xref: /titanic_51/usr/src/uts/common/fs/zfs/dmu_objset.c (revision 177d5b5f8c0e969013441207a0a705ae66b08cf7)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
26  */
27 
28 /* Portions Copyright 2010 Robert Milkowski */
29 
30 #include <sys/cred.h>
31 #include <sys/zfs_context.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_prop.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dsl_deleg.h>
39 #include <sys/dnode.h>
40 #include <sys/dbuf.h>
41 #include <sys/zvol.h>
42 #include <sys/dmu_tx.h>
43 #include <sys/zap.h>
44 #include <sys/zil.h>
45 #include <sys/dmu_impl.h>
46 #include <sys/zfs_ioctl.h>
47 #include <sys/sa.h>
48 #include <sys/zfs_onexit.h>
49 #include <sys/dsl_destroy.h>
50 
51 /*
52  * Needed to close a window in dnode_move() that allows the objset to be freed
53  * before it can be safely accessed.
54  */
55 krwlock_t os_lock;
56 
57 void
58 dmu_objset_init(void)
59 {
60 	rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
61 }
62 
63 void
64 dmu_objset_fini(void)
65 {
66 	rw_destroy(&os_lock);
67 }
68 
69 spa_t *
70 dmu_objset_spa(objset_t *os)
71 {
72 	return (os->os_spa);
73 }
74 
75 zilog_t *
76 dmu_objset_zil(objset_t *os)
77 {
78 	return (os->os_zil);
79 }
80 
81 dsl_pool_t *
82 dmu_objset_pool(objset_t *os)
83 {
84 	dsl_dataset_t *ds;
85 
86 	if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
87 		return (ds->ds_dir->dd_pool);
88 	else
89 		return (spa_get_dsl(os->os_spa));
90 }
91 
92 dsl_dataset_t *
93 dmu_objset_ds(objset_t *os)
94 {
95 	return (os->os_dsl_dataset);
96 }
97 
98 dmu_objset_type_t
99 dmu_objset_type(objset_t *os)
100 {
101 	return (os->os_phys->os_type);
102 }
103 
104 void
105 dmu_objset_name(objset_t *os, char *buf)
106 {
107 	dsl_dataset_name(os->os_dsl_dataset, buf);
108 }
109 
110 uint64_t
111 dmu_objset_id(objset_t *os)
112 {
113 	dsl_dataset_t *ds = os->os_dsl_dataset;
114 
115 	return (ds ? ds->ds_object : 0);
116 }
117 
118 zfs_sync_type_t
119 dmu_objset_syncprop(objset_t *os)
120 {
121 	return (os->os_sync);
122 }
123 
124 zfs_logbias_op_t
125 dmu_objset_logbias(objset_t *os)
126 {
127 	return (os->os_logbias);
128 }
129 
130 static void
131 checksum_changed_cb(void *arg, uint64_t newval)
132 {
133 	objset_t *os = arg;
134 
135 	/*
136 	 * Inheritance should have been done by now.
137 	 */
138 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
139 
140 	os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
141 }
142 
143 static void
144 compression_changed_cb(void *arg, uint64_t newval)
145 {
146 	objset_t *os = arg;
147 
148 	/*
149 	 * Inheritance and range checking should have been done by now.
150 	 */
151 	ASSERT(newval != ZIO_COMPRESS_INHERIT);
152 
153 	os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
154 }
155 
156 static void
157 copies_changed_cb(void *arg, uint64_t newval)
158 {
159 	objset_t *os = arg;
160 
161 	/*
162 	 * Inheritance and range checking should have been done by now.
163 	 */
164 	ASSERT(newval > 0);
165 	ASSERT(newval <= spa_max_replication(os->os_spa));
166 
167 	os->os_copies = newval;
168 }
169 
170 static void
171 dedup_changed_cb(void *arg, uint64_t newval)
172 {
173 	objset_t *os = arg;
174 	spa_t *spa = os->os_spa;
175 	enum zio_checksum checksum;
176 
177 	/*
178 	 * Inheritance should have been done by now.
