xref: /titanic_41/usr/src/uts/common/fs/zfs/dmu_objset.c (revision bafec74292ca6805e5acb387856f4e60a5314b37)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/cred.h>
27 #include <sys/zfs_context.h>
28 #include <sys/dmu_objset.h>
29 #include <sys/dsl_dir.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_prop.h>
32 #include <sys/dsl_pool.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dsl_deleg.h>
35 #include <sys/dnode.h>
36 #include <sys/dbuf.h>
37 #include <sys/zvol.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/zap.h>
40 #include <sys/zil.h>
41 #include <sys/dmu_impl.h>
42 #include <sys/zfs_ioctl.h>
43 
44 spa_t *
45 dmu_objset_spa(objset_t *os)
46 {
47 	return (os->os_spa);
48 }
49 
50 zilog_t *
51 dmu_objset_zil(objset_t *os)
52 {
53 	return (os->os_zil);
54 }
55 
56 dsl_pool_t *
57 dmu_objset_pool(objset_t *os)
58 {
59 	dsl_dataset_t *ds;
60 
61 	if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
62 		return (ds->ds_dir->dd_pool);
63 	else
64 		return (spa_get_dsl(os->os_spa));
65 }
66 
67 dsl_dataset_t *
68 dmu_objset_ds(objset_t *os)
69 {
70 	return (os->os_dsl_dataset);
71 }
72 
73 dmu_objset_type_t
74 dmu_objset_type(objset_t *os)
75 {
76 	return (os->os_phys->os_type);
77 }
78 
79 void
80 dmu_objset_name(objset_t *os, char *buf)
81 {
82 	dsl_dataset_name(os->os_dsl_dataset, buf);
83 }
84 
85 uint64_t
86 dmu_objset_id(objset_t *os)
87 {
88 	dsl_dataset_t *ds = os->os_dsl_dataset;
89 
90 	return (ds ? ds->ds_object : 0);
91 }
92 
93 uint64_t
94 dmu_objset_logbias(objset_t *os)
95 {
96 	return (os->os_logbias);
97 }
98 
99 static void
100 checksum_changed_cb(void *arg, uint64_t newval)
101 {
102 	objset_t *os = arg;
103 
104 	/*
105 	 * Inheritance should have been done by now.
106 	 */
107 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
108 
109 	os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
110 }
111 
112 static void
113 compression_changed_cb(void *arg, uint64_t newval)
114 {
115 	objset_t *os = arg;
116 
117 	/*
118 	 * Inheritance and range checking should have been done by now.
119 	 */
120 	ASSERT(newval != ZIO_COMPRESS_INHERIT);
121 
122 	os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
123 }
124 
125 static void
126 copies_changed_cb(void *arg, uint64_t newval)
127 {
128 	objset_t *os = arg;
129 
130 	/*
131 	 * Inheritance and range checking should have been done by now.
132 	 */
133 	ASSERT(newval > 0);
134 	ASSERT(newval <= spa_max_replication(os->os_spa));
135 
136 	os->os_copies = newval;
137 }
138 
139 static void
140 dedup_changed_cb(void *arg, uint64_t newval)
141 {
142 	objset_t *os = arg;
143 	spa_t *spa = os->os_spa;
144 	enum zio_checksum checksum;
145 
146 	/*
147 	 * Inheritance should have been done by now.
148 	 */
149 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
150 
151 	checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
152 
153 	os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
154 	os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
155 }
156 
157 static void
158 primary_cache_changed_cb(void *arg, uint64_t newval)
159 {
160 	objset_t *os = arg;
161 
162 	/*
163 	 * Inheritance and range checking should have been done by now.
164 	 */
165 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
166 	    newval == ZFS_CACHE_METADATA);
167 
168 	os->os_primary_cache = newval;
169 }
170 
171 static void
172 secondary_cache_changed_cb(void *arg, uint64_t newval)
173 {
174 	objset_t *os = arg;
175 
176 	/*
177 	 * Inheritance and range checking should have been done by now.
178 	 */
179 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
180 	    newval == ZFS_CACHE_METADATA);
181 
182 	os->os_secondary_cache = newval;
183 }
184 
185 static void
186 logbias_changed_cb(void *arg, uint64_t newval)
187 {
188 	objset_t *os = arg;
189 
190 	ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
191 	    newval == ZFS_LOGBIAS_THROUGHPUT);
192 	os->os_logbias = newval;
193 	if (os->os_zil)
194 		zil_set_logbias(os->os_zil, newval);
195 }
196 
197 void
198 dmu_objset_byteswap(void *buf, size_t size)
199 {
200 	objset_phys_t *osp = buf;
201 
202 	ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
203 	dnode_byteswap(&osp->os_meta_dnode);
204 	byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
205 	osp->os_type = BSWAP_64(osp->os_type);
206 	osp->os_flags = BSWAP_64(osp->os_flags);
207 	if (size == sizeof (objset_phys_t)) {
208 		dnode_byteswap(&osp->os_userused_dnode);
209 		dnode_byteswap(&osp->os_groupused_dnode);
210 	}
211 }
212 
213 int
214 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
215     objset_t **osp)
216 {
217 	objset_t *os;
218 	int i, err;
219 
220 	ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
221 
222 	os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
223 	os->os_dsl_dataset = ds;
224 	os->os_spa = spa;
225 	os->os_rootbp = bp;
226 	if (!BP_IS_HOLE(os->os_rootbp)) {
227 		uint32_t aflags = ARC_WAIT;
228 		zbookmark_t zb;
229 		SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
230 		    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
231 
232 		if (DMU_OS_IS_L2CACHEABLE(os))
233 			aflags |= ARC_L2CACHE;
234 
235 		dprintf_bp(os->os_rootbp, "reading %s", "");
236 		/*
237 		 * NB: when bprewrite scrub can change the bp,
238 		 * and this is called from dmu_objset_open_ds_os, the bp
239 		 * could change, and we'll need a lock.
