter(&ds->ds_lock);
484 	while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
485 		rw_exit(&dp->dp_config_rwlock);
486 		cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
487 		if (DSL_DATASET_IS_DESTROYED(ds)) {
488 			mutex_exit(&ds->ds_lock);
489 			dsl_dataset_drop_ref(ds, tag);
490 			rw_enter(&dp->dp_config_rwlock, RW_READER);
491 			return (ENOENT);
492 		}
493 		rw_enter(&dp->dp_config_rwlock, RW_READER);
494 	}
495 	mutex_exit(&ds->ds_lock);
496 	return (0);
497 }
498 
499 int
500 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
501     dsl_dataset_t **dsp)
502 {
503 	int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
504 
505 	if (err)
506 		return (err);
507 	return (dsl_dataset_hold_ref(*dsp, tag));
508 }
509 
510 int
511 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, int flags, void *owner,
512     dsl_dataset_t **dsp)
513 {
514 	int err = dsl_dataset_hold_obj(dp, dsobj, owner, dsp);
515 
516 	ASSERT(DS_MODE_TYPE(flags) != DS_MODE_USER);
517 
518 	if (err)
519 		return (err);
520 	if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
521 		dsl_dataset_rele(*dsp, owner);
522 		return (EBUSY);
523 	}
524 	return (0);
525 }
526 
527 int
528 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
529 {
530 	dsl_dir_t *dd;
531 	dsl_pool_t *dp;
532 	const char *snapname;
533 	uint64_t obj;
534 	int err = 0;
535 
536 	err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
537 	if (err)
538 		return (err);
539 
540 	dp = dd->dd_pool;
541 	obj = dd->dd_phys->dd_head_dataset_obj;
542 	rw_enter(&dp->dp_config_rwlock, RW_READER);
543 	if (obj)
544 		err = dsl_dataset_get_ref(dp, obj, tag, dsp);
545 	else
546 		err = ENOENT;
547 	if (err)
548 		goto out;
549 
550 	err = dsl_dataset_hold_ref(*dsp, tag);
551 
552 	/* we may be looking for a snapshot */
553 	if (err == 0 && snapname != NULL) {
554 		dsl_dataset_t *ds = NULL;
555 
556 		if (*snapname++ != '@') {
557 			dsl_dataset_rele(*dsp, tag);
558 			err = ENOENT;
559 			goto out;
560 		}
561 
562 		dprintf("looking for snapshot '%s'\n", snapname);
563 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
564 		if (err == 0)
565 			err = dsl_dataset_get_ref(dp, obj, tag, &ds);
566 		dsl_dataset_rele(*dsp, tag);
567 
568 		ASSERT3U((err == 0), ==, (ds != NULL));
569 
570 		if (ds) {
571 			mutex_enter(&ds->ds_lock);
572 			if (ds->ds_snapname[0] == 0)
573 				(void) strlcpy(ds->ds_snapname, snapname,
574 				    sizeof (ds->ds_snapname));
575 			mutex_exit(&ds->ds_lock);
576 			err = dsl_dataset_hold_ref(ds, tag);
577 			*dsp = err ? NULL : ds;
578 		}
579 	}
580 out:
581 	rw_exit(&dp->dp_config_rwlock);
582 	dsl_dir_close(dd, FTAG);
583 	return (err);
584 }
585 
586 int
587 dsl_dataset_own(const char *name, int flags, void *owner, dsl_dataset_t **dsp)
588 {
589 	int err = dsl_dataset_hold(name, owner, dsp);
590 	if (err)
591 		return (err);
592 	if ((*dsp)->ds_phys->ds_num_children > 0 &&
593 	    !DS_MODE_IS_READONLY(flags)) {
594 		dsl_dataset_rele(*dsp, owner);
595 		return (EROFS);
596 	}
597 	if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
598 		dsl_dataset_rele(*dsp, owner);
599 		return (EBUSY);
600 	}
601 	return (0);
602 }
603 
604 void
605 dsl_dataset_name(dsl_dataset_t *ds, char *name)
606 {
607 	if (ds == NULL) {
608 		(void) strcpy(name, "mos");
609 	} else {
610 		dsl_dir_name(ds->ds_dir, name);
611 		VERIFY(0 == dsl_dataset_get_snapname(ds));
612 		if (ds->ds_snapname[0]) {
613 			(void) strcat(name, "@");
614 			/*
615 			 * We use a "recursive" mutex so that we
616 			 * can call dprintf_ds() with ds_lock held.
617 			 */
618 			if (!MUTEX_HELD(&ds->ds_lock)) {
619 				mutex_enter(&ds->ds_lock);
620 				(void) strcat(name, ds->ds_snapname);
621 				mutex_exit(&ds->ds_lock);
622 			} else {
623 				(void) strcat(name, ds->ds_snapname);
624 			}
625 		}
626 	}
627 }
628 
629 static int
630 dsl_dataset_namelen(dsl_dataset_t *ds)
631 {
632 	int result;
633 
634 	if (ds == NULL) {
635 		result = 3;	/* "mos" */
636 	} else {
637 		result = dsl_dir_namelen(ds->ds_dir);
638 		VERIFY(0 == dsl_dataset_get_snapname(ds));
639 		if (ds->ds_snapname[0]) {
640 			++result;	/* adding one for the @-sign */
641 			if (!MUTEX_HELD(&ds->ds_lock)) {
642 				mutex_enter(&ds->ds_lock);
643 				result += strlen(ds->ds_snapname);
644 				mutex_exit(&ds->ds_lock);
645 			} else {
646 				result += strlen(ds->ds_snapname);
647 			}
648 		}
649 	}
650 
651 	return (result);
652 }
653 
654 static void
655 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
656 {
657 	dmu_buf_rele(ds->ds_dbuf, tag);
658 }
659 
660 void
661 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
662 {
663 	ASSERT(ds->ds_owner != tag);
664 	if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
665 		rw_exit(&ds->ds_rwlock);
666 	}
667 	dsl_dataset_drop_ref(ds, tag);
668 }
669 
670 void
671 dsl_dataset_disown(dsl_dataset_t *ds, void *owner)
672 {
673 	ASSERT((ds->ds_owner == owner && ds->ds_dbuf) ||
674 	    (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
675 
676 	mutex_enter(&ds->ds_lock);
677 	ds->ds_owner = NULL;
678 	if (RW_WRITE_HELD(&ds->ds_rwlock)) {
679 		rw_exit(&ds->ds_rwlock);
680 		cv_broadcast(&ds->ds_exclusive_cv);
681 	}
682 	mutex_exit(&ds->ds_lock);
683 	if (ds->ds_dbuf)
684 		dsl_dataset_drop_ref(ds, owner);
685 	else
686 		dsl_dataset_evict(ds->ds_dbuf, ds);
687 }
688 
689 boolean_t
690 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *owner)
691 {
692 	boolean_t gotit = FALSE;
693 
694 	mutex_enter(&ds->ds_lock);
695 	if (ds->ds_owner == NULL &&
696 	    (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
697 		ds->ds_owner = owner;
698 		if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
699 			rw_exit(&ds->ds_rwlock);
700 		gotit = TRUE;
701 	}
702 	mutex_exit(&ds->ds_lock);
703 	return (gotit);
704 }
705 
706 void
707 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
708 {
709 	ASSERT3P(owner, ==, ds->ds_owner);
710 	if (!RW_WRITE_HELD(&ds->ds_rwlock))
711 		rw_enter(&ds->ds_rwlock, RW_WRITER);
712 }
713 
714 void
715 dsl_dataset_create_root(dsl_pool_t *dp, uint64_t *ddobjp, dmu_tx_t *tx)
716 {
717 	objset_t *mos = dp->dp_meta_objset;
718 	dmu_buf_t *dbuf;
719 	dsl_dataset_phys_t *dsphys;
720 	dsl_dataset_t *ds;
721 	uint64_t dsobj;
722 	dsl_dir_t *dd;
723 
724 	dsl_dir_create_root(mos, ddobjp, tx);
725 	VERIFY(0 == dsl_dir_open_obj(dp, *ddobjp, NULL, FTAG, &dd));
726 
727 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
728 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
729 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
730 	dmu_buf_will_dirty(dbuf, tx);
731 	dsphys = dbuf->db_data;
732 	dsphys->ds_dir_obj = dd->dd_object;
733 	dsphys->ds_fsid_guid = unique_create();
734 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
735 	    sizeof (dsphys->ds_guid));
736 	dsphys->ds_snapnames_zapobj =
737 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
738 	    DMU_OT_NONE, 0, tx);
739 	dsphys->ds_creation_time = gethrestime_sec();
740 	dsphys->ds_creation_txg = tx->tx_txg;
741 	dsphys->ds_deadlist_obj =
742 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
743 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
744 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
745 	dmu_buf_rele(dbuf, FTAG);
746 
747 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
748 	dd->dd_phys->dd_head_dataset_obj = dsobj;
749 	dsl_dir_close(dd, FTAG);
750 
751 	VERIFY(0 == dsl_dataset_get_ref(dp, dsobj, FTAG, &ds));
752 	(void) dmu_objset_create_impl(dp->dp_spa, ds,
753 	    &ds->ds_phys->ds_bp, DMU_OST_ZFS, tx);
754 	dsl_dataset_drop_ref(ds, FTAG);
755 }
756 
757 uint64_t
758 dsl_dataset_create_sync_impl(dsl_dir_t *dd, dsl_dataset_t *origin,
759     uint64_t flags, dmu_tx_t *tx)
760 {
761 	dsl_pool_t *dp = dd->dd_pool;
762 	dmu_buf_t *dbuf;
763 	dsl_dataset_phys_t *dsphys;
764 	uint64_t dsobj;
765 	objset_t *mos = dp->dp_meta_objset;
766 
767 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
768 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
769 	ASSERT(dmu_tx_is_syncing(tx));
770 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
771 
772 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
773 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
774 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
775 	dmu_buf_will_dirty(dbuf, tx);
776 	dsphys = dbuf->db_data;
777 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
778 	dsphys->ds_dir_obj = dd->dd_object;
779 	dsphys->ds_flags = flags;
780 	dsphys->ds_fsid_guid = unique_create();
781 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
782 	    sizeof (dsphys->ds_guid));
783 	dsphys->ds_snapnames_zapobj =
784 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
785 	    DMU_OT_NONE, 0, tx);
786 	dsphys->ds_creation_time = gethrestime_sec();
787 	dsphys->ds_creation_txg = tx->tx_txg;