179 	 */
180 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
181 
182 	checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
183 
184 	os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
185 	os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
186 }
187 
188 static void
189 primary_cache_changed_cb(void *arg, uint64_t newval)
190 {
191 	objset_t *os = arg;
192 
193 	/*
194 	 * Inheritance and range checking should have been done by now.
195 	 */
196 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
197 	    newval == ZFS_CACHE_METADATA);
198 
199 	os->os_primary_cache = newval;
200 }
201 
202 static void
203 secondary_cache_changed_cb(void *arg, uint64_t newval)
204 {
205 	objset_t *os = arg;
206 
207 	/*
208 	 * Inheritance and range checking should have been done by now.
209 	 */
210 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
211 	    newval == ZFS_CACHE_METADATA);
212 
213 	os->os_secondary_cache = newval;
214 }
215 
216 static void
217 sync_changed_cb(void *arg, uint64_t newval)
218 {
219 	objset_t *os = arg;
220 
221 	/*
222 	 * Inheritance and range checking should have been done by now.
223 	 */
224 	ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
225 	    newval == ZFS_SYNC_DISABLED);
226 
227 	os->os_sync = newval;
228 	if (os->os_zil)
229 		zil_set_sync(os->os_zil, newval);
230 }
231 
232 static void
233 redundant_metadata_changed_cb(void *arg, uint64_t newval)
234 {
235 	objset_t *os = arg;
236 
237 	/*
238 	 * Inheritance and range checking should have been done by now.
239 	 */
240 	ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
241 	    newval == ZFS_REDUNDANT_METADATA_MOST);
242 
243 	os->os_redundant_metadata = newval;
244 }
245 
246 static void
247 logbias_changed_cb(void *arg, uint64_t newval)
248 {
249 	objset_t *os = arg;
250 
251 	ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
252 	    newval == ZFS_LOGBIAS_THROUGHPUT);
253 	os->os_logbias = newval;
254 	if (os->os_zil)
255 		zil_set_logbias(os->os_zil, newval);
256 }
257 
258 static void
259 recordsize_changed_cb(void *arg, uint64_t newval)
260 {
261 	objset_t *os = arg;
262 
263 	os->os_recordsize = newval;
264 }
265 
266 void
267 dmu_objset_byteswap(void *buf, size_t size)
268 {
269 	objset_phys_t *osp = buf;
270 
271 	ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
272 	dnode_byteswap(&osp->os_meta_dnode);
273 	byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
274 	osp->os_type = BSWAP_64(osp->os_type);
275 	osp->os_flags = BSWAP_64(osp->os_flags);
276 	if (size == sizeof (objset_phys_t)) {
277 		dnode_byteswap(&osp->os_userused_dnode);
278 		dnode_byteswap(&osp->os_groupused_dnode);
279 	}
280 }
281 
282 int
283 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
284     objset_t **osp)
285 {
286 	objset_t *os;
287 	int i, err;
288 
289 	ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
290 
291 	os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
292 	os->os_dsl_dataset = ds;
293 	os->os_spa = spa;
294 	os->os_rootbp = bp;
295 	if (!BP_IS_HOLE(os->os_rootbp)) {
296 		uint32_t aflags = ARC_WAIT;
297 		zbookmark_phys_t zb;
298 		SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
299 		    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
300 
301 		if (DMU_OS_IS_L2CACHEABLE(os))
302 			aflags |= ARC_L2CACHE;
303 		if (DMU_OS_IS_L2COMPRESSIBLE(os))
304 			aflags |= ARC_L2COMPRESS;
305 
306 		dprintf_bp(os->os_rootbp, "reading %s", "");
307 		err = arc_read(NULL, spa, os->os_rootbp,
308 		    arc_getbuf_func, &os->os_phys_buf,
309 		    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
310 		if (err != 0) {
311 			kmem_free(os, sizeof (objset_t));
312 			/* convert checksum errors into IO errors */
313 			if (err == ECKSUM)
314 				err = SET_ERROR(EIO);
315 			return (err);
316 		}
317 
318 		/* Increase the blocksize if we are permitted. */
319 		if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
320 		    arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
321 			arc_buf_t *buf = arc_buf_alloc(spa,
322 			    sizeof (objset_phys_t), &os->os_phys_buf,
323 			    ARC_BUFC_METADATA);
324 			bzero(buf->b_data, sizeof (objset_phys_t));
325 			bcopy(os->os_phys_buf->b_data, buf->b_data,
326 			    arc_buf_size(os->os_phys_buf));
327 			(void) arc_buf_remove_ref(os->os_phys_buf,
328 			    &os->os_phys_buf);
329 			os->os_phys_buf = buf;
330 		}
331 
332 		os->os_phys = os->os_phys_buf->b_data;
333 		os->os_flags = os->os_phys->os_flags;
334 	} else {
335 		int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
336 		    sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
337 		os->os_phys_buf = arc_buf_alloc(spa, size,
338 		    &os->os_phys_buf, ARC_BUFC_METADATA);
339 		os->os_phys = os->os_phys_buf->b_data;
340 		bzero(os->os_phys, size);
341 	}
342 
343 	/*
344 	 * Note: the changed_cb will be called once before the register
345 	 * func returns, thus changing the checksum/compression from the
346 	 * default (fletcher2/off).  Snapshots don't need to know about
347 	 * checksum/compression/copies.
