xref: /titanic_51/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision e3c2d6aa3bc760b22fad3c83f876553f0d2c5b66)
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 2007 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 dsl_checkfunc_t dsl_dataset_destroy_begin_check;
45 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
46 static dsl_checkfunc_t dsl_dataset_rollback_check;
47 static dsl_syncfunc_t dsl_dataset_rollback_sync;
48 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
49 
50 #define	DS_REF_MAX	(1ULL << 62)
51 
52 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
53 
54 /*
55  * We use weighted reference counts to express the various forms of exclusion
56  * between different open modes.  A STANDARD open is 1 point, an EXCLUSIVE open
57  * is DS_REF_MAX, and a PRIMARY open is little more than half of an EXCLUSIVE.
58  * This makes the exclusion logic simple: the total refcnt for all opens cannot
59  * exceed DS_REF_MAX.  For example, EXCLUSIVE opens are exclusive because their
60  * weight (DS_REF_MAX) consumes the entire refcnt space.  PRIMARY opens consume
61  * just over half of the refcnt space, so there can't be more than one, but it
62  * can peacefully coexist with any number of STANDARD opens.
63  */
64 static uint64_t ds_refcnt_weight[DS_MODE_LEVELS] = {
65 	0,			/* DS_MODE_NONE - invalid		*/
66 	1,			/* DS_MODE_STANDARD - unlimited number	*/
67 	(DS_REF_MAX >> 1) + 1,	/* DS_MODE_PRIMARY - only one of these	*/
68 	DS_REF_MAX		/* DS_MODE_EXCLUSIVE - no other opens	*/
69 };
70 
71 /*
72  * Figure out how much of this delta should be propogated to the dsl_dir
73  * layer.  If there's a refreservation, that space has already been
74  * partially accounted for in our ancestors.
75  */
76 static int64_t
77 parent_delta(dsl_dataset_t *ds, int64_t delta)
78 {
79 	uint64_t old_bytes, new_bytes;
80 
81 	if (ds->ds_reserved == 0)
82 		return (delta);
83 
84 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
85 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
86 
87 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
88 	return (new_bytes - old_bytes);
89 }
90 
91 void
92 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
93 {
94 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
95 	int compressed = BP_GET_PSIZE(bp);
96 	int uncompressed = BP_GET_UCSIZE(bp);
97 	int64_t delta;
98 
99 	dprintf_bp(bp, "born, ds=%p\n", ds);
100 
101 	ASSERT(dmu_tx_is_syncing(tx));
102 	/* It could have been compressed away to nothing */
103 	if (BP_IS_HOLE(bp))
104 		return;
105 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
106 	ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
107 	if (ds == NULL) {
108 		/*
109 		 * Account for the meta-objset space in its placeholder
110 		 * dsl_dir.
111 		 */
112 		ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
113 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
114 		    used, compressed, uncompressed, tx);
115 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
116 		return;
117 	}
118 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
119 	mutex_enter(&ds->ds_lock);
120 	delta = parent_delta(ds, used);
121 	ds->ds_phys->ds_used_bytes += used;
122 	ds->ds_phys->ds_compressed_bytes += compressed;
123 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
124 	ds->ds_phys->ds_unique_bytes += used;
125 	mutex_exit(&ds->ds_lock);
126 	dsl_dir_diduse_space(ds->ds_dir, delta, compressed, uncompressed, tx);
127 }
128 
129 void
130 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
131     dmu_tx_t *tx)
132 {
133 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
134 	int compressed = BP_GET_PSIZE(bp);
135 	int uncompressed = BP_GET_UCSIZE(bp);
136 
137 	ASSERT(dmu_tx_is_syncing(tx));
138 	/* No block pointer => nothing to free */
139 	if (BP_IS_HOLE(bp))
140 		return;
141 
142 	ASSERT(used > 0);
143 	if (ds == NULL) {
144 		int err;
145 		/*
146 		 * Account for the meta-objset space in its placeholder
147 		 * dataset.
148 		 */
149 		err = arc_free(pio, tx->tx_pool->dp_spa,
150 		    tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
151 		ASSERT(err == 0);
152 
153 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
154 		    -used, -compressed, -uncompressed, tx);
155 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
156 		return;
157 	}
158 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
159 
160 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
161 
162 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
163 		int err;
164 		int64_t delta;
165 
166 		dprintf_bp(bp, "freeing: %s", "");
167 		err = arc_free(pio, tx->tx_pool->dp_spa,
168 		    tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
169 		ASSERT(err == 0);
170 
171 		mutex_enter(&ds->ds_lock);
172 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
173 		    !DS_UNIQUE_IS_ACCURATE(ds));
174 		delta = parent_delta(ds, -used);
175 		ds->ds_phys->ds_unique_bytes -= used;
176 		mutex_exit(&ds->ds_lock);
177 		dsl_dir_diduse_space(ds->ds_dir,
178 		    delta, -compressed, -uncompressed, tx);
179 	} else {
180 		dprintf_bp(bp, "putting on dead list: %s", "");
181 		VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
182 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
183 		if (ds->ds_phys->ds_prev_snap_obj != 0) {
184 			ASSERT3U(ds->ds_prev->ds_object, ==,
185 			    ds->ds_phys->ds_prev_snap_obj);
186 			ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
187 			if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
188 			    ds->ds_object && bp->blk_birth >
189 			    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
190 				dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
191 				mutex_enter(&ds->ds_prev->ds_lock);
192 				ds->ds_prev->ds_phys->ds_unique_bytes +=
193 				    used;
194 				mutex_exit(&ds->ds_prev->ds_lock);
195 			}
196 		}
197 	}
198 	mutex_enter(&ds->ds_lock);
199 	ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
200 	ds->ds_phys->ds_used_bytes -= used;
201 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
202 	ds->ds_phys->ds_compressed_bytes -= compressed;
203 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
204 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
205 	mutex_exit(&ds->ds_lock);
206 }
207 
208 uint64_t
209 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
210 {
211 	uint64_t trysnap = 0;
212 
213 	if (ds == NULL)
214 		return (0);
215 	/*
216 	 * The snapshot creation could fail, but that would cause an
217 	 * incorrect FALSE return, which would only result in an
218 	 * overestimation of the amount of space that an operation would
219 	 * consume, which is OK.
220 	 *
221 	 * There's also a small window where we could miss a pending
222 	 * snapshot, because we could set the sync task in the quiescing
223 	 * phase.  So this should only be used as a guess.
224 	 */
225 	if (ds->ds_trysnap_txg >
226 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
227 		trysnap = ds->ds_trysnap_txg;
228 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
229 }
230 
231 int
232 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
233 {
234 	return (blk_birth > dsl_dataset_prev_snap_txg(ds));
235 }
236 
237 /* ARGSUSED */
238 static void
239 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
240 {
241 	dsl_dataset_t *ds = dsv;
242 
243 	/* open_refcount == DS_REF_MAX when deleting */
244 	ASSERT(ds->ds_open_refcount == 0 ||
245 	    ds->ds_open_refcount == DS_REF_MAX);
246 
247 	dprintf_ds(ds, "evicting %s\n", "");
248 
249 	unique_remove(ds->ds_fsid_guid);
250 
251 	if (ds->ds_user_ptr != NULL)
252 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
253 
254 	if (ds->ds_prev) {
255 		dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds);
256 		ds->ds_prev = NULL;
257 	}
258 
259 	bplist_close(&ds->ds_deadlist);
260 	dsl_dir_close(ds->ds_dir, ds);
261 
262 	ASSERT(!list_link_active(&ds->ds_synced_link));
263 
264 	mutex_destroy(&ds->ds_lock);
265 	mutex_destroy(&ds->ds_opening_lock);
266 	mutex_destroy(&ds->ds_deadlist.bpl_lock);
267 
268 	kmem_free(ds, sizeof (dsl_dataset_t));
269 }
270 
271 static int
272 dsl_dataset_get_snapname(dsl_dataset_t *ds)
273 {
274 	dsl_dataset_phys_t *headphys;
275 	int err;
276 	dmu_buf_t *headdbuf;
277 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
278 	objset_t *mos = dp->dp_meta_objset;
279 
280 	if (ds->ds_snapname[0])
281 		return (0);
282 	if (ds->ds_phys->ds_next_snap_obj == 0)
283 		return (0);
284 
285 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
286 	    FTAG, &headdbuf);
287 	if (err)
288 		return (err);
289 	headphys = headdbuf->db_data;
290 	err = zap_value_search(dp->dp_meta_objset,
291 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
292 	dmu_buf_rele(headdbuf, FTAG);
293 	return (err);
294 }
295 
296 int
297 dsl_dataset_open_obj(dsl_pool_t *dp, uint64_t dsobj, const char *snapname,
298     int mode, void *tag, dsl_dataset_t **dsp)
299 {
300 	uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)];
301 	objset_t *mos = dp->dp_meta_objset;
302 	dmu_buf_t *dbuf;
303 	dsl_dataset_t *ds;
304 	int err;
305 
306 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
307 	    dsl_pool_sync_context(dp));
308 
309 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
310 	if (err)
311 		return (err);
312 	ds = dmu_buf_get_user(dbuf);
313 	if (ds == NULL) {
314 		dsl_dataset_t *winner;
315 
316 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
317 		ds->ds_dbuf = dbuf;
318 		ds->ds_object = dsobj;
319 		ds->ds_phys = dbuf->db_data;
320 
321 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
322 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
323 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
324 		    NULL);
325 
326 		err = bplist_open(&ds->ds_deadlist,
327 		    mos, ds->ds_phys->ds_deadlist_obj);
328 		if (err == 0) {
329 			err = dsl_dir_open_obj(dp,
330 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
331 		}
332 		if (err) {
333 			/*
334 			 * we don't really need to close the blist if we
335 			 * just opened it.
