xref: /titanic_52/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision a83cadce5d3331b64803bfc641036cec23602c74)
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_checkfunc_t dsl_dataset_destroy_check;
49 static dsl_syncfunc_t dsl_dataset_destroy_sync;
50 
51 #define	DS_REF_MAX	(1ULL << 62)
52 
53 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
54 
55 /*
56  * We use weighted reference counts to express the various forms of exclusion
57  * between different open modes.  A STANDARD open is 1 point, an EXCLUSIVE open
58  * is DS_REF_MAX, and a PRIMARY open is little more than half of an EXCLUSIVE.
59  * This makes the exclusion logic simple: the total refcnt for all opens cannot
60  * exceed DS_REF_MAX.  For example, EXCLUSIVE opens are exclusive because their
61  * weight (DS_REF_MAX) consumes the entire refcnt space.  PRIMARY opens consume
62  * just over half of the refcnt space, so there can't be more than one, but it
63  * can peacefully coexist with any number of STANDARD opens.
64  */
65 static uint64_t ds_refcnt_weight[DS_MODE_LEVELS] = {
66 	0,			/* DS_MODE_NONE - invalid		*/
67 	1,			/* DS_MODE_STANDARD - unlimited number	*/
68 	(DS_REF_MAX >> 1) + 1,	/* DS_MODE_PRIMARY - only one of these	*/
69 	DS_REF_MAX		/* DS_MODE_EXCLUSIVE - no other opens	*/
70 };
71 
72 
73 void
74 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
75 {
76 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
77 	int compressed = BP_GET_PSIZE(bp);
78 	int uncompressed = BP_GET_UCSIZE(bp);
79 
80 	dprintf_bp(bp, "born, ds=%p\n", ds);
81 
82 	ASSERT(dmu_tx_is_syncing(tx));
83 	/* It could have been compressed away to nothing */
84 	if (BP_IS_HOLE(bp))
85 		return;
86 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
87 	ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
88 	if (ds == NULL) {
89 		/*
90 		 * Account for the meta-objset space in its placeholder
91 		 * dsl_dir.
92 		 */
93 		ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
94 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
95 		    used, compressed, uncompressed, tx);
96 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
97 		return;
98 	}
99 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
100 	mutex_enter(&ds->ds_lock);
101 	ds->ds_phys->ds_used_bytes += used;
102 	ds->ds_phys->ds_compressed_bytes += compressed;
103 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
104 	ds->ds_phys->ds_unique_bytes += used;
105 	mutex_exit(&ds->ds_lock);
106 	dsl_dir_diduse_space(ds->ds_dir,
107 	    used, compressed, uncompressed, tx);
108 }
109 
110 void
111 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
112     dmu_tx_t *tx)
113 {
114 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
115 	int compressed = BP_GET_PSIZE(bp);
116 	int uncompressed = BP_GET_UCSIZE(bp);
117 
118 	ASSERT(dmu_tx_is_syncing(tx));
119 	/* No block pointer => nothing to free */
120 	if (BP_IS_HOLE(bp))
121 		return;
122 
123 	ASSERT(used > 0);
124 	if (ds == NULL) {
125 		int err;
126 		/*
127 		 * Account for the meta-objset space in its placeholder
128 		 * dataset.
129 		 */
130 		err = arc_free(pio, tx->tx_pool->dp_spa,
131 		    tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
132 		ASSERT(err == 0);
133 
134 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
135 		    -used, -compressed, -uncompressed, tx);
136 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
137 		return;
138 	}
139 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
140 
141 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
142 
143 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
144 		int err;
145 
146 		dprintf_bp(bp, "freeing: %s", "");
147 		err = arc_free(pio, tx->tx_pool->dp_spa,
148 		    tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
149 		ASSERT(err == 0);
150 
151 		mutex_enter(&ds->ds_lock);
152 		/* XXX unique_bytes is not accurate for head datasets */
153 		/* ASSERT3U(ds->ds_phys->ds_unique_bytes, >=, used); */
154 		ds->ds_phys->ds_unique_bytes -= used;
155 		mutex_exit(&ds->ds_lock);
156 		dsl_dir_diduse_space(ds->ds_dir,
157 		    -used, -compressed, -uncompressed, tx);
158 	} else {
159 		dprintf_bp(bp, "putting on dead list: %s", "");
160 		VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
161 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
162 		if (ds->ds_phys->ds_prev_snap_obj != 0) {
163 			ASSERT3U(ds->ds_prev->ds_object, ==,
164 			    ds->ds_phys->ds_prev_snap_obj);
165 			ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
166 			if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
167 			    ds->ds_object && bp->blk_birth >
168 			    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
169 				dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
170 				mutex_enter(&ds->ds_prev->ds_lock);
171 				ds->ds_prev->ds_phys->ds_unique_bytes +=
172 				    used;
173 				mutex_exit(&ds->ds_prev->ds_lock);
174 			}
175 		}
176 	}
177 	mutex_enter(&ds->ds_lock);
178 	ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
179 	ds->ds_phys->ds_used_bytes -= used;
180 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
181 	ds->ds_phys->ds_compressed_bytes -= compressed;
182 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
183 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
184 	mutex_exit(&ds->ds_lock);
185 }
186 
187 uint64_t
188 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
189 {
190 	uint64_t trysnap = 0;
191 
192 	if (ds == NULL)
193 		return (0);
194 	/*
195 	 * The snapshot creation could fail, but that would cause an
196 	 * incorrect FALSE return, which would only result in an
197 	 * overestimation of the amount of space that an operation would
198 	 * consume, which is OK.
199 	 *
200 	 * There's also a small window where we could miss a pending
201 	 * snapshot, because we could set the sync task in the quiescing
202 	 * phase.  So this should only be used as a guess.
203 	 */
204 	if (ds->ds_trysnap_txg >
205 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
206 		trysnap = ds->ds_trysnap_txg;
207 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
208 }
209 
210 int
211 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
212 {
213 	return (blk_birth > dsl_dataset_prev_snap_txg(ds));
214 }
215 
216 /* ARGSUSED */
217 static void
218 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
219 {
220 	dsl_dataset_t *ds = dsv;
221 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
222 
223 	/* open_refcount == DS_REF_MAX when deleting */
224 	ASSERT(ds->ds_open_refcount == 0 ||
225 	    ds->ds_open_refcount == DS_REF_MAX);
226 
227 	dprintf_ds(ds, "evicting %s\n", "");
228 
229 	unique_remove(ds->ds_phys->ds_fsid_guid);
230 
231 	if (ds->ds_user_ptr != NULL)
232 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
233 
234 	if (ds->ds_prev) {
235 		dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds);
236 		ds->ds_prev = NULL;
237 	}
238 
239 	bplist_close(&ds->ds_deadlist);
240 	dsl_dir_close(ds->ds_dir, ds);
241 
242 	if (list_link_active(&ds->ds_synced_link))
243 		list_remove(&dp->dp_synced_objsets, ds);
244 
245 	mutex_destroy(&ds->ds_lock);
246 	mutex_destroy(&ds->ds_deadlist.bpl_lock);
247 
248 	kmem_free(ds, sizeof (dsl_dataset_t));
249 }
250 
251 static int
252 dsl_dataset_get_snapname(dsl_dataset_t *ds)
253 {
254 	dsl_dataset_phys_t *headphys;
255 	int err;
256 	dmu_buf_t *headdbuf;
257 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
258 	objset_t *mos = dp->dp_meta_objset;
259 
260 	if (ds->ds_snapname[0])
261 		return (0);
262 	if (ds->ds_phys->ds_next_snap_obj == 0)
263 		return (0);
264 
265 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
266 	    FTAG, &headdbuf);
267 	if (err)
268 		return (err);
269 	headphys = headdbuf->db_data;
270 	err = zap_value_search(dp->dp_meta_objset,
271 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
272 	dmu_buf_rele(headdbuf, FTAG);
273 	return (err);
274 }
275 
276 int
277 dsl_dataset_open_obj(dsl_pool_t *dp, uint64_t dsobj, const char *snapname,
278     int mode, void *tag, dsl_dataset_t **dsp)
279 {
280 	uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)];
281 	objset_t *mos = dp->dp_meta_objset;
282 	dmu_buf_t *dbuf;
283 	dsl_dataset_t *ds;
284 	int err;
285 
286 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
287 	    dsl_pool_sync_context(dp));
288 
289 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
290 	if (err)
291 		return (err);
292 	ds = dmu_buf_get_user(dbuf);
293 	if (ds == NULL) {
294 		dsl_dataset_t *winner;
295 
296 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
297 		ds->ds_dbuf = dbuf;
298 		ds->ds_object = dsobj;
299 		ds->ds_phys = dbuf->db_data;
300 
301 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
302 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
303 		    NULL);
304 
305 		err = bplist_open(&ds->ds_deadlist,
306 		    mos, ds->ds_phys->ds_deadlist_obj);
307 		if (err == 0) {
308 			err = dsl_dir_open_obj(dp,
309 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
310 		}
311 		if (err) {
312 			/*
313 			 * we don't really need to close the blist if we
314 			 * just opened it.