348 	 */
349 	if (ds != NULL) {
350 		err = dsl_prop_register(ds,
351 		    zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
352 		    primary_cache_changed_cb, os);
353 		if (err == 0) {
354 			err = dsl_prop_register(ds,
355 			    zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
356 			    secondary_cache_changed_cb, os);
357 		}
358 		if (!dsl_dataset_is_snapshot(ds)) {
359 			if (err == 0) {
360 				err = dsl_prop_register(ds,
361 				    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
362 				    checksum_changed_cb, os);
363 			}
364 			if (err == 0) {
365 				err = dsl_prop_register(ds,
366 				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
367 				    compression_changed_cb, os);
368 			}
369 			if (err == 0) {
370 				err = dsl_prop_register(ds,
371 				    zfs_prop_to_name(ZFS_PROP_COPIES),
372 				    copies_changed_cb, os);
373 			}
374 			if (err == 0) {
375 				err = dsl_prop_register(ds,
376 				    zfs_prop_to_name(ZFS_PROP_DEDUP),
377 				    dedup_changed_cb, os);
378 			}
379 			if (err == 0) {
380 				err = dsl_prop_register(ds,
381 				    zfs_prop_to_name(ZFS_PROP_LOGBIAS),
382 				    logbias_changed_cb, os);
383 			}
384 			if (err == 0) {
385 				err = dsl_prop_register(ds,
386 				    zfs_prop_to_name(ZFS_PROP_SYNC),
387 				    sync_changed_cb, os);
388 			}
389 			if (err == 0) {
390 				err = dsl_prop_register(ds,
391 				    zfs_prop_to_name(
392 				    ZFS_PROP_REDUNDANT_METADATA),
393 				    redundant_metadata_changed_cb, os);
394 			}
395 			if (err == 0) {
396 				err = dsl_prop_register(ds,
397 				    zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
398 				    recordsize_changed_cb, os);
399 			}
400 		}
401 		if (err != 0) {
402 			VERIFY(arc_buf_remove_ref(os->os_phys_buf,
403 			    &os->os_phys_buf));
404 			kmem_free(os, sizeof (objset_t));
405 			return (err);
406 		}
407 	} else {
408 		/* It's the meta-objset. */
409 		os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
410 		os->os_compress = ZIO_COMPRESS_LZJB;
411 		os->os_copies = spa_max_replication(spa);
412 		os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
413 		os->os_dedup_verify = B_FALSE;
414 		os->os_logbias = ZFS_LOGBIAS_LATENCY;
415 		os->os_sync = ZFS_SYNC_STANDARD;
416 		os->os_primary_cache = ZFS_CACHE_ALL;
417 		os->os_secondary_cache = ZFS_CACHE_ALL;
418 	}
419 
420 	if (ds == NULL || !dsl_dataset_is_snapshot(ds))
421 		os->os_zil_header = os->os_phys->os_zil_header;
422 	os->os_zil = zil_alloc(os, &os->os_zil_header);
423 
424 	for (i = 0; i < TXG_SIZE; i++) {
425 		list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
426 		    offsetof(dnode_t, dn_dirty_link[i]));
427 		list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
428 		    offsetof(dnode_t, dn_dirty_link[i]));
429 	}
430 	list_create(&os->os_dnodes, sizeof (dnode_t),
431 	    offsetof(dnode_t, dn_link));
432 	list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
433 	    offsetof(dmu_buf_impl_t, db_link));
434 
435 	mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
436 	mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
437 	mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
438 
439 	DMU_META_DNODE(os) = dnode_special_open(os,