336 			 */
337 			mutex_destroy(&ds->ds_lock);
338 			mutex_destroy(&ds->ds_opening_lock);
339 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
340 			kmem_free(ds, sizeof (dsl_dataset_t));
341 			dmu_buf_rele(dbuf, tag);
342 			return (err);
343 		}
344 
345 		if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) {
346 			ds->ds_snapname[0] = '\0';
347 			if (ds->ds_phys->ds_prev_snap_obj) {
348 				err = dsl_dataset_open_obj(dp,
349 				    ds->ds_phys->ds_prev_snap_obj, NULL,
350 				    DS_MODE_NONE, ds, &ds->ds_prev);
351 			}
352 		} else {
353 			if (snapname) {
354 #ifdef ZFS_DEBUG
355 				dsl_dataset_phys_t *headphys;
356 				dmu_buf_t *headdbuf;
357 				err = dmu_bonus_hold(mos,
358 				    ds->ds_dir->dd_phys->dd_head_dataset_obj,
359 				    FTAG, &headdbuf);
360 				if (err == 0) {
361 					headphys = headdbuf->db_data;
362 					uint64_t foundobj;
363 					err = zap_lookup(dp->dp_meta_objset,
364 					    headphys->ds_snapnames_zapobj,
365 					    snapname, sizeof (foundobj), 1,
366 					    &foundobj);
367 					ASSERT3U(foundobj, ==, dsobj);
368 					dmu_buf_rele(headdbuf, FTAG);
369 				}
370 #endif
371 				(void) strcat(ds->ds_snapname, snapname);
372 			} else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
373 				err = dsl_dataset_get_snapname(ds);
374 			}
375 		}
376 
377 		if (err == 0) {
378 			winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
379 			    dsl_dataset_evict);
380 		}
381 		if (err || winner) {
382 			bplist_close(&ds->ds_deadlist);
383 			if (ds->ds_prev) {
384 				dsl_dataset_close(ds->ds_prev,
385 				    DS_MODE_NONE, ds);
386 			}
387 			dsl_dir_close(ds->ds_dir, ds);
388 			mutex_destroy(&ds->ds_lock);
389 			mutex_destroy(&ds->ds_opening_lock);
390 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
391 			kmem_free(ds, sizeof (dsl_dataset_t));
392 			if (err) {
393 				dmu_buf_rele(dbuf, tag);
394 				return (err);
395 			}
396 			ds = winner;
397 		} else {
398 			ds->ds_fsid_guid =
399 			    unique_insert(ds->ds_phys->ds_fsid_guid);
400 		}
401 
402 		if (!dsl_dataset_is_snapshot(ds)) {
403 			boolean_t need_lock =
404 			    !RW_LOCK_HELD(&dp->dp_config_rwlock);
405 
406 			if (need_lock)
407 				rw_enter(&dp->dp_config_rwlock, RW_READER);
408 			VERIFY(0 == dsl_prop_get_ds_locked(ds->ds_dir,
409 			    "refreservation", sizeof (uint64_t), 1,
410 			    &ds->ds_reserved, NULL));
411 			VERIFY(0 == dsl_prop_get_ds_locked(ds->ds_dir,
412 			    "refquota", sizeof (uint64_t), 1, &ds->ds_quota,
413 			    NULL));
414 			if (need_lock)
415 				rw_exit(&dp->dp_config_rwlock);
416 		} else {
417 			ds->ds_reserved = ds->ds_quota = 0;
418 		}
419 	}
420 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
421 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
422 
423 	mutex_enter(&ds->ds_lock);
424 	if ((DS_MODE_LEVEL(mode) == DS_MODE_PRIMARY &&
425 	    (ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) &&
426 	    !DS_MODE_IS_INCONSISTENT(mode)) ||
427 	    (ds->ds_open_refcount + weight > DS_REF_MAX)) {
428 		mutex_exit(&ds->ds_lock);
429 		dsl_dataset_close(ds, DS_MODE_NONE, tag);
430 		return (EBUSY);
431 	}
432 	ds->ds_open_refcount += weight;
433 	mutex_exit(&ds->ds_lock);
434 
435 	*dsp = ds;
436 	return (0);
437 }
438 
439 int
440 dsl_dataset_open_spa(spa_t *spa, const char *name, int mode,
441     void *tag, dsl_dataset_t **dsp)
442 {
443 	dsl_dir_t *dd;
444 	dsl_pool_t *dp;
445 	const char *tail;
446 	uint64_t obj;
447 	dsl_dataset_t *ds = NULL;
448 	int err = 0;
449 
450 	err = dsl_dir_open_spa(spa, name, FTAG, &dd, &tail);
451 	if (err)
452 		return (err);
453 
454 	dp = dd->dd_pool;
455 	obj = dd->dd_phys->dd_head_dataset_obj;
456 	rw_enter(&dp->dp_config_rwlock, RW_READER);
457 	if (obj == 0) {
458 		/* A dataset with no associated objset */
459 		err = ENOENT;
460 		goto out;
461 	}
462 
463 	if (tail != NULL) {
464 		objset_t *mos = dp->dp_meta_objset;
465 
466 		err = dsl_dataset_open_obj(dp, obj, NULL,
467 		    DS_MODE_NONE, tag, &ds);
468 		if (err)
469 			goto out;
470 		obj = ds->ds_phys->ds_snapnames_zapobj;
471 		dsl_dataset_close(ds, DS_MODE_NONE, tag);
472 		ds = NULL;
473 
474 		if (tail[0] != '@') {
475 			err = ENOENT;
476 			goto out;
477 		}
478 		tail++;
479 
480 		/* Look for a snapshot */
481 		if (!DS_MODE_IS_READONLY(mode)) {
482 			err = EROFS;
483 			goto out;
484 		}
485 		dprintf("looking for snapshot '%s'\n", tail);
486 		err = zap_lookup(mos, obj, tail, 8, 1, &obj);
487 		if (err)
488 			goto out;
489 	}
490 	err = dsl_dataset_open_obj(dp, obj, tail, mode, tag, &ds);
491 
492 out:
493 	rw_exit(&dp->dp_config_rwlock);
494 	dsl_dir_close(dd, FTAG);
495 
496 	ASSERT3U((err == 0), ==, (ds != NULL));
497 	/* ASSERT(ds == NULL || strcmp(name, ds->ds_name) == 0); */
498 
499 	*dsp = ds;
500 	return (err);
501 }
502 
503 int
504 dsl_dataset_open(const char *name, int mode, void *tag, dsl_dataset_t **dsp)
505 {
506 	return (dsl_dataset_open_spa(NULL, name, mode, tag, dsp));
507 }
508 
509 void
510 dsl_dataset_name(dsl_dataset_t *ds, char *name)
511 {
512 	if (ds == NULL) {
513 		(void) strcpy(name, "mos");
514 	} else {
515 		dsl_dir_name(ds->ds_dir, name);
516 		VERIFY(0 == dsl_dataset_get_snapname(ds));
517 		if (ds->ds_snapname[0]) {
518 			(void) strcat(name, "@");
519 			if (!MUTEX_HELD(&ds->ds_lock)) {
520 				/*
521 				 * We use a "recursive" mutex so that we
522 				 * can call dprintf_ds() with ds_lock held.
523 				 */
524 				mutex_enter(&ds->ds_lock);
525 				(void) strcat(name, ds->ds_snapname);
526 				mutex_exit(&ds->ds_lock);
527 			} else {
528 				(void) strcat(name, ds->ds_snapname);
529 			}
530 		}
531 	}
532 }
533 
534 static int
535 dsl_dataset_namelen(dsl_dataset_t *ds)
536 {
537 	int result;
538 
539 	if (ds == NULL) {
540 		result = 3;	/* "mos" */
541 	} else {
542 		result = dsl_dir_namelen(ds->ds_dir);
543 		VERIFY(0 == dsl_dataset_get_snapname(ds));
544 		if (ds->ds_snapname[0]) {
545 			++result;	/* adding one for the @-sign */
546 			if (!MUTEX_HELD(&ds->ds_lock)) {
547 				/* see dsl_datset_name */
548 				mutex_enter(&ds->ds_lock);
549 				result += strlen(ds->ds_snapname);
550 				mutex_exit(&ds->ds_lock);
551 			} else {
552 				result += strlen(ds->ds_snapname);
553 			}
554 		}
555 	}
556 
557 	return (result);
558 }
559 
560 void
561 dsl_dataset_close(dsl_dataset_t *ds, int mode, void *tag)
562 {
563 	uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)];
564 	mutex_enter(&ds->ds_lock);
565 	ASSERT3U(ds->ds_open_refcount, >=, weight);
566 	ds->ds_open_refcount -= weight;
567 	dprintf_ds(ds, "closing mode %u refcount now 0x%llx\n",
568 	    mode, ds->ds_open_refcount);
569 	mutex_exit(&ds->ds_lock);
570 
571 	dmu_buf_rele(ds->ds_dbuf, tag);
572 }
573 
574 void
575 dsl_dataset_downgrade(dsl_dataset_t *ds, int oldmode, int newmode)
576 {
577 	uint64_t oldweight = ds_refcnt_weight[DS_MODE_LEVEL(oldmode)];
578 	uint64_t newweight = ds_refcnt_weight[DS_MODE_LEVEL(newmode)];
579 	mutex_enter(&ds->ds_lock);
580 	ASSERT3U(ds->ds_open_refcount, >=, oldweight);
581 	ASSERT3U(oldweight, >=, newweight);
582 	ds->ds_open_refcount -= oldweight;
583 	ds->ds_open_refcount += newweight;
584 	mutex_exit(&ds->ds_lock);
585 }
586 
587 boolean_t
588 dsl_dataset_tryupgrade(dsl_dataset_t *ds, int oldmode, int newmode)
589 {
590 	boolean_t rv;
591 	uint64_t oldweight = ds_refcnt_weight[DS_MODE_LEVEL(oldmode)];
592 	uint64_t newweight = ds_refcnt_weight[DS_MODE_LEVEL(newmode)];
593 	mutex_enter(&ds->ds_lock);
594 	ASSERT3U(ds->ds_open_refcount, >=, oldweight);
595 	ASSERT3U(newweight, >=, oldweight);
596 	if (ds->ds_open_refcount - oldweight + newweight > DS_REF_MAX) {
597 		rv = B_FALSE;
598 	} else {
599 		ds->ds_open_refcount -= oldweight;
600 		ds->ds_open_refcount += newweight;
601 		rv = B_TRUE;
602 	}
603 	mutex_exit(&ds->ds_lock);
604 	return (rv);
605 }
606 
607 void
608 dsl_dataset_create_root(dsl_pool_t *dp, uint64_t *ddobjp, dmu_tx_t *tx)
609 {
610 	objset_t *mos = dp->dp_meta_objset;
611 	dmu_buf_t *dbuf;
612 	dsl_dataset_phys_t *dsphys;
613 	dsl_dataset_t *ds;
614 	uint64_t dsobj;
615 	dsl_dir_t *dd;
616 
617 	dsl_dir_create_root(mos, ddobjp, tx);
618 	VERIFY(0 == dsl_dir_open_obj(dp, *ddobjp, NULL, FTAG, &dd));
619 
620 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
621 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
622 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
623 	dmu_buf_will_dirty(dbuf, tx);
624 	dsphys = dbuf->db_data;
625 	dsphys->ds_dir_obj = dd->dd_object;
626 	dsphys->ds_fsid_guid = unique_create();
627 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
628 	    sizeof (dsphys->ds_guid));
629 	dsphys->ds_snapnames_zapobj =
630 	    zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx);
631 	dsphys->ds_creation_time = gethrestime_sec();
632 	dsphys->ds_creation_txg = tx->tx_txg;
633 	dsphys->ds_deadlist_obj =
634 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
635 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
636 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
637 	dmu_buf_rele(dbuf, FTAG);
638 
639 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
640 	dd->dd_phys->dd_head_dataset_obj = dsobj;
641 	dsl_dir_close(dd, FTAG);
642 
643 	VERIFY(0 ==
644 	    dsl_dataset_open_obj(dp, dsobj, NULL, DS_MODE_NONE, FTAG, &ds));
645 	(void) dmu_objset_create_impl(dp->dp_spa, ds,
646 	    &ds->ds_phys->ds_bp, DMU_OST_ZFS, tx);
647 	dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
648 }
649 
650 uint64_t
651 dsl_dataset_create_sync_impl(dsl_dir_t *dd, dsl_dataset_t *origin, dmu_tx_t *tx)
652 {
653 	dsl_pool_t *dp = dd->dd_pool;
654 	dmu_buf_t *dbuf;
655 	dsl_dataset_phys_t *dsphys;
656 	uint64_t dsobj;
657 	objset_t *mos = dp->dp_meta_objset;
658 
659 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
660 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
661 	ASSERT(dmu_tx_is_syncing(tx));
662 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
663 
664 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
665 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
666 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
667 	dmu_buf_will_dirty(dbuf, tx);
668 	dsphys = dbuf->db_data;
669 	dsphys->ds_dir_obj = dd->dd_object;
670 	dsphys->ds_fsid_guid = unique_create();
671 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
672 	    sizeof (dsphys->ds_guid));
673 	dsphys->ds_snapnames_zapobj =
674 	    zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx);
675 	dsphys->ds_creation_time = gethrestime_sec();
676 	dsphys->ds_creation_txg = tx->tx_txg;
677 	dsphys->ds_deadlist_obj =
678 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
679 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
680 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
681 
682 	if (origin) {
683 		dsphys->ds_prev_snap_obj = origin->ds_object;
684 		dsphys->ds_prev_snap_txg =
685 		    origin->ds_phys->ds_creation_txg;
686 		dsphys->ds_used_bytes =
687 		    origin->ds_phys->ds_used_bytes;
688 		dsphys->ds_compressed_bytes =
689 		    origin->ds_phys->ds_compressed_bytes;
690 		dsphys->ds_uncompressed_bytes =
691 		    origin->ds_phys->ds_uncompressed_bytes;
692 		dsphys->ds_bp = origin->ds_phys->ds_bp;
693 
694 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
695 		origin->ds_phys->ds_num_children++;
696 
697 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
698 		dd->dd_phys->dd_origin_obj = origin->ds_object;
699 	}
700 	dmu_buf_rele(dbuf, FTAG);
701 
702 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
703 	dd->dd_phys->dd_head_dataset_obj = dsobj;
704 
705 	return (dsobj);
706 }
707 
708 uint64_t
709 dsl_dataset_create_sync(dsl_dir_t *pdd,
710     const char *lastname, dsl_dataset_t *origin, cred_t *cr, dmu_tx_t *tx)
711 {
712 	dsl_pool_t *dp = pdd->dd_pool;
713 	uint64_t dsobj, ddobj;
714 	dsl_dir_t *dd;
715 
716 	ASSERT(lastname[0] != '@');
717 
718 	ddobj = dsl_dir_create_sync(pdd, lastname, tx);
719 	VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
720 
721 	dsobj = dsl_dataset_create_sync_impl(dd, origin, tx);
722 
723 	dsl_deleg_set_create_perms(dd, tx, cr);
724 
725 	dsl_dir_close(dd, FTAG);
726 
727 	return (dsobj);
728 }
729 
730 struct destroyarg {
731 	dsl_sync_task_group_t *dstg;
732 	char *snapname;
733 	char *failed;
734 };
735 
736 static int
737 dsl_snapshot_destroy_one(char *name, void *arg)
738 {
739 	struct destroyarg *da = arg;
740 	dsl_dataset_t *ds;
741 	char *cp;
742 	int err;
743 
744 	(void) strcat(name, "@");
745 	(void) strcat(name, da->snapname);
746 	err = dsl_dataset_open(name,
747 	    DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT,
748 	    da->dstg, &ds);
749 	cp = strchr(name, '@');
750 	*cp = '\0';
751 	if (err == ENOENT)
752 		return (0);
753 	if (err) {
754 		(void) strcpy(da->failed, name);
755 		return (err);
756 	}
757 
758 	dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
759 	    dsl_dataset_destroy_sync, ds, da->dstg, 0);
760 	return (0);
761 }
762 
763 /*
764  * Destroy 'snapname' in all descendants of 'fsname'.