315 			 */
316 			mutex_destroy(&ds->ds_lock);
317 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
318 			kmem_free(ds, sizeof (dsl_dataset_t));
319 			dmu_buf_rele(dbuf, tag);
320 			return (err);
321 		}
322 
323 		if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) {
324 			ds->ds_snapname[0] = '\0';
325 			if (ds->ds_phys->ds_prev_snap_obj) {
326 				err = dsl_dataset_open_obj(dp,
327 				    ds->ds_phys->ds_prev_snap_obj, NULL,
328 				    DS_MODE_NONE, ds, &ds->ds_prev);
329 			}
330 		} else {
331 			if (snapname) {
332 #ifdef ZFS_DEBUG
333 				dsl_dataset_phys_t *headphys;
334 				dmu_buf_t *headdbuf;
335 				err = dmu_bonus_hold(mos,
336 				    ds->ds_dir->dd_phys->dd_head_dataset_obj,
337 				    FTAG, &headdbuf);
338 				if (err == 0) {
339 					headphys = headdbuf->db_data;
340 					uint64_t foundobj;
341 					err = zap_lookup(dp->dp_meta_objset,
342 					    headphys->ds_snapnames_zapobj,
343 					    snapname, sizeof (foundobj), 1,
344 					    &foundobj);
345 					ASSERT3U(foundobj, ==, dsobj);
346 					dmu_buf_rele(headdbuf, FTAG);
347 				}
348 #endif
349 				(void) strcat(ds->ds_snapname, snapname);
350 			} else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
351 				err = dsl_dataset_get_snapname(ds);
352 			}
353 		}
354 
355 		if (err == 0) {
356 			winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
357 			    dsl_dataset_evict);
358 		}
359 		if (err || winner) {
360 			bplist_close(&ds->ds_deadlist);
361 			if (ds->ds_prev) {
362 				dsl_dataset_close(ds->ds_prev,
363 				    DS_MODE_NONE, ds);
364 			}
365 			dsl_dir_close(ds->ds_dir, ds);
366 			mutex_destroy(&ds->ds_lock);
367 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
368 			kmem_free(ds, sizeof (dsl_dataset_t));
369 			if (err) {
370 				dmu_buf_rele(dbuf, tag);
371 				return (err);
372 			}
373 			ds = winner;
374 		} else {
375 			uint64_t new =
376 			    unique_insert(ds->ds_phys->ds_fsid_guid);
377 			if (new != ds->ds_phys->ds_fsid_guid) {
378 				/* XXX it won't necessarily be synced... */
379 				ds->ds_phys->ds_fsid_guid = new;
380 			}
381 		}
382 	}
383 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
384 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
385 
386 	mutex_enter(&ds->ds_lock);
387 	if ((DS_MODE_LEVEL(mode) == DS_MODE_PRIMARY &&
388 	    (ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) &&
389 	    !DS_MODE_IS_INCONSISTENT(mode)) ||
390 	    (ds->ds_open_refcount + weight > DS_REF_MAX)) {
391 		mutex_exit(&ds->ds_lock);
392 		dsl_dataset_close(ds, DS_MODE_NONE, tag);
393 		return (EBUSY);
394 	}
395 	ds->ds_open_refcount += weight;
396 	mutex_exit(&ds->ds_lock);
397 
398 	*dsp = ds;
399 	return (0);
400 }
401 
402 int
403 dsl_dataset_open_spa(spa_t *spa, const char *name, int mode,
404     void *tag, dsl_dataset_t **dsp)
405 {
406 	dsl_dir_t *dd;
407 	dsl_pool_t *dp;
408 	const char *tail;
409 	uint64_t obj;
410 	dsl_dataset_t *ds = NULL;
411 	int err = 0;
412 
413 	err = dsl_dir_open_spa(spa, name, FTAG, &dd, &tail);
414 	if (err)
415 		return (err);
416 
417 	dp = dd->dd_pool;
418 	obj = dd->dd_phys->dd_head_dataset_obj;
419 	rw_enter(&dp->dp_config_rwlock, RW_READER);
420 	if (obj == 0) {
421 		/* A dataset with no associated objset */
422 		err = ENOENT;
423 		goto out;
424 	}
425 
426 	if (tail != NULL) {
427 		objset_t *mos = dp->dp_meta_objset;
428 
429 		err = dsl_dataset_open_obj(dp, obj, NULL,
430 		    DS_MODE_NONE, tag, &ds);
431 		if (err)
432 			goto out;
433 		obj = ds->ds_phys->ds_snapnames_zapobj;
434 		dsl_dataset_close(ds, DS_MODE_NONE, tag);
435 		ds = NULL;
436 
437 		if (tail[0] != '@') {
438 			err = ENOENT;
439 			goto out;
440 		}
441 		tail++;
442 
443 		/* Look for a snapshot */
444 		if (!DS_MODE_IS_READONLY(mode)) {
445 			err = EROFS;
446 			goto out;
447 		}
448 		dprintf("looking for snapshot '%s'\n", tail);
449 		err = zap_lookup(mos, obj, tail, 8, 1, &obj);
450 		if (err)
451 			goto out;
452 	}
453 	err = dsl_dataset_open_obj(dp, obj, tail, mode, tag, &ds);
454 
455 out:
456 	rw_exit(&dp->dp_config_rwlock);
457 	dsl_dir_close(dd, FTAG);
458 
459 	ASSERT3U((err == 0), ==, (ds != NULL));
460 	/* ASSERT(ds == NULL || strcmp(name, ds->ds_name) == 0); */
461 
462 	*dsp = ds;
463 	return (err);
464 }
465 
466 int
467 dsl_dataset_open(const char *name, int mode, void *tag, dsl_dataset_t **dsp)
468 {
469 	return (dsl_dataset_open_spa(NULL, name, mode, tag, dsp));
470 }
471 
472 void
473 dsl_dataset_name(dsl_dataset_t *ds, char *name)
474 {
475 	if (ds == NULL) {
476 		(void) strcpy(name, "mos");
477 	} else {
478 		dsl_dir_name(ds->ds_dir, name);
479 		VERIFY(0 == dsl_dataset_get_snapname(ds));
480 		if (ds->ds_snapname[0]) {
481 			(void) strcat(name, "@");
482 			if (!MUTEX_HELD(&ds->ds_lock)) {
483 				/*
484 				 * We use a "recursive" mutex so that we
485 				 * can call dprintf_ds() with ds_lock held.