440 	    &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
441 	    &os->os_meta_dnode);
442 	if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
443 		DMU_USERUSED_DNODE(os) = dnode_special_open(os,
444 		    &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
445 		    &os->os_userused_dnode);
446 		DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
447 		    &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
448 		    &os->os_groupused_dnode);
449 	}
450 
451 	*osp = os;
452 	return (0);
453 }
454 
455 int
456 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
457 {
458 	int err = 0;
459 
460 	mutex_enter(&ds->ds_opening_lock);
461 	if (ds->ds_objset == NULL) {
462 		objset_t *os;
463 		err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
464 		    ds, dsl_dataset_get_blkptr(ds), &os);
465 
466 		if (err == 0) {
467 			mutex_enter(&ds->ds_lock);
468 			ASSERT(ds->ds_objset == NULL);
469 			ds->ds_objset = os;
470 			mutex_exit(&ds->ds_lock);
471 		}
472 	}
473 	*osp = ds->ds_objset;
474 	mutex_exit(&ds->ds_opening_lock);
475 	return (err);
476 }
477 
478 /*
479  * Holds the pool while the objset is held.  Therefore only one objset
480  * can be held at a time.
481  */
482 int
483 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
484 {
485 	dsl_pool_t *dp;
486 	dsl_dataset_t *ds;
487 	int err;
488 
489 	err = dsl_pool_hold(name, tag, &dp);
490 	if (err != 0)
491 		return (err);
492 	err = dsl_dataset_hold(dp, name, tag, &ds);
493 	if (err != 0) {
494 		dsl_pool_rele(dp, tag);
495 		return (err);
496 	}
497 
498 	err = dmu_objset_from_ds(ds, osp);
499 	if (err != 0) {
500 		dsl_dataset_rele(ds, tag);
501 		dsl_pool_rele(dp, tag);
502 	}
503 
504 	return (err);
505 }
506 
507 /*
508  * dsl_pool must not be held when this is called.
509  * Upon successful return, there will be a longhold on the dataset,
510  * and the dsl_pool will not be held.
511  */
512 int
513 dmu_objset_own(const char *name, dmu_objset_type_t type,
514     boolean_t readonly, void *tag, objset_t **osp)
515 {
516 	dsl_pool_t *dp;
517 	dsl_dataset_t *ds;
518 	int err;
519 
520 	err = dsl_pool_hold(name, FTAG, &dp);
521 	if (err != 0)
522 		return (err);
523 	err = dsl_dataset_own(dp, name, tag, &ds);
524 	if (err != 0) {
525 		dsl_pool_rele(dp, FTAG);
526 		return (err);
527 	}
528 
529 	err = dmu_objset_from_ds(ds, osp);
530 	dsl_pool_rele(dp, FTAG);
531 	if (err != 0) {
532 		dsl_dataset_disown(ds, tag);
533 	} else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
534 		dsl_dataset_disown(ds, tag);
535 		return (SET_ERROR(EINVAL));
536 	} else if (!readonly && dsl_dataset_is_snapshot(ds)) {
537 		dsl_dataset_disown(ds, tag);
538 		return (SET_ERROR(EROFS));
539 	}
540 	return (err);
541 }
542 
543 void
544 dmu_objset_rele(objset_t *os, void *tag)
545 {
546 	dsl_pool_t *dp = dmu_objset_pool(os);
547 	dsl_dataset_rele(os->os_dsl_dataset, tag);
548 	dsl_pool_rele(dp, tag);
549 }
550 
551 /*
552  * When we are called, os MUST refer to an objset associated with a dataset
553  * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
554  * == tag.  We will then release and reacquire ownership of the dataset while
555  * holding the pool config_rwlock to avoid intervening namespace or ownership
556  * changes may occur.