765  */
766 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
767 int
768 dsl_snapshots_destroy(char *fsname, char *snapname)
769 {
770 	int err;
771 	struct destroyarg da;
772 	dsl_sync_task_t *dst;
773 	spa_t *spa;
774 
775 	err = spa_open(fsname, &spa, FTAG);
776 	if (err)
777 		return (err);
778 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
779 	da.snapname = snapname;
780 	da.failed = fsname;
781 
782 	err = dmu_objset_find(fsname,
783 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
784 
785 	if (err == 0)
786 		err = dsl_sync_task_group_wait(da.dstg);
787 
788 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
789 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
790 		dsl_dataset_t *ds = dst->dst_arg1;
791 		if (dst->dst_err) {
792 			dsl_dataset_name(ds, fsname);
793 			*strchr(fsname, '@') = '\0';
794 		}
795 		/*
796 		 * If it was successful, destroy_sync would have
797 		 * closed the ds
798 		 */
799 		if (err)
800 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, da.dstg);
801 	}
802 
803 	dsl_sync_task_group_destroy(da.dstg);
804 	spa_close(spa, FTAG);
805 	return (err);
806 }
807 
808 /*
809  * ds must be opened EXCLUSIVE or PRIMARY.  on return (whether
810  * successful or not), ds will be closed and caller can no longer
811  * dereference it.
812  */
813 int
814 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
815 {
816 	int err;
817 	dsl_sync_task_group_t *dstg;
818 	objset_t *os;
819 	dsl_dir_t *dd;
820 	uint64_t obj;
821 
822 	if (ds->ds_open_refcount != DS_REF_MAX) {
823 		if (dsl_dataset_tryupgrade(ds, DS_MODE_PRIMARY,
824 		    DS_MODE_EXCLUSIVE) == 0) {
825 			dsl_dataset_close(ds, DS_MODE_PRIMARY, tag);
826 			return (EBUSY);
827 		}
828 	}
829 
830 	if (dsl_dataset_is_snapshot(ds)) {
831 		/* Destroying a snapshot is simpler */
832 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
833 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
834 		    ds, tag, 0);
835 		goto out;
836 	}
837 
838 	dd = ds->ds_dir;
839 
840 	/*
841 	 * Check for errors and mark this ds as inconsistent, in
842 	 * case we crash while freeing the objects.
843 	 */
844 	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
845 	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
846 	if (err)
847 		goto out;
848 
849 	err = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
850 	if (err)
851 		goto out;
852 
853 	/*
854 	 * remove the objects in open context, so that we won't
855 	 * have too much to do in syncing context.
856 	 */
857 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
858 	    ds->ds_phys->ds_prev_snap_txg)) {
859 		dmu_tx_t *tx = dmu_tx_create(os);
860 		dmu_tx_hold_free(tx, obj, 0, DMU_OBJECT_END);
861 		dmu_tx_hold_bonus(tx, obj);
862 		err = dmu_tx_assign(tx, TXG_WAIT);
863 		if (err) {
864 			/*
865 			 * Perhaps there is not enough disk
866 			 * space.  Just deal with it from
867 			 * dsl_dataset_destroy_sync().
868 			 */
869 			dmu_tx_abort(tx);
870 			continue;
871 		}
872 		VERIFY(0 == dmu_object_free(os, obj, tx));
873 		dmu_tx_commit(tx);
874 	}
875 	/* Make sure it's not dirty before we finish destroying it. */
876 	txg_wait_synced(dd->dd_pool, 0);
877 
878 	dmu_objset_close(os);
879 	if (err != ESRCH)
880 		goto out;
881 
882 	if (ds->ds_user_ptr) {
883 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
884 		ds->ds_user_ptr = NULL;
885 	}
886 
887 	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
888 	err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
889 	rw_exit(&dd->dd_pool->dp_config_rwlock);
890 
891 	if (err)
892 		goto out;
893 
894 	/*
895 	 * Blow away the dsl_dir + head dataset.
896 	 */
897 	dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
898 	dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
899 	    dsl_dataset_destroy_sync, ds, tag, 0);
900 	dsl_sync_task_create(dstg, dsl_dir_destroy_check,
901 	    dsl_dir_destroy_sync, dd, FTAG, 0);
902 	err = dsl_sync_task_group_wait(dstg);
903 	dsl_sync_task_group_destroy(dstg);
904 	/* if it is successful, *destroy_sync will close the ds+dd */
905 	if (err)
906 		dsl_dir_close(dd, FTAG);
907 out:
908 	if (err)
909 		dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag);
910 	return (err);
911 }
912 
913 int
914 dsl_dataset_rollback(dsl_dataset_t *ds, dmu_objset_type_t ost)
915 {
916 	ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX);
917 
918 	return (dsl_sync_task_do(ds->ds_dir->dd_pool,
919 	    dsl_dataset_rollback_check, dsl_dataset_rollback_sync,
920 	    ds, &ost, 0));
921 }
922 
923 void *
924 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
925     void *p, dsl_dataset_evict_func_t func)
926 {
927 	void *old;
928 
929 	mutex_enter(&ds->ds_lock);
930 	old = ds->ds_user_ptr;
931 	if (old == NULL) {
932 		ds->ds_user_ptr = p;
933 		ds->ds_user_evict_func = func;
934 	}
935 	mutex_exit(&ds->ds_lock);
936 	return (old);
937 }
938 
939 void *
940 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
941 {
942 	return (ds->ds_user_ptr);
943 }
944 
945 
946 blkptr_t *
947 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
948 {
949 	return (&ds->ds_phys->ds_bp);
950 }
951 
952 void
953 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
954 {
955 	ASSERT(dmu_tx_is_syncing(tx));
956 	/* If it's the meta-objset, set dp_meta_rootbp */
957 	if (ds == NULL) {
958 		tx->tx_pool->dp_meta_rootbp = *bp;
959 	} else {
960 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
961 		ds->ds_phys->ds_bp = *bp;
962 	}
963 }
964 
965 spa_t *
966 dsl_dataset_get_spa(dsl_dataset_t *ds)
967 {
968 	return (ds->ds_dir->dd_pool->dp_spa);
969 }
970 
971 void
972 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
973 {
974 	dsl_pool_t *dp;
975 
976 	if (ds == NULL) /* this is the meta-objset */
977 		return;
978 
979 	ASSERT(ds->ds_user_ptr != NULL);
980 
981 	if (ds->ds_phys->ds_next_snap_obj != 0)
982 		panic("dirtying snapshot!");
983 
984 	dp = ds->ds_dir->dd_pool;
985 
986 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
987 		/* up the hold count until we can be written out */
988 		dmu_buf_add_ref(ds->ds_dbuf, ds);
989 	}
990 }
991 
992 /*
993  * The unique space in the head dataset can be calculated by subtracting
994  * the space used in the most recent snapshot, that is still being used
995  * in this file system, from the space currently in use.  To figure out
996  * the space in the most recent snapshot still in use, we need to take
997  * the total space used in the snapshot and subtract out the space that
998  * has been freed up since the snapshot was taken.
999  */
1000 static void
1001 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1002 {
1003 	uint64_t mrs_used;
1004 	uint64_t dlused, dlcomp, dluncomp;
1005 
1006 	ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj);
1007 
1008 	if (ds->ds_phys->ds_prev_snap_obj != 0)
1009 		mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1010 	else
1011 		mrs_used = 0;
1012 
1013 	VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp,
1014 	    &dluncomp));
1015 
1016 	ASSERT3U(dlused, <=, mrs_used);
1017 	ds->ds_phys->ds_unique_bytes =
1018 	    ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1019 
1020 	if (!DS_UNIQUE_IS_ACCURATE(ds) &&
1021 	    spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1022 	    SPA_VERSION_UNIQUE_ACCURATE)
1023 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1024 }
1025 
1026 static uint64_t
1027 dsl_dataset_unique(dsl_dataset_t *ds)
1028 {
1029 	if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds))
1030 		dsl_dataset_recalc_head_uniq(ds);
1031 
1032 	return (ds->ds_phys->ds_unique_bytes);
1033 }
1034 
1035 struct killarg {
1036 	int64_t *usedp;
1037 	int64_t *compressedp;
1038 	int64_t *uncompressedp;
1039 	zio_t *zio;
1040 	dmu_tx_t *tx;
1041 };
1042 
1043 static int
1044 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
1045 {
1046 	struct killarg *ka = arg;
1047 	blkptr_t *bp = &bc->bc_blkptr;
1048 
1049 	ASSERT3U(bc->bc_errno, ==, 0);
1050 
1051 	/*
1052 	 * Since this callback is not called concurrently, no lock is
1053 	 * needed on the accounting values.