486 				 */
487 				mutex_enter(&ds->ds_lock);
488 				(void) strcat(name, ds->ds_snapname);
489 				mutex_exit(&ds->ds_lock);
490 			} else {
491 				(void) strcat(name, ds->ds_snapname);
492 			}
493 		}
494 	}
495 }
496 
497 static int
498 dsl_dataset_namelen(dsl_dataset_t *ds)
499 {
500 	int result;
501 
502 	if (ds == NULL) {
503 		result = 3;	/* "mos" */
504 	} else {
505 		result = dsl_dir_namelen(ds->ds_dir);
506 		VERIFY(0 == dsl_dataset_get_snapname(ds));
507 		if (ds->ds_snapname[0]) {
508 			++result;	/* adding one for the @-sign */
509 			if (!MUTEX_HELD(&ds->ds_lock)) {
510 				/* see dsl_datset_name */
511 				mutex_enter(&ds->ds_lock);
512 				result += strlen(ds->ds_snapname);
513 				mutex_exit(&ds->ds_lock);
514 			} else {
515 				result += strlen(ds->ds_snapname);
516 			}
517 		}
518 	}
519 
520 	return (result);
521 }
522 
523 void
524 dsl_dataset_close(dsl_dataset_t *ds, int mode, void *tag)
525 {
526 	uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)];
527 	mutex_enter(&ds->ds_lock);
528 	ASSERT3U(ds->ds_open_refcount, >=, weight);
529 	ds->ds_open_refcount -= weight;
530 	dprintf_ds(ds, "closing mode %u refcount now 0x%llx\n",
531 	    mode, ds->ds_open_refcount);
532 	mutex_exit(&ds->ds_lock);
533 
534 	dmu_buf_rele(ds->ds_dbuf, tag);
535 }
536 
537 void
538 dsl_dataset_create_root(dsl_pool_t *dp, uint64_t *ddobjp, dmu_tx_t *tx)
539 {
540 	objset_t *mos = dp->dp_meta_objset;
541 	dmu_buf_t *dbuf;
542 	dsl_dataset_phys_t *dsphys;
543 	dsl_dataset_t *ds;
544 	uint64_t dsobj;
545 	dsl_dir_t *dd;
546 
547 	dsl_dir_create_root(mos, ddobjp, tx);
548 	VERIFY(0 == dsl_dir_open_obj(dp, *ddobjp, NULL, FTAG, &dd));
549 
550 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
551 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
552 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
553 	dmu_buf_will_dirty(dbuf, tx);
554 	dsphys = dbuf->db_data;
555 	dsphys->ds_dir_obj = dd->dd_object;
556 	dsphys->ds_fsid_guid = unique_create();
557 	unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */
558 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
559 	    sizeof (dsphys->ds_guid));
560 	dsphys->ds_snapnames_zapobj =
561 	    zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx);
562 	dsphys->ds_creation_time = gethrestime_sec();
563 	dsphys->ds_creation_txg = tx->tx_txg;
564 	dsphys->ds_deadlist_obj =
565 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
566 	dmu_buf_rele(dbuf, FTAG);
567 
568 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
569 	dd->dd_phys->dd_head_dataset_obj = dsobj;
570 	dsl_dir_close(dd, FTAG);
571 
572 	VERIFY(0 ==
573 	    dsl_dataset_open_obj(dp, dsobj, NULL, DS_MODE_NONE, FTAG, &ds));
574 	(void) dmu_objset_create_impl(dp->dp_spa, ds,
575 	    &ds->ds_phys->ds_bp, DMU_OST_ZFS, tx);
576 	dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
577 }
578 
579 uint64_t
580 dsl_dataset_create_sync(dsl_dir_t *pdd,
581     const char *lastname, dsl_dataset_t *clone_parent, dmu_tx_t *tx)
582 {
583 	dsl_pool_t *dp = pdd->dd_pool;
584 	dmu_buf_t *dbuf;
585 	dsl_dataset_phys_t *dsphys;
586 	uint64_t dsobj, ddobj;
587 	objset_t *mos = dp->dp_meta_objset;
588 	dsl_dir_t *dd;
589 
590 	ASSERT(clone_parent == NULL || clone_parent->ds_dir->dd_pool == dp);
591 	ASSERT(clone_parent == NULL ||
592 	    clone_parent->ds_phys->ds_num_children > 0);
593 	ASSERT(lastname[0] != '@');
594 	ASSERT(dmu_tx_is_syncing(tx));
595 
596 	ddobj = dsl_dir_create_sync(pdd, lastname, tx);
597 	VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
598 
599 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
600 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
601 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
602 	dmu_buf_will_dirty(dbuf, tx);
603 	dsphys = dbuf->db_data;
604 	dsphys->ds_dir_obj = dd->dd_object;
605 	dsphys->ds_fsid_guid = unique_create();
606 	unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */
607 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
608 	    sizeof (dsphys->ds_guid));
609 	dsphys->ds_snapnames_zapobj =
610 	    zap_create(mos, DMU_OT_DSL_DS_SNAP_MAP, DMU_OT_NONE, 0, tx);
611 	dsphys->ds_creation_time = gethrestime_sec();
612 	dsphys->ds_creation_txg = tx->tx_txg;
613 	dsphys->ds_deadlist_obj =
614 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
615 	if (clone_parent) {
616 		dsphys->ds_prev_snap_obj = clone_parent->ds_object;
617 		dsphys->ds_prev_snap_txg =
618 		    clone_parent->ds_phys->ds_creation_txg;
619 		dsphys->ds_used_bytes =
620 		    clone_parent->ds_phys->ds_used_bytes;
621 		dsphys->ds_compressed_bytes =
622 		    clone_parent->ds_phys->ds_compressed_bytes;
623 		dsphys->ds_uncompressed_bytes =
624 		    clone_parent->ds_phys->ds_uncompressed_bytes;
625 		dsphys->ds_bp = clone_parent->ds_phys->ds_bp;
626 
627 		dmu_buf_will_dirty(clone_parent->ds_dbuf, tx);
628 		clone_parent->ds_phys->ds_num_children++;
629 
630 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
631 		dd->dd_phys->dd_clone_parent_obj = clone_parent->ds_object;
632 	}
633 	dmu_buf_rele(dbuf, FTAG);
634 
635 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
636 	dd->dd_phys->dd_head_dataset_obj = dsobj;
637 	dsl_dir_close(dd, FTAG);
638 
639 	return (dsobj);
640 }
641 
642 struct destroyarg {
643 	dsl_sync_task_group_t *dstg;
644 	char *snapname;
645 	char *failed;
646 };
647 
648 static int
649 dsl_snapshot_destroy_one(char *name, void *arg)
650 {
651 	struct destroyarg *da = arg;
652 	dsl_dataset_t *ds;
653 	char *cp;
654 	int err;
655 
656 	(void) strcat(name, "@");
657 	(void) strcat(name, da->snapname);
658 	err = dsl_dataset_open(name,
659 	    DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT,
660 	    da->dstg, &ds);
661 	cp = strchr(name, '@');
662 	*cp = '\0';
663 	if (err == ENOENT)
664 		return (0);
665 	if (err) {
666 		(void) strcpy(da->failed, name);
667 		return (err);
668 	}
669 
670 	dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
671 	    dsl_dataset_destroy_sync, ds, da->dstg, 0);
672 	return (0);
673 }
674 
675 /*
676  * Destroy 'snapname' in all descendants of 'fsname'.
677  */
678 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
679 int
680 dsl_snapshots_destroy(char *fsname, char *snapname)
681 {
682 	int err;
683 	struct destroyarg da;
684 	dsl_sync_task_t *dst;
685 	spa_t *spa;
686 
687 	err = spa_open(fsname, &spa, FTAG);
688 	if (err)
689 		return (err);
690 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
691 	da.snapname = snapname;
692 	da.failed = fsname;
693 
694 	err = dmu_objset_find(fsname,
695 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
696 
697 	if (err == 0)
698 		err = dsl_sync_task_group_wait(da.dstg);
699 
700 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
701 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
702 		dsl_dataset_t *ds = dst->dst_arg1;
703 		if (dst->dst_err) {
704 			dsl_dataset_name(ds, fsname);
705 			*strchr(fsname, '@') = '\0';
706 		}
707 		/*
708 		 * If it was successful, destroy_sync would have
709 		 * closed the ds
710 		 */
711 		if (err)
712 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, da.dstg);
713 	}
714 
715 	dsl_sync_task_group_destroy(da.dstg);
716 	spa_close(spa, FTAG);
717 	return (err);
718 }
719 
720 int
721 dsl_dataset_destroy(const char *name)
722 {
723 	int err;
724 	dsl_sync_task_group_t *dstg;
725 	objset_t *os;
726 	dsl_dataset_t *ds;
727 	dsl_dir_t *dd;
728 	uint64_t obj;
729 
730 	if (strchr(name, '@')) {
731 		/* Destroying a snapshot is simpler */
732 		err = dsl_dataset_open(name,
733 		    DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT,
734 		    FTAG, &ds);
735 		if (err)
736 			return (err);
737 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
738 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
739 		    ds, FTAG, 0);
740 		if (err)
741 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
742 		return (err);
743 	}
744 
745 	err = dmu_objset_open(name, DMU_OST_ANY,
746 	    DS_MODE_EXCLUSIVE | DS_MODE_INCONSISTENT, &os);
747 	if (err)
748 		return (err);
749 	ds = os->os->os_dsl_dataset;
750 	dd = ds->ds_dir;
751 
752 	/*
753 	 * Check for errors and mark this ds as inconsistent, in
754 	 * case we crash while freeing the objects.
755 	 */
756 	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
757 	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
758 	if (err) {
759 		dmu_objset_close(os);
760 		return (err);
761 	}
762 
763 	/*
764 	 * remove the objects in open context, so that we won't
765 	 * have too much to do in syncing context.
766 	 */
767 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
768 	    ds->ds_phys->ds_prev_snap_txg)) {
769 		dmu_tx_t *tx = dmu_tx_create(os);
770 		dmu_tx_hold_free(tx, obj, 0, DMU_OBJECT_END);
771 		dmu_tx_hold_bonus(tx, obj);
772 		err = dmu_tx_assign(tx, TXG_WAIT);
773 		if (err) {
774 			/*
775 			 * Perhaps there is not enough disk
776 			 * space.  Just deal with it from
777 			 * dsl_dataset_destroy_sync().
778 			 */
779 			dmu_tx_abort(tx);
780 			continue;
781 		}
782 		VERIFY(0 == dmu_object_free(os, obj, tx));
783 		dmu_tx_commit(tx);
784 	}
785 	/* Make sure it's not dirty before we finish destroying it. */
786 	txg_wait_synced(dd->dd_pool, 0);
787 
788 	dmu_objset_close(os);
789 	if (err != ESRCH)
790 		return (err);
791 
792 	err = dsl_dataset_open(name,
793 	    DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT,
794 	    FTAG, &ds);
795 	if (err)
796 		return (err);
797 
798 	err = dsl_dir_open(name, FTAG, &dd, NULL);
799 	if (err) {
800 		dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
801 		return (err);
802 	}
803 
804 	/*
805 	 * Blow away the dsl_dir + head dataset.