557  *
558  * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
559  * release the hold on its dataset and acquire a new one on the dataset of the
560  * same name so that it can be partially torn down and reconstructed.
561  */
562 void
563 dmu_objset_refresh_ownership(objset_t *os, void *tag)
564 {
565 	dsl_pool_t *dp;
566 	dsl_dataset_t *ds, *newds;
567 	char name[MAXNAMELEN];
568 
569 	ds = os->os_dsl_dataset;
570 	VERIFY3P(ds, !=, NULL);
571 	VERIFY3P(ds->ds_owner, ==, tag);
572 	VERIFY(dsl_dataset_long_held(ds));
573 
574 	dsl_dataset_name(ds, name);
575 	dp = dmu_objset_pool(os);
576 	dsl_pool_config_enter(dp, FTAG);
577 	dmu_objset_disown(os, tag);
578 	VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
579 	VERIFY3P(newds, ==, os->os_dsl_dataset);
580 	dsl_pool_config_exit(dp, FTAG);
581 }
582 
583 void
584 dmu_objset_disown(objset_t *os, void *tag)
585 {
586 	dsl_dataset_disown(os->os_dsl_dataset, tag);
587 }
588 
589 void
590 dmu_objset_evict_dbufs(objset_t *os)
591 {
592 	dnode_t *dn;
593 
594 	mutex_enter(&os->os_lock);
595 
596 	/* process the mdn last, since the other dnodes have holds on it */
597 	list_remove(&os->os_dnodes, DMU_META_DNODE(os));
598 	list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
599 
600 	/*
601 	 * Find the first dnode with holds.  We have to do this dance
602 	 * because dnode_add_ref() only works if you already have a
603 	 * hold.  If there are no holds then it has no dbufs so OK to
604 	 * skip.
605 	 */
606 	for (dn = list_head(&os->os_dnodes);
607 	    dn && !dnode_add_ref(dn, FTAG);
608 	    dn = list_next(&os->os_dnodes, dn))
609 		continue;
610 
611 	while (dn) {
612 		dnode_t *next_dn = dn;
613 
614 		do {
615 			next_dn = list_next(&os->os_dnodes, next_dn);
616 		} while (next_dn && !dnode_add_ref(next_dn, FTAG));
617 
618 		mutex_exit(&os->os_lock);
619 		dnode_evict_dbufs(dn);
620 		dnode_rele(dn, FTAG);
621 		mutex_enter(&os->os_lock);
622 		dn = next_dn;
623 	}
624 	mutex_exit(&os->os_lock);
625 }
626 
627 void
628 dmu_objset_evict(objset_t *os)
629 {
630 	dsl_dataset_t *ds = os->os_dsl_dataset;
631 
632 	for (int t = 0; t < TXG_SIZE; t++)
633 		ASSERT(!dmu_objset_is_dirty(os, t));
634 
635 	if (ds) {
636 		if (!dsl_dataset_is_snapshot(ds)) {
637 			VERIFY0(dsl_prop_unregister(ds,
638 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
639 			    checksum_changed_cb, os));
640 			VERIFY0(dsl_prop_unregister(ds,
641 			    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
642 			    compression_changed_cb, os));
643 			VERIFY0(dsl_prop_unregister(ds,
644 			    zfs_prop_to_name(ZFS_PROP_COPIES),
645 			    copies_changed_cb, os));
646 			VERIFY0(dsl_prop_unregister(ds,
647 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
648 			    dedup_changed_cb, os));
649 			VERIFY0(dsl_prop_unregister(ds,
650 			    zfs_prop_to_name(ZFS_PROP_LOGBIAS),
651 			    logbias_changed_cb, os));
652 			VERIFY0(dsl_prop_unregister(ds,
653 			    zfs_prop_to_name(ZFS_PROP_SYNC),
654 			    sync_changed_cb, os));
655 			VERIFY0(dsl_prop_unregister(ds,
656 			    zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA),
657 			    redundant_metadata_changed_cb, os));
658 			VERIFY0(dsl_prop_unregister(ds,
659 			    zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
660 			    recordsize_changed_cb, os));
661 		}
662 		VERIFY0(dsl_prop_unregister(ds,
663 		    zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
664 		    primary_cache_changed_cb, os));
665 		VERIFY0(dsl_prop_unregister(ds,
666 		    zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
667 		    secondary_cache_changed_cb, os));
668 	}
669 
670 	if (os->os_sa)
671 		sa_tear_down(os);
672 
673 	dmu_objset_evict_dbufs(os);
674 
675 	dnode_special_close(&os->os_meta_dnode);
676 	if (DMU_USERUSED_DNODE(os)) {
677 		dnode_special_close(&os->os_userused_dnode);
678 		dnode_special_close(&os->os_groupused_dnode);
679 	}
680 	zil_free(os->os_zil);
681 
682 	ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
683 
684 	VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
685 
686 	/*
687 	 * This is a barrier to prevent the objset from going away in
688 	 * dnode_move() until we can safely ensure that the objset is still in
689 	 * use. We consider the objset valid before the barrier and invalid
690 	 * after the barrier.