1054 	 */
1055 	*ka->usedp += bp_get_dasize(spa, bp);
1056 	*ka->compressedp += BP_GET_PSIZE(bp);
1057 	*ka->uncompressedp += BP_GET_UCSIZE(bp);
1058 	/* XXX check for EIO? */
1059 	(void) arc_free(ka->zio, spa, ka->tx->tx_txg, bp, NULL, NULL,
1060 	    ARC_NOWAIT);
1061 	return (0);
1062 }
1063 
1064 /* ARGSUSED */
1065 static int
1066 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
1067 {
1068 	dsl_dataset_t *ds = arg1;
1069 	dmu_objset_type_t *ost = arg2;
1070 
1071 	/*
1072 	 * We can only roll back to emptyness if it is a ZPL objset.
1073 	 */
1074 	if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0)
1075 		return (EINVAL);
1076 
1077 	/*
1078 	 * This must not be a snapshot.
1079 	 */
1080 	if (ds->ds_phys->ds_next_snap_obj != 0)
1081 		return (EINVAL);
1082 
1083 	/*
1084 	 * If we made changes this txg, traverse_dsl_dataset won't find
1085 	 * them.  Try again.
1086 	 */
1087 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1088 		return (EAGAIN);
1089 
1090 	return (0);
1091 }
1092 
1093 /* ARGSUSED */
1094 static void
1095 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1096 {
1097 	dsl_dataset_t *ds = arg1;
1098 	dmu_objset_type_t *ost = arg2;
1099 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1100 
1101 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1102 
1103 	/*
1104 	 * Before the roll back destroy the zil.
1105 	 */
1106 	if (ds->ds_user_ptr != NULL) {
1107 		zil_rollback_destroy(
1108 		    ((objset_impl_t *)ds->ds_user_ptr)->os_zil, tx);
1109 
1110 		/*
1111 		 * We need to make sure that the objset_impl_t is reopened after
1112 		 * we do the rollback, otherwise it will have the wrong
1113 		 * objset_phys_t.  Normally this would happen when this
1114 		 * DS_MODE_EXCLUSIVE dataset-open is closed, thus causing the
1115 		 * dataset to be immediately evicted.  But when doing "zfs recv
1116 		 * -F", we reopen the objset before that, so that there is no
1117 		 * window where the dataset is closed and inconsistent.
1118 		 */
1119 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1120 		ds->ds_user_ptr = NULL;
1121 	}
1122 
1123 	/* Zero out the deadlist. */
1124 	bplist_close(&ds->ds_deadlist);
1125 	bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1126 	ds->ds_phys->ds_deadlist_obj =
1127 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1128 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1129 	    ds->ds_phys->ds_deadlist_obj));
1130 
1131 	{
1132 		/* Free blkptrs that we gave birth to */
1133 		zio_t *zio;
1134 		int64_t used = 0, compressed = 0, uncompressed = 0;
1135 		struct killarg ka;
1136 
1137 		zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
1138 		    ZIO_FLAG_MUSTSUCCEED);
1139 		ka.usedp = &used;
1140 		ka.compressedp = &compressed;
1141 		ka.uncompressedp = &uncompressed;
1142 		ka.zio = zio;
1143 		ka.tx = tx;
1144 		(void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1145 		    ADVANCE_POST, kill_blkptr, &ka);
1146 		(void) zio_wait(zio);
1147 
1148 		dsl_dir_diduse_space(ds->ds_dir,
1149 		    -used, -compressed, -uncompressed, tx);
1150 	}
1151 
1152 	if (ds->ds_prev) {
1153 		/* Change our contents to that of the prev snapshot */
1154 		ASSERT3U(ds->ds_prev->ds_object, ==,
1155 		    ds->ds_phys->ds_prev_snap_obj);
1156 		ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1157 		ds->ds_phys->ds_used_bytes =
1158 		    ds->ds_prev->ds_phys->ds_used_bytes;
1159 		ds->ds_phys->ds_compressed_bytes =
1160 		    ds->ds_prev->ds_phys->ds_compressed_bytes;
1161 		ds->ds_phys->ds_uncompressed_bytes =
1162 		    ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1163 		ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1164 		ds->ds_phys->ds_unique_bytes = 0;
1165 
1166 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1167 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1168 			ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1169 		}
1170 	} else {
1171 		/* Zero out our contents, recreate objset */
1172 		bzero(&ds->ds_phys->ds_bp, sizeof (blkptr_t));
1173 		ds->ds_phys->ds_used_bytes = 0;
1174 		ds->ds_phys->ds_compressed_bytes = 0;
1175 		ds->ds_phys->ds_uncompressed_bytes = 0;
1176 		ds->ds_phys->ds_flags = 0;
1177 		ds->ds_phys->ds_unique_bytes = 0;
1178 		(void) dmu_objset_create_impl(ds->ds_dir->dd_pool->dp_spa, ds,
1179 		    &ds->ds_phys->ds_bp, *ost, tx);
1180 	}
1181 
1182 	spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa,
1183 	    tx, cr, "dataset = %llu", ds->ds_object);
1184 }
1185 
1186 /* ARGSUSED */
1187 static int
1188 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1189 {
1190 	dsl_dataset_t *ds = arg1;
1191 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1192 	uint64_t count;
1193 	int err;
1194 
1195 	/*
1196 	 * Can't delete a head dataset if there are snapshots of it.
1197 	 * (Except if the only snapshots are from the branch we cloned
1198 	 * from.)
1199 	 */
1200 	if (ds->ds_prev != NULL &&
1201 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1202 		return (EINVAL);
1203 
1204 	/*
1205 	 * This is really a dsl_dir thing, but check it here so that
1206 	 * we'll be less likely to leave this dataset inconsistent &
1207 	 * nearly destroyed.
1208 	 */
1209 	err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1210 	if (err)
1211 		return (err);
1212 	if (count != 0)
1213 		return (EEXIST);
1214 
1215 	return (0);
1216 }
1217 
1218 /* ARGSUSED */
1219 static void
1220 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1221 {
1222 	dsl_dataset_t *ds = arg1;
1223 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1224 
1225 	/* Mark it as inconsistent on-disk, in case we crash */
1226 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1227 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1228 
1229 	spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1230 	    cr, "dataset = %llu", ds->ds_object);
1231 }
1232 
1233 /* ARGSUSED */
1234 int
1235 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1236 {
1237 	dsl_dataset_t *ds = arg1;
1238 
1239 	/* Can't delete a branch point. */
1240 	if (ds->ds_phys->ds_num_children > 1)
1241 		return (EEXIST);
1242 
1243 	/*
1244 	 * Can't delete a head dataset if there are snapshots of it.
1245 	 * (Except if the only snapshots are from the branch we cloned
1246 	 * from.)
1247 	 */
1248 	if (ds->ds_prev != NULL &&
1249 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1250 		return (EINVAL);
1251 
1252 	/*
1253 	 * If we made changes this txg, traverse_dsl_dataset won't find
1254 	 * them.  Try again.
1255 	 */
1256 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1257 		return (EAGAIN);
1258 
1259 	/* XXX we should do some i/o error checking... */
1260 	return (0);
1261 }
1262 
1263 void
1264 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1265 {
1266 	dsl_dataset_t *ds = arg1;
1267 	int64_t used = 0, compressed = 0, uncompressed = 0;
1268 	zio_t *zio;
1269 	int err;
1270 	int after_branch_point = FALSE;
1271 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1272 	objset_t *mos = dp->dp_meta_objset;
1273 	dsl_dataset_t *ds_prev = NULL;
1274 	uint64_t obj;
1275 
1276 	ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX);
1277 	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1278 	ASSERT(ds->ds_prev == NULL ||
1279 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1280 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1281 
1282 	/* Remove our reservation */
1283 	if (ds->ds_reserved != 0) {
1284 		uint64_t val = 0;
1285 		dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1286 		ASSERT3U(ds->ds_reserved, ==, 0);
1287 	}
1288 
1289 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1290 
1291 	obj = ds->ds_object;
1292 
1293 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1294 		if (ds->ds_prev) {
1295 			ds_prev = ds->ds_prev;
1296 		} else {
1297 			VERIFY(0 == dsl_dataset_open_obj(dp,
1298 			    ds->ds_phys->ds_prev_snap_obj, NULL,
1299 			    DS_MODE_NONE, FTAG, &ds_prev));
1300 		}
1301 		after_branch_point =
1302 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1303 
1304 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1305 		if (after_branch_point &&
1306 		    ds->ds_phys->ds_next_snap_obj == 0) {
1307 			/* This clone is toast. */
1308 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1309 			ds_prev->ds_phys->ds_num_children--;
1310 		} else if (!after_branch_point) {
1311 			ds_prev->ds_phys->ds_next_snap_obj =
1312 			    ds->ds_phys->ds_next_snap_obj;
1313 		}
1314 	}
1315 
1316 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1317 
1318 	if (ds->ds_phys->ds_next_snap_obj != 0) {
1319 		blkptr_t bp;
1320 		dsl_dataset_t *ds_next;
1321 		uint64_t itor = 0;
1322 		uint64_t old_unique;
1323 
1324 		spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1325 
1326 		VERIFY(0 == dsl_dataset_open_obj(dp,
1327 		    ds->ds_phys->ds_next_snap_obj, NULL,
1328 		    DS_MODE_NONE, FTAG, &ds_next));
1329 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1330 
1331 		old_unique = dsl_dataset_unique(ds_next);
1332 
1333 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1334 		ds_next->ds_phys->ds_prev_snap_obj =
1335 		    ds->ds_phys->ds_prev_snap_obj;
1336 		ds_next->ds_phys->ds_prev_snap_txg =
1337 		    ds->ds_phys->ds_prev_snap_txg;
1338 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1339 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1340 
1341 		/*
1342 		 * Transfer to our deadlist (which will become next's
1343 		 * new deadlist) any entries from next's current
1344 		 * deadlist which were born before prev, and free the
1345 		 * other entries.
1346 		 *
1347 		 * XXX we're doing this long task with the config lock held
1348 		 */
1349 		while (bplist_iterate(&ds_next->ds_deadlist, &itor,
1350 		    &bp) == 0) {
1351 			if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1352 				VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1353 				    &bp, tx));
1354 				if (ds_prev && !after_branch_point &&
1355 				    bp.blk_birth >
1356 				    ds_prev->ds_phys->ds_prev_snap_txg) {
1357 					ds_prev->ds_phys->ds_unique_bytes +=
1358 					    bp_get_dasize(dp->dp_spa, &bp);
1359 				}
1360 			} else {
1361 				used += bp_get_dasize(dp->dp_spa, &bp);
1362 				compressed += BP_GET_PSIZE(&bp);
1363 				uncompressed += BP_GET_UCSIZE(&bp);
1364 				/* XXX check return value? */
1365 				(void) arc_free(zio, dp->dp_spa, tx->tx_txg,
1366 				    &bp, NULL, NULL, ARC_NOWAIT);
1367 			}
1368 		}
1369 
1370 		/* free next's deadlist */
1371 		bplist_close(&ds_next->ds_deadlist);
1372 		bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1373 
1374 		/* set next's deadlist to our deadlist */
1375 		ds_next->ds_phys->ds_deadlist_obj =
1376 		    ds->ds_phys->ds_deadlist_obj;
1377 		VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1378 		    ds_next->ds_phys->ds_deadlist_obj));
1379 		ds->ds_phys->ds_deadlist_obj = 0;
1380 
1381 		if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1382 			/*
1383 			 * Update next's unique to include blocks which
1384 			 * were previously shared by only this snapshot
1385 			 * and it.  Those blocks will be born after the
1386 			 * prev snap and before this snap, and will have
1387 			 * died after the next snap and before the one
1388 			 * after that (ie. be on the snap after next's
1389 			 * deadlist).