806 	 */
807 	dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
808 	dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
809 	    dsl_dataset_destroy_sync, ds, FTAG, 0);
810 	dsl_sync_task_create(dstg, dsl_dir_destroy_check,
811 	    dsl_dir_destroy_sync, dd, FTAG, 0);
812 	err = dsl_sync_task_group_wait(dstg);
813 	dsl_sync_task_group_destroy(dstg);
814 	/* if it is successful, *destroy_sync will close the ds+dd */
815 	if (err) {
816 		dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
817 		dsl_dir_close(dd, FTAG);
818 	}
819 	return (err);
820 }
821 
822 int
823 dsl_dataset_rollback(dsl_dataset_t *ds)
824 {
825 	ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX);
826 	return (dsl_sync_task_do(ds->ds_dir->dd_pool,
827 	    dsl_dataset_rollback_check, dsl_dataset_rollback_sync,
828 	    ds, NULL, 0));
829 }
830 
831 void *
832 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
833     void *p, dsl_dataset_evict_func_t func)
834 {
835 	void *old;
836 
837 	mutex_enter(&ds->ds_lock);
838 	old = ds->ds_user_ptr;
839 	if (old == NULL) {
840 		ds->ds_user_ptr = p;
841 		ds->ds_user_evict_func = func;
842 	}
843 	mutex_exit(&ds->ds_lock);
844 	return (old);
845 }
846 
847 void *
848 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
849 {
850 	return (ds->ds_user_ptr);
851 }
852 
853 
854 blkptr_t *
855 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
856 {
857 	return (&ds->ds_phys->ds_bp);
858 }
859 
860 void
861 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
862 {
863 	ASSERT(dmu_tx_is_syncing(tx));
864 	/* If it's the meta-objset, set dp_meta_rootbp */
865 	if (ds == NULL) {
866 		tx->tx_pool->dp_meta_rootbp = *bp;
867 	} else {
868 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
869 		ds->ds_phys->ds_bp = *bp;
870 	}
871 }
872 
873 spa_t *
874 dsl_dataset_get_spa(dsl_dataset_t *ds)
875 {
876 	return (ds->ds_dir->dd_pool->dp_spa);
877 }
878 
879 void
880 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
881 {
882 	dsl_pool_t *dp;
883 
884 	if (ds == NULL) /* this is the meta-objset */
885 		return;
886 
887 	ASSERT(ds->ds_user_ptr != NULL);
888 
889 	if (ds->ds_phys->ds_next_snap_obj != 0)
890 		panic("dirtying snapshot!");
891 
892 	dp = ds->ds_dir->dd_pool;
893 
894 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
895 		/* up the hold count until we can be written out */
896 		dmu_buf_add_ref(ds->ds_dbuf, ds);
897 	}
898 }
899 
900 struct killarg {
901 	uint64_t *usedp;
902 	uint64_t *compressedp;
903 	uint64_t *uncompressedp;
904 	zio_t *zio;
905 	dmu_tx_t *tx;
906 };
907 
908 static int
909 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
910 {
911 	struct killarg *ka = arg;
912 	blkptr_t *bp = &bc->bc_blkptr;
913 
914 	ASSERT3U(bc->bc_errno, ==, 0);
915 
916 	/*
917 	 * Since this callback is not called concurrently, no lock is
918 	 * needed on the accounting values.
919 	 */
920 	*ka->usedp += bp_get_dasize(spa, bp);
921 	*ka->compressedp += BP_GET_PSIZE(bp);
922 	*ka->uncompressedp += BP_GET_UCSIZE(bp);
923 	/* XXX check for EIO? */
924 	(void) arc_free(ka->zio, spa, ka->tx->tx_txg, bp, NULL, NULL,
925 	    ARC_NOWAIT);
926 	return (0);
927 }
928 
929 /* ARGSUSED */
930 static int
931 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
932 {
933 	dsl_dataset_t *ds = arg1;
934 
935 	/*
936 	 * There must be a previous snapshot.  I suppose we could roll
937 	 * it back to being empty (and re-initialize the upper (ZPL)
938 	 * layer).  But for now there's no way to do this via the user
939 	 * interface.
940 	 */
941 	if (ds->ds_phys->ds_prev_snap_txg == 0)
942 		return (EINVAL);
943 
944 	/*
945 	 * This must not be a snapshot.
946 	 */
947 	if (ds->ds_phys->ds_next_snap_obj != 0)
948 		return (EINVAL);
949 
950 	/*
951 	 * If we made changes this txg, traverse_dsl_dataset won't find
952 	 * them.  Try again.
953 	 */
954 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
955 		return (EAGAIN);
956 
957 	return (0);
958 }
959 
960 /* ARGSUSED */
961 static void
962 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
963 {
964 	dsl_dataset_t *ds = arg1;
965 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
966 
967 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
968 
969 	/* Zero out the deadlist. */
970 	bplist_close(&ds->ds_deadlist);
971 	bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
972 	ds->ds_phys->ds_deadlist_obj =
973 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
974 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
975 	    ds->ds_phys->ds_deadlist_obj));
976 
977 	{
978 		/* Free blkptrs that we gave birth to */
979 		zio_t *zio;
980 		uint64_t used = 0, compressed = 0, uncompressed = 0;
981 		struct killarg ka;
982 
983 		zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
984 		    ZIO_FLAG_MUSTSUCCEED);
985 		ka.usedp = &used;
986 		ka.compressedp = &compressed;
987 		ka.uncompressedp = &uncompressed;
988 		ka.zio = zio;
989 		ka.tx = tx;
990 		(void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
991 		    ADVANCE_POST, kill_blkptr, &ka);
992 		(void) zio_wait(zio);
993 
994 		dsl_dir_diduse_space(ds->ds_dir,
995 		    -used, -compressed, -uncompressed, tx);
996 	}
997 
998 	/* Change our contents to that of the prev snapshot */
999 	ASSERT3U(ds->ds_prev->ds_object, ==, ds->ds_phys->ds_prev_snap_obj);
1000 	ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1001 	ds->ds_phys->ds_used_bytes = ds->ds_prev->ds_phys->ds_used_bytes;
1002 	ds->ds_phys->ds_compressed_bytes =
1003 	    ds->ds_prev->ds_phys->ds_compressed_bytes;
1004 	ds->ds_phys->ds_uncompressed_bytes =
1005 	    ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1006 	ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1007 	ds->ds_phys->ds_unique_bytes = 0;
1008 
1009 	if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1010 		dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1011 		ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1012 	}
1013 
1014 	spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa,
1015 	    tx, cr, "dataset = %llu", ds->ds_object);
1016 }
1017 
1018 /* ARGSUSED */
1019 static int
1020 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1021 {
1022 	dsl_dataset_t *ds = arg1;
1023 
1024 	/*
1025 	 * Can't delete a head dataset if there are snapshots of it.
1026 	 * (Except if the only snapshots are from the branch we cloned
1027 	 * from.)
1028 	 */
1029 	if (ds->ds_prev != NULL &&
1030 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1031 		return (EINVAL);
1032 
1033 	return (0);
1034 }
1035 
1036 /* ARGSUSED */
1037 static void
1038 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1039 {
1040 	dsl_dataset_t *ds = arg1;
1041 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1042 
1043 	/* Mark it as inconsistent on-disk, in case we crash */
1044 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1045 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1046 
1047 	spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1048 	    cr, "dataset = %llu", ds->ds_object);
1049 }
1050 
1051 /* ARGSUSED */
1052 static int
1053 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1054 {
1055 	dsl_dataset_t *ds = arg1;
1056 
1057 	/* Can't delete a branch point. */
1058 	if (ds->ds_phys->ds_num_children > 1)
1059 		return (EEXIST);
1060 
1061 	/*
1062 	 * Can't delete a head dataset if there are snapshots of it.
1063 	 * (Except if the only snapshots are from the branch we cloned
1064 	 * from.)
1065 	 */
1066 	if (ds->ds_prev != NULL &&
1067 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1068 		return (EINVAL);
1069 
1070 	/*
1071 	 * If we made changes this txg, traverse_dsl_dataset won't find
1072 	 * them.  Try again.