691 	 */
692 	rw_enter(&os_lock, RW_READER);
693 	rw_exit(&os_lock);
694 
695 	mutex_destroy(&os->os_lock);
696 	mutex_destroy(&os->os_obj_lock);
697 	mutex_destroy(&os->os_user_ptr_lock);
698 	kmem_free(os, sizeof (objset_t));
699 }
700 
701 timestruc_t
702 dmu_objset_snap_cmtime(objset_t *os)
703 {
704 	return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
705 }
706 
707 /* called from dsl for meta-objset */
708 objset_t *
709 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
710     dmu_objset_type_t type, dmu_tx_t *tx)
711 {
712 	objset_t *os;
713 	dnode_t *mdn;
714 
715 	ASSERT(dmu_tx_is_syncing(tx));
716 
717 	if (ds != NULL)
718 		VERIFY0(dmu_objset_from_ds(ds, &os));
719 	else
720 		VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
721 
722 	mdn = DMU_META_DNODE(os);
723 
724 	dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
725 	    DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
726 
727 	/*
728 	 * We don't want to have to increase the meta-dnode's nlevels
729 	 * later, because then we could do it in quescing context while
730 	 * we are also accessing it in open context.
731 	 *
732 	 * This precaution is not necessary for the MOS (ds == NULL),
733 	 * because the MOS is only updated in syncing context.
734 	 * This is most fortunate: the MOS is the only objset that
735 	 * needs to be synced multiple times as spa_sync() iterates
736 	 * to convergence, so minimizing its dn_nlevels matters.
737 	 */
738 	if (ds != NULL) {
739 		int levels = 1;
740 
741 		/*
742 		 * Determine the number of levels necessary for the meta-dnode
743 		 * to contain DN_MAX_OBJECT dnodes.
744 		 */
745 		while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
746 		    (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
747 		    DN_MAX_OBJECT * sizeof (dnode_phys_t))
748 			levels++;
749 
750 		mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
751 		    mdn->dn_nlevels = levels;
752 	}
753 
754 	ASSERT(type != DMU_OST_NONE);
755 	ASSERT(type != DMU_OST_ANY);
756 	ASSERT(type < DMU_OST_NUMTYPES);
757 	os->os_phys->os_type = type;
758 	if (dmu_objset_userused_enabled(os)) {
759 		os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
760 		os->os_flags = os->os_phys->os_flags;
761 	}
762 
763 	dsl_dataset_dirty(ds, tx);
764 
765 	return (os);
766 }
767 
768 typedef struct dmu_objset_create_arg {
769 	const char *doca_name;
770 	cred_t *doca_cred;
771 	void (*doca_userfunc)(objset_t *os, void *arg,
772 	    cred_t *cr, dmu_tx_t *tx);
773 	void *doca_userarg;
774 	dmu_objset_type_t doca_type;
775 	uint64_t doca_flags;
776 } dmu_objset_create_arg_t;
777 
778 /*ARGSUSED*/
779 static int
780 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
781 {
782 	dmu_objset_create_arg_t *doca = arg;
783 	dsl_pool_t *dp = dmu_tx_pool(tx);
784 	dsl_dir_t *pdd;
785 	const char *tail;
786 	int error;
787 
788 	if (strchr(doca->doca_name, '@') != NULL)
789 		return (SET_ERROR(EINVAL));
790 
791 	error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
792 	if (error != 0)
793 		return (error);
794 	if (tail == NULL) {
795 		dsl_dir_rele(pdd, FTAG);
796 		return (SET_ERROR(EEXIST));
797 	}
798 	error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
799 	    doca->doca_cred);
800 	dsl_dir_rele(pdd, FTAG);
801 
802 	return (error);
803 }
804 
805 static void
806 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