240 		 */
241 		err = arc_read_nolock(NULL, spa, os->os_rootbp,
242 		    arc_getbuf_func, &os->os_phys_buf,
243 		    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
244 		if (err) {
245 			kmem_free(os, sizeof (objset_t));
246 			/* convert checksum errors into IO errors */
247 			if (err == ECKSUM)
248 				err = EIO;
249 			return (err);
250 		}
251 
252 		/* Increase the blocksize if we are permitted. */
253 		if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
254 		    arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
255 			arc_buf_t *buf = arc_buf_alloc(spa,
256 			    sizeof (objset_phys_t), &os->os_phys_buf,
257 			    ARC_BUFC_METADATA);
258 			bzero(buf->b_data, sizeof (objset_phys_t));
259 			bcopy(os->os_phys_buf->b_data, buf->b_data,
260 			    arc_buf_size(os->os_phys_buf));
261 			(void) arc_buf_remove_ref(os->os_phys_buf,
262 			    &os->os_phys_buf);
263 			os->os_phys_buf = buf;
264 		}
265 
266 		os->os_phys = os->os_phys_buf->b_data;
267 		os->os_flags = os->os_phys->os_flags;
268 	} else {
269 		int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
270 		    sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
271 		os->os_phys_buf = arc_buf_alloc(spa, size,
272 		    &os->os_phys_buf, ARC_BUFC_METADATA);
273 		os->os_phys = os->os_phys_buf->b_data;
274 		bzero(os->os_phys, size);
275 	}
276 
277 	/*
278 	 * Note: the changed_cb will be called once before the register
279 	 * func returns, thus changing the checksum/compression from the
280 	 * default (fletcher2/off).  Snapshots don't need to know about
281 	 * checksum/compression/copies.
282 	 */
283 	if (ds) {
284 		err = dsl_prop_register(ds, "primarycache",
285 		    primary_cache_changed_cb, os);
286 		if (err == 0)
287 			err = dsl_prop_register(ds, "secondarycache",
288 			    secondary_cache_changed_cb, os);
289 		if (!dsl_dataset_is_snapshot(ds)) {
290 			if (err == 0)
291 				err = dsl_prop_register(ds, "checksum",
292 				    checksum_changed_cb, os);
293 			if (err == 0)
294 				err = dsl_prop_register(ds, "compression",
295 				    compression_changed_cb, os);
296 			if (err == 0)
297 				err = dsl_prop_register(ds, "copies",
298 				    copies_changed_cb, os);
299 			if (err == 0)
300 				err = dsl_prop_register(ds, "dedup",
301 				    dedup_changed_cb, os);
302 			if (err == 0)
303 				err = dsl_prop_register(ds, "logbias",
304 				    logbias_changed_cb, os);
305 		}
306 		if (err) {
307 			VERIFY(arc_buf_remove_ref(os->os_phys_buf,
308 			    &os->os_phys_buf) == 1);
309 			kmem_free(os, sizeof (objset_t));
310 			return (err);
311 		}
312 	} else if (ds == NULL) {
313 		/* It's the meta-objset. */
314 		os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
315 		os->os_compress = ZIO_COMPRESS_LZJB;
316 		os->os_copies = spa_max_replication(spa);
317 		os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
318 		os->os_dedup_verify = 0;
319 		os->os_logbias = 0;
320 		os->os_primary_cache = ZFS_CACHE_ALL;
321 		os->os_secondary_cache = ZFS_CACHE_ALL;
322 	}
323 
324 	os->os_zil_header = os->os_phys->os_zil_header;
325 	os->os_zil = zil_alloc(os, &os->os_zil_header);
326 
327 	for (i = 0; i < TXG_SIZE; i++) {
328 		list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
329 		    offsetof(dnode_t, dn_dirty_link[i]));
330 		list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
331 		    offsetof(dnode_t, dn_dirty_link[i]));
332 	}
333 	list_create(&os->os_dnodes, sizeof (dnode_t),
334 	    offsetof(dnode_t, dn_link));
335 	list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
336 	    offsetof(dmu_buf_impl_t, db_link));
337 
338 	mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
339 	mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
340 	mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
341 
342 	os->os_meta_dnode = dnode_special_open(os,
343 	    &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT);
344 	if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
345 		os->os_userused_dnode = dnode_special_open(os,
346 		    &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT);
347 		os->os_groupused_dnode = dnode_special_open(os,
348 		    &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT);
349 	}
350 
351 	/*
352 	 * We should be the only thread trying to do this because we
353 	 * have ds_opening_lock
354 	 */
355 	if (ds) {
356 		mutex_enter(&ds->ds_lock);
357 		ASSERT(ds->ds_objset == NULL);
358 		ds->ds_objset = os;
359 		mutex_exit(&ds->ds_lock);
360 	}
361 
362 	*osp = os;
363 	return (0);
364 }
365 
366 int
367 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
368 {
369 	int err = 0;
370 
371 	mutex_enter(&ds->ds_opening_lock);
372 	*osp = ds->ds_objset;
373 	if (*osp == NULL) {
374 		err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
375 		    ds, &ds->ds_phys->ds_bp, osp);
376 	}
377 	mutex_exit(&ds->ds_opening_lock);
378 	return (err);
379 }
380 
381 /* called from zpl */
382 int
383 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
384 {
385 	dsl_dataset_t *ds;
386 	int err;
387 
388 	err = dsl_dataset_hold(name, tag, &ds);
389 	if (err)
390 		return (err);
391 
392 	err = dmu_objset_from_ds(ds, osp);
393 	if (err)
394 		dsl_dataset_rele(ds, tag);
395 
396 	return (err);
397 }
398 
399 /* called from zpl */
400 int
401 dmu_objset_own(const char *name, dmu_objset_type_t type,
402     boolean_t readonly, void *tag, objset_t **osp)
403 {
404 	dsl_dataset_t *ds;
405 	int err;
406 
407 	err = dsl_dataset_own(name, B_FALSE, tag, &ds);
408 	if (err)
409 		return (err);
410 
411 	err = dmu_objset_from_ds(ds, osp);
412 	if (err) {
413 		dsl_dataset_disown(ds, tag);
414 	} else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
415 		dmu_objset_disown(*osp, tag);
416 		return (EINVAL);
417 	} else if (!readonly && dsl_dataset_is_snapshot(ds)) {
418 		dmu_objset_disown(*osp, tag);
419 		return (EROFS);
420 	}
421 	return (err);
422 }
423 
424 void
425 dmu_objset_rele(objset_t *os, void *tag)
426 {
427 	dsl_dataset_rele(os->os_dsl_dataset, tag);
428 }
429 
430 void
431 dmu_objset_disown(objset_t *os, void *tag)
432 {
433 	dsl_dataset_disown(os->os_dsl_dataset, tag);
434 }
435 
436 int
437 dmu_objset_evict_dbufs(objset_t *os)
438 {
439 	dnode_t *dn;
440 
441 	mutex_enter(&os->os_lock);
442 
443 	/* process the mdn last, since the other dnodes have holds on it */
444 	list_remove(&os->os_dnodes, os->os_meta_dnode);
445 	list_insert_tail(&os->os_dnodes, os->os_meta_dnode);
446 
447 	/*
448 	 * Find the first dnode with holds.  We have to do this dance
449 	 * because dnode_add_ref() only works if you already have a
450 	 * hold.  If there are no holds then it has no dbufs so OK to
451 	 * skip.