788 	dsphys->ds_deadlist_obj =
789 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
790 
791 	if (origin) {
792 		dsphys->ds_prev_snap_obj = origin->ds_object;
793 		dsphys->ds_prev_snap_txg =
794 		    origin->ds_phys->ds_creation_txg;
795 		dsphys->ds_used_bytes =
796 		    origin->ds_phys->ds_used_bytes;
797 		dsphys->ds_compressed_bytes =
798 		    origin->ds_phys->ds_compressed_bytes;
799 		dsphys->ds_uncompressed_bytes =
800 		    origin->ds_phys->ds_uncompressed_bytes;
801 		dsphys->ds_bp = origin->ds_phys->ds_bp;
802 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
803 
804 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
805 		origin->ds_phys->ds_num_children++;
806 
807 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
808 		dd->dd_phys->dd_origin_obj = origin->ds_object;
809 	}
810 
811 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
812 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
813 
814 	dmu_buf_rele(dbuf, FTAG);
815 
816 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
817 	dd->dd_phys->dd_head_dataset_obj = dsobj;
818 
819 	return (dsobj);
820 }
821 
822 uint64_t
823 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
824     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
825 {
826 	dsl_pool_t *dp = pdd->dd_pool;
827 	uint64_t dsobj, ddobj;
828 	dsl_dir_t *dd;
829 
830 	ASSERT(lastname[0] != '@');
831 
832 	ddobj = dsl_dir_create_sync(pdd, lastname, tx);
833 	VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
834 
835 	dsobj = dsl_dataset_create_sync_impl(dd, origin, flags, tx);
836 
837 	dsl_deleg_set_create_perms(dd, tx, cr);
838 
839 	dsl_dir_close(dd, FTAG);
840 
841 	return (dsobj);
842 }
843 
844 struct destroyarg {
845 	dsl_sync_task_group_t *dstg;
846 	char *snapname;
847 	char *failed;
848 };
849 
850 static int
851 dsl_snapshot_destroy_one(char *name, void *arg)
852 {
853 	struct destroyarg *da = arg;
854 	dsl_dataset_t *ds;
855 	char *cp;
856 	int err;
857 
858 	(void) strcat(name, "@");
859 	(void) strcat(name, da->snapname);
860 	err = dsl_dataset_own(name, DS_MODE_READONLY | DS_MODE_INCONSISTENT,
861 	    da->dstg, &ds);
862 	cp = strchr(name, '@');
863 	*cp = '\0';
864 	if (err == 0) {
865 		dsl_dataset_make_exclusive(ds, da->dstg);
866 		dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
867 		    dsl_dataset_destroy_sync, ds, da->dstg, 0);
868 	} else if (err == ENOENT) {
869 		err = 0;
870 	} else {
871 		(void) strcpy(da->failed, name);
872 	}
873 	return (err);
874 }
875 
876 /*
877  * Destroy 'snapname' in all descendants of 'fsname'.
878  */
879 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
880 int
881 dsl_snapshots_destroy(char *fsname, char *snapname)
882 {
883 	int err;
884 	struct destroyarg da;
885 	dsl_sync_task_t *dst;
886 	spa_t *spa;
887 
888 	err = spa_open(fsname, &spa, FTAG);
889 	if (err)
890 		return (err);
891 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
892 	da.snapname = snapname;
893 	da.failed = fsname;
894 
895 	err = dmu_objset_find(fsname,
896 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
897 
898 	if (err == 0)
899 		err = dsl_sync_task_group_wait(da.dstg);
900 
901 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
902 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
903 		dsl_dataset_t *ds = dst->dst_arg1;
904 		/*
905 		 * Return the file system name that triggered the error
906 		 */
907 		if (dst->dst_err) {
908 			dsl_dataset_name(ds, fsname);
909 			*strchr(fsname, '@') = '\0';
910 		}
911 		dsl_dataset_disown(ds, da.dstg);
912 	}
913 
914 	dsl_sync_task_group_destroy(da.dstg);
915 	spa_close(spa, FTAG);
916 	return (err);
917 }
918 
919 /*
920  * ds must be opened as OWNER.  On return (whether successful or not),
921  * ds will be closed and caller can no longer dereference it.
922  */
923 int
924 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
925 {
926 	int err;
927 	dsl_sync_task_group_t *dstg;
928 	objset_t *os;
929 	dsl_dir_t *dd;
930 	uint64_t obj;
931 
932 	if (dsl_dataset_is_snapshot(ds)) {
933 		/* Destroying a snapshot is simpler */
934 		dsl_dataset_make_exclusive(ds, tag);
935 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
936 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
937 		    ds, tag, 0);
938 		goto out;
939 	}
940 
941 	dd = ds->ds_dir;
942 
943 	/*
944 	 * Check for errors and mark this ds as inconsistent, in
945 	 * case we crash while freeing the objects.
946 	 */
947 	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
948 	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
949 	if (err)
950 		goto out;
951 
952 	err = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
953 	if (err)
954 		goto out;
955 
956 	/*
957 	 * remove the objects in open context, so that we won't
958 	 * have too much to do in syncing context.
959 	 */
960 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
961 	    ds->ds_phys->ds_prev_snap_txg)) {
962 		/*
963 		 * Ignore errors, if there is not enough disk space
964 		 * we will deal with it in dsl_dataset_destroy_sync().
965 		 */
966 		(void) dmu_free_object(os, obj);
967 	}
968 
969 	dmu_objset_close(os);
970 	if (err != ESRCH)
971 		goto out;
972 
973 	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
974 	err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
975 	rw_exit(&dd->dd_pool->dp_config_rwlock);
976 
977 	if (err)
978 		goto out;
979 
980 	if (ds->ds_user_ptr) {
981 		/*
982 		 * We need to sync out all in-flight IO before we try
983 		 * to evict (the dataset evict func is trying to clear
984 		 * the cached entries for this dataset in the ARC).
985 		 */
986 		txg_wait_synced(dd->dd_pool, 0);
987 	}
988 
989 	/*
990 	 * Blow away the dsl_dir + head dataset.
991 	 */
992 	dsl_dataset_make_exclusive(ds, tag);
993 	if (ds->ds_user_ptr) {
994 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
995 		ds->ds_user_ptr = NULL;
996 	}
997 	dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
998 	dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
999 	    dsl_dataset_destroy_sync, ds, tag, 0);
1000 	dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1001 	    dsl_dir_destroy_sync, dd, FTAG, 0);
1002 	err = dsl_sync_task_group_wait(dstg);
1003 	dsl_sync_task_group_destroy(dstg);
1004 	/* if it is successful, dsl_dir_destroy_sync will close the dd */
1005 	if (err)
1006 		dsl_dir_close(dd, FTAG);
1007 out:
1008 	dsl_dataset_disown(ds, tag);
1009 	return (err);
1010 }
1011 
1012 int
1013 dsl_dataset_rollback(dsl_dataset_t *ds, dmu_objset_type_t ost)
1014 {
1015 	ASSERT(ds->ds_owner);
1016 
1017 	return (dsl_sync_task_do(ds->ds_dir->dd_pool,
1018 	    dsl_dataset_rollback_check, dsl_dataset_rollback_sync,
1019 	    ds, &ost, 0));
1020 }
1021 
1022 void *
1023 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
1024     void *p, dsl_dataset_evict_func_t func)
1025 {
1026 	void *old;
1027 
1028 	mutex_enter(&ds->ds_lock);
1029 	old = ds->ds_user_ptr;
1030 	if (old == NULL) {
1031 		ds->ds_user_ptr = p;
1032 		ds->ds_user_evict_func = func;
1033 	}
1034 	mutex_exit(&ds->ds_lock);
1035 	return (old);
1036 }
1037 
1038 void *
1039 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
1040 {
1041 	return (ds->ds_user_ptr);
1042 }
1043 
1044 
1045 blkptr_t *
1046 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1047 {
1048 	return (&ds->ds_phys->ds_bp);
1049 }
1050 
1051 void
1052 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1053 {
1054 	ASSERT(dmu_tx_is_syncing(tx));
1055 	/* If it's the meta-objset, set dp_meta_rootbp */
1056 	if (ds == NULL) {
1057 		tx->tx_pool->dp_meta_rootbp = *bp;
1058 	} else {
1059 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1060 		ds->ds_phys->ds_bp = *bp;
1061 	}
1062 }
1063 
1064 spa_t *
1065 dsl_dataset_get_spa(dsl_dataset_t *ds)
1066 {
1067 	return (ds->ds_dir->dd_pool->dp_spa);
1068 }
1069 
1070 void
1071 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1072 {
1073 	dsl_pool_t *dp;
1074 
1075 	if (ds == NULL) /* this is the meta-objset */
1076 		return;
1077 
1078 	ASSERT(ds->ds_user_ptr != NULL);
1079 
1080 	if (ds->ds_phys->ds_next_snap_obj != 0)
1081 		panic("dirtying snapshot!");
1082 
1083 	dp = ds->ds_dir->dd_pool;
1084 
1085 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1086 		/* up the hold count until we can be written out */
1087 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1088 	}
1089 }
1090 
1091 /*
1092  * The unique space in the head dataset can be calculated by subtracting
1093  * the space used in the most recent snapshot, that is still being used
1094  * in this file system, from the space currently in use.  To figure out
1095  * the space in the most recent snapshot still in use, we need to take
1096  * the total space used in the snapshot and subtract out the space that
1097  * has been freed up since the snapshot was taken.