1390 			 *
1391 			 * XXX we're doing this long task with the
1392 			 * config lock held
1393 			 */
1394 			dsl_dataset_t *ds_after_next;
1395 
1396 			VERIFY(0 == dsl_dataset_open_obj(dp,
1397 			    ds_next->ds_phys->ds_next_snap_obj, NULL,
1398 			    DS_MODE_NONE, FTAG, &ds_after_next));
1399 			itor = 0;
1400 			while (bplist_iterate(&ds_after_next->ds_deadlist,
1401 			    &itor, &bp) == 0) {
1402 				if (bp.blk_birth >
1403 				    ds->ds_phys->ds_prev_snap_txg &&
1404 				    bp.blk_birth <=
1405 				    ds->ds_phys->ds_creation_txg) {
1406 					ds_next->ds_phys->ds_unique_bytes +=
1407 					    bp_get_dasize(dp->dp_spa, &bp);
1408 				}
1409 			}
1410 
1411 			dsl_dataset_close(ds_after_next, DS_MODE_NONE, FTAG);
1412 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1413 		} else {
1414 			ASSERT3P(ds_next->ds_prev, ==, ds);
1415 			dsl_dataset_close(ds_next->ds_prev, DS_MODE_NONE,
1416 			    ds_next);
1417 			if (ds_prev) {
1418 				VERIFY(0 == dsl_dataset_open_obj(dp,
1419 				    ds->ds_phys->ds_prev_snap_obj, NULL,
1420 				    DS_MODE_NONE, ds_next, &ds_next->ds_prev));
1421 			} else {
1422 				ds_next->ds_prev = NULL;
1423 			}
1424 
1425 			dsl_dataset_recalc_head_uniq(ds_next);
1426 
1427 			/*
1428 			 * Reduce the amount of our unconsmed refreservation
1429 			 * being charged to our parent by the amount of
1430 			 * new unique data we have gained.
1431 			 */
1432 			if (old_unique < ds_next->ds_reserved) {
1433 				int64_t mrsdelta;
1434 				uint64_t new_unique =
1435 				    ds_next->ds_phys->ds_unique_bytes;
1436 
1437 				ASSERT(old_unique <= new_unique);
1438 				mrsdelta = MIN(new_unique - old_unique,
1439 				    ds_next->ds_reserved - old_unique);
1440 				dsl_dir_diduse_space(ds->ds_dir, -mrsdelta,
1441 				    0, 0, tx);
1442 			}
1443 		}
1444 		dsl_dataset_close(ds_next, DS_MODE_NONE, FTAG);
1445 
1446 		/*
1447 		 * NB: unique_bytes might not be accurate for the head objset.
1448 		 * Before SPA_VERSION 9, we didn't update its value when we
1449 		 * deleted the most recent snapshot.
1450 		 */
1451 		ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1452 	} else {
1453 		/*
1454 		 * There's no next snapshot, so this is a head dataset.
1455 		 * Destroy the deadlist.  Unless it's a clone, the
1456 		 * deadlist should be empty.  (If it's a clone, it's
1457 		 * safe to ignore the deadlist contents.)
1458 		 */
1459 		struct killarg ka;
1460 
1461 		ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1462 		bplist_close(&ds->ds_deadlist);
1463 		bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1464 		ds->ds_phys->ds_deadlist_obj = 0;
1465 
1466 		/*
1467 		 * Free everything that we point to (that's born after
1468 		 * the previous snapshot, if we are a clone)
1469 		 *
1470 		 * XXX we're doing this long task with the config lock held
1471 		 */
1472 		ka.usedp = &used;
1473 		ka.compressedp = &compressed;
1474 		ka.uncompressedp = &uncompressed;
1475 		ka.zio = zio;
1476 		ka.tx = tx;
1477 		err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1478 		    ADVANCE_POST, kill_blkptr, &ka);
1479 		ASSERT3U(err, ==, 0);
1480 		ASSERT(spa_version(dp->dp_spa) <
1481 		    SPA_VERSION_UNIQUE_ACCURATE ||
1482 		    used == ds->ds_phys->ds_unique_bytes);
1483 	}
1484 
1485 	err = zio_wait(zio);
1486 	ASSERT3U(err, ==, 0);
1487 
1488 	dsl_dir_diduse_space(ds->ds_dir, -used, -compressed, -uncompressed, tx);
1489 
1490 	if (ds->ds_phys->ds_snapnames_zapobj) {
1491 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1492 		ASSERT(err == 0);
1493 	}
1494 
1495 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1496 		/* Erase the link in the dataset */
1497 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1498 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1499 		/*
1500 		 * dsl_dir_sync_destroy() called us, they'll destroy
1501 		 * the dataset.
1502 		 */
1503 	} else {
1504 		/* remove from snapshot namespace */
1505 		dsl_dataset_t *ds_head;
1506 		VERIFY(0 == dsl_dataset_open_obj(dp,
1507 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, NULL,
1508 		    DS_MODE_NONE, FTAG, &ds_head));
1509 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1510 #ifdef ZFS_DEBUG
1511 		{
1512 			uint64_t val;
1513 			err = zap_lookup(mos,
1514 			    ds_head->ds_phys->ds_snapnames_zapobj,
1515 			    ds->ds_snapname, 8, 1, &val);
1516 			ASSERT3U(err, ==, 0);
1517 			ASSERT3U(val, ==, obj);
1518 		}
1519 #endif
1520 		err = zap_remove(mos, ds_head->ds_phys->ds_snapnames_zapobj,
1521 		    ds->ds_snapname, tx);
1522 		ASSERT(err == 0);
1523 		dsl_dataset_close(ds_head, DS_MODE_NONE, FTAG);
1524 	}
1525 
1526 	if (ds_prev && ds->ds_prev != ds_prev)
1527 		dsl_dataset_close(ds_prev, DS_MODE_NONE, FTAG);
1528 
1529 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1530 	spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1531 	    cr, "dataset = %llu", ds->ds_object);
1532 
1533 	dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag);
1534 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1535 
1536 }
1537 
1538 static int
1539 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1540 {
1541 	uint64_t asize;
1542 
1543 	if (!dmu_tx_is_syncing(tx))
1544 		return (0);
1545 
1546 	/*
1547 	 * If there's an fs-only reservation, any blocks that might become
1548 	 * owned by the snapshot dataset must be accommodated by space
1549 	 * outside of the reservation.
1550 	 */
1551 	asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1552 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1553 		return (ENOSPC);
1554 
1555 	/*
1556 	 * Propogate any reserved space for this snapshot to other
1557 	 * snapshot checks in this sync group.
1558 	 */
1559 	if (asize > 0)
1560 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1561 
1562 	return (0);
1563 }
1564 
1565 /* ARGSUSED */
1566 int
1567 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1568 {
1569 	dsl_dataset_t *ds = arg1;
1570 	const char *snapname = arg2;
1571 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1572 	int err;
1573 	uint64_t value;
1574 
1575 	/*
1576 	 * We don't allow multiple snapshots of the same txg.  If there
1577 	 * is already one, try again.
1578 	 */
1579 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1580 		return (EAGAIN);
1581 
1582 	/*
1583 	 * Check for conflicting name snapshot name.
1584 	 */
1585 	err = zap_lookup(mos, ds->ds_phys->ds_snapnames_zapobj,
1586 	    snapname, 8, 1, &value);
1587 	if (err == 0)
1588 		return (EEXIST);
1589 	if (err != ENOENT)
1590 		return (err);
1591 
1592 	/*
1593 	 * Check that the dataset's name is not too long.  Name consists
1594 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1595 	 */
1596 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1597 		return (ENAMETOOLONG);
1598 
1599 	err = dsl_dataset_snapshot_reserve_space(ds, tx);
1600 	if (err)
1601 		return (err);
1602 
1603 	ds->ds_trysnap_txg = tx->tx_txg;
1604 	return (0);
1605 }
1606 
1607 void
1608 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1609 {
1610 	dsl_dataset_t *ds = arg1;
1611 	const char *snapname = arg2;
1612 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1613 	dmu_buf_t *dbuf;
1614 	dsl_dataset_phys_t *dsphys;
1615 	uint64_t dsobj;
1616 	objset_t *mos = dp->dp_meta_objset;
1617 	int err;
1618 
1619 	spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1620 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1621 
1622 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1623 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1624 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1625 	dmu_buf_will_dirty(dbuf, tx);
1626 	dsphys = dbuf->db_data;
1627 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1628 	dsphys->ds_fsid_guid = unique_create();
1629 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1630 	    sizeof (dsphys->ds_guid));
1631 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1632 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1633 	dsphys->ds_next_snap_obj = ds->ds_object;
1634 	dsphys->ds_num_children = 1;
1635 	dsphys->ds_creation_time = gethrestime_sec();
1636 	dsphys->ds_creation_txg = tx->tx_txg;
1637 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1638 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1639 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1640 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1641 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1642 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1643 	dmu_buf_rele(dbuf, FTAG);
1644 
1645 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1646 	if (ds->ds_prev) {
1647 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1648 		    ds->ds_object ||
1649 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1650 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1651 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1652 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1653 			    ds->ds_prev->ds_phys->ds_creation_txg);
1654 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1655 		}
1656 	}
1657 
1658 	/*
1659 	 * If we have a reference-reservation on this dataset, we will
1660 	 * need to increase the amount of refreservation being charged
1661 	 * since our unique space is going to zero.
1662 	 */
1663 	if (ds->ds_reserved) {
1664 		int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1665 		dsl_dir_diduse_space(ds->ds_dir, add, 0, 0, tx);
1666 	}
1667 
1668 	bplist_close(&ds->ds_deadlist);
1669 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1670 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, dsphys->ds_creation_txg);
1671 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1672 	ds->ds_phys->ds_prev_snap_txg = dsphys->ds_creation_txg;
1673 	ds->ds_phys->ds_unique_bytes = 0;
1674 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1675 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1676 	ds->ds_phys->ds_deadlist_obj =
1677 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1678 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1679 	    ds->ds_phys->ds_deadlist_obj));
1680 
1681 	dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1682 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1683 	    snapname, 8, 1, &dsobj, tx);
1684 	ASSERT(err == 0);
1685 
1686 	if (ds->ds_prev)
1687 		dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds);
1688 	VERIFY(0 == dsl_dataset_open_obj(dp,
1689 	    ds->ds_phys->ds_prev_snap_obj, snapname,
1690 	    DS_MODE_NONE, ds, &ds->ds_prev));
1691 
1692 	spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1693 	    "dataset = %llu", dsobj);
1694 }
1695 
1696 void
1697 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1698 {
1699 	ASSERT(dmu_tx_is_syncing(tx));
1700 	ASSERT(ds->ds_user_ptr != NULL);
1701 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1702 
1703 	/*
1704 	 * in case we had to change ds_fsid_guid when we opened it,
1705 	 * sync it out now.