807 {
808 	dmu_objset_create_arg_t *doca = arg;
809 	dsl_pool_t *dp = dmu_tx_pool(tx);
810 	dsl_dir_t *pdd;
811 	const char *tail;
812 	dsl_dataset_t *ds;
813 	uint64_t obj;
814 	blkptr_t *bp;
815 	objset_t *os;
816 
817 	VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
818 
819 	obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
820 	    doca->doca_cred, tx);
821 
822 	VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
823 	bp = dsl_dataset_get_blkptr(ds);
824 	os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
825 	    ds, bp, doca->doca_type, tx);
826 
827 	if (doca->doca_userfunc != NULL) {
828 		doca->doca_userfunc(os, doca->doca_userarg,
829 		    doca->doca_cred, tx);
830 	}
831 
832 	spa_history_log_internal_ds(ds, "create", tx, "");
833 	dsl_dataset_rele(ds, FTAG);
834 	dsl_dir_rele(pdd, FTAG);
835 }
836 
837 int
838 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
839     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
840 {
841 	dmu_objset_create_arg_t doca;
842 
843 	doca.doca_name = name;
844 	doca.doca_cred = CRED();
845 	doca.doca_flags = flags;
846 	doca.doca_userfunc = func;
847 	doca.doca_userarg = arg;
848 	doca.doca_type = type;
849 
850 	return (dsl_sync_task(name,
851 	    dmu_objset_create_check, dmu_objset_create_sync, &doca,
852 	    5, ZFS_SPACE_CHECK_NORMAL));
853 }
854 
855 typedef struct dmu_objset_clone_arg {
856 	const char *doca_clone;
857 	const char *doca_origin;
858 	cred_t *doca_cred;
859 } dmu_objset_clone_arg_t;
860 
861 /*ARGSUSED*/
862 static int
863 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
864 {
865 	dmu_objset_clone_arg_t *doca = arg;
866 	dsl_dir_t *pdd;
867 	const char *tail;
868 	int error;
869 	dsl_dataset_t *origin;
870 	dsl_pool_t *dp = dmu_tx_pool(tx);
871 
872 	if (strchr(doca->doca_clone, '@') != NULL)
873 		return (SET_ERROR(EINVAL));
874 
875 	error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
876 	if (error != 0)
877 		return (error);
878 	if (tail == NULL) {
879 		dsl_dir_rele(pdd, FTAG);
880 		return (SET_ERROR(EEXIST));
881 	}
882 	/* You can't clone across pools. */
883 	if (pdd->dd_pool != dp) {
884 		dsl_dir_rele(pdd, FTAG);
885 		return (SET_ERROR(EXDEV));
886 	}
887 	error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
888 	    doca->doca_cred);
889 	if (error != 0) {
890 		dsl_dir_rele(pdd, FTAG);
891 		return (SET_ERROR(EDQUOT));
892 	}
893 	dsl_dir_rele(pdd, FTAG);
894 
895 	error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
896 	if (error != 0)
897 		return (error);
898 
899 	/* You can't clone across pools. */
900 	if (origin->ds_dir->dd_pool != dp) {
901 		dsl_dataset_rele(origin, FTAG);
902 		return (SET_ERROR(EXDEV));
903 	}
904 
905 	/* You can only clone snapshots, not the head datasets. */
906 	if (!dsl_dataset_is_snapshot(origin)) {
907 		dsl_dataset_rele(origin, FTAG);
908 		return (SET_ERROR(EINVAL));
909 	}
910 	dsl_dataset_rele(origin, FTAG);
911 
912 	return (0);
913 }
914 
915 static void
916 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
917 {
918 	dmu_objset_clone_arg_t *doca = arg;
919 	dsl_pool_t *dp = dmu_tx_pool(tx);
920 	dsl_dir_t *pdd;
921 	const char *tail;
922 	dsl_dataset_t *origin, *ds;
923 	uint64_t obj;
924 	char namebuf[MAXNAMELEN];
925 
926 	VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
927 	VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
928 