1098  */
1099 static void
1100 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1101 {
1102 	uint64_t mrs_used;
1103 	uint64_t dlused, dlcomp, dluncomp;
1104 
1105 	ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj);
1106 
1107 	if (ds->ds_phys->ds_prev_snap_obj != 0)
1108 		mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1109 	else
1110 		mrs_used = 0;
1111 
1112 	VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp,
1113 	    &dluncomp));
1114 
1115 	ASSERT3U(dlused, <=, mrs_used);
1116 	ds->ds_phys->ds_unique_bytes =
1117 	    ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1118 
1119 	if (!DS_UNIQUE_IS_ACCURATE(ds) &&
1120 	    spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1121 	    SPA_VERSION_UNIQUE_ACCURATE)
1122 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1123 }
1124 
1125 static uint64_t
1126 dsl_dataset_unique(dsl_dataset_t *ds)
1127 {
1128 	if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds))
1129 		dsl_dataset_recalc_head_uniq(ds);
1130 
1131 	return (ds->ds_phys->ds_unique_bytes);
1132 }
1133 
1134 struct killarg {
1135 	int64_t *usedp;
1136 	int64_t *compressedp;
1137 	int64_t *uncompressedp;
1138 	zio_t *zio;
1139 	dmu_tx_t *tx;
1140 };
1141 
1142 static int
1143 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
1144 {
1145 	struct killarg *ka = arg;
1146 	blkptr_t *bp = &bc->bc_blkptr;
1147 
1148 	ASSERT3U(bc->bc_errno, ==, 0);
1149 
1150 	/*
1151 	 * Since this callback is not called concurrently, no lock is
1152 	 * needed on the accounting values.
1153 	 */
1154 	*ka->usedp += bp_get_dasize(spa, bp);
1155 	*ka->compressedp += BP_GET_PSIZE(bp);
1156 	*ka->uncompressedp += BP_GET_UCSIZE(bp);
1157 	/* XXX check for EIO? */
1158 	(void) arc_free(ka->zio, spa, ka->tx->tx_txg, bp, NULL, NULL,
1159 	    ARC_NOWAIT);
1160 	return (0);
1161 }
1162 
1163 /* ARGSUSED */
1164 static int
1165 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
1166 {
1167 	dsl_dataset_t *ds = arg1;
1168 	dmu_objset_type_t *ost = arg2;
1169 
1170 	/*
1171 	 * We can only roll back to emptyness if it is a ZPL objset.
1172 	 */
1173 	if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0)
1174 		return (EINVAL);
1175 
1176 	/*
1177 	 * This must not be a snapshot.
1178 	 */
1179 	if (ds->ds_phys->ds_next_snap_obj != 0)
1180 		return (EINVAL);
1181 
1182 	/*
1183 	 * If we made changes this txg, traverse_dsl_dataset won't find
1184 	 * them.  Try again.
1185 	 */
1186 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1187 		return (EAGAIN);
1188 
1189 	return (0);
1190 }
1191 
1192 /* ARGSUSED */
1193 static void
1194 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1195 {
1196 	dsl_dataset_t *ds = arg1;
1197 	dmu_objset_type_t *ost = arg2;
1198 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1199 
1200 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1201 
1202 	/*
1203 	 * Before the roll back destroy the zil.
1204 	 */
1205 	if (ds->ds_user_ptr != NULL) {
1206 		zil_rollback_destroy(
1207 		    ((objset_impl_t *)ds->ds_user_ptr)->os_zil, tx);
1208 
1209 		/*
1210 		 * We need to make sure that the objset_impl_t is reopened after
1211 		 * we do the rollback, otherwise it will have the wrong
1212 		 * objset_phys_t.  Normally this would happen when this
1213 		 * dataset-open is closed, thus causing the
1214 		 * dataset to be immediately evicted.  But when doing "zfs recv
1215 		 * -F", we reopen the objset before that, so that there is no
1216 		 * window where the dataset is closed and inconsistent.
1217 		 */
1218 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1219 		ds->ds_user_ptr = NULL;
1220 	}
1221 
1222 	/* Zero out the deadlist. */
1223 	bplist_close(&ds->ds_deadlist);
1224 	bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1225 	ds->ds_phys->ds_deadlist_obj =
1226 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1227 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1228 	    ds->ds_phys->ds_deadlist_obj));
1229 
1230 	{
1231 		/* Free blkptrs that we gave birth to */
1232 		zio_t *zio;
1233 		int64_t used = 0, compressed = 0, uncompressed = 0;
1234 		struct killarg ka;
1235 		int64_t delta;
1236 
1237 		zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
1238 		    ZIO_FLAG_MUSTSUCCEED);
1239 		ka.usedp = &used;
1240 		ka.compressedp = &compressed;
1241 		ka.uncompressedp = &uncompressed;
1242 		ka.zio = zio;
1243 		ka.tx = tx;
1244 		(void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1245 		    ADVANCE_POST, kill_blkptr, &ka);
1246 		(void) zio_wait(zio);
1247 
1248 		/* only deduct space beyond any refreservation */
1249 		delta = parent_delta(ds, -used);
1250 		dsl_dir_diduse_space(ds->ds_dir,
1251 		    delta, -compressed, -uncompressed, tx);
1252 	}
1253 
1254 	if (ds->ds_prev) {
1255 		/* Change our contents to that of the prev snapshot */
1256 		ASSERT3U(ds->ds_prev->ds_object, ==,
1257 		    ds->ds_phys->ds_prev_snap_obj);
1258 		ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1259 		ds->ds_phys->ds_used_bytes =
1260 		    ds->ds_prev->ds_phys->ds_used_bytes;
1261 		ds->ds_phys->ds_compressed_bytes =
1262 		    ds->ds_prev->ds_phys->ds_compressed_bytes;
1263 		ds->ds_phys->ds_uncompressed_bytes =
1264 		    ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1265 		ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1266 		ds->ds_phys->ds_unique_bytes = 0;
1267 
1268 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1269 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1270 			ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1271 		}
1272 	} else {
1273 		/* Zero out our contents, recreate objset */
1274 		bzero(&ds->ds_phys->ds_bp, sizeof (blkptr_t));
1275 		ds->ds_phys->ds_used_bytes = 0;
1276 		ds->ds_phys->ds_compressed_bytes = 0;
1277 		ds->ds_phys->ds_uncompressed_bytes = 0;
1278 		ds->ds_phys->ds_flags = 0;
1279 		ds->ds_phys->ds_unique_bytes = 0;
1280 		(void) dmu_objset_create_impl(ds->ds_dir->dd_pool->dp_spa, ds,
1281 		    &ds->ds_phys->ds_bp, *ost, tx);
1282 	}
1283 
1284 	spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa,
1285 	    tx, cr, "dataset = %llu", ds->ds_object);
1286 }
1287 
1288 /* ARGSUSED */
1289 static int
1290 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1291 {
1292 	dsl_dataset_t *ds = arg1;
1293 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1294 	uint64_t count;
1295 	int err;
1296 
1297 	/*
1298 	 * Can't delete a head dataset if there are snapshots of it.
1299 	 * (Except if the only snapshots are from the branch we cloned
1300 	 * from.)
1301 	 */
1302 	if (ds->ds_prev != NULL &&
1303 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1304 		return (EINVAL);
1305 
1306 	/*
1307 	 * This is really a dsl_dir thing, but check it here so that
1308 	 * we'll be less likely to leave this dataset inconsistent &
1309 	 * nearly destroyed.
1310 	 */
1311 	err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1312 	if (err)
1313 		return (err);
1314 	if (count != 0)
1315 		return (EEXIST);
1316 
1317 	return (0);
1318 }
1319 
1320 /* ARGSUSED */
1321 static void
1322 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1323 {
1324 	dsl_dataset_t *ds = arg1;
1325 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1326 
1327 	/* Mark it as inconsistent on-disk, in case we crash */
1328 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1329 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1330 
1331 	spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1332 	    cr, "dataset = %llu", ds->ds_object);
1333 }
1334 
1335 /* ARGSUSED */
1336 int
1337 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1338 {
1339 	dsl_dataset_t *ds = arg1;
1340 
1341 	/* we have an owner hold, so noone else can destroy us */
1342 	ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1343 
1344 	/* Can't delete a branch point. */
1345 	if (ds->ds_phys->ds_num_children > 1)
1346 		return (EEXIST);
1347 
1348 	/*
1349 	 * Can't delete a head dataset if there are snapshots of it.
1350 	 * (Except if the only snapshots are from the branch we cloned
1351 	 * from.)
1352 	 */
1353 	if (ds->ds_prev != NULL &&
1354 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1355 		return (EINVAL);
1356 
1357 	/*
1358 	 * If we made changes this txg, traverse_dsl_dataset won't find
1359 	 * them.  Try again.