452 	 */
453 	for (dn = list_head(&os->os_dnodes);
454 	    dn && !dnode_add_ref(dn, FTAG);
455 	    dn = list_next(&os->os_dnodes, dn))
456 		continue;
457 
458 	while (dn) {
459 		dnode_t *next_dn = dn;
460 
461 		do {
462 			next_dn = list_next(&os->os_dnodes, next_dn);
463 		} while (next_dn && !dnode_add_ref(next_dn, FTAG));
464 
465 		mutex_exit(&os->os_lock);
466 		dnode_evict_dbufs(dn);
467 		dnode_rele(dn, FTAG);
468 		mutex_enter(&os->os_lock);
469 		dn = next_dn;
470 	}
471 	mutex_exit(&os->os_lock);
472 	return (list_head(&os->os_dnodes) != os->os_meta_dnode);
473 }
474 
475 void
476 dmu_objset_evict(objset_t *os)
477 {
478 	dsl_dataset_t *ds = os->os_dsl_dataset;
479 
480 	for (int t = 0; t < TXG_SIZE; t++)
481 		ASSERT(!dmu_objset_is_dirty(os, t));
482 
483 	if (ds) {
484 		if (!dsl_dataset_is_snapshot(ds)) {
485 			VERIFY(0 == dsl_prop_unregister(ds, "checksum",
486 			    checksum_changed_cb, os));
487 			VERIFY(0 == dsl_prop_unregister(ds, "compression",
488 			    compression_changed_cb, os));
489 			VERIFY(0 == dsl_prop_unregister(ds, "copies",
490 			    copies_changed_cb, os));
491 			VERIFY(0 == dsl_prop_unregister(ds, "dedup",
492 			    dedup_changed_cb, os));
493 			VERIFY(0 == dsl_prop_unregister(ds, "logbias",
494 			    logbias_changed_cb, os));
495 		}
496 		VERIFY(0 == dsl_prop_unregister(ds, "primarycache",
497 		    primary_cache_changed_cb, os));
498 		VERIFY(0 == dsl_prop_unregister(ds, "secondarycache",
499 		    secondary_cache_changed_cb, os));
500 	}
501 
502 	/*
503 	 * We should need only a single pass over the dnode list, since
504 	 * nothing can be added to the list at this point.
505 	 */
506 	(void) dmu_objset_evict_dbufs(os);
507 
508 	dnode_special_close(os->os_meta_dnode);
509 	if (os->os_userused_dnode) {
510 		dnode_special_close(os->os_userused_dnode);
511 		dnode_special_close(os->os_groupused_dnode);
512 	}
513 	zil_free(os->os_zil);
514 
515 	ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
516 
517 	VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf) == 1);
518 	mutex_destroy(&os->os_lock);
519 	mutex_destroy(&os->os_obj_lock);
520 	mutex_destroy(&os->os_user_ptr_lock);
521 	kmem_free(os, sizeof (objset_t));
522 }
523 
524 timestruc_t
525 dmu_objset_snap_cmtime(objset_t *os)
526 {
527 	return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
528 }
529 
530 /* called from dsl for meta-objset */
531 objset_t *
532 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
533     dmu_objset_type_t type, dmu_tx_t *tx)
534 {
535 	objset_t *os;
536 	dnode_t *mdn;
537 
538 	ASSERT(dmu_tx_is_syncing(tx));
539 	if (ds)
540 		mutex_enter(&ds->ds_opening_lock);
541 	VERIFY(0 == dmu_objset_open_impl(spa, ds, bp, &os));
542 	if (ds)
543 		mutex_exit(&ds->ds_opening_lock);
544 	mdn = os->os_meta_dnode;
545 
546 	dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
547 	    DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
548 
549 	/*
550 	 * We don't want to have to increase the meta-dnode's nlevels
551 	 * later, because then we could do it in quescing context while
552 	 * we are also accessing it in open context.
553 	 *
554 	 * This precaution is not necessary for the MOS (ds == NULL),
555 	 * because the MOS is only updated in syncing context.
556 	 * This is most fortunate: the MOS is the only objset that
557 	 * needs to be synced multiple times as spa_sync() iterates
558 	 * to convergence, so minimizing its dn_nlevels matters.
559 	 */
560 	if (ds != NULL) {
561 		int levels = 1;
562 
563 		/*
564 		 * Determine the number of levels necessary for the meta-dnode
565 		 * to contain DN_MAX_OBJECT dnodes.