1706 	 */
1707 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1708 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1709 
1710 	dsl_dir_dirty(ds->ds_dir, tx);
1711 	dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1712 }
1713 
1714 void
1715 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1716 {
1717 	uint64_t refd, avail, uobjs, aobjs;
1718 
1719 	dsl_dir_stats(ds->ds_dir, nv);
1720 
1721 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1722 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1723 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1724 
1725 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1726 	    ds->ds_phys->ds_creation_time);
1727 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1728 	    ds->ds_phys->ds_creation_txg);
1729 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1730 	    ds->ds_quota);
1731 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1732 	    ds->ds_reserved);
1733 
1734 	if (ds->ds_phys->ds_next_snap_obj) {
1735 		/*
1736 		 * This is a snapshot; override the dd's space used with
1737 		 * our unique space and compression ratio.
1738 		 */
1739 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1740 		    ds->ds_phys->ds_unique_bytes);
1741 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1742 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1743 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1744 		    ds->ds_phys->ds_compressed_bytes));
1745 	}
1746 }
1747 
1748 void
1749 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1750 {
1751 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1752 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1753 	stat->dds_guid = ds->ds_phys->ds_guid;
1754 	if (ds->ds_phys->ds_next_snap_obj) {
1755 		stat->dds_is_snapshot = B_TRUE;
1756 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1757 	}
1758 
1759 	/* clone origin is really a dsl_dir thing... */
1760 	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1761 	if (ds->ds_dir->dd_phys->dd_origin_obj) {
1762 		dsl_dataset_t *ods;
1763 
1764 		VERIFY(0 == dsl_dataset_open_obj(ds->ds_dir->dd_pool,
1765 		    ds->ds_dir->dd_phys->dd_origin_obj,
1766 		    NULL, DS_MODE_NONE, FTAG, &ods));
1767 		dsl_dataset_name(ods, stat->dds_origin);
1768 		dsl_dataset_close(ods, DS_MODE_NONE, FTAG);
1769 	}
1770 	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1771 }
1772 
1773 uint64_t
1774 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1775 {
1776 	return (ds->ds_fsid_guid);
1777 }
1778 
1779 void
1780 dsl_dataset_space(dsl_dataset_t *ds,
1781     uint64_t *refdbytesp, uint64_t *availbytesp,
1782     uint64_t *usedobjsp, uint64_t *availobjsp)
1783 {
1784 	*refdbytesp = ds->ds_phys->ds_used_bytes;
1785 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1786 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1787 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1788 	if (ds->ds_quota != 0) {
1789 		/*
1790 		 * Adjust available bytes according to refquota
1791 		 */
1792 		if (*refdbytesp < ds->ds_quota)
1793 			*availbytesp = MIN(*availbytesp,
1794 			    ds->ds_quota - *refdbytesp);
1795 		else
1796 			*availbytesp = 0;
1797 	}
1798 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1799 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1800 }
1801 
1802 boolean_t
1803 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
1804 {
1805 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1806 
1807 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
1808 	    dsl_pool_sync_context(dp));
1809 	if (ds->ds_prev == NULL)
1810 		return (B_FALSE);
1811 	if (ds->ds_phys->ds_bp.blk_birth >
1812 	    ds->ds_prev->ds_phys->ds_creation_txg)
1813 		return (B_TRUE);
1814 	return (B_FALSE);
1815 }
1816 
1817 /* ARGSUSED */
1818 static int
1819 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1820 {
1821 	dsl_dataset_t *ds = arg1;
1822 	char *newsnapname = arg2;
1823 	dsl_dir_t *dd = ds->ds_dir;
1824 	objset_t *mos = dd->dd_pool->dp_meta_objset;
1825 	dsl_dataset_t *hds;
1826 	uint64_t val;
1827 	int err;
1828 
1829 	err = dsl_dataset_open_obj(dd->dd_pool,
1830 	    dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds);
1831 	if (err)
1832 		return (err);
1833 
1834 	/* new name better not be in use */
1835 	err = zap_lookup(mos, hds->ds_phys->ds_snapnames_zapobj,
1836 	    newsnapname, 8, 1, &val);
1837 	dsl_dataset_close(hds, DS_MODE_NONE, FTAG);
1838 
1839 	if (err == 0)
1840 		err = EEXIST;
1841 	else if (err == ENOENT)
1842 		err = 0;
1843 
1844 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1845 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
1846 		err = ENAMETOOLONG;
1847 
1848 	return (err);
1849 }
1850 
1851 static void
1852 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
1853     cred_t *cr, dmu_tx_t *tx)
1854 {
1855 	dsl_dataset_t *ds = arg1;
1856 	const char *newsnapname = arg2;
1857 	dsl_dir_t *dd = ds->ds_dir;
1858 	objset_t *mos = dd->dd_pool->dp_meta_objset;
1859 	dsl_dataset_t *hds;
1860 	int err;
1861 
1862 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
1863 
1864 	VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool,
1865 	    dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds));
1866 
1867 	VERIFY(0 == dsl_dataset_get_snapname(ds));
1868 	err = zap_remove(mos, hds->ds_phys->ds_snapnames_zapobj,
1869 	    ds->ds_snapname, tx);
1870 	ASSERT3U(err, ==, 0);
1871 	mutex_enter(&ds->ds_lock);
1872 	(void) strcpy(ds->ds_snapname, newsnapname);
1873 	mutex_exit(&ds->ds_lock);
1874 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
1875 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
1876 	ASSERT3U(err, ==, 0);
1877 
1878 	spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
1879 	    cr, "dataset = %llu", ds->ds_object);
1880 	dsl_dataset_close(hds, DS_MODE_NONE, FTAG);
1881 }
1882 
1883 struct renamesnaparg {
1884 	dsl_sync_task_group_t *dstg;
1885 	char failed[MAXPATHLEN];
1886 	char *oldsnap;
1887 	char *newsnap;
1888 };
1889 
1890 static int
1891 dsl_snapshot_rename_one(char *name, void *arg)
1892 {
1893 	struct renamesnaparg *ra = arg;
1894 	dsl_dataset_t *ds = NULL;
1895 	char *cp;
1896 	int err;
1897 
1898 	cp = name + strlen(name);
1899 	*cp = '@';
1900 	(void) strcpy(cp + 1, ra->oldsnap);
1901 
1902 	/*
1903 	 * For recursive snapshot renames the parent won't be changing
1904 	 * so we just pass name for both the to/from argument.
1905 	 */
1906 	if (err = zfs_secpolicy_rename_perms(name, name, CRED())) {
1907 		(void) strcpy(ra->failed, name);
1908 		return (err);
1909 	}
1910 
1911 	err = dsl_dataset_open(name, DS_MODE_READONLY | DS_MODE_STANDARD,
1912 	    ra->dstg, &ds);
1913 	if (err == ENOENT) {
1914 		*cp = '\0';
1915 		return (0);
1916 	}
1917 	if (err) {
1918 		(void) strcpy(ra->failed, name);
1919 		*cp = '\0';
1920 		dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg);
1921 		return (err);
1922 	}
1923 
1924 #ifdef _KERNEL
1925 	/* for all filesystems undergoing rename, we'll need to unmount it */
1926 	(void) zfs_unmount_snap(name, NULL);
1927 #endif
1928 
1929 	*cp = '\0';
1930 
1931 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
1932 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
1933 
1934 	return (0);
1935 }
1936 
1937 static int
1938 dsl_recursive_rename(char *oldname, const char *newname)
1939 {
1940 	int err;
1941 	struct renamesnaparg *ra;
1942 	dsl_sync_task_t *dst;
1943 	spa_t *spa;
1944 	char *cp, *fsname = spa_strdup(oldname);
1945 	int len = strlen(oldname);
1946 
1947 	/* truncate the snapshot name to get the fsname */
1948 	cp = strchr(fsname, '@');
1949 	*cp = '\0';
1950 
1951 	err = spa_open(fsname, &spa, FTAG);
1952 	if (err) {
1953 		kmem_free(fsname, len + 1);
1954 		return (err);
1955 	}
1956 	ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
1957 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
1958 
1959 	ra->oldsnap = strchr(oldname, '@') + 1;
1960 	ra->newsnap = strchr(newname, '@') + 1;
1961 	*ra->failed = '\0';
1962 
1963 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
1964 	    DS_FIND_CHILDREN);
1965 	kmem_free(fsname, len + 1);
1966 
1967 	if (err == 0) {
1968 		err = dsl_sync_task_group_wait(ra->dstg);
1969 	}
1970 
1971 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
1972 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
1973 		dsl_dataset_t *ds = dst->dst_arg1;
1974 		if (dst->dst_err) {
1975 			dsl_dir_name(ds->ds_dir, ra->failed);
1976 			(void) strcat(ra->failed, "@");
1977 			(void) strcat(ra->failed, ra->newsnap);
1978 		}
1979 		dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg);
1980 	}
1981 
1982 	if (err)
1983 		(void) strcpy(oldname, ra->failed);
1984 
1985 	dsl_sync_task_group_destroy(ra->dstg);
1986 	kmem_free(ra, sizeof (struct renamesnaparg));
1987 	spa_close(spa, FTAG);
1988 	return (err);
1989 }
1990 
1991 static int
1992 dsl_valid_rename(char *oldname, void *arg)
1993 {
1994 	int delta = *(int *)arg;
1995 
1996 	if (strlen(oldname) + delta >= MAXNAMELEN)
1997 		return (ENAMETOOLONG);
1998 
1999 	return (0);
2000 }
2001 
2002 #pragma weak dmu_objset_rename = dsl_dataset_rename
2003 int
2004 dsl_dataset_rename(char *oldname, const char *newname,
2005     boolean_t recursive)
2006 {
2007 	dsl_dir_t *dd;
2008 	dsl_dataset_t *ds;
2009 	const char *tail;
2010 	int err;
2011 
2012 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2013 	if (err)
2014 		return (err);
2015 	if (tail == NULL) {
2016 		int delta = strlen(newname) - strlen(oldname);
2017 
2018 		/* if we're growing, validate child size lengths */
2019 		if (delta > 0)
2020 			err = dmu_objset_find(oldname, dsl_valid_rename,
2021 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2022 
2023 		if (!err)
2024 			err = dsl_dir_rename(dd, newname);
2025 		dsl_dir_close(dd, FTAG);