1360 	 */
1361 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1362 		return (EAGAIN);
1363 
1364 	/* XXX we should do some i/o error checking... */
1365 	return (0);
1366 }
1367 
1368 struct refsarg {
1369 	kmutex_t lock;
1370 	boolean_t gone;
1371 	kcondvar_t cv;
1372 };
1373 
1374 /* ARGSUSED */
1375 static void
1376 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1377 {
1378 	struct refsarg *arg = argv;
1379 
1380 	mutex_enter(&arg->lock);
1381 	arg->gone = TRUE;
1382 	cv_signal(&arg->cv);
1383 	mutex_exit(&arg->lock);
1384 }
1385 
1386 static void
1387 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1388 {
1389 	struct refsarg arg;
1390 
1391 	mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1392 	cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1393 	arg.gone = FALSE;
1394 	(void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1395 	    dsl_dataset_refs_gone);
1396 	dmu_buf_rele(ds->ds_dbuf, tag);
1397 	mutex_enter(&arg.lock);
1398 	while (!arg.gone)
1399 		cv_wait(&arg.cv, &arg.lock);
1400 	ASSERT(arg.gone);
1401 	mutex_exit(&arg.lock);
1402 	ds->ds_dbuf = NULL;
1403 	ds->ds_phys = NULL;
1404 	mutex_destroy(&arg.lock);
1405 	cv_destroy(&arg.cv);
1406 }
1407 
1408 void
1409 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1410 {
1411 	dsl_dataset_t *ds = arg1;
1412 	int64_t used = 0, compressed = 0, uncompressed = 0;
1413 	zio_t *zio;
1414 	int err;
1415 	int after_branch_point = FALSE;
1416 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1417 	objset_t *mos = dp->dp_meta_objset;
1418 	dsl_dataset_t *ds_prev = NULL;
1419 	uint64_t obj;
1420 
1421 	ASSERT(ds->ds_owner);
1422 	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1423 	ASSERT(ds->ds_prev == NULL ||
1424 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1425 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1426 
1427 	/* signal any waiters that this dataset is going away */
1428 	mutex_enter(&ds->ds_lock);
1429 	ds->ds_owner = dsl_reaper;
1430 	cv_broadcast(&ds->ds_exclusive_cv);
1431 	mutex_exit(&ds->ds_lock);
1432 
1433 	/* Remove our reservation */
1434 	if (ds->ds_reserved != 0) {
1435 		uint64_t val = 0;
1436 		dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1437 		ASSERT3U(ds->ds_reserved, ==, 0);
1438 	}
1439 
1440 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1441 
1442 	obj = ds->ds_object;
1443 
1444 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1445 		if (ds->ds_prev) {
1446 			ds_prev = ds->ds_prev;
1447 		} else {
1448 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1449 			    ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1450 		}
1451 		after_branch_point =
1452 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1453 
1454 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1455 		if (after_branch_point &&
1456 		    ds->ds_phys->ds_next_snap_obj == 0) {
1457 			/* This clone is toast. */
1458 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1459 			ds_prev->ds_phys->ds_num_children--;
1460 		} else if (!after_branch_point) {
1461 			ds_prev->ds_phys->ds_next_snap_obj =
1462 			    ds->ds_phys->ds_next_snap_obj;
1463 		}
1464 	}
1465 
1466 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1467 
1468 	if (ds->ds_phys->ds_next_snap_obj != 0) {
1469 		blkptr_t bp;
1470 		dsl_dataset_t *ds_next;
1471 		uint64_t itor = 0;
1472 		uint64_t old_unique;
1473 
1474 		spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1475 
1476 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1477 		    ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1478 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1479 
1480 		old_unique = dsl_dataset_unique(ds_next);
1481 
1482 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1483 		ds_next->ds_phys->ds_prev_snap_obj =
1484 		    ds->ds_phys->ds_prev_snap_obj;
1485 		ds_next->ds_phys->ds_prev_snap_txg =
1486 		    ds->ds_phys->ds_prev_snap_txg;
1487 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1488 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1489 
1490 		/*
1491 		 * Transfer to our deadlist (which will become next's
1492 		 * new deadlist) any entries from next's current
1493 		 * deadlist which were born before prev, and free the
1494 		 * other entries.
1495 		 *
1496 		 * XXX we're doing this long task with the config lock held
1497 		 */
1498 		while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) {
1499 			if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1500 				VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1501 				    &bp, tx));
1502 				if (ds_prev && !after_branch_point &&
1503 				    bp.blk_birth >
1504 				    ds_prev->ds_phys->ds_prev_snap_txg) {
1505 					ds_prev->ds_phys->ds_unique_bytes +=
1506 					    bp_get_dasize(dp->dp_spa, &bp);
1507 				}
1508 			} else {
1509 				used += bp_get_dasize(dp->dp_spa, &bp);
1510 				compressed += BP_GET_PSIZE(&bp);
1511 				uncompressed += BP_GET_UCSIZE(&bp);
1512 				/* XXX check return value? */
1513 				(void) arc_free(zio, dp->dp_spa, tx->tx_txg,
1514 				    &bp, NULL, NULL, ARC_NOWAIT);
1515 			}
1516 		}
1517 
1518 		/* free next's deadlist */
1519 		bplist_close(&ds_next->ds_deadlist);
1520 		bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1521 
1522 		/* set next's deadlist to our deadlist */
1523 		bplist_close(&ds->ds_deadlist);
1524 		ds_next->ds_phys->ds_deadlist_obj =
1525 		    ds->ds_phys->ds_deadlist_obj;
1526 		VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1527 		    ds_next->ds_phys->ds_deadlist_obj));
1528 		ds->ds_phys->ds_deadlist_obj = 0;
1529 
1530 		if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1531 			/*
1532 			 * Update next's unique to include blocks which
1533 			 * were previously shared by only this snapshot
1534 			 * and it.  Those blocks will be born after the
1535 			 * prev snap and before this snap, and will have
1536 			 * died after the next snap and before the one
1537 			 * after that (ie. be on the snap after next's
1538 			 * deadlist).
1539 			 *
1540 			 * XXX we're doing this long task with the
1541 			 * config lock held
1542 			 */
1543 			dsl_dataset_t *ds_after_next;
1544 
1545 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1546 			    ds_next->ds_phys->ds_next_snap_obj,
1547 			    FTAG, &ds_after_next));
1548 			itor = 0;
1549 			while (bplist_iterate(&ds_after_next->ds_deadlist,
1550 			    &itor, &bp) == 0) {
1551 				if (bp.blk_birth >
1552 				    ds->ds_phys->ds_prev_snap_txg &&
1553 				    bp.blk_birth <=
1554 				    ds->ds_phys->ds_creation_txg) {
1555 					ds_next->ds_phys->ds_unique_bytes +=
1556 					    bp_get_dasize(dp->dp_spa, &bp);
1557 				}
1558 			}
1559 
1560 			dsl_dataset_rele(ds_after_next, FTAG);
1561 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1562 		} else {
1563 			ASSERT3P(ds_next->ds_prev, ==, ds);
1564 			dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1565 			ds_next->ds_prev = NULL;
1566 			if (ds_prev) {
1567 				VERIFY(0 == dsl_dataset_get_ref(dp,
1568 				    ds->ds_phys->ds_prev_snap_obj,
1569 				    ds_next, &ds_next->ds_prev));
1570 			}
1571 
1572 			dsl_dataset_recalc_head_uniq(ds_next);
1573 
1574 			/*
1575 			 * Reduce the amount of our unconsmed refreservation
1576 			 * being charged to our parent by the amount of
1577 			 * new unique data we have gained.
1578 			 */
1579 			if (old_unique < ds_next->ds_reserved) {
1580 				int64_t mrsdelta;
1581 				uint64_t new_unique =
1582 				    ds_next->ds_phys->ds_unique_bytes;
1583 
1584 				ASSERT(old_unique <= new_unique);
1585 				mrsdelta = MIN(new_unique - old_unique,
1586 				    ds_next->ds_reserved - old_unique);
1587 				dsl_dir_diduse_space(ds->ds_dir, -mrsdelta,
1588 				    0, 0, tx);
1589 			}
1590 		}
1591 		dsl_dataset_rele(ds_next, FTAG);
1592 
1593 		/*
1594 		 * NB: unique_bytes might not be accurate for the head objset.
1595 		 * Before SPA_VERSION 9, we didn't update its value when we
1596 		 * deleted the most recent snapshot.
1597 		 */
1598 		ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1599 	} else {
1600 		/*
1601 		 * There's no next snapshot, so this is a head dataset.
1602 		 * Destroy the deadlist.  Unless it's a clone, the
1603 		 * deadlist should be empty.  (If it's a clone, it's
1604 		 * safe to ignore the deadlist contents.)
1605 		 */
1606 		struct killarg ka;
1607 
1608 		ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1609 		bplist_close(&ds->ds_deadlist);
1610 		bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1611 		ds->ds_phys->ds_deadlist_obj = 0;
1612 
1613 		/*
1614 		 * Free everything that we point to (that's born after
1615 		 * the previous snapshot, if we are a clone)
1616 		 *
1617 		 * XXX we're doing this long task with the config lock held
1618 		 */
1619 		ka.usedp = &used;
1620 		ka.compressedp = &compressed;
1621 		ka.uncompressedp = &uncompressed;
1622 		ka.zio = zio;
1623 		ka.tx = tx;
1624 		err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1625 		    ADVANCE_POST, kill_blkptr, &ka);
1626 		ASSERT3U(err, ==, 0);
1627 		ASSERT(spa_version(dp->dp_spa) <
1628 		    SPA_VERSION_UNIQUE_ACCURATE ||
1629 		    used == ds->ds_phys->ds_unique_bytes);
1630 	}
1631 
1632 	err = zio_wait(zio);
1633 	ASSERT3U(err, ==, 0);
1634 
1635 	dsl_dir_diduse_space(ds->ds_dir, -used, -compressed, -uncompressed, tx);
1636 
1637 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1638 		/* Erase the link in the dir */
1639 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1640 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1641 		ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1642 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1643 		ASSERT(err == 0);
1644 	} else {
1645 		/* remove from snapshot namespace */
1646 		dsl_dataset_t *ds_head;
1647 		ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1648 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1649 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1650 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1651 #ifdef ZFS_DEBUG
1652 		{
1653 			uint64_t val;
1654 
1655 			err = dsl_dataset_snap_lookup(ds_head,
1656 			    ds->ds_snapname, &val);
1657 			ASSERT3U(err, ==, 0);
1658 			ASSERT3U(val, ==, obj);
1659 		}
1660 #endif
1661 		err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1662 		ASSERT(err == 0);
1663 		dsl_dataset_rele(ds_head, FTAG);
1664 	}
1665 
1666 	if (ds_prev && ds->ds_prev != ds_prev)
1667 		dsl_dataset_rele(ds_prev, FTAG);
1668 
1669 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1670 	spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1671 	    cr, "dataset = %llu", ds->ds_object);
1672 
1673 	dsl_dir_close(ds->ds_dir, ds);
1674 	ds->ds_dir = NULL;
1675 	dsl_dataset_drain_refs(ds, tag);
1676 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1677 }
1678 
1679 static int
1680 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1681 {
1682 	uint64_t asize;
1683 
1684 	if (!dmu_tx_is_syncing(tx))
1685 		return (0);
1686 
1687 	/*
1688 	 * If there's an fs-only reservation, any blocks that might become
1689 	 * owned by the snapshot dataset must be accommodated by space
1690 	 * outside of the reservation.