1073 	 */
1074 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1075 		return (EAGAIN);
1076 
1077 	/* XXX we should do some i/o error checking... */
1078 	return (0);
1079 }
1080 
1081 static void
1082 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1083 {
1084 	dsl_dataset_t *ds = arg1;
1085 	uint64_t used = 0, compressed = 0, uncompressed = 0;
1086 	zio_t *zio;
1087 	int err;
1088 	int after_branch_point = FALSE;
1089 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1090 	objset_t *mos = dp->dp_meta_objset;
1091 	dsl_dataset_t *ds_prev = NULL;
1092 	uint64_t obj;
1093 
1094 	ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX);
1095 	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1096 	ASSERT(ds->ds_prev == NULL ||
1097 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1098 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1099 
1100 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1101 
1102 	obj = ds->ds_object;
1103 
1104 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1105 		if (ds->ds_prev) {
1106 			ds_prev = ds->ds_prev;
1107 		} else {
1108 			VERIFY(0 == dsl_dataset_open_obj(dp,
1109 			    ds->ds_phys->ds_prev_snap_obj, NULL,
1110 			    DS_MODE_NONE, FTAG, &ds_prev));
1111 		}
1112 		after_branch_point =
1113 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1114 
1115 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1116 		if (after_branch_point &&
1117 		    ds->ds_phys->ds_next_snap_obj == 0) {
1118 			/* This clone is toast. */
1119 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1120 			ds_prev->ds_phys->ds_num_children--;
1121 		} else if (!after_branch_point) {
1122 			ds_prev->ds_phys->ds_next_snap_obj =
1123 			    ds->ds_phys->ds_next_snap_obj;
1124 		}
1125 	}
1126 
1127 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1128 
1129 	if (ds->ds_phys->ds_next_snap_obj != 0) {
1130 		blkptr_t bp;
1131 		dsl_dataset_t *ds_next;
1132 		uint64_t itor = 0;
1133 
1134 		spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1135 
1136 		VERIFY(0 == dsl_dataset_open_obj(dp,
1137 		    ds->ds_phys->ds_next_snap_obj, NULL,
1138 		    DS_MODE_NONE, FTAG, &ds_next));
1139 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1140 
1141 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1142 		ds_next->ds_phys->ds_prev_snap_obj =
1143 		    ds->ds_phys->ds_prev_snap_obj;
1144 		ds_next->ds_phys->ds_prev_snap_txg =
1145 		    ds->ds_phys->ds_prev_snap_txg;
1146 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1147 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1148 
1149 		/*
1150 		 * Transfer to our deadlist (which will become next's
1151 		 * new deadlist) any entries from next's current
1152 		 * deadlist which were born before prev, and free the
1153 		 * other entries.
1154 		 *
1155 		 * XXX we're doing this long task with the config lock held
1156 		 */
1157 		while (bplist_iterate(&ds_next->ds_deadlist, &itor,
1158 		    &bp) == 0) {
1159 			if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1160 				VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1161 				    &bp, tx));
1162 				if (ds_prev && !after_branch_point &&
1163 				    bp.blk_birth >
1164 				    ds_prev->ds_phys->ds_prev_snap_txg) {
1165 					ds_prev->ds_phys->ds_unique_bytes +=
1166 					    bp_get_dasize(dp->dp_spa, &bp);
1167 				}
1168 			} else {
1169 				used += bp_get_dasize(dp->dp_spa, &bp);
1170 				compressed += BP_GET_PSIZE(&bp);
1171 				uncompressed += BP_GET_UCSIZE(&bp);
1172 				/* XXX check return value? */
1173 				(void) arc_free(zio, dp->dp_spa, tx->tx_txg,
1174 				    &bp, NULL, NULL, ARC_NOWAIT);
1175 			}
1176 		}
1177 
1178 		/* free next's deadlist */
1179 		bplist_close(&ds_next->ds_deadlist);
1180 		bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1181 
1182 		/* set next's deadlist to our deadlist */
1183 		ds_next->ds_phys->ds_deadlist_obj =
1184 		    ds->ds_phys->ds_deadlist_obj;
1185 		VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1186 		    ds_next->ds_phys->ds_deadlist_obj));
1187 		ds->ds_phys->ds_deadlist_obj = 0;
1188 
1189 		if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1190 			/*
1191 			 * Update next's unique to include blocks which
1192 			 * were previously shared by only this snapshot
1193 			 * and it.  Those blocks will be born after the
1194 			 * prev snap and before this snap, and will have
1195 			 * died after the next snap and before the one
1196 			 * after that (ie. be on the snap after next's
1197 			 * deadlist).
1198 			 *
1199 			 * XXX we're doing this long task with the
1200 			 * config lock held
1201 			 */
1202 			dsl_dataset_t *ds_after_next;
1203 
1204 			VERIFY(0 == dsl_dataset_open_obj(dp,
1205 			    ds_next->ds_phys->ds_next_snap_obj, NULL,
1206 			    DS_MODE_NONE, FTAG, &ds_after_next));
1207 			itor = 0;
1208 			while (bplist_iterate(&ds_after_next->ds_deadlist,
1209 			    &itor, &bp) == 0) {
1210 				if (bp.blk_birth >
1211 				    ds->ds_phys->ds_prev_snap_txg &&
1212 				    bp.blk_birth <=
1213 				    ds->ds_phys->ds_creation_txg) {
1214 					ds_next->ds_phys->ds_unique_bytes +=
1215 					    bp_get_dasize(dp->dp_spa, &bp);
1216 				}
1217 			}
1218 
1219 			dsl_dataset_close(ds_after_next, DS_MODE_NONE, FTAG);
1220 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1221 		} else {
1222 			/*
1223 			 * It would be nice to update the head dataset's
1224 			 * unique.  To do so we would have to traverse
1225 			 * it for blocks born after ds_prev, which is
1226 			 * pretty expensive just to maintain something
1227 			 * for debugging purposes.
1228 			 */
1229 			ASSERT3P(ds_next->ds_prev, ==, ds);
1230 			dsl_dataset_close(ds_next->ds_prev, DS_MODE_NONE,
1231 			    ds_next);
1232 			if (ds_prev) {
1233 				VERIFY(0 == dsl_dataset_open_obj(dp,
1234 				    ds->ds_phys->ds_prev_snap_obj, NULL,
1235 				    DS_MODE_NONE, ds_next, &ds_next->ds_prev));
1236 			} else {
1237 				ds_next->ds_prev = NULL;
1238 			}
1239 		}
1240 		dsl_dataset_close(ds_next, DS_MODE_NONE, FTAG);
1241 
1242 		/*
1243 		 * NB: unique_bytes is not accurate for head objsets
1244 		 * because we don't update it when we delete the most
1245 		 * recent snapshot -- see above comment.
1246 		 */
1247 		ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1248 	} else {
1249 		/*
1250 		 * There's no next snapshot, so this is a head dataset.
1251 		 * Destroy the deadlist.  Unless it's a clone, the
1252 		 * deadlist should be empty.  (If it's a clone, it's
1253 		 * safe to ignore the deadlist contents.)
1254 		 */
1255 		struct killarg ka;
1256 
1257 		ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1258 		bplist_close(&ds->ds_deadlist);
1259 		bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1260 		ds->ds_phys->ds_deadlist_obj = 0;
1261 
1262 		/*
1263 		 * Free everything that we point to (that's born after
1264 		 * the previous snapshot, if we are a clone)
1265 		 *
1266 		 * XXX we're doing this long task with the config lock held
1267 		 */
1268 		ka.usedp = &used;
1269 		ka.compressedp = &compressed;
1270 		ka.uncompressedp = &uncompressed;
1271 		ka.zio = zio;
1272 		ka.tx = tx;
1273 		err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1274 		    ADVANCE_POST, kill_blkptr, &ka);
1275 		ASSERT3U(err, ==, 0);
1276 	}
1277 
1278 	err = zio_wait(zio);
1279 	ASSERT3U(err, ==, 0);
1280 
1281 	dsl_dir_diduse_space(ds->ds_dir, -used, -compressed, -uncompressed, tx);
1282 
1283 	if (ds->ds_phys->ds_snapnames_zapobj) {
1284 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1285 		ASSERT(err == 0);
1286 	}
1287 
1288 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1289 		/* Erase the link in the dataset */
1290 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1291 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1292 		/*
1293 		 * dsl_dir_sync_destroy() called us, they'll destroy
1294 		 * the dataset.
1295 		 */
1296 	} else {
1297 		/* remove from snapshot namespace */
1298 		dsl_dataset_t *ds_head;
1299 		VERIFY(0 == dsl_dataset_open_obj(dp,
1300 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, NULL,
1301 		    DS_MODE_NONE, FTAG, &ds_head));
1302 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1303 #ifdef ZFS_DEBUG
1304 		{
1305 			uint64_t val;
1306 			err = zap_lookup(mos,
1307 			    ds_head->ds_phys->ds_snapnames_zapobj,
1308 			    ds->ds_snapname, 8, 1, &val);
1309 			ASSERT3U(err, ==, 0);
1310 			ASSERT3U(val, ==, obj);
1311 		}
1312 #endif
1313 		err = zap_remove(mos, ds_head->ds_phys->ds_snapnames_zapobj,
1314 		    ds->ds_snapname, tx);
1315 		ASSERT(err == 0);
1316 		dsl_dataset_close(ds_head, DS_MODE_NONE, FTAG);
1317 	}
1318 
1319 	if (ds_prev && ds->ds_prev != ds_prev)
1320 		dsl_dataset_close(ds_prev, DS_MODE_NONE, FTAG);
1321 
1322 	spa_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1323 	spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1324 	    cr, "dataset = %llu", ds->ds_object);
1325 
1326 	dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag);
1327 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1328 
1329 }
1330 
1331 /* ARGSUSED */
1332 int
1333 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1334 {
1335 	objset_t *os = arg1;
1336 	dsl_dataset_t *ds = os->os->os_dsl_dataset;
1337 	const char *snapname = arg2;
1338 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1339 	int err;
1340 	uint64_t value;
1341 
1342 	/*
1343 	 * We don't allow multiple snapshots of the same txg.  If there
1344 	 * is already one, try again.