566 		 */
567 		while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
568 		    (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
569 		    DN_MAX_OBJECT * sizeof (dnode_phys_t))
570 			levels++;
571 
572 		mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
573 		    mdn->dn_nlevels = levels;
574 	}
575 
576 	ASSERT(type != DMU_OST_NONE);
577 	ASSERT(type != DMU_OST_ANY);
578 	ASSERT(type < DMU_OST_NUMTYPES);
579 	os->os_phys->os_type = type;
580 	if (dmu_objset_userused_enabled(os)) {
581 		os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
582 		os->os_flags = os->os_phys->os_flags;
583 	}
584 
585 	dsl_dataset_dirty(ds, tx);
586 
587 	return (os);
588 }
589 
590 struct oscarg {
591 	void (*userfunc)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx);
592 	void *userarg;
593 	dsl_dataset_t *clone_origin;
594 	const char *lastname;
595 	dmu_objset_type_t type;
596 	uint64_t flags;
597 };
598 
599 /*ARGSUSED*/
600 static int
601 dmu_objset_create_check(void *arg1, void *arg2, dmu_tx_t *tx)
602 {
603 	dsl_dir_t *dd = arg1;
604 	struct oscarg *oa = arg2;
605 	objset_t *mos = dd->dd_pool->dp_meta_objset;
606 	int err;
607 	uint64_t ddobj;
608 
609 	err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
610 	    oa->lastname, sizeof (uint64_t), 1, &ddobj);
611 	if (err != ENOENT)
612 		return (err ? err : EEXIST);
613 
614 	if (oa->clone_origin != NULL) {
615 		/* You can't clone across pools. */
616 		if (oa->clone_origin->ds_dir->dd_pool != dd->dd_pool)
617 			return (EXDEV);
618 
619 		/* You can only clone snapshots, not the head datasets. */
620 		if (!dsl_dataset_is_snapshot(oa->clone_origin))
621 			return (EINVAL);
622 	}
623 
624 	return (0);
625 }
626 
627 static void
628 dmu_objset_create_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
629 {
630 	dsl_dir_t *dd = arg1;
631 	struct oscarg *oa = arg2;
632 	uint64_t dsobj;
633 
634 	ASSERT(dmu_tx_is_syncing(tx));
635 
636 	dsobj = dsl_dataset_create_sync(dd, oa->lastname,
637 	    oa->clone_origin, oa->flags, cr, tx);
638 
639 	if (oa->clone_origin == NULL) {
640 		dsl_dataset_t *ds;
641 		blkptr_t *bp;
642 		objset_t *os;
643 
644 		VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, dsobj,
645 		    FTAG, &ds));
646 		bp = dsl_dataset_get_blkptr(ds);
647 		ASSERT(BP_IS_HOLE(bp));
648 
649 		os = dmu_objset_create_impl(dsl_dataset_get_spa(ds),
650 		    ds, bp, oa->type, tx);
651 
652 		if (oa->userfunc)
653 			oa->userfunc(os, oa->userarg, cr, tx);
654 		dsl_dataset_rele(ds, FTAG);
655 	}
656 
657 	spa_history_internal_log(LOG_DS_CREATE, dd->dd_pool->dp_spa,
658 	    tx, cr, "dataset = %llu", dsobj);
659 }
660 
661 int
662 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
663     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
664 {
665 	dsl_dir_t *pdd;
666 	const char *tail;
667 	int err = 0;
668 	struct oscarg oa = { 0 };
669 
670 	ASSERT(strchr(name, '@') == NULL);
671 	err = dsl_dir_open(name, FTAG, &pdd, &tail);
672 	if (err)
673 		return (err);
674 	if (tail == NULL) {
675 		dsl_dir_close(pdd, FTAG);
676 		return (EEXIST);
677 	}
678 
679 	oa.userfunc = func;
680 	oa.userarg = arg;
681 	oa.lastname = tail;
682 	oa.type = type;
683 	oa.flags = flags;
684 
685 	err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
686 	    dmu_objset_create_sync, pdd, &oa, 5);
687 	dsl_dir_close(pdd, FTAG);
688 	return (err);
689 }
690 
691 int
692 dmu_objset_clone(const char *name, dsl_dataset_t *clone_origin, uint64_t flags)
693 {
694 	dsl_dir_t *pdd;
695 	const char *tail;
696 	int err = 0;
697 	struct oscarg oa = { 0 };
698 
699 	ASSERT(strchr(name, '@') == NULL);
700 	err = dsl_dir_open(name, FTAG, &pdd, &tail);
701 	if (err)
702 		return (err);
703 	if (tail == NULL) {
704 		dsl_dir_close(pdd, FTAG);
705 		return (EEXIST);
706 	}
707 
708 	oa.lastname = tail;
709 	oa.clone_origin = clone_origin;
710 	oa.flags = flags;
711 
712 	err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
713 	    dmu_objset_create_sync, pdd, &oa, 5);
714 	dsl_dir_close(pdd, FTAG);
715 	return (err);
716 }
717 
718 int
719 dmu_objset_destroy(const char *name, boolean_t defer)
720 {
721 	dsl_dataset_t *ds;
722 	int error;
723 
724 	/*
725 	 * dsl_dataset_destroy() can free any claimed-but-unplayed
726 	 * intent log, but if there is an active log, it has blocks that
727 	 * are allocated, but may not yet be reflected in the on-disk
728 	 * structure.  Only the ZIL knows how to free them, so we have
729 	 * to call into it here.