1691 	 */
1692 	asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1693 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1694 		return (ENOSPC);
1695 
1696 	/*
1697 	 * Propogate any reserved space for this snapshot to other
1698 	 * snapshot checks in this sync group.
1699 	 */
1700 	if (asize > 0)
1701 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1702 
1703 	return (0);
1704 }
1705 
1706 /* ARGSUSED */
1707 int
1708 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1709 {
1710 	dsl_dataset_t *ds = arg1;
1711 	const char *snapname = arg2;
1712 	int err;
1713 	uint64_t value;
1714 
1715 	/*
1716 	 * We don't allow multiple snapshots of the same txg.  If there
1717 	 * is already one, try again.
1718 	 */
1719 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1720 		return (EAGAIN);
1721 
1722 	/*
1723 	 * Check for conflicting name snapshot name.
1724 	 */
1725 	err = dsl_dataset_snap_lookup(ds, snapname, &value);
1726 	if (err == 0)
1727 		return (EEXIST);
1728 	if (err != ENOENT)
1729 		return (err);
1730 
1731 	/*
1732 	 * Check that the dataset's name is not too long.  Name consists
1733 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1734 	 */
1735 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1736 		return (ENAMETOOLONG);
1737 
1738 	err = dsl_dataset_snapshot_reserve_space(ds, tx);
1739 	if (err)
1740 		return (err);
1741 
1742 	ds->ds_trysnap_txg = tx->tx_txg;
1743 	return (0);
1744 }
1745 
1746 void
1747 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1748 {
1749 	dsl_dataset_t *ds = arg1;
1750 	const char *snapname = arg2;
1751 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1752 	dmu_buf_t *dbuf;
1753 	dsl_dataset_phys_t *dsphys;
1754 	uint64_t dsobj;
1755 	objset_t *mos = dp->dp_meta_objset;
1756 	int err;
1757 
1758 	spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1759 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1760 
1761 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1762 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1763 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1764 	dmu_buf_will_dirty(dbuf, tx);
1765 	dsphys = dbuf->db_data;
1766 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1767 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1768 	dsphys->ds_fsid_guid = unique_create();
1769 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1770 	    sizeof (dsphys->ds_guid));
1771 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1772 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1773 	dsphys->ds_next_snap_obj = ds->ds_object;
1774 	dsphys->ds_num_children = 1;
1775 	dsphys->ds_creation_time = gethrestime_sec();
1776 	dsphys->ds_creation_txg = tx->tx_txg;
1777 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1778 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1779 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1780 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1781 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1782 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1783 	dmu_buf_rele(dbuf, FTAG);
1784 
1785 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1786 	if (ds->ds_prev) {
1787 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1788 		    ds->ds_object ||
1789 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1790 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1791 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1792 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1793 			    ds->ds_prev->ds_phys->ds_creation_txg);
1794 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1795 		}
1796 	}
1797 
1798 	/*
1799 	 * If we have a reference-reservation on this dataset, we will
1800 	 * need to increase the amount of refreservation being charged
1801 	 * since our unique space is going to zero.
1802 	 */
1803 	if (ds->ds_reserved) {
1804 		int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1805 		dsl_dir_diduse_space(ds->ds_dir, add, 0, 0, tx);
1806 	}
1807 
1808 	bplist_close(&ds->ds_deadlist);
1809 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1810 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1811 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1812 	ds->ds_phys->ds_prev_snap_txg = tx->tx_txg;
1813 	ds->ds_phys->ds_unique_bytes = 0;
1814 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1815 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1816 	ds->ds_phys->ds_deadlist_obj =
1817 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1818 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1819 	    ds->ds_phys->ds_deadlist_obj));
1820 
1821 	dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1822 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1823 	    snapname, 8, 1, &dsobj, tx);
1824 	ASSERT(err == 0);
1825 
1826 	if (ds->ds_prev)
1827 		dsl_dataset_drop_ref(ds->ds_prev, ds);
1828 	VERIFY(0 == dsl_dataset_get_ref(dp,
1829 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1830 
1831 	spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1832 	    "dataset = %llu", dsobj);
1833 }
1834 
1835 void
1836 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1837 {
1838 	ASSERT(dmu_tx_is_syncing(tx));
1839 	ASSERT(ds->ds_user_ptr != NULL);
1840 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1841 
1842 	/*
1843 	 * in case we had to change ds_fsid_guid when we opened it,
1844 	 * sync it out now.
1845 	 */
1846 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1847 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1848 
1849 	dsl_dir_dirty(ds->ds_dir, tx);
1850 	dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1851 }
1852 
1853 void
1854 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1855 {
1856 	uint64_t refd, avail, uobjs, aobjs;
1857 
1858 	dsl_dir_stats(ds->ds_dir, nv);
1859 
1860 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1861 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1862 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1863 
1864 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1865 	    ds->ds_phys->ds_creation_time);
1866 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1867 	    ds->ds_phys->ds_creation_txg);
1868 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1869 	    ds->ds_quota);
1870 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1871 	    ds->ds_reserved);
1872 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1873 	    ds->ds_phys->ds_guid);
1874 
1875 	if (ds->ds_phys->ds_next_snap_obj) {
1876 		/*
1877 		 * This is a snapshot; override the dd's space used with
1878 		 * our unique space and compression ratio.
1879 		 */
1880 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1881 		    ds->ds_phys->ds_unique_bytes);
1882 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1883 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1884 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1885 		    ds->ds_phys->ds_compressed_bytes));
1886 	}
1887 }
1888 
1889 void
1890 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1891 {
1892 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1893 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1894 	stat->dds_guid = ds->ds_phys->ds_guid;
1895 	if (ds->ds_phys->ds_next_snap_obj) {
1896 		stat->dds_is_snapshot = B_TRUE;
1897 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1898 	}
1899 
1900 	/* clone origin is really a dsl_dir thing... */
1901 	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1902 	if (ds->ds_dir->dd_phys->dd_origin_obj) {
1903 		dsl_dataset_t *ods;
1904 
1905 		VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
1906 		    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1907 		dsl_dataset_name(ods, stat->dds_origin);
1908 		dsl_dataset_drop_ref(ods, FTAG);
1909 	}
1910 	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1911 }
1912 
1913 uint64_t
1914 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1915 {
1916 	return (ds->ds_fsid_guid);
1917 }
1918 
1919 void
1920 dsl_dataset_space(dsl_dataset_t *ds,
1921     uint64_t *refdbytesp, uint64_t *availbytesp,
1922     uint64_t *usedobjsp, uint64_t *availobjsp)
1923 {
1924 	*refdbytesp = ds->ds_phys->ds_used_bytes;
1925 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1926 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1927 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1928 	if (ds->ds_quota != 0) {
1929 		/*
1930 		 * Adjust available bytes according to refquota
1931 		 */
1932 		if (*refdbytesp < ds->ds_quota)
1933 			*availbytesp = MIN(*availbytesp,
1934 			    ds->ds_quota - *refdbytesp);
1935 		else
1936 			*availbytesp = 0;
1937 	}
1938 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1939 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1940 }
1941 
1942 boolean_t
1943 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
1944 {
1945 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1946 
1947 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
1948 	    dsl_pool_sync_context(dp));
1949 	if (ds->ds_prev == NULL)
1950 		return (B_FALSE);
1951 	if (ds->ds_phys->ds_bp.blk_birth >
1952 	    ds->ds_prev->ds_phys->ds_creation_txg)
1953 		return (B_TRUE);
1954 	return (B_FALSE);
1955 }
1956 
1957 /* ARGSUSED */
1958 static int
1959 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1960 {
1961 	dsl_dataset_t *ds = arg1;
1962 	char *newsnapname = arg2;
1963 	dsl_dir_t *dd = ds->ds_dir;
1964 	dsl_dataset_t *hds;
1965 	uint64_t val;
1966 	int err;
1967 
1968 	err = dsl_dataset_hold_obj(dd->dd_pool,
1969 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
1970 	if (err)
1971 		return (err);
1972 
1973 	/* new name better not be in use */
1974 	err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
1975 	dsl_dataset_rele(hds, FTAG);
1976 
1977 	if (err == 0)
1978 		err = EEXIST;
1979 	else if (err == ENOENT)
1980 		err = 0;
1981 
1982 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1983 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
1984 		err = ENAMETOOLONG;
1985 
1986 	return (err);
1987 }
1988 
1989 static void
1990 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
1991     cred_t *cr, dmu_tx_t *tx)
1992 {
1993 	dsl_dataset_t *ds = arg1;
1994 	const char *newsnapname = arg2;
1995 	dsl_dir_t *dd = ds->ds_dir;
1996 	objset_t *mos = dd->dd_pool->dp_meta_objset;
1997 	dsl_dataset_t *hds;
1998 	int err;
1999 
2000 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2001 
2002 	VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2003 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2004 
2005 	VERIFY(0 == dsl_dataset_get_snapname(ds));
2006 	err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2007 	ASSERT3U(err, ==, 0);
2008 	mutex_enter(&ds->ds_lock);
2009 	(void) strcpy(ds->ds_snapname, newsnapname);
2010 	mutex_exit(&ds->ds_lock);
2011 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2012 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2013 	ASSERT3U(err, ==, 0);
2014 
2015 	spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2016 	    cr, "dataset = %llu", ds->ds_object);
2017 	dsl_dataset_rele(hds, FTAG);
2018 }
2019 
2020 struct renamesnaparg {
2021 	dsl_sync_task_group_t *dstg;
2022 	char failed[MAXPATHLEN];
2023 	char *oldsnap;
2024 	char *newsnap;
2025 };
2026 
2027 static int
2028 dsl_snapshot_rename_one(char *name, void *arg)
2029 {
2030 	struct renamesnaparg *ra = arg;
2031 	dsl_dataset_t *ds = NULL;
2032 	char *cp;
2033 	int err;
2034 
2035 	cp = name + strlen(name);
2036 	*cp = '@';
2037 	(void) strcpy(cp + 1, ra->oldsnap);
2038 
2039 	/*
2040 	 * For recursive snapshot renames the parent won't be changing
2041 	 * so we just pass name for both the to/from argument.