2026 		return (err);
2027 	}
2028 	if (tail[0] != '@') {
2029 		/* the name ended in a nonexistant component */
2030 		dsl_dir_close(dd, FTAG);
2031 		return (ENOENT);
2032 	}
2033 
2034 	dsl_dir_close(dd, FTAG);
2035 
2036 	/* new name must be snapshot in same filesystem */
2037 	tail = strchr(newname, '@');
2038 	if (tail == NULL)
2039 		return (EINVAL);
2040 	tail++;
2041 	if (strncmp(oldname, newname, tail - newname) != 0)
2042 		return (EXDEV);
2043 
2044 	if (recursive) {
2045 		err = dsl_recursive_rename(oldname, newname);
2046 	} else {
2047 		err = dsl_dataset_open(oldname,
2048 		    DS_MODE_READONLY | DS_MODE_STANDARD, FTAG, &ds);
2049 		if (err)
2050 			return (err);
2051 
2052 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2053 		    dsl_dataset_snapshot_rename_check,
2054 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2055 
2056 		dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);
2057 	}
2058 
2059 	return (err);
2060 }
2061 
2062 struct promotearg {
2063 	uint64_t used, comp, uncomp, unique;
2064 	uint64_t newnext_obj, snapnames_obj;
2065 };
2066 
2067 /* ARGSUSED */
2068 static int
2069 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2070 {
2071 	dsl_dataset_t *hds = arg1;
2072 	struct promotearg *pa = arg2;
2073 	dsl_dir_t *dd = hds->ds_dir;
2074 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2075 	dsl_dir_t *odd = NULL;
2076 	dsl_dataset_t *ds = NULL;
2077 	dsl_dataset_t *origin_ds = NULL;
2078 	dsl_dataset_t *newnext_ds = NULL;
2079 	int err;
2080 	char *name = NULL;
2081 	uint64_t itor = 0;
2082 	blkptr_t bp;
2083 
2084 	bzero(pa, sizeof (*pa));
2085 
2086 	/* Check that it is a clone */
2087 	if (dd->dd_phys->dd_origin_obj == 0)
2088 		return (EINVAL);
2089 
2090 	/* Since this is so expensive, don't do the preliminary check */
2091 	if (!dmu_tx_is_syncing(tx))
2092 		return (0);
2093 
2094 	if (err = dsl_dataset_open_obj(dp, dd->dd_phys->dd_origin_obj,
2095 	    NULL, DS_MODE_EXCLUSIVE, FTAG, &origin_ds))
2096 		goto out;
2097 	odd = origin_ds->ds_dir;
2098 
2099 	{
2100 		dsl_dataset_t *phds;
2101 		if (err = dsl_dataset_open_obj(dd->dd_pool,
2102 		    odd->dd_phys->dd_head_dataset_obj,
2103 		    NULL, DS_MODE_NONE, FTAG, &phds))
2104 			goto out;
2105 		pa->snapnames_obj = phds->ds_phys->ds_snapnames_zapobj;
2106 		dsl_dataset_close(phds, DS_MODE_NONE, FTAG);
2107 	}
2108 
2109 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
2110 		err = EXDEV;
2111 		goto out;
2112 	}
2113 
2114 	/* find origin's new next ds */
2115 	VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, hds->ds_object,
2116 	    NULL, DS_MODE_NONE, FTAG, &newnext_ds));
2117 	while (newnext_ds->ds_phys->ds_prev_snap_obj != origin_ds->ds_object) {
2118 		dsl_dataset_t *prev;
2119 
2120 		if (err = dsl_dataset_open_obj(dd->dd_pool,
2121 		    newnext_ds->ds_phys->ds_prev_snap_obj,
2122 		    NULL, DS_MODE_NONE, FTAG, &prev))
2123 			goto out;
2124 		dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG);
2125 		newnext_ds = prev;
2126 	}
2127 	pa->newnext_obj = newnext_ds->ds_object;
2128 
2129 	/* compute origin's new unique space */
2130 	while ((err = bplist_iterate(&newnext_ds->ds_deadlist,
2131 	    &itor, &bp)) == 0) {
2132 		if (bp.blk_birth > origin_ds->ds_phys->ds_prev_snap_txg)
2133 			pa->unique += bp_get_dasize(dd->dd_pool->dp_spa, &bp);
2134 	}
2135 	if (err != ENOENT)
2136 		goto out;
2137 
2138 	/* Walk the snapshots that we are moving */
2139 	name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2140 	ds = origin_ds;
2141 	/* CONSTCOND */
2142 	while (TRUE) {
2143 		uint64_t val, dlused, dlcomp, dluncomp;
2144 		dsl_dataset_t *prev;
2145 
2146 		/* Check that the snapshot name does not conflict */
2147 		dsl_dataset_name(ds, name);
2148 		err = zap_lookup(dd->dd_pool->dp_meta_objset,
2149 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2150 		    8, 1, &val);
2151 		if (err != ENOENT) {
2152 			if (err == 0)
2153 				err = EEXIST;
2154 			goto out;
2155 		}
2156 
2157 		/*
2158 		 * compute space to transfer.  Each snapshot gave birth to:
2159 		 * (my used) - (prev's used) + (deadlist's used)
2160 		 */
2161 		pa->used += ds->ds_phys->ds_used_bytes;
2162 		pa->comp += ds->ds_phys->ds_compressed_bytes;
2163 		pa->uncomp += ds->ds_phys->ds_uncompressed_bytes;
2164 
2165 		/* If we reach the first snapshot, we're done. */
2166 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2167 			break;
2168 
2169 		if (err = bplist_space(&ds->ds_deadlist,
2170 		    &dlused, &dlcomp, &dluncomp))
2171 			goto out;
2172 		if (err = dsl_dataset_open_obj(dd->dd_pool,
2173 		    ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE,
2174 		    FTAG, &prev))
2175 			goto out;
2176 		pa->used += dlused - prev->ds_phys->ds_used_bytes;
2177 		pa->comp += dlcomp - prev->ds_phys->ds_compressed_bytes;
2178 		pa->uncomp += dluncomp - prev->ds_phys->ds_uncompressed_bytes;
2179 
2180 		/*
2181 		 * We could be a clone of a clone.  If we reach our
2182 		 * parent's branch point, we're done.
2183 		 */
2184 		if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
2185 			dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG);
2186 			break;
2187 		}
2188 		if (ds != origin_ds)
2189 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2190 		ds = prev;
2191 	}
2192 
2193 	/* Check that there is enough space here */
2194 	err = dsl_dir_transfer_possible(odd, dd, pa->used);
2195 
2196 out:
2197 	if (ds && ds != origin_ds)
2198 		dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2199 	if (origin_ds)
2200 		dsl_dataset_close(origin_ds, DS_MODE_EXCLUSIVE, FTAG);
2201 	if (newnext_ds)
2202 		dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG);
2203 	if (name)
2204 		kmem_free(name, MAXPATHLEN);
2205 	return (err);
2206 }
2207 
2208 static void
2209 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2210 {
2211 	dsl_dataset_t *hds = arg1;
2212 	struct promotearg *pa = arg2;
2213 	dsl_dir_t *dd = hds->ds_dir;
2214 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2215 	dsl_dir_t *odd = NULL;
2216 	dsl_dataset_t *ds, *origin_ds;
2217 	char *name;
2218 
2219 	ASSERT(dd->dd_phys->dd_origin_obj != 0);
2220 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2221 
2222 	VERIFY(0 == dsl_dataset_open_obj(dp, dd->dd_phys->dd_origin_obj,
2223 	    NULL, DS_MODE_EXCLUSIVE, FTAG, &origin_ds));
2224 	/*
2225 	 * We need to explicitly open odd, since origin_ds's dd will be
2226 	 * changing.
2227 	 */
2228 	VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2229 	    NULL, FTAG, &odd));
2230 
2231 	/* move snapshots to this dir */
2232 	name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2233 	ds = origin_ds;
2234 	/* CONSTCOND */
2235 	while (TRUE) {
2236 		dsl_dataset_t *prev;
2237 
2238 		/* move snap name entry */
2239 		dsl_dataset_name(ds, name);
2240 		VERIFY(0 == zap_remove(dp->dp_meta_objset,
2241 		    pa->snapnames_obj, ds->ds_snapname, tx));
2242 		VERIFY(0 == zap_add(dp->dp_meta_objset,
2243 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2244 		    8, 1, &ds->ds_object, tx));
2245 
2246 		/* change containing dsl_dir */
2247 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2248 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2249 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2250 		ASSERT3P(ds->ds_dir, ==, odd);
2251 		dsl_dir_close(ds->ds_dir, ds);
2252 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2253 		    NULL, ds, &ds->ds_dir));
2254 
2255 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2256 
2257 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2258 			break;
2259 
2260 		VERIFY(0 == dsl_dataset_open_obj(dp,
2261 		    ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE,
2262 		    FTAG, &prev));
2263 
2264 		if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
2265 			dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG);
2266 			break;
2267 		}
2268 		if (ds != origin_ds)
2269 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2270 		ds = prev;
2271 	}
2272 	if (ds != origin_ds)
2273 		dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2274 
2275 	/* change origin's next snap */
2276 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2277 	origin_ds->ds_phys->ds_next_snap_obj = pa->newnext_obj;
2278 
2279 	/* change origin */
2280 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2281 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2282 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2283 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2284 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2285 
2286 	/* change space accounting */
2287 	dsl_dir_diduse_space(odd, -pa->used, -pa->comp, -pa->uncomp, tx);
2288 	dsl_dir_diduse_space(dd, pa->used, pa->comp, pa->uncomp, tx);
2289 	origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2290 
2291 	/* log history record */
2292 	spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2293 	    cr, "dataset = %llu", ds->ds_object);
2294 
2295 	dsl_dir_close(odd, FTAG);
2296 	dsl_dataset_close(origin_ds, DS_MODE_EXCLUSIVE, FTAG);
2297 	kmem_free(name, MAXPATHLEN);
2298 }
2299 
2300 int
2301 dsl_dataset_promote(const char *name)
2302 {
2303 	dsl_dataset_t *ds;
2304 	int err;
2305 	dmu_object_info_t doi;
2306 	struct promotearg pa;
2307 
2308 	err = dsl_dataset_open(name, DS_MODE_NONE, FTAG, &ds);
2309 	if (err)
2310 		return (err);
2311 
2312 	err = dmu_object_info(ds->ds_dir->dd_pool->dp_meta_objset,
2313 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2314 	if (err) {
2315 		dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2316 		return (err);
2317 	}
2318 
2319 	/*
2320 	 * Add in 128x the snapnames zapobj size, since we will be moving
2321 	 * a bunch of snapnames to the promoted ds, and dirtying their
2322 	 * bonus buffers.