730 	 */
731 	error = dsl_dataset_own(name, B_TRUE, FTAG, &ds);
732 	if (error == 0) {
733 		objset_t *os;
734 		if (dmu_objset_from_ds(ds, &os) == 0)
735 			zil_destroy(dmu_objset_zil(os), B_FALSE);
736 		error = dsl_dataset_destroy(ds, FTAG, defer);
737 		/* dsl_dataset_destroy() closes the ds. */
738 	}
739 
740 	return (error);
741 }
742 
743 struct snaparg {
744 	dsl_sync_task_group_t *dstg;
745 	char *snapname;
746 	char failed[MAXPATHLEN];
747 	boolean_t checkperms;
748 	nvlist_t *props;
749 };
750 
751 static int
752 snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
753 {
754 	objset_t *os = arg1;
755 	struct snaparg *sn = arg2;
756 
757 	/* The props have already been checked by zfs_check_userprops(). */
758 
759 	return (dsl_dataset_snapshot_check(os->os_dsl_dataset,
760 	    sn->snapname, tx));
761 }
762 
763 static void
764 snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
765 {
766 	objset_t *os = arg1;
767 	dsl_dataset_t *ds = os->os_dsl_dataset;
768 	struct snaparg *sn = arg2;
769 
770 	dsl_dataset_snapshot_sync(ds, sn->snapname, cr, tx);
771 
772 	if (sn->props)
773 		dsl_props_set_sync(ds->ds_prev, sn->props, cr, tx);
774 }
775 
776 static int
777 dmu_objset_snapshot_one(char *name, void *arg)
778 {
779 	struct snaparg *sn = arg;
780 	objset_t *os;
781 	int err;
782 
783 	(void) strcpy(sn->failed, name);
784 
785 	/*
786 	 * Check permissions only when requested.  This only applies when
787 	 * doing a recursive snapshot.  The permission checks for the starting
788 	 * dataset have already been performed in zfs_secpolicy_snapshot()
789 	 */
790 	if (sn->checkperms == B_TRUE &&
791 	    (err = zfs_secpolicy_snapshot_perms(name, CRED())))
792 		return (err);
793 
794 	err = dmu_objset_hold(name, sn, &os);
795 	if (err != 0)
796 		return (err);
797 
798 	/* If the objset is in an inconsistent state, return busy */
799 	if (os->os_dsl_dataset->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) {
800 		dmu_objset_rele(os, sn);
801 		return (EBUSY);
802 	}
803 
804 	/*
805 	 * NB: we need to wait for all in-flight changes to get to disk,
806 	 * so that we snapshot those changes.  zil_suspend does this as
807 	 * a side effect.
808 	 */
809 	err = zil_suspend(dmu_objset_zil(os));
810 	if (err == 0) {
811 		dsl_sync_task_create(sn->dstg, snapshot_check,
812 		    snapshot_sync, os, sn, 3);
813 	} else {
814 		dmu_objset_rele(os, sn);
815 	}
816 
817 	return (err);
818 }
819 
820 int
821 dmu_objset_snapshot(char *fsname, char *snapname,
822     nvlist_t *props, boolean_t recursive)
823 {
824 	dsl_sync_task_t *dst;
825 	struct snaparg sn;
826 	spa_t *spa;
827 	int err;
828 
829 	(void) strcpy(sn.failed, fsname);
830 
831 	err = spa_open(fsname, &spa, FTAG);
832 	if (err)
833 		return (err);
834 
835 	sn.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
836 	sn.snapname = snapname;
837 	sn.props = props;
838 
839 	if (recursive) {
840 		sn.checkperms = B_TRUE;
841 		err = dmu_objset_find(fsname,
842 		    dmu_objset_snapshot_one, &sn, DS_FIND_CHILDREN);
843 	} else {
844 		sn.checkperms = B_FALSE;
845 		err = dmu_objset_snapshot_one(fsname, &sn);
846 	}
847 
848 	if (err == 0)
849 		err = dsl_sync_task_group_wait(sn.dstg);
850 
851 	for (dst = list_head(&sn.dstg->dstg_tasks); dst;
852 	    dst = list_next(&sn.dstg->dstg_tasks, dst)) {
853 		objset_t *os = dst->dst_arg1;
854 		dsl_dataset_t *ds = os->os_dsl_dataset;
855 		if (dst->dst_err)
856 			dsl_dataset_name(ds, sn.failed);
857 		zil_resume(dmu_objset_zil(os));
858 		dmu_objset_rele(os, &sn);
859 	}
860 
861 	if (err)
862 		(void) strcpy(fsname, sn.failed);
863 	dsl_sync_task_group_destroy(sn.dstg);
864 	spa_close(spa, FTAG);
865 	return (err);
866 }
867 
868 static void
869 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
870 {
871 	dnode_t *dn;
872 
873 	while (dn = list_head(list)) {
874 		ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
875 		ASSERT(dn->dn_dbuf->db_data_pending);
876 		/*
877 		 * Initialize dn_zio outside dnode_sync() because the
878 		 * meta-dnode needs to set it ouside dnode_sync().
879 		 */
880 		dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
881 		ASSERT(dn->dn_zio);
882 
883 		ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
884 		list_remove(list, dn);
885 
886 		if (newlist) {
887 			(void) dnode_add_ref(dn, newlist);
888 			list_insert_tail(newlist, dn);
889 		}
890 
891 		dnode_sync(dn, tx);
892 	}
893 }
894 
895 /* ARGSUSED */
896 static void
897 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
898 {
899 	blkptr_t *bp = zio->io_bp;
900 	objset_t *os = arg;
901 	dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
902 
903 	ASSERT(bp == os->os_rootbp);
904 	ASSERT(BP_GET_TYPE(bp) == DMU_OT_OBJSET);
905 	ASSERT(BP_GET_LEVEL(bp) == 0);
906 
907 	/*
908 	 * Update rootbp fill count: it should be the number of objects
909 	 * allocated in the object set (not counting the "special"
910 	 * objects that are stored in the objset_phys_t -- the meta
911 	 * dnode and user/group accounting objects).
912 	 */
913 	bp->blk_fill = 0;
914 	for (int i = 0; i < dnp->dn_nblkptr; i++)
915 		bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
916 }
917 
918 /* ARGSUSED */
919 static void
920 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
921 {
922 	blkptr_t *bp = zio->io_bp;
923 	blkptr_t *bp_orig = &zio->io_bp_orig;
924 	objset_t *os = arg;
925 
926 	if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
927 		ASSERT(BP_EQUAL(bp, bp_orig));
928 	} else {
929 		dsl_dataset_t *ds = os->os_dsl_dataset;
930 		dmu_tx_t *tx = os->os_synctx;
931 
932 		(void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
933 		dsl_dataset_block_born(ds, bp, tx);
934 	}
935 }
936 
937 /* called from dsl */
938 void
939 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
940 {
941 	int txgoff;
942 	zbookmark_t zb;
943 	zio_prop_t zp;
944 	zio_t *zio;
945 	list_t *list;
946 	list_t *newlist = NULL;
947 	dbuf_dirty_record_t *dr;
948 
949 	dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
950 
951 	ASSERT(dmu_tx_is_syncing(tx));
952 	/* XXX the write_done callback should really give us the tx... */
953 	os->os_synctx = tx;
954 
955 	if (os->os_dsl_dataset == NULL) {
956 		/*
957 		 * This is the MOS.  If we have upgraded,
958 		 * spa_max_replication() could change, so reset
959 		 * os_copies here.