2042 	 */
2043 	if (err = zfs_secpolicy_rename_perms(name, name, CRED())) {
2044 		(void) strcpy(ra->failed, name);
2045 		return (err);
2046 	}
2047 
2048 #ifdef _KERNEL
2049 	/*
2050 	 * For all filesystems undergoing rename, we'll need to unmount it.
2051 	 */
2052 	(void) zfs_unmount_snap(name, NULL);
2053 #endif
2054 	err = dsl_dataset_hold(name, ra->dstg, &ds);
2055 	*cp = '\0';
2056 	if (err == ENOENT) {
2057 		return (0);
2058 	} else if (err) {
2059 		(void) strcpy(ra->failed, name);
2060 		return (err);
2061 	}
2062 
2063 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2064 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2065 
2066 	return (0);
2067 }
2068 
2069 static int
2070 dsl_recursive_rename(char *oldname, const char *newname)
2071 {
2072 	int err;
2073 	struct renamesnaparg *ra;
2074 	dsl_sync_task_t *dst;
2075 	spa_t *spa;
2076 	char *cp, *fsname = spa_strdup(oldname);
2077 	int len = strlen(oldname);
2078 
2079 	/* truncate the snapshot name to get the fsname */
2080 	cp = strchr(fsname, '@');
2081 	*cp = '\0';
2082 
2083 	err = spa_open(fsname, &spa, FTAG);
2084 	if (err) {
2085 		kmem_free(fsname, len + 1);
2086 		return (err);
2087 	}
2088 	ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2089 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2090 
2091 	ra->oldsnap = strchr(oldname, '@') + 1;
2092 	ra->newsnap = strchr(newname, '@') + 1;
2093 	*ra->failed = '\0';
2094 
2095 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2096 	    DS_FIND_CHILDREN);
2097 	kmem_free(fsname, len + 1);
2098 
2099 	if (err == 0) {
2100 		err = dsl_sync_task_group_wait(ra->dstg);
2101 	}
2102 
2103 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2104 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2105 		dsl_dataset_t *ds = dst->dst_arg1;
2106 		if (dst->dst_err) {
2107 			dsl_dir_name(ds->ds_dir, ra->failed);
2108 			(void) strcat(ra->failed, "@");
2109 			(void) strcat(ra->failed, ra->newsnap);
2110 		}
2111 		dsl_dataset_rele(ds, ra->dstg);
2112 	}
2113 
2114 	if (err)
2115 		(void) strcpy(oldname, ra->failed);
2116 
2117 	dsl_sync_task_group_destroy(ra->dstg);
2118 	kmem_free(ra, sizeof (struct renamesnaparg));
2119 	spa_close(spa, FTAG);
2120 	return (err);
2121 }
2122 
2123 static int
2124 dsl_valid_rename(char *oldname, void *arg)
2125 {
2126 	int delta = *(int *)arg;
2127 
2128 	if (strlen(oldname) + delta >= MAXNAMELEN)
2129 		return (ENAMETOOLONG);
2130 
2131 	return (0);
2132 }
2133 
2134 #pragma weak dmu_objset_rename = dsl_dataset_rename
2135 int
2136 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2137 {
2138 	dsl_dir_t *dd;
2139 	dsl_dataset_t *ds;
2140 	const char *tail;
2141 	int err;
2142 
2143 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2144 	if (err)
2145 		return (err);
2146 	if (tail == NULL) {
2147 		int delta = strlen(newname) - strlen(oldname);
2148 
2149 		/* if we're growing, validate child size lengths */
2150 		if (delta > 0)
2151 			err = dmu_objset_find(oldname, dsl_valid_rename,
2152 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2153 
2154 		if (!err)
2155 			err = dsl_dir_rename(dd, newname);
2156 		dsl_dir_close(dd, FTAG);
2157 		return (err);
2158 	}
2159 	if (tail[0] != '@') {
2160 		/* the name ended in a nonexistant component */
2161 		dsl_dir_close(dd, FTAG);
2162 		return (ENOENT);
2163 	}
2164 
2165 	dsl_dir_close(dd, FTAG);
2166 
2167 	/* new name must be snapshot in same filesystem */
2168 	tail = strchr(newname, '@');
2169 	if (tail == NULL)
2170 		return (EINVAL);
2171 	tail++;
2172 	if (strncmp(oldname, newname, tail - newname) != 0)
2173 		return (EXDEV);
2174 
2175 	if (recursive) {
2176 		err = dsl_recursive_rename(oldname, newname);
2177 	} else {
2178 		err = dsl_dataset_hold(oldname, FTAG, &ds);
2179 		if (err)
2180 			return (err);
2181 
2182 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2183 		    dsl_dataset_snapshot_rename_check,
2184 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2185 
2186 		dsl_dataset_rele(ds, FTAG);
2187 	}
2188 
2189 	return (err);
2190 }
2191 
2192 struct promotedsarg {
2193 	list_node_t link;
2194 	dsl_dataset_t *ds;
2195 };
2196 
2197 struct promotearg {
2198 	list_t snap_list;
2199 	dsl_dataset_t *clone_origin, *old_head;
2200 	uint64_t used, comp, uncomp, unique;
2201 	uint64_t newnext_obj;
2202 };
2203 
2204 /* ARGSUSED */
2205 static int
2206 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2207 {
2208 	dsl_dataset_t *hds = arg1;
2209 	struct promotearg *pa = arg2;
2210 	struct promotedsarg *snap = list_head(&pa->snap_list);
2211 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2212 	dsl_dataset_t *origin_ds = snap->ds;
2213 	dsl_dataset_t *newnext_ds;
2214 	char *name;
2215 	uint64_t itor = 0;
2216 	blkptr_t bp;
2217 	int err;
2218 
2219 	/* Check that it is a clone */
2220 	if (hds->ds_dir->dd_phys->dd_origin_obj == 0)
2221 		return (EINVAL);
2222 
2223 	/* Since this is so expensive, don't do the preliminary check */
2224 	if (!dmu_tx_is_syncing(tx))
2225 		return (0);
2226 
2227 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2228 		return (EXDEV);
2229 
2230 	/* find origin's new next ds */
2231 	newnext_ds = hds;
2232 	while (newnext_ds->ds_phys->ds_prev_snap_obj != origin_ds->ds_object) {
2233 		dsl_dataset_t *prev;
2234 
2235 		err = dsl_dataset_hold_obj(dp,
2236 		    newnext_ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2237 		if (newnext_ds != hds)
2238 			dsl_dataset_rele(newnext_ds, FTAG);
2239 		if (err)
2240 			return (err);
2241 		newnext_ds = prev;
2242 	}
2243 	pa->newnext_obj = newnext_ds->ds_object;
2244 
2245 	/* compute origin's new unique space */
2246 	pa->unique = 0;
2247 	while ((err = bplist_iterate(&newnext_ds->ds_deadlist,
2248 	    &itor, &bp)) == 0) {
2249 		if (bp.blk_birth > origin_ds->ds_phys->ds_prev_snap_txg)
2250 			pa->unique += bp_get_dasize(dp->dp_spa, &bp);
2251 	}
2252 	if (newnext_ds != hds)
2253 		dsl_dataset_rele(newnext_ds, FTAG);
2254 	if (err != ENOENT)
2255 		return (err);
2256 
2257 	name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2258 
2259 	/*
2260 	 * Walk the snapshots that we are moving
2261 	 *
2262 	 * Compute space to transfer.  Each snapshot gave birth to:
2263 	 * (my used) - (prev's used) + (deadlist's used)
2264 	 * So a sequence would look like:
2265 	 * uN - u(N-1) + dN + ... + u1 - u0 + d1 + u0 - 0 + d0
2266 	 * Which simplifies to:
2267 	 * uN + dN + ... + d1 + d0
2268 	 * Note however, if we stop before we reach the ORIGIN we get:
2269 	 * uN + dN + ... + dM - uM-1
2270 	 */
2271 	pa->used = origin_ds->ds_phys->ds_used_bytes;
2272 	pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2273 	pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2274 	do {
2275 		uint64_t val, dlused, dlcomp, dluncomp;
2276 		dsl_dataset_t *ds = snap->ds;
2277 
2278 		/* Check that the snapshot name does not conflict */
2279 		dsl_dataset_name(ds, name);
2280 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2281 		if (err == 0)
2282 			err = EEXIST;
2283 		if (err != ENOENT)
2284 			break;
2285 		err = 0;
2286 
2287 		/* The very first snapshot does not have a deadlist */
2288 		if (ds->ds_phys->ds_prev_snap_obj != 0) {
2289 			if (err = bplist_space(&ds->ds_deadlist,
2290 			    &dlused, &dlcomp, &dluncomp))
2291 				break;
2292 			pa->used += dlused;
2293 			pa->comp += dlcomp;
2294 			pa->uncomp += dluncomp;
2295 		}
2296 	} while (snap = list_next(&pa->snap_list, snap));
2297 
2298 	/*
2299 	 * If we are a clone of a clone then we never reached ORIGIN,
2300 	 * so we need to subtract out the clone origin's used space.
2301 	 */
2302 	if (pa->clone_origin) {
2303 		pa->used -= pa->clone_origin->ds_phys->ds_used_bytes;
2304 		pa->comp -= pa->clone_origin->ds_phys->ds_compressed_bytes;
2305 		pa->uncomp -= pa->clone_origin->ds_phys->ds_uncompressed_bytes;
2306 	}
2307 
2308 	kmem_free(name, MAXPATHLEN);
2309 
2310 	/* Check that there is enough space here */
2311 	if (err == 0) {
2312 		dsl_dir_t *odd = origin_ds->ds_dir;
2313 		err = dsl_dir_transfer_possible(odd, hds->ds_dir, pa->used);
2314 	}
2315 
2316 	return (err);
2317 }
2318 
2319 static void
2320 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2321 {
2322 	dsl_dataset_t *hds = arg1;
2323 	struct promotearg *pa = arg2;
2324 	struct promotedsarg *snap = list_head(&pa->snap_list);
2325 	dsl_dataset_t *origin_ds = snap->ds;
2326 	dsl_dir_t *dd = hds->ds_dir;
2327 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2328 	dsl_dir_t *odd = NULL;
2329 	char *name;
2330 
2331 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2332 
2333 	/*
2334 	 * We need to explicitly open odd, since origin_ds's dd will be
2335 	 * changing.