1345 	 */
1346 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1347 		return (EAGAIN);
1348 
1349 	/*
1350 	 * Check for conflicting name snapshot name.
1351 	 */
1352 	err = zap_lookup(mos, ds->ds_phys->ds_snapnames_zapobj,
1353 	    snapname, 8, 1, &value);
1354 	if (err == 0)
1355 		return (EEXIST);
1356 	if (err != ENOENT)
1357 		return (err);
1358 
1359 	/*
1360 	 * Check that the dataset's name is not too long.  Name consists
1361 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1362 	 */
1363 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1364 		return (ENAMETOOLONG);
1365 
1366 	ds->ds_trysnap_txg = tx->tx_txg;
1367 	return (0);
1368 }
1369 
1370 void
1371 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1372 {
1373 	objset_t *os = arg1;
1374 	dsl_dataset_t *ds = os->os->os_dsl_dataset;
1375 	const char *snapname = arg2;
1376 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1377 	dmu_buf_t *dbuf;
1378 	dsl_dataset_phys_t *dsphys;
1379 	uint64_t dsobj;
1380 	objset_t *mos = dp->dp_meta_objset;
1381 	int err;
1382 
1383 	spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1384 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1385 
1386 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1387 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1388 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1389 	dmu_buf_will_dirty(dbuf, tx);
1390 	dsphys = dbuf->db_data;
1391 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1392 	dsphys->ds_fsid_guid = unique_create();
1393 	unique_remove(dsphys->ds_fsid_guid); /* it isn't open yet */
1394 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1395 	    sizeof (dsphys->ds_guid));
1396 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1397 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1398 	dsphys->ds_next_snap_obj = ds->ds_object;
1399 	dsphys->ds_num_children = 1;
1400 	dsphys->ds_creation_time = gethrestime_sec();
1401 	dsphys->ds_creation_txg = tx->tx_txg;
1402 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1403 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1404 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1405 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1406 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1407 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1408 	dmu_buf_rele(dbuf, FTAG);
1409 
1410 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1411 	if (ds->ds_prev) {
1412 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1413 		    ds->ds_object ||
1414 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1415 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1416 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1417 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1418 			    ds->ds_prev->ds_phys->ds_creation_txg);
1419 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1420 		}
1421 	}
1422 
1423 	bplist_close(&ds->ds_deadlist);
1424 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1425 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, dsphys->ds_creation_txg);
1426 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1427 	ds->ds_phys->ds_prev_snap_txg = dsphys->ds_creation_txg;
1428 	ds->ds_phys->ds_unique_bytes = 0;
1429 	ds->ds_phys->ds_deadlist_obj =
1430 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1431 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1432 	    ds->ds_phys->ds_deadlist_obj));
1433 
1434 	dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1435 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1436 	    snapname, 8, 1, &dsobj, tx);
1437 	ASSERT(err == 0);
1438 
1439 	if (ds->ds_prev)
1440 		dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds);
1441 	VERIFY(0 == dsl_dataset_open_obj(dp,
1442 	    ds->ds_phys->ds_prev_snap_obj, snapname,
1443 	    DS_MODE_NONE, ds, &ds->ds_prev));
1444 
1445 	spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1446 	    "dataset = %llu", dsobj);
1447 }
1448 
1449 void
1450 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1451 {
1452 	ASSERT(dmu_tx_is_syncing(tx));
1453 	ASSERT(ds->ds_user_ptr != NULL);
1454 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1455 
1456 	dsl_dir_dirty(ds->ds_dir, tx);
1457 	dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1458 	/* Unneeded? bplist_close(&ds->ds_deadlist); */
1459 }
1460 
1461 void
1462 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1463 {
1464 	dsl_dir_stats(ds->ds_dir, nv);
1465 
1466 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1467 	    ds->ds_phys->ds_creation_time);
1468 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1469 	    ds->ds_phys->ds_creation_txg);
1470 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED,
1471 	    ds->ds_phys->ds_used_bytes);
1472 
1473 	if (ds->ds_phys->ds_next_snap_obj) {
1474 		/*
1475 		 * This is a snapshot; override the dd's space used with
1476 		 * our unique space and compression ratio.
1477 		 */
1478 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1479 		    ds->ds_phys->ds_unique_bytes);
1480 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1481 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1482 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1483 		    ds->ds_phys->ds_compressed_bytes));
1484 	}
1485 }
1486 
1487 void
1488 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1489 {
1490 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1491 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1492 	if (ds->ds_phys->ds_next_snap_obj) {
1493 		stat->dds_is_snapshot = B_TRUE;
1494 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1495 	}
1496 
1497 	/* clone origin is really a dsl_dir thing... */
1498 	if (ds->ds_dir->dd_phys->dd_clone_parent_obj) {
1499 		dsl_dataset_t *ods;
1500 
1501 		rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1502 		VERIFY(0 == dsl_dataset_open_obj(ds->ds_dir->dd_pool,
1503 		    ds->ds_dir->dd_phys->dd_clone_parent_obj,
1504 		    NULL, DS_MODE_NONE, FTAG, &ods));
1505 		dsl_dataset_name(ods, stat->dds_clone_of);
1506 		dsl_dataset_close(ods, DS_MODE_NONE, FTAG);
1507 		rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1508 	}
1509 }
1510 
1511 uint64_t
1512 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1513 {
1514 	return (ds->ds_phys->ds_fsid_guid);
1515 }
1516 
1517 void
1518 dsl_dataset_space(dsl_dataset_t *ds,
1519     uint64_t *refdbytesp, uint64_t *availbytesp,
1520     uint64_t *usedobjsp, uint64_t *availobjsp)
1521 {
1522 	*refdbytesp = ds->ds_phys->ds_used_bytes;
1523 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1524 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1525 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1526 }
1527 
1528 /* ARGSUSED */
1529 static int
1530 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1531 {
1532 	dsl_dataset_t *ds = arg1;
1533 	char *newsnapname = arg2;
1534 	dsl_dir_t *dd = ds->ds_dir;
1535 	objset_t *mos = dd->dd_pool->dp_meta_objset;
1536 	dsl_dataset_t *hds;
1537 	uint64_t val;
1538 	int err;
1539 
1540 	err = dsl_dataset_open_obj(dd->dd_pool,
1541 	    dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds);
1542 	if (err)
1543 		return (err);
1544 
1545 	/* new name better not be in use */
1546 	err = zap_lookup(mos, hds->ds_phys->ds_snapnames_zapobj,
1547 	    newsnapname, 8, 1, &val);
1548 	dsl_dataset_close(hds, DS_MODE_NONE, FTAG);
1549 
1550 	if (err == 0)
1551 		err = EEXIST;
1552 	else if (err == ENOENT)
1553 		err = 0;
1554 
1555 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1556 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
1557 		err = ENAMETOOLONG;
1558 
1559 	return (err);
1560 }
1561 
1562 static void
1563 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
1564     cred_t *cr, dmu_tx_t *tx)
1565 {
1566 	dsl_dataset_t *ds = arg1;
1567 	const char *newsnapname = arg2;
1568 	dsl_dir_t *dd = ds->ds_dir;
1569 	objset_t *mos = dd->dd_pool->dp_meta_objset;
1570 	dsl_dataset_t *hds;
1571 	int err;
1572 
1573 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
1574 
1575 	VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool,
1576 	    dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds));
1577 
1578 	VERIFY(0 == dsl_dataset_get_snapname(ds));
1579 	err = zap_remove(mos, hds->ds_phys->ds_snapnames_zapobj,
1580 	    ds->ds_snapname, tx);
1581 	ASSERT3U(err, ==, 0);
1582 	mutex_enter(&ds->ds_lock);
1583 	(void) strcpy(ds->ds_snapname, newsnapname);
1584 	mutex_exit(&ds->ds_lock);
1585 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
1586 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
1587 	ASSERT3U(err, ==, 0);
1588 
1589 	spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
1590 	    cr, "dataset = %llu", ds->ds_object);
1591 	dsl_dataset_close(hds, DS_MODE_NONE, FTAG);
1592 }
1593 
1594 struct renamearg {
1595 	dsl_sync_task_group_t *dstg;
1596 	char failed[MAXPATHLEN];
1597 	char *oldsnap;
1598 	char *newsnap;
1599 };
1600 
1601 static int
1602 dsl_snapshot_rename_one(char *name, void *arg)
1603 {
1604 	struct renamearg *ra = arg;
1605 	dsl_dataset_t *ds = NULL;
1606 	char *cp;
1607 	int err;
1608 
1609 	cp = name + strlen(name);
1610 	*cp = '@';
1611 	(void) strcpy(cp + 1, ra->oldsnap);
1612 
1613 	/*
1614 	 * For recursive snapshot renames the parent won't be changing
1615 	 * so we just pass name for both the to/from argument.