960 		 */
961 		os->os_copies = spa_max_replication(os->os_spa);
962 	}
963 
964 	/*
965 	 * Create the root block IO
966 	 */
967 	arc_release(os->os_phys_buf, &os->os_phys_buf);
968 
969 	SET_BOOKMARK(&zb, os->os_dsl_dataset ?
970 	    os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
971 	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
972 
973 	dmu_write_policy(os, NULL, 0, 0, &zp);
974 
975 	zio = arc_write(pio, os->os_spa, tx->tx_txg,
976 	    os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), &zp,
977 	    dmu_objset_write_ready, dmu_objset_write_done, os,
978 	    ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
979 
980 	/*
981 	 * Sync special dnodes - the parent IO for the sync is the root block
982 	 */
983 	os->os_meta_dnode->dn_zio = zio;
984 	dnode_sync(os->os_meta_dnode, tx);
985 
986 	os->os_phys->os_flags = os->os_flags;
987 
988 	if (os->os_userused_dnode &&
989 	    os->os_userused_dnode->dn_type != DMU_OT_NONE) {
990 		os->os_userused_dnode->dn_zio = zio;
991 		dnode_sync(os->os_userused_dnode, tx);
992 		os->os_groupused_dnode->dn_zio = zio;
993 		dnode_sync(os->os_groupused_dnode, tx);
994 	}
995 
996 	txgoff = tx->tx_txg & TXG_MASK;
997 
998 	if (dmu_objset_userused_enabled(os)) {
999 		newlist = &os->os_synced_dnodes;
1000 		/*
1001 		 * We must create the list here because it uses the
1002 		 * dn_dirty_link[] of this txg.
1003 		 */
1004 		list_create(newlist, sizeof (dnode_t),
1005 		    offsetof(dnode_t, dn_dirty_link[txgoff]));
1006 	}
1007 
1008 	dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1009 	dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1010 
1011 	list = &os->os_meta_dnode->dn_dirty_records[txgoff];
1012 	while (dr = list_head(list)) {
1013 		ASSERT(dr->dr_dbuf->db_level == 0);
1014 		list_remove(list, dr);
1015 		if (dr->dr_zio)
1016 			zio_nowait(dr->dr_zio);
1017 	}
1018 	/*
1019 	 * Free intent log blocks up to this tx.
1020 	 */
1021 	zil_sync(os->os_zil, tx);
1022 	os->os_phys->os_zil_header = os->os_zil_header;
1023 	zio_nowait(zio);
1024 }
1025 
1026 boolean_t
1027 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1028 {
1029 	return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1030 	    !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1031 }
1032 
1033 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1034 
1035 void
1036 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1037 {
1038 	used_cbs[ost] = cb;
1039 }
1040 
1041 boolean_t
1042 dmu_objset_userused_enabled(objset_t *os)
1043 {
1044 	return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1045 	    used_cbs[os->os_phys->os_type] &&
1046 	    os->os_userused_dnode);
1047 }
1048 
1049 static void
1050 do_userquota_callback(objset_t *os, dnode_phys_t *dnp,
1051     boolean_t subtract, dmu_tx_t *tx)
1052 {
1053 	static const char zerobuf[DN_MAX_BONUSLEN] = {0};
1054 	uint64_t user, group;
1055 
1056 	ASSERT(dnp->dn_type != 0 ||
1057 	    (bcmp(DN_BONUS(dnp), zerobuf, DN_MAX_BONUSLEN) == 0 &&
1058 	    DN_USED_BYTES(dnp) == 0));
1059 
1060 	if ((dnp->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) &&
1061 	    0 == used_cbs[os->os_phys->os_type](dnp->dn_bonustype,
1062 	    DN_BONUS(dnp), &user, &group)) {
1063 		int64_t delta = DNODE_SIZE + DN_USED_BYTES(dnp);
1064 		if (subtract)
1065 			delta = -delta;
1066 		VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1067 		    user, delta, tx));
1068 		VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1069 		    group, delta, tx));
1070 	}
1071 }
1072 
1073 void
1074 dmu_objset_do_userquota_callbacks(objset_t *os, dmu_tx_t *tx)
1075 {
1076 	dnode_t *dn;
1077 	list_t *list = &os->os_synced_dnodes;
1078 
1079 	ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1080 
1081 	while (dn = list_head(list)) {
1082 		ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1083 		ASSERT(dn->dn_oldphys);
1084 		ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1085 		    dn->dn_phys->dn_flags &
1086 		    DNODE_FLAG_USERUSED_ACCOUNTED);
1087 
1088 		/* Allocate the user/groupused objects if necessary. */
1089 		if (os->os_userused_dnode->dn_type == DMU_OT_NONE) {
1090 			VERIFY(0 == zap_create_claim(os,
1091 			    DMU_USERUSED_OBJECT,
1092 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1093 			VERIFY(0 == zap_create_claim(os,
1094 			    DMU_GROUPUSED_OBJECT,
1095 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1096 		}
1097 
1098 		/*
1099 		 * We intentionally modify the zap object even if the
1100 		 * net delta (due to phys-oldphys) is zero.  Otherwise
1101 		 * the block of the zap obj could be shared between
1102 		 * datasets but need to be different between them after
1103 		 * a bprewrite.
1104 		 */
1105 		do_userquota_callback(os, dn->dn_oldphys, B_TRUE, tx);
1106 		do_userquota_callback(os, dn->dn_phys, B_FALSE, tx);
1107 
1108 		/*
1109 		 * The mutex is needed here for interlock with dnode_allocate.