2336 	 */
2337 	VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2338 	    NULL, FTAG, &odd));
2339 
2340 	/* change origin's next snap */
2341 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2342 	origin_ds->ds_phys->ds_next_snap_obj = pa->newnext_obj;
2343 
2344 	/* change origin */
2345 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2346 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2347 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2348 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2349 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2350 
2351 	/* move snapshots to this dir */
2352 	name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2353 	do {
2354 		dsl_dataset_t *ds = snap->ds;
2355 
2356 		/* move snap name entry */
2357 		dsl_dataset_name(ds, name);
2358 		VERIFY(0 == dsl_dataset_snap_remove(pa->old_head,
2359 		    ds->ds_snapname, tx));
2360 		VERIFY(0 == zap_add(dp->dp_meta_objset,
2361 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2362 		    8, 1, &ds->ds_object, tx));
2363 
2364 		/* change containing dsl_dir */
2365 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2366 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2367 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2368 		ASSERT3P(ds->ds_dir, ==, odd);
2369 		dsl_dir_close(ds->ds_dir, ds);
2370 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2371 		    NULL, ds, &ds->ds_dir));
2372 
2373 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2374 	} while (snap = list_next(&pa->snap_list, snap));
2375 
2376 	/* change space accounting */
2377 	dsl_dir_diduse_space(odd, -pa->used, -pa->comp, -pa->uncomp, tx);
2378 	dsl_dir_diduse_space(dd, pa->used, pa->comp, pa->uncomp, tx);
2379 	origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2380 
2381 	/* log history record */
2382 	spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2383 	    cr, "dataset = %llu", hds->ds_object);
2384 
2385 	dsl_dir_close(odd, FTAG);
2386 	kmem_free(name, MAXPATHLEN);
2387 }
2388 
2389 int
2390 dsl_dataset_promote(const char *name)
2391 {
2392 	dsl_dataset_t *ds;
2393 	dsl_dir_t *dd;
2394 	dsl_pool_t *dp;
2395 	dmu_object_info_t doi;
2396 	struct promotearg pa;
2397 	struct promotedsarg *snap;
2398 	uint64_t snap_obj;
2399 	uint64_t last_snap = 0;
2400 	int err;
2401 
2402 	err = dsl_dataset_hold(name, FTAG, &ds);
2403 	if (err)
2404 		return (err);
2405 	dd = ds->ds_dir;
2406 	dp = dd->dd_pool;
2407 
2408 	err = dmu_object_info(dp->dp_meta_objset,
2409 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2410 	if (err) {
2411 		dsl_dataset_rele(ds, FTAG);
2412 		return (err);
2413 	}
2414 
2415 	/*
2416 	 * We are going to inherit all the snapshots taken before our
2417 	 * origin (i.e., our new origin will be our parent's origin).
2418 	 * Take ownership of them so that we can rename them into our
2419 	 * namespace.
2420 	 */
2421 	pa.clone_origin = NULL;
2422 	list_create(&pa.snap_list,
2423 	    sizeof (struct promotedsarg), offsetof(struct promotedsarg, link));
2424 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2425 	ASSERT(dd->dd_phys->dd_origin_obj != 0);
2426 	snap_obj = dd->dd_phys->dd_origin_obj;
2427 	while (snap_obj) {
2428 		snap = kmem_alloc(sizeof (struct promotedsarg), KM_SLEEP);
2429 		err = dsl_dataset_own_obj(dp, snap_obj, 0, FTAG, &snap->ds);
2430 		if (err == ENOENT) {
2431 			/* lost race with snapshot destroy */
2432 			struct promotedsarg *last = list_tail(&pa.snap_list);
2433 			ASSERT(snap_obj != last->ds->ds_phys->ds_prev_snap_obj);
2434 			snap_obj = last->ds->ds_phys->ds_prev_snap_obj;
2435 			kmem_free(snap, sizeof (struct promotedsarg));
2436 			continue;
2437 		} else if (err) {
2438 			kmem_free(snap, sizeof (struct promotedsarg));
2439 			rw_exit(&dp->dp_config_rwlock);
2440 			goto out;
2441 		}
2442 		/*
2443 		 * We could be a clone of a clone.  If we reach our
2444 		 * parent's branch point, we're done.
2445 		 */
2446 		if (last_snap &&
2447 		    snap->ds->ds_phys->ds_next_snap_obj != last_snap) {
2448 			pa.clone_origin = snap->ds;
2449 			kmem_free(snap, sizeof (struct promotedsarg));
2450 			snap_obj = 0;
2451 		} else {
2452 			list_insert_tail(&pa.snap_list, snap);
2453 			last_snap = snap_obj;
2454 			snap_obj = snap->ds->ds_phys->ds_prev_snap_obj;
2455 		}
2456 	}
2457 	snap = list_head(&pa.snap_list);
2458 	ASSERT(snap != NULL);
2459 	err = dsl_dataset_hold_obj(dp,
2460 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &pa.old_head);
2461 	rw_exit(&dp->dp_config_rwlock);
2462 
2463 	if (err)
2464 		goto out;
2465 
2466 	/*
2467 	 * Add in 128x the snapnames zapobj size, since we will be moving
2468 	 * a bunch of snapnames to the promoted ds, and dirtying their
2469 	 * bonus buffers.
2470 	 */
2471 	err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2472 	    dsl_dataset_promote_sync, ds, &pa, 2 + 2 * doi.doi_physical_blks);
2473 
2474 	dsl_dataset_rele(pa.old_head, FTAG);
2475 out:
2476 	while ((snap = list_tail(&pa.snap_list)) != NULL) {
2477 		list_remove(&pa.snap_list, snap);
2478 		dsl_dataset_disown(snap->ds, FTAG);
2479 		kmem_free(snap, sizeof (struct promotedsarg));
2480 	}
2481 	list_destroy(&pa.snap_list);
2482 	if (pa.clone_origin)
2483 		dsl_dataset_disown(pa.clone_origin, FTAG);
2484 	dsl_dataset_rele(ds, FTAG);
2485 	return (err);
2486 }
2487 
2488 struct cloneswaparg {
2489 	dsl_dataset_t *cds; /* clone dataset */
2490 	dsl_dataset_t *ohds; /* origin's head dataset */
2491 	boolean_t force;
2492 	int64_t unused_refres_delta; /* change in unconsumed refreservation */
2493 };
2494 
2495 /* ARGSUSED */
2496 static int
2497 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2498 {
2499 	struct cloneswaparg *csa = arg1;
2500 
2501 	/* they should both be heads */
2502 	if (dsl_dataset_is_snapshot(csa->cds) ||
2503 	    dsl_dataset_is_snapshot(csa->ohds))
2504 		return (EINVAL);
2505 
2506 	/* the branch point should be just before them */
2507 	if (csa->cds->ds_prev != csa->ohds->ds_prev)
2508 		return (EINVAL);
2509 
2510 	/* cds should be the clone */
2511 	if (csa->cds->ds_prev->ds_phys->ds_next_snap_obj !=
2512 	    csa->ohds->ds_object)
2513 		return (EINVAL);
2514 
2515 	/* the clone should be a child of the origin */
2516 	if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2517 		return (EINVAL);
2518 
2519 	/* ohds shouldn't be modified unless 'force' */
2520 	if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2521 		return (ETXTBSY);
2522 
2523 	/* adjust amount of any unconsumed refreservation */
2524 	csa->unused_refres_delta =
2525 	    (int64_t)MIN(csa->ohds->ds_reserved,
2526 	    csa->ohds->ds_phys->ds_unique_bytes) -
2527 	    (int64_t)MIN(csa->ohds->ds_reserved,
2528 	    csa->cds->ds_phys->ds_unique_bytes);
2529 
2530 	if (csa->unused_refres_delta > 0 &&
2531 	    csa->unused_refres_delta >
2532 	    dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
2533 		return (ENOSPC);
2534 
2535 	return (0);
2536 }
2537 
2538 /* ARGSUSED */
2539 static void
2540 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2541 {
2542 	struct cloneswaparg *csa = arg1;
2543 	dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2544 	uint64_t itor = 0;
2545 	blkptr_t bp;
2546 	uint64_t unique = 0;
2547 	int err;
2548 
2549 	ASSERT(csa->cds->ds_reserved == 0);
2550 	ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota);
2551 
2552 	dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2553 	dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2554 	dmu_buf_will_dirty(csa->cds->ds_prev->ds_dbuf, tx);
2555 
2556 	if (csa->cds->ds_user_ptr != NULL) {
2557 		csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr);
2558 		csa->cds->ds_user_ptr = NULL;
2559 	}
2560 
2561 	if (csa->ohds->ds_user_ptr != NULL) {
2562 		csa->ohds->ds_user_evict_func(csa->ohds,
2563 		    csa->ohds->ds_user_ptr);
2564 		csa->ohds->ds_user_ptr = NULL;
2565 	}
2566 
2567 	/* compute unique space */
2568 	while ((err = bplist_iterate(&csa->cds->ds_deadlist,
2569 	    &itor, &bp)) == 0) {
2570 		if (bp.blk_birth > csa->cds->ds_prev->ds_phys->ds_prev_snap_txg)
2571 			unique += bp_get_dasize(dp->dp_spa, &bp);
2572 	}
2573 	VERIFY(err == ENOENT);
2574 
2575 	/* reset origin's unique bytes */
2576 	csa->cds->ds_prev->ds_phys->ds_unique_bytes = unique;
2577 
2578 	/* swap blkptrs */
2579 	{
2580 		blkptr_t tmp;
2581 		tmp = csa->ohds->ds_phys->ds_bp;
2582 		csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2583 		csa->cds->ds_phys->ds_bp = tmp;
2584 	}
2585 
2586 	/* set dd_*_bytes */
2587 	{
2588 		int64_t dused, dcomp, duncomp;
2589 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2590 		uint64_t odl_used, odl_comp, odl_uncomp;
2591 
2592 		VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2593 		    &cdl_comp, &cdl_uncomp));
2594 		VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2595 		    &odl_comp, &odl_uncomp));
2596 		dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2597 		    (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2598 		dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2599 		    (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2600 		duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2601 		    cdl_uncomp -
2602 		    (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2603 
2604 		dsl_dir_diduse_space(csa->ohds->ds_dir,
2605 		    dused, dcomp, duncomp, tx);
2606 		dsl_dir_diduse_space(csa->cds->ds_dir,
2607 		    -dused, -dcomp, -duncomp, tx);
2608 	}
2609 
2610 #define	SWITCH64(x, y) \
2611 	{ \
2612 		uint64_t __tmp = (x); \
2613 		(x) = (y); \
2614 		(y) = __tmp; \
2615 	}
2616 
2617 	/* swap ds_*_bytes */
2618 	SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2619 	    csa->cds->ds_phys->ds_used_bytes);
2620 	SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2621 	    csa->cds->ds_phys->ds_compressed_bytes);
2622 	SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2623 	    csa->cds->ds_phys->ds_uncompressed_bytes);
2624 	SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
2625 	    csa->cds->ds_phys->ds_unique_bytes);
2626 
2627 	/* apply any parent delta for change in unconsumed refreservation */
2628 	dsl_dir_diduse_space(csa->ohds->ds_dir, csa->unused_refres_delta,
2629 	    0, 0, tx);
2630 
2631 	/* swap deadlists */
2632 	bplist_close(&csa->cds->ds_deadlist);
2633 	bplist_close(&csa->ohds->ds_deadlist);
2634 	SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2635 	    csa->cds->ds_phys->ds_deadlist_obj);
2636 	VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2637 	    csa->cds->ds_phys->ds_deadlist_obj));
2638 	VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2639 	    csa->ohds->ds_phys->ds_deadlist_obj));
2640 }
2641 
2642 /*
2643  * Swap 'clone' with its origin head file system.  Used at the end
2644  * of "online recv" to swizzle the file system to the new version.