1616 	 */
1617 	if (err = zfs_secpolicy_rename_perms(name, name, CRED())) {
1618 		(void) strcpy(ra->failed, name);
1619 		return (err);
1620 	}
1621 
1622 	err = dsl_dataset_open(name, DS_MODE_READONLY | DS_MODE_STANDARD,
1623 	    ra->dstg, &ds);
1624 	if (err == ENOENT) {
1625 		*cp = '\0';
1626 		return (0);
1627 	}
1628 	if (err) {
1629 		(void) strcpy(ra->failed, name);
1630 		*cp = '\0';
1631 		dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg);
1632 		return (err);
1633 	}
1634 
1635 #ifdef _KERNEL
1636 	/* for all filesystems undergoing rename, we'll need to unmount it */
1637 	(void) zfs_unmount_snap(name, NULL);
1638 #endif
1639 
1640 	*cp = '\0';
1641 
1642 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
1643 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
1644 
1645 	return (0);
1646 }
1647 
1648 static int
1649 dsl_recursive_rename(char *oldname, const char *newname)
1650 {
1651 	int err;
1652 	struct renamearg *ra;
1653 	dsl_sync_task_t *dst;
1654 	spa_t *spa;
1655 	char *cp, *fsname = spa_strdup(oldname);
1656 	int len = strlen(oldname);
1657 
1658 	/* truncate the snapshot name to get the fsname */
1659 	cp = strchr(fsname, '@');
1660 	*cp = '\0';
1661 
1662 	err = spa_open(fsname, &spa, FTAG);
1663 	if (err) {
1664 		kmem_free(fsname, len + 1);
1665 		return (err);
1666 	}
1667 	ra = kmem_alloc(sizeof (struct renamearg), KM_SLEEP);
1668 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
1669 
1670 	ra->oldsnap = strchr(oldname, '@') + 1;
1671 	ra->newsnap = strchr(newname, '@') + 1;
1672 	*ra->failed = '\0';
1673 
1674 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
1675 	    DS_FIND_CHILDREN);
1676 	kmem_free(fsname, len + 1);
1677 
1678 	if (err == 0) {
1679 		err = dsl_sync_task_group_wait(ra->dstg);
1680 	}
1681 
1682 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
1683 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
1684 		dsl_dataset_t *ds = dst->dst_arg1;
1685 		if (dst->dst_err) {
1686 			dsl_dir_name(ds->ds_dir, ra->failed);
1687 			(void) strcat(ra->failed, "@");
1688 			(void) strcat(ra->failed, ra->newsnap);
1689 		}
1690 		dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg);
1691 	}
1692 
1693 	if (err)
1694 		(void) strcpy(oldname, ra->failed);
1695 
1696 	dsl_sync_task_group_destroy(ra->dstg);
1697 	kmem_free(ra, sizeof (struct renamearg));
1698 	spa_close(spa, FTAG);
1699 	return (err);
1700 }
1701 
1702 static int
1703 dsl_valid_rename(char *oldname, void *arg)
1704 {
1705 	int delta = *(int *)arg;
1706 
1707 	if (strlen(oldname) + delta >= MAXNAMELEN)
1708 		return (ENAMETOOLONG);
1709 
1710 	return (0);
1711 }
1712 
1713 #pragma weak dmu_objset_rename = dsl_dataset_rename
1714 int
1715 dsl_dataset_rename(char *oldname, const char *newname,
1716     boolean_t recursive)
1717 {
1718 	dsl_dir_t *dd;
1719 	dsl_dataset_t *ds;
1720 	const char *tail;
1721 	int err;
1722 
1723 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
1724 	if (err)
1725 		return (err);
1726 	if (tail == NULL) {
1727 		int delta = strlen(newname) - strlen(oldname);
1728 
1729 		/* if we're growing, validate child size lengths */
1730 		if (delta > 0)
1731 			err = dmu_objset_find(oldname, dsl_valid_rename,
1732 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
1733 
1734 		if (!err)
1735 			err = dsl_dir_rename(dd, newname);
1736 		dsl_dir_close(dd, FTAG);
1737 		return (err);
1738 	}
1739 	if (tail[0] != '@') {
1740 		/* the name ended in a nonexistant component */
1741 		dsl_dir_close(dd, FTAG);
1742 		return (ENOENT);
1743 	}
1744 
1745 	dsl_dir_close(dd, FTAG);
1746 
1747 	/* new name must be snapshot in same filesystem */
1748 	tail = strchr(newname, '@');
1749 	if (tail == NULL)
1750 		return (EINVAL);
1751 	tail++;
1752 	if (strncmp(oldname, newname, tail - newname) != 0)
1753 		return (EXDEV);
1754 
1755 	if (recursive) {
1756 		err = dsl_recursive_rename(oldname, newname);
1757 	} else {
1758 		err = dsl_dataset_open(oldname,
1759 		    DS_MODE_READONLY | DS_MODE_STANDARD, FTAG, &ds);
1760 		if (err)
1761 			return (err);
1762 
1763 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1764 		    dsl_dataset_snapshot_rename_check,
1765 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
1766 
1767 		dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);
1768 	}
1769 
1770 	return (err);
1771 }
1772 
1773 struct promotearg {
1774 	uint64_t used, comp, uncomp, unique;
1775 	uint64_t newnext_obj, snapnames_obj;
1776 };
1777 
1778 /* ARGSUSED */
1779 static int
1780 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
1781 {
1782 	dsl_dataset_t *hds = arg1;
1783 	struct promotearg *pa = arg2;
1784 	dsl_dir_t *dd = hds->ds_dir;
1785 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
1786 	dsl_dir_t *pdd = NULL;
1787 	dsl_dataset_t *ds = NULL;
1788 	dsl_dataset_t *pivot_ds = NULL;
1789 	dsl_dataset_t *newnext_ds = NULL;
1790 	int err;
1791 	char *name = NULL;
1792 	uint64_t itor = 0;
1793 	blkptr_t bp;
1794 
1795 	bzero(pa, sizeof (*pa));
1796 
1797 	/* Check that it is a clone */
1798 	if (dd->dd_phys->dd_clone_parent_obj == 0)
1799 		return (EINVAL);
1800 
1801 	/* Since this is so expensive, don't do the preliminary check */
1802 	if (!dmu_tx_is_syncing(tx))
1803 		return (0);
1804 
1805 	if (err = dsl_dataset_open_obj(dp,
1806 	    dd->dd_phys->dd_clone_parent_obj,
1807 	    NULL, DS_MODE_EXCLUSIVE, FTAG, &pivot_ds))
1808 		goto out;
1809 	pdd = pivot_ds->ds_dir;
1810 
1811 	{
1812 		dsl_dataset_t *phds;
1813 		if (err = dsl_dataset_open_obj(dd->dd_pool,
1814 		    pdd->dd_phys->dd_head_dataset_obj,
1815 		    NULL, DS_MODE_NONE, FTAG, &phds))
1816 			goto out;
1817 		pa->snapnames_obj = phds->ds_phys->ds_snapnames_zapobj;
1818 		dsl_dataset_close(phds, DS_MODE_NONE, FTAG);
1819 	}
1820 
1821 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
1822 		err = EXDEV;
1823 		goto out;
1824 	}
1825 
1826 	/* find pivot point's new next ds */
1827 	VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, hds->ds_object,
1828 	    NULL, DS_MODE_NONE, FTAG, &newnext_ds));
1829 	while (newnext_ds->ds_phys->ds_prev_snap_obj != pivot_ds->ds_object) {
1830 		dsl_dataset_t *prev;
1831 
1832 		if (err = dsl_dataset_open_obj(dd->dd_pool,
1833 		    newnext_ds->ds_phys->ds_prev_snap_obj,
1834 		    NULL, DS_MODE_NONE, FTAG, &prev))
1835 			goto out;
1836 		dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG);
1837 		newnext_ds = prev;
1838 	}
1839 	pa->newnext_obj = newnext_ds->ds_object;
1840 
1841 	/* compute pivot point's new unique space */
1842 	while ((err = bplist_iterate(&newnext_ds->ds_deadlist,
1843 	    &itor, &bp)) == 0) {
1844 		if (bp.blk_birth > pivot_ds->ds_phys->ds_prev_snap_txg)
1845 			pa->unique += bp_get_dasize(dd->dd_pool->dp_spa, &bp);
1846 	}
1847 	if (err != ENOENT)
1848 		goto out;
1849 
1850 	/* Walk the snapshots that we are moving */
1851 	name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1852 	ds = pivot_ds;
1853 	/* CONSTCOND */
1854 	while (TRUE) {
1855 		uint64_t val, dlused, dlcomp, dluncomp;
1856 		dsl_dataset_t *prev;
1857 
1858 		/* Check that the snapshot name does not conflict */
1859 		dsl_dataset_name(ds, name);
1860 		err = zap_lookup(dd->dd_pool->dp_meta_objset,
1861 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
1862 		    8, 1, &val);
1863 		if (err != ENOENT) {
1864 			if (err == 0)
1865 				err = EEXIST;
1866 			goto out;
1867 		}
1868 
1869 		/*
1870 		 * compute space to transfer.  Each snapshot gave birth to:
1871 		 * (my used) - (prev's used) + (deadlist's used)
1872 		 */
1873 		pa->used += ds->ds_phys->ds_used_bytes;
1874 		pa->comp += ds->ds_phys->ds_compressed_bytes;
1875 		pa->uncomp += ds->ds_phys->ds_uncompressed_bytes;
1876 
1877 		/* If we reach the first snapshot, we're done. */
1878 		if (ds->ds_phys->ds_prev_snap_obj == 0)
1879 			break;
1880 
1881 		if (err = bplist_space(&ds->ds_deadlist,
1882 		    &dlused, &dlcomp, &dluncomp))
1883 			goto out;
1884 		if (err = dsl_dataset_open_obj(dd->dd_pool,
1885 		    ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE,
1886 		    FTAG, &prev))
1887 			goto out;
1888 		pa->used += dlused - prev->ds_phys->ds_used_bytes;
1889 		pa->comp += dlcomp - prev->ds_phys->ds_compressed_bytes;
1890 		pa->uncomp += dluncomp - prev->ds_phys->ds_uncompressed_bytes;
1891 
1892 		/*
1893 		 * We could be a clone of a clone.  If we reach our
1894 		 * parent's branch point, we're done.