1110 		 */
1111 		mutex_enter(&dn->dn_mtx);
1112 		zio_buf_free(dn->dn_oldphys, sizeof (dnode_phys_t));
1113 		dn->dn_oldphys = NULL;
1114 		mutex_exit(&dn->dn_mtx);
1115 
1116 		list_remove(list, dn);
1117 		dnode_rele(dn, list);
1118 	}
1119 }
1120 
1121 boolean_t
1122 dmu_objset_userspace_present(objset_t *os)
1123 {
1124 	return (os->os_phys->os_flags &
1125 	    OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1126 }
1127 
1128 int
1129 dmu_objset_userspace_upgrade(objset_t *os)
1130 {
1131 	uint64_t obj;
1132 	int err = 0;
1133 
1134 	if (dmu_objset_userspace_present(os))
1135 		return (0);
1136 	if (!dmu_objset_userused_enabled(os))
1137 		return (ENOTSUP);
1138 	if (dmu_objset_is_snapshot(os))
1139 		return (EINVAL);
1140 
1141 	/*
1142 	 * We simply need to mark every object dirty, so that it will be
1143 	 * synced out and now accounted.  If this is called
1144 	 * concurrently, or if we already did some work before crashing,
1145 	 * that's fine, since we track each object's accounted state
1146 	 * independently.
1147 	 */
1148 
1149 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1150 		dmu_tx_t *tx;
1151 		dmu_buf_t *db;
1152 		int objerr;
1153 
1154 		if (issig(JUSTLOOKING) && issig(FORREAL))
1155 			return (EINTR);
1156 
1157 		objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1158 		if (objerr)
1159 			continue;
1160 		tx = dmu_tx_create(os);
1161 		dmu_tx_hold_bonus(tx, obj);
1162 		objerr = dmu_tx_assign(tx, TXG_WAIT);
1163 		if (objerr) {
1164 			dmu_tx_abort(tx);
1165 			continue;
1166 		}
1167 		dmu_buf_will_dirty(db, tx);
1168 		dmu_buf_rele(db, FTAG);
1169 		dmu_tx_commit(tx);
1170 	}
1171 
1172 	os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1173 	txg_wait_synced(dmu_objset_pool(os), 0);
1174 	return (0);
1175 }
1176 
1177 void
1178 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1179     uint64_t *usedobjsp, uint64_t *availobjsp)
1180 {
1181 	dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1182 	    usedobjsp, availobjsp);
1183 }
1184 
1185 uint64_t
1186 dmu_objset_fsid_guid(objset_t *os)
1187 {
1188 	return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1189 }
1190 
1191 void
1192 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1193 {
1194 	stat->dds_type = os->os_phys->os_type;
1195 	if (os->os_dsl_dataset)
1196 		dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1197 }
1198 
1199 void
1200 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1201 {
1202 	ASSERT(os->os_dsl_dataset ||
1203 	    os->os_phys->os_type == DMU_OST_META);
1204 
1205 	if (os->os_dsl_dataset != NULL)
1206 		dsl_dataset_stats(os->os_dsl_dataset, nv);
1207 
1208 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1209 	    os->os_phys->os_type);
1210 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1211 	    dmu_objset_userspace_present(os));
1212 }
1213 
1214 int
1215 dmu_objset_is_snapshot(objset_t *os)
1216 {
1217 	if (os->os_dsl_dataset != NULL)
1218 		return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1219 	else
1220 		return (B_FALSE);
1221 }
1222 
1223 int
1224 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1225     boolean_t *conflict)
1226 {
1227 	dsl_dataset_t *ds = os->os_dsl_dataset;
1228 	uint64_t ignored;
1229 
1230 	if (ds->ds_phys->ds_snapnames_zapobj == 0)
1231 		return (ENOENT);
1232 
1233 	return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1234 	    ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1235 	    real, maxlen, conflict));
1236 }
1237 
1238 int
1239 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1240     uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1241 {
1242 	dsl_dataset_t *ds = os->os_dsl_dataset;
1243 	zap_cursor_t cursor;
1244 	zap_attribute_t attr;
1245 
1246 	if (ds->ds_phys->ds_snapnames_zapobj == 0)
1247 		return (ENOENT);
1248 
1249 	zap_cursor_init_serialized(&cursor,
1250 	    ds->ds_dir->dd_pool->dp_meta_objset,
1251 	    ds->ds_phys->ds_snapnames_zapobj, *offp);
1252 
1253 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1254 		zap_cursor_fini(&cursor);
1255 		return (ENOENT);
1256 	}
1257 
1258 	if (strlen(attr.za_name) + 1 > namelen) {
1259 		zap_cursor_fini(&cursor);
1260 		return (ENAMETOOLONG);
1261 	}
1262 
1263 	(void) strcpy(name, attr.za_name);
1264 	if (idp)
1265 		*idp = attr.za_first_integer;
1266 	if (case_conflict)
1267 		*case_conflict = attr.za_normalization_conflict;
1268 	zap_cursor_advance(&cursor);
1269 	*offp = zap_cursor_serialize(&cursor);
1270 	zap_cursor_fini(&cursor);
1271 
1272 	return (0);
1273 }
1274 
1275 int
1276 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1277     uint64_t *idp, uint64_t *offp)
1278 {
1279 	dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1280 	zap_cursor_t cursor;
1281 	zap_attribute_t attr;
1282 
1283 	/* there is no next dir on a snapshot! */
1284 	if (os->os_dsl_dataset->ds_object !=
1285 	    dd->dd_phys->dd_head_dataset_obj)
1286 		return (ENOENT);
1287 
1288 	zap_cursor_init_serialized(&cursor,
1289 	    dd->dd_pool->dp_meta_objset,
1290 	    dd->dd_phys->dd_child_dir_zapobj, *offp);
1291 
1292 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1293 		zap_cursor_fini(&cursor);
1294 		return (ENOENT);
1295 	}
1296 
1297 	if (strlen(attr.za_name) + 1 > namelen) {
1298 		zap_cursor_fini(&cursor);
1299 		return (ENAMETOOLONG);
1300 	}
1301 
1302 	(void) strcpy(name, attr.za_name);
1303 	if (idp)
1304 		*idp = attr.za_first_integer;
1305 	zap_cursor_advance(&cursor);
1306 	*offp = zap_cursor_serialize(&cursor);
1307 	zap_cursor_fini(&cursor);
1308 
1309 	return (0);
1310 }
1311 
1312 struct findarg {
1313 	int (*func)(char *, void *);
1314 	void *arg;
1315 };
1316 
1317 /* ARGSUSED */
1318 static int
1319 findfunc(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
1320 {
1321 	struct findarg *fa = arg;
1322 	return (fa->func((char *)dsname, fa->arg));
1323 }
1324 
1325 /*
1326  * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1327  * Perhaps change all callers to use dmu_objset_find_spa()?