2323 	 */
2324 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2325 	    dsl_dataset_promote_check,
2326 	    dsl_dataset_promote_sync, ds, &pa, 2 + 2 * doi.doi_physical_blks);
2327 	dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2328 	return (err);
2329 }
2330 
2331 struct cloneswaparg {
2332 	dsl_dataset_t *cds; /* clone dataset */
2333 	dsl_dataset_t *ohds; /* origin's head dataset */
2334 	boolean_t force;
2335 };
2336 
2337 /* ARGSUSED */
2338 static int
2339 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2340 {
2341 	struct cloneswaparg *csa = arg1;
2342 
2343 	if (csa->ohds->ds_reserved != 0)
2344 		return (EINVAL);
2345 
2346 	/* they should both be heads */
2347 	if (dsl_dataset_is_snapshot(csa->cds) ||
2348 	    dsl_dataset_is_snapshot(csa->ohds))
2349 		return (EINVAL);
2350 
2351 	/* the branch point should be just before them */
2352 	if (csa->cds->ds_prev != csa->ohds->ds_prev)
2353 		return (EINVAL);
2354 
2355 	/* cds should be the clone */
2356 	if (csa->cds->ds_prev->ds_phys->ds_next_snap_obj !=
2357 	    csa->ohds->ds_object)
2358 		return (EINVAL);
2359 
2360 	/* the clone should be a child of the origin */
2361 	if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2362 		return (EINVAL);
2363 
2364 	/* ohds shouldn't be modified unless 'force' */
2365 	if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2366 		return (ETXTBSY);
2367 	return (0);
2368 }
2369 
2370 /* ARGSUSED */
2371 static void
2372 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2373 {
2374 	struct cloneswaparg *csa = arg1;
2375 	dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2376 	uint64_t itor = 0;
2377 	blkptr_t bp;
2378 	uint64_t unique = 0;
2379 	int err;
2380 
2381 	dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2382 	dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2383 	dmu_buf_will_dirty(csa->cds->ds_prev->ds_dbuf, tx);
2384 
2385 	if (csa->cds->ds_user_ptr != NULL) {
2386 		csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr);
2387 		csa->cds->ds_user_ptr = NULL;
2388 	}
2389 
2390 	if (csa->ohds->ds_user_ptr != NULL) {
2391 		csa->ohds->ds_user_evict_func(csa->ohds,
2392 		    csa->ohds->ds_user_ptr);
2393 		csa->ohds->ds_user_ptr = NULL;
2394 	}
2395 
2396 	/* compute unique space */
2397 	while ((err = bplist_iterate(&csa->cds->ds_deadlist,
2398 	    &itor, &bp)) == 0) {
2399 		if (bp.blk_birth > csa->cds->ds_prev->ds_phys->ds_prev_snap_txg)
2400 			unique += bp_get_dasize(dp->dp_spa, &bp);
2401 	}
2402 	VERIFY(err == ENOENT);
2403 
2404 	/* undo any accounting due to a refreservation */
2405 	if (csa->ohds->ds_reserved > csa->ohds->ds_phys->ds_unique_bytes) {
2406 		dsl_dir_diduse_space(csa->ohds->ds_dir,
2407 		    csa->ohds->ds_phys->ds_unique_bytes -
2408 		    csa->ohds->ds_reserved, 0, 0, tx);
2409 	}
2410 
2411 	/* reset origin's unique bytes */
2412 	csa->cds->ds_prev->ds_phys->ds_unique_bytes = unique;
2413 
2414 	/* swap blkptrs */
2415 	{
2416 		blkptr_t tmp;
2417 		tmp = csa->ohds->ds_phys->ds_bp;
2418 		csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2419 		csa->cds->ds_phys->ds_bp = tmp;
2420 	}
2421 
2422 	/* set dd_*_bytes */
2423 	{
2424 		int64_t dused, dcomp, duncomp;
2425 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2426 		uint64_t odl_used, odl_comp, odl_uncomp;
2427 
2428 		VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2429 		    &cdl_comp, &cdl_uncomp));
2430 		VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2431 		    &odl_comp, &odl_uncomp));
2432 		dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2433 		    (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2434 		dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2435 		    (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2436 		duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2437 		    cdl_uncomp -
2438 		    (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2439 
2440 		dsl_dir_diduse_space(csa->ohds->ds_dir,
2441 		    dused, dcomp, duncomp, tx);
2442 		dsl_dir_diduse_space(csa->cds->ds_dir,
2443 		    -dused, -dcomp, -duncomp, tx);
2444 	}
2445 
2446 #define	SWITCH64(x, y) \
2447 	{ \
2448 		uint64_t __tmp = (x); \
2449 		(x) = (y); \
2450 		(y) = __tmp; \
2451 	}
2452 
2453 	/* redo any accounting due to a refreservation */
2454 	if (csa->ohds->ds_reserved > csa->ohds->ds_phys->ds_unique_bytes) {
2455 		dsl_dir_diduse_space(csa->ohds->ds_dir,
2456 		    csa->ohds->ds_reserved -
2457 		    csa->ohds->ds_phys->ds_unique_bytes, 0, 0, tx);
2458 	}
2459 
2460 	/* swap ds_*_bytes */
2461 	SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2462 	    csa->cds->ds_phys->ds_used_bytes);
2463 	SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2464 	    csa->cds->ds_phys->ds_compressed_bytes);
2465 	SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2466 	    csa->cds->ds_phys->ds_uncompressed_bytes);
2467 
2468 	/* swap deadlists */
2469 	bplist_close(&csa->cds->ds_deadlist);
2470 	bplist_close(&csa->ohds->ds_deadlist);
2471 	SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2472 	    csa->cds->ds_phys->ds_deadlist_obj);
2473 	VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2474 	    csa->cds->ds_phys->ds_deadlist_obj));
2475 	VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2476 	    csa->ohds->ds_phys->ds_deadlist_obj));
2477 	/* fix up clone's unique */
2478 	dsl_dataset_recalc_head_uniq(csa->cds);
2479 
2480 }
2481 
2482 /*
2483  * Swap the clone "cosname" with its origin head file system.
2484  */
2485 int
2486 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2487     boolean_t force)
2488 {
2489 	struct cloneswaparg csa;
2490 
2491 	ASSERT(clone->ds_open_refcount == DS_REF_MAX);
2492 	ASSERT(origin_head->ds_open_refcount == DS_REF_MAX);
2493 
2494 	csa.cds = clone;
2495 	csa.ohds = origin_head;
2496 	csa.force = force;
2497 	return (dsl_sync_task_do(clone->ds_dir->dd_pool,
2498 	    dsl_dataset_clone_swap_check,
2499 	    dsl_dataset_clone_swap_sync, &csa, NULL, 9));
2500 }
2501 
2502 /*
2503  * Given a pool name and a dataset object number in that pool,
2504  * return the name of that dataset.
2505  */
2506 int
2507 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2508 {
2509 	spa_t *spa;
2510 	dsl_pool_t *dp;
2511 	dsl_dataset_t *ds = NULL;
2512 	int error;
2513 
2514 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
2515 		return (error);
2516 	dp = spa_get_dsl(spa);
2517 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2518 	if ((error = dsl_dataset_open_obj(dp, obj,
2519 	    NULL, DS_MODE_NONE, FTAG, &ds)) != 0) {
2520 		rw_exit(&dp->dp_config_rwlock);
2521 		spa_close(spa, FTAG);
2522 		return (error);
2523 	}
2524 	dsl_dataset_name(ds, buf);
2525 	dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2526 	rw_exit(&dp->dp_config_rwlock);
2527 	spa_close(spa, FTAG);
2528 
2529 	return (0);
2530 }
2531 
2532 int
2533 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2534     uint64_t asize, uint64_t inflight, uint64_t *used)
2535 {
2536 	int error = 0;
2537 
2538 	ASSERT3S(asize, >, 0);
2539 
2540 	mutex_enter(&ds->ds_lock);
2541 	/*
2542 	 * Make a space adjustment for reserved bytes.
2543 	 */
2544 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2545 		ASSERT3U(*used, >=,
2546 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2547 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2548 	}
2549 
2550 	if (!check_quota || ds->ds_quota == 0) {
2551 		mutex_exit(&ds->ds_lock);
2552 		return (0);
2553 	}
2554 	/*
2555 	 * If they are requesting more space, and our current estimate
2556 	 * is over quota, they get to try again unless the actual
2557 	 * on-disk is over quota and there are no pending changes (which
2558 	 * may free up space for us).
2559 	 */
2560 	if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
2561 		if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
2562 			error = ERESTART;
2563 		else
2564 			error = EDQUOT;
2565 	}
2566 	mutex_exit(&ds->ds_lock);
2567 
2568 	return (error);
2569 }
2570 
2571 /* ARGSUSED */
2572 static int
2573 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
2574 {
2575 	dsl_dataset_t *ds = arg1;
2576 	uint64_t *quotap = arg2;
2577 	uint64_t new_quota = *quotap;
2578 
2579 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
2580 		return (ENOTSUP);
2581 
2582 	if (new_quota == 0)
2583 		return (0);
2584 
2585 	if (new_quota < ds->ds_phys->ds_used_bytes ||
2586 	    new_quota < ds->ds_reserved)
2587 		return (ENOSPC);
2588 
2589 	return (0);
2590 }
2591 
2592 /* ARGSUSED */
2593 void
2594 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2595 {
2596 	dsl_dataset_t *ds = arg1;
2597 	uint64_t *quotap = arg2;
2598 	uint64_t new_quota = *quotap;
2599 
2600 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2601 
2602 	mutex_enter(&ds->ds_lock);
2603 	ds->ds_quota = new_quota;
2604 	mutex_exit(&ds->ds_lock);
2605 
2606 	dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
2607 
2608 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
2609 	    tx, cr, "%lld dataset = %llu ",
2610 	    (longlong_t)new_quota, ds->ds_dir->dd_phys->dd_head_dataset_obj);
2611 }
2612 
2613 int
2614 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
2615 {
2616 	dsl_dataset_t *ds;
2617 	int err;
2618 
2619 	err = dsl_dataset_open(dsname, DS_MODE_STANDARD, FTAG, &ds);
2620 	if (err)
2621 		return (err);
2622 
2623 	/*
2624 	 * If someone removes a file, then tries to set the quota, we
2625 	 * want to make sure the file freeing takes effect.
2626 	 */
2627 	txg_wait_open(ds->ds_dir->dd_pool, 0);
2628 
2629 	err = dsl_sync_task_do(ds->ds_dir->dd_pool, dsl_dataset_set_quota_check,
2630 	    dsl_dataset_set_quota_sync, ds, &quota, 0);
2631 	dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);
2632 	return (err);
2633 }
2634 
2635 static int
2636 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
2637 {
2638 	dsl_dataset_t *ds = arg1;
2639 	uint64_t *reservationp = arg2;
2640 	uint64_t new_reservation = *reservationp;
2641 	int64_t delta;
2642 	uint64_t unique;
2643 
2644 	if (new_reservation > INT64_MAX)
2645 		return (EOVERFLOW);
2646 
2647 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
2648 	    SPA_VERSION_REFRESERVATION)
2649 		return (ENOTSUP);
2650 
2651 	if (dsl_dataset_is_snapshot(ds))
2652 		return (EINVAL);
2653 
2654 	/*
2655 	 * If we are doing the preliminary check in open context, the
2656 	 * space estimates may be inaccurate.
2657 	 */
2658 	if (!dmu_tx_is_syncing(tx))
2659 		return (0);
2660 
2661 	mutex_enter(&ds->ds_lock);
2662 	unique = dsl_dataset_unique(ds);
2663 	delta = MAX(unique, new_reservation) - MAX(unique, ds->ds_reserved);
2664 	mutex_exit(&ds->ds_lock);
2665 
2666 	if (delta > 0 &&
2667 	    delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2668 		return (ENOSPC);
2669 	if (delta > 0 && ds->ds_quota > 0 &&
2670 	    new_reservation > ds->ds_quota)
2671 		return (ENOSPC);
2672 
2673 	return (0);
2674 }
2675 
2676 /* ARGSUSED */
2677 static void
2678 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
2679     dmu_tx_t *tx)
2680 {
2681 	dsl_dataset_t *ds = arg1;
2682 	uint64_t *reservationp = arg2;
2683 	uint64_t new_reservation = *reservationp;
2684 	uint64_t unique;
2685 	int64_t delta;
2686 
2687 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2688 
2689 	mutex_enter(&ds->ds_lock);
2690 	unique = dsl_dataset_unique(ds);
2691 	delta = MAX(0, (int64_t)(new_reservation - unique)) -
2692 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
2693 	ds->ds_reserved = new_reservation;
2694 	mutex_exit(&ds->ds_lock);
2695 
2696 	dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
2697 	    new_reservation, cr, tx);
2698 
2699 	dsl_dir_diduse_space(ds->ds_dir, delta, 0, 0, tx);
2700 
2701 	spa_history_internal_log(LOG_DS_REFRESERV,
2702 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
2703 	    (longlong_t)new_reservation,
2704 	    ds->ds_dir->dd_phys->dd_head_dataset_obj);
2705 }
2706 
2707 int
2708 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
2709 {
2710 	dsl_dataset_t *ds;
2711 	int err;
2712 
2713 	err = dsl_dataset_open(dsname, DS_MODE_STANDARD, FTAG, &ds);
2714 	if (err)
2715 		return (err);
2716 
2717 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2718 	    dsl_dataset_set_reservation_check,
2719 	    dsl_dataset_set_reservation_sync, ds, &reservation, 0);
2720 	dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);
2721 	return (err);
2722 }
2723