1895 		 */
1896 		if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
1897 			dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG);
1898 			break;
1899 		}
1900 		if (ds != pivot_ds)
1901 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
1902 		ds = prev;
1903 	}
1904 
1905 	/* Check that there is enough space here */
1906 	err = dsl_dir_transfer_possible(pdd, dd, pa->used);
1907 
1908 out:
1909 	if (ds && ds != pivot_ds)
1910 		dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
1911 	if (pivot_ds)
1912 		dsl_dataset_close(pivot_ds, DS_MODE_EXCLUSIVE, FTAG);
1913 	if (newnext_ds)
1914 		dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG);
1915 	if (name)
1916 		kmem_free(name, MAXPATHLEN);
1917 	return (err);
1918 }
1919 
1920 static void
1921 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1922 {
1923 	dsl_dataset_t *hds = arg1;
1924 	struct promotearg *pa = arg2;
1925 	dsl_dir_t *dd = hds->ds_dir;
1926 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
1927 	dsl_dir_t *pdd = NULL;
1928 	dsl_dataset_t *ds, *pivot_ds;
1929 	char *name;
1930 
1931 	ASSERT(dd->dd_phys->dd_clone_parent_obj != 0);
1932 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
1933 
1934 	VERIFY(0 == dsl_dataset_open_obj(dp,
1935 	    dd->dd_phys->dd_clone_parent_obj,
1936 	    NULL, DS_MODE_EXCLUSIVE, FTAG, &pivot_ds));
1937 	/*
1938 	 * We need to explicitly open pdd, since pivot_ds's pdd will be
1939 	 * changing.
1940 	 */
1941 	VERIFY(0 == dsl_dir_open_obj(dp, pivot_ds->ds_dir->dd_object,
1942 	    NULL, FTAG, &pdd));
1943 
1944 	/* move snapshots to this dir */
1945 	name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1946 	ds = pivot_ds;
1947 	/* CONSTCOND */
1948 	while (TRUE) {
1949 		dsl_dataset_t *prev;
1950 
1951 		/* move snap name entry */
1952 		dsl_dataset_name(ds, name);
1953 		VERIFY(0 == zap_remove(dp->dp_meta_objset,
1954 		    pa->snapnames_obj, ds->ds_snapname, tx));
1955 		VERIFY(0 == zap_add(dp->dp_meta_objset,
1956 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
1957 		    8, 1, &ds->ds_object, tx));
1958 
1959 		/* change containing dsl_dir */
1960 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1961 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, pdd->dd_object);
1962 		ds->ds_phys->ds_dir_obj = dd->dd_object;
1963 		ASSERT3P(ds->ds_dir, ==, pdd);
1964 		dsl_dir_close(ds->ds_dir, ds);
1965 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
1966 		    NULL, ds, &ds->ds_dir));
1967 
1968 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
1969 
1970 		if (ds->ds_phys->ds_prev_snap_obj == 0)
1971 			break;
1972 
1973 		VERIFY(0 == dsl_dataset_open_obj(dp,
1974 		    ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE,
1975 		    FTAG, &prev));
1976 
1977 		if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
1978 			dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG);
1979 			break;
1980 		}
1981 		if (ds != pivot_ds)
1982 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
1983 		ds = prev;
1984 	}
1985 	if (ds != pivot_ds)
1986 		dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
1987 
1988 	/* change pivot point's next snap */
1989 	dmu_buf_will_dirty(pivot_ds->ds_dbuf, tx);
1990 	pivot_ds->ds_phys->ds_next_snap_obj = pa->newnext_obj;
1991 
1992 	/* change clone_parent-age */
1993 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1994 	ASSERT3U(dd->dd_phys->dd_clone_parent_obj, ==, pivot_ds->ds_object);
1995 	dd->dd_phys->dd_clone_parent_obj = pdd->dd_phys->dd_clone_parent_obj;
1996 	dmu_buf_will_dirty(pdd->dd_dbuf, tx);
1997 	pdd->dd_phys->dd_clone_parent_obj = pivot_ds->ds_object;
1998 
1999 	/* change space accounting */
2000 	dsl_dir_diduse_space(pdd, -pa->used, -pa->comp, -pa->uncomp, tx);
2001 	dsl_dir_diduse_space(dd, pa->used, pa->comp, pa->uncomp, tx);
2002 	pivot_ds->ds_phys->ds_unique_bytes = pa->unique;
2003 
2004 	/* log history record */
2005 	spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2006 	    cr, "dataset = %llu", ds->ds_object);
2007 
2008 	dsl_dir_close(pdd, FTAG);
2009 	dsl_dataset_close(pivot_ds, DS_MODE_EXCLUSIVE, FTAG);
2010 	kmem_free(name, MAXPATHLEN);
2011 }
2012 
2013 int
2014 dsl_dataset_promote(const char *name)
2015 {
2016 	dsl_dataset_t *ds;
2017 	int err;
2018 	dmu_object_info_t doi;
2019 	struct promotearg pa;
2020 
2021 	err = dsl_dataset_open(name, DS_MODE_NONE, FTAG, &ds);
2022 	if (err)
2023 		return (err);
2024 
2025 	err = dmu_object_info(ds->ds_dir->dd_pool->dp_meta_objset,
2026 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2027 	if (err) {
2028 		dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2029 		return (err);
2030 	}
2031 
2032 	/*
2033 	 * Add in 128x the snapnames zapobj size, since we will be moving
2034 	 * a bunch of snapnames to the promoted ds, and dirtying their
2035 	 * bonus buffers.
2036 	 */
2037 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2038 	    dsl_dataset_promote_check,
2039 	    dsl_dataset_promote_sync, ds, &pa, 2 + 2 * doi.doi_physical_blks);
2040 	dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2041 	return (err);
2042 }
2043 
2044 /*
2045  * Given a pool name and a dataset object number in that pool,
2046  * return the name of that dataset.
2047  */
2048 int
2049 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2050 {
2051 	spa_t *spa;
2052 	dsl_pool_t *dp;
2053 	dsl_dataset_t *ds = NULL;
2054 	int error;
2055 
2056 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
2057 		return (error);
2058 	dp = spa_get_dsl(spa);
2059 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2060 	if ((error = dsl_dataset_open_obj(dp, obj,
2061 	    NULL, DS_MODE_NONE, FTAG, &ds)) != 0) {
2062 		rw_exit(&dp->dp_config_rwlock);
2063 		spa_close(spa, FTAG);
2064 		return (error);
2065 	}
2066 	dsl_dataset_name(ds, buf);
2067 	dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2068 	rw_exit(&dp->dp_config_rwlock);
2069 	spa_close(spa, FTAG);
2070 
2071 	return (0);
2072 }
2073