1328  */
1329 int
1330 dmu_objset_find(char *name, int func(char *, void *), void *arg, int flags)
1331 {
1332 	struct findarg fa;
1333 	fa.func = func;
1334 	fa.arg = arg;
1335 	return (dmu_objset_find_spa(NULL, name, findfunc, &fa, flags));
1336 }
1337 
1338 /*
1339  * Find all objsets under name, call func on each
1340  */
1341 int
1342 dmu_objset_find_spa(spa_t *spa, const char *name,
1343     int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags)
1344 {
1345 	dsl_dir_t *dd;
1346 	dsl_pool_t *dp;
1347 	dsl_dataset_t *ds;
1348 	zap_cursor_t zc;
1349 	zap_attribute_t *attr;
1350 	char *child;
1351 	uint64_t thisobj;
1352 	int err;
1353 
1354 	if (name == NULL)
1355 		name = spa_name(spa);
1356 	err = dsl_dir_open_spa(spa, name, FTAG, &dd, NULL);
1357 	if (err)
1358 		return (err);
1359 
1360 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1361 	if (dd->dd_myname[0] == '$') {
1362 		dsl_dir_close(dd, FTAG);
1363 		return (0);
1364 	}
1365 
1366 	thisobj = dd->dd_phys->dd_head_dataset_obj;
1367 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1368 	dp = dd->dd_pool;
1369 
1370 	/*
1371 	 * Iterate over all children.
1372 	 */
1373 	if (flags & DS_FIND_CHILDREN) {
1374 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
1375 		    dd->dd_phys->dd_child_dir_zapobj);
1376 		    zap_cursor_retrieve(&zc, attr) == 0;
1377 		    (void) zap_cursor_advance(&zc)) {
1378 			ASSERT(attr->za_integer_length == sizeof (uint64_t));
1379 			ASSERT(attr->za_num_integers == 1);
1380 
1381 			child = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1382 			(void) strcpy(child, name);
1383 			(void) strcat(child, "/");
1384 			(void) strcat(child, attr->za_name);
1385 			err = dmu_objset_find_spa(spa, child, func, arg, flags);
1386 			kmem_free(child, MAXPATHLEN);
1387 			if (err)
1388 				break;
1389 		}
1390 		zap_cursor_fini(&zc);
1391 
1392 		if (err) {
1393 			dsl_dir_close(dd, FTAG);
1394 			kmem_free(attr, sizeof (zap_attribute_t));
1395 			return (err);
1396 		}
1397 	}
1398 
1399 	/*
1400 	 * Iterate over all snapshots.
1401 	 */
1402 	if (flags & DS_FIND_SNAPSHOTS) {
1403 		if (!dsl_pool_sync_context(dp))
1404 			rw_enter(&dp->dp_config_rwlock, RW_READER);
1405 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1406 		if (!dsl_pool_sync_context(dp))
1407 			rw_exit(&dp->dp_config_rwlock);
1408 
1409 		if (err == 0) {
1410 			uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1411 			dsl_dataset_rele(ds, FTAG);
1412 
1413 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1414 			    zap_cursor_retrieve(&zc, attr) == 0;
1415 			    (void) zap_cursor_advance(&zc)) {
1416 				ASSERT(attr->za_integer_length ==
1417 				    sizeof (uint64_t));
1418 				ASSERT(attr->za_num_integers == 1);
1419 
1420 				child = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1421 				(void) strcpy(child, name);
1422 				(void) strcat(child, "@");
1423 				(void) strcat(child, attr->za_name);
1424 				err = func(spa, attr->za_first_integer,
1425 				    child, arg);
1426 				kmem_free(child, MAXPATHLEN);
1427 				if (err)
1428 					break;
1429 			}
1430 			zap_cursor_fini(&zc);
1431 		}
1432 	}
1433 
1434 	dsl_dir_close(dd, FTAG);
1435 	kmem_free(attr, sizeof (zap_attribute_t));
1436 
1437 	if (err)
1438 		return (err);
1439 
1440 	/*
1441 	 * Apply to self if appropriate.
1442 	 */
1443 	err = func(spa, thisobj, name, arg);
1444 	return (err);
1445 }
1446 
1447 /* ARGSUSED */
1448 int
1449 dmu_objset_prefetch(char *name, void *arg)
1450 {
1451 	dsl_dataset_t *ds;
1452 
1453 	if (dsl_dataset_hold(name, FTAG, &ds))
1454 		return (0);
1455 
1456 	if (!BP_IS_HOLE(&ds->ds_phys->ds_bp)) {
1457 		mutex_enter(&ds->ds_opening_lock);
1458 		if (ds->ds_objset == NULL) {
1459 			uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
1460 			zbookmark_t zb;
1461 
1462 			SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT,
1463 			    ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1464 
1465 			(void) arc_read_nolock(NULL, dsl_dataset_get_spa(ds),
1466 			    &ds->ds_phys->ds_bp, NULL, NULL,
1467 			    ZIO_PRIORITY_ASYNC_READ,
1468 			    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
1469 			    &aflags, &zb);
1470 		}
1471 		mutex_exit(&ds->ds_opening_lock);
1472 	}
1473 
1474 	dsl_dataset_rele(ds, FTAG);
1475 	return (0);
1476 }
1477 
1478 void
1479 dmu_objset_set_user(objset_t *os, void *user_ptr)
1480 {
1481 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1482 	os->os_user_ptr = user_ptr;
1483 }
1484 
1485 void *
1486 dmu_objset_get_user(objset_t *os)
1487 {
1488 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1489 	return (os->os_user_ptr);
1490 }
1491