2645  */
2646 int
2647 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2648     boolean_t force)
2649 {
2650 	struct cloneswaparg csa;
2651 	int error;
2652 
2653 	ASSERT(clone->ds_owner);
2654 	ASSERT(origin_head->ds_owner);
2655 retry:
2656 	/* Need exclusive access for the swap */
2657 	rw_enter(&clone->ds_rwlock, RW_WRITER);
2658 	if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
2659 		rw_exit(&clone->ds_rwlock);
2660 		rw_enter(&origin_head->ds_rwlock, RW_WRITER);
2661 		if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
2662 			rw_exit(&origin_head->ds_rwlock);
2663 			goto retry;
2664 		}
2665 	}
2666 	csa.cds = clone;
2667 	csa.ohds = origin_head;
2668 	csa.force = force;
2669 	error = dsl_sync_task_do(clone->ds_dir->dd_pool,
2670 	    dsl_dataset_clone_swap_check,
2671 	    dsl_dataset_clone_swap_sync, &csa, NULL, 9);
2672 	return (error);
2673 }
2674 
2675 /*
2676  * Given a pool name and a dataset object number in that pool,
2677  * return the name of that dataset.
2678  */
2679 int
2680 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2681 {
2682 	spa_t *spa;
2683 	dsl_pool_t *dp;
2684 	dsl_dataset_t *ds;
2685 	int error;
2686 
2687 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
2688 		return (error);
2689 	dp = spa_get_dsl(spa);
2690 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2691 	if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
2692 		dsl_dataset_name(ds, buf);
2693 		dsl_dataset_rele(ds, FTAG);
2694 	}
2695 	rw_exit(&dp->dp_config_rwlock);
2696 	spa_close(spa, FTAG);
2697 
2698 	return (error);
2699 }
2700 
2701 int
2702 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2703     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2704 {
2705 	int error = 0;
2706 
2707 	ASSERT3S(asize, >, 0);
2708 
2709 	/*
2710 	 * *ref_rsrv is the portion of asize that will come from any
2711 	 * unconsumed refreservation space.
2712 	 */
2713 	*ref_rsrv = 0;
2714 
2715 	mutex_enter(&ds->ds_lock);
2716 	/*
2717 	 * Make a space adjustment for reserved bytes.
2718 	 */
2719 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2720 		ASSERT3U(*used, >=,
2721 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2722 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2723 		*ref_rsrv =
2724 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2725 	}
2726 
2727 	if (!check_quota || ds->ds_quota == 0) {
2728 		mutex_exit(&ds->ds_lock);
2729 		return (0);
2730 	}
2731 	/*
2732 	 * If they are requesting more space, and our current estimate
2733 	 * is over quota, they get to try again unless the actual
2734 	 * on-disk is over quota and there are no pending changes (which
2735 	 * may free up space for us).
2736 	 */
2737 	if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
2738 		if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
2739 			error = ERESTART;
2740 		else
2741 			error = EDQUOT;
2742 	}
2743 	mutex_exit(&ds->ds_lock);
2744 
2745 	return (error);
2746 }
2747 
2748 /* ARGSUSED */
2749 static int
2750 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
2751 {
2752 	dsl_dataset_t *ds = arg1;
2753 	uint64_t *quotap = arg2;
2754 	uint64_t new_quota = *quotap;
2755 
2756 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
2757 		return (ENOTSUP);
2758 
2759 	if (new_quota == 0)
2760 		return (0);
2761 
2762 	if (new_quota < ds->ds_phys->ds_used_bytes ||
2763 	    new_quota < ds->ds_reserved)
2764 		return (ENOSPC);
2765 
2766 	return (0);
2767 }
2768 
2769 /* ARGSUSED */
2770 void
2771 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2772 {
2773 	dsl_dataset_t *ds = arg1;
2774 	uint64_t *quotap = arg2;
2775 	uint64_t new_quota = *quotap;
2776 
2777 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2778 
2779 	ds->ds_quota = new_quota;
2780 
2781 	dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
2782 
2783 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
2784 	    tx, cr, "%lld dataset = %llu ",
2785 	    (longlong_t)new_quota, ds->ds_object);
2786 }
2787 
2788 int
2789 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
2790 {
2791 	dsl_dataset_t *ds;
2792 	int err;
2793 
2794 	err = dsl_dataset_hold(dsname, FTAG, &ds);
2795 	if (err)
2796 		return (err);
2797 
2798 	if (quota != ds->ds_quota) {
2799 		/*
2800 		 * If someone removes a file, then tries to set the quota, we
2801 		 * want to make sure the file freeing takes effect.
2802 		 */
2803 		txg_wait_open(ds->ds_dir->dd_pool, 0);
2804 
2805 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2806 		    dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
2807 		    ds, &quota, 0);
2808 	}
2809 	dsl_dataset_rele(ds, FTAG);
2810 	return (err);
2811 }
2812 
2813 static int
2814 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
2815 {
2816 	dsl_dataset_t *ds = arg1;
2817 	uint64_t *reservationp = arg2;
2818 	uint64_t new_reservation = *reservationp;
2819 	int64_t delta;
2820 	uint64_t unique;
2821 
2822 	if (new_reservation > INT64_MAX)
2823 		return (EOVERFLOW);
2824 
2825 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
2826 	    SPA_VERSION_REFRESERVATION)
2827 		return (ENOTSUP);
2828 
2829 	if (dsl_dataset_is_snapshot(ds))
2830 		return (EINVAL);
2831 
2832 	/*
2833 	 * If we are doing the preliminary check in open context, the
2834 	 * space estimates may be inaccurate.
2835 	 */
2836 	if (!dmu_tx_is_syncing(tx))
2837 		return (0);
2838 
2839 	mutex_enter(&ds->ds_lock);
2840 	unique = dsl_dataset_unique(ds);
2841 	delta = MAX(unique, new_reservation) - MAX(unique, ds->ds_reserved);
2842 	mutex_exit(&ds->ds_lock);
2843 
2844 	if (delta > 0 &&
2845 	    delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2846 		return (ENOSPC);
2847 	if (delta > 0 && ds->ds_quota > 0 &&
2848 	    new_reservation > ds->ds_quota)
2849 		return (ENOSPC);
2850 
2851 	return (0);
2852 }
2853 
2854 /* ARGSUSED */
2855 static void
2856 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
2857     dmu_tx_t *tx)
2858 {
2859 	dsl_dataset_t *ds = arg1;
2860 	uint64_t *reservationp = arg2;
2861 	uint64_t new_reservation = *reservationp;
2862 	uint64_t unique;
2863 	int64_t delta;
2864 
2865 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2866 
2867 	mutex_enter(&ds->ds_lock);
2868 	unique = dsl_dataset_unique(ds);
2869 	delta = MAX(0, (int64_t)(new_reservation - unique)) -
2870 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
2871 	ds->ds_reserved = new_reservation;
2872 	mutex_exit(&ds->ds_lock);
2873 
2874 	dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
2875 	    new_reservation, cr, tx);
2876 
2877 	dsl_dir_diduse_space(ds->ds_dir, delta, 0, 0, tx);
2878 
2879 	spa_history_internal_log(LOG_DS_REFRESERV,
2880 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
2881 	    (longlong_t)new_reservation,
2882 	    ds->ds_dir->dd_phys->dd_head_dataset_obj);
2883 }
2884 
2885 int
2886 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
2887 {
2888 	dsl_dataset_t *ds;
2889 	int err;
2890 
2891 	err = dsl_dataset_hold(dsname, FTAG, &ds);
2892 	if (err)
2893 		return (err);
2894 
2895 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2896 	    dsl_dataset_set_reservation_check,
2897 	    dsl_dataset_set_reservation_sync, ds, &reservation, 0);
2898 	dsl_dataset_rele(ds, FTAG);
2899 	return (err);
2900 }
2901