929 	obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
930 	    doca->doca_cred, tx);
931 
932 	VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
933 	dsl_dataset_name(origin, namebuf);
934 	spa_history_log_internal_ds(ds, "clone", tx,
935 	    "origin=%s (%llu)", namebuf, origin->ds_object);
936 	dsl_dataset_rele(ds, FTAG);
937 	dsl_dataset_rele(origin, FTAG);
938 	dsl_dir_rele(pdd, FTAG);
939 }
940 
941 int
942 dmu_objset_clone(const char *clone, const char *origin)
943 {
944 	dmu_objset_clone_arg_t doca;
945 
946 	doca.doca_clone = clone;
947 	doca.doca_origin = origin;
948 	doca.doca_cred = CRED();
949 
950 	return (dsl_sync_task(clone,
951 	    dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
952 	    5, ZFS_SPACE_CHECK_NORMAL));
953 }
954 
955 int
956 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
957 {
958 	int err;
959 	char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
960 	nvlist_t *snaps = fnvlist_alloc();
961 
962 	fnvlist_add_boolean(snaps, longsnap);
963 	strfree(longsnap);
964 	err = dsl_dataset_snapshot(snaps, NULL, NULL);
965 	fnvlist_free(snaps);
966 	return (err);
967 }
968 
969 static void
970 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
971 {
972 	dnode_t *dn;
973 
974 	while (dn = list_head(list)) {
975 		ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
976 		ASSERT(dn->dn_dbuf->db_data_pending);
977 		/*
978 		 * Initialize dn_zio outside dnode_sync() because the
979 		 * meta-dnode needs to set it ouside dnode_sync().
980 		 */
981 		dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
982 		ASSERT(dn->dn_zio);
983 
984 		ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
985 		list_remove(list, dn);
986 
987 		if (newlist) {
988 			(void) dnode_add_ref(dn, newlist);
989 			list_insert_tail(newlist, dn);
990 		}
991 
992 		dnode_sync(dn, tx);
993 	}
994 }
995 
996 /* ARGSUSED */
997 static void
998 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
999 {
1000 	blkptr_t *bp = zio->io_bp;
1001 	objset_t *os = arg;
1002 	dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1003 
1004 	ASSERT(!BP_IS_EMBEDDED(bp));
1005 	ASSERT3P(bp, ==, os->os_rootbp);
1006 	ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1007 	ASSERT0(BP_GET_LEVEL(bp));
1008 
1009 	/*
1010 	 * Update rootbp fill count: it should be the number of objects
1011 	 * allocated in the object set (not counting the "special"
1012 	 * objects that are stored in the objset_phys_t -- the meta
1013 	 * dnode and user/group accounting objects).
1014 	 */
1015 	bp->blk_fill = 0;
1016 	for (int i = 0; i < dnp->dn_nblkptr; i++)
1017 		bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1018 }
1019 
1020 /* ARGSUSED */
1021 static void
1022 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1023 {
1024 	blkptr_t *bp = zio->io_bp;
1025 	blkptr_t *bp_orig = &zio->io_bp_orig;
1026 	objset_t *os = arg;
1027 
1028 	if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1029 		ASSERT(BP_EQUAL(bp, bp_orig));
1030 	} else {
1031 		dsl_dataset_t *ds = os->os_dsl_dataset;
1032 		dmu_tx_t *tx = os->os_synctx;
1033 
1034 		(void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1035 		dsl_dataset_block_born(ds, bp, tx);
1036 	}
1037 }
1038 
1039 /* called from dsl */
1040 void
1041 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1042 {
1043 	int txgoff;
1044 	zbookmark_phys_t zb;
1045 	zio_prop_t zp;
1046 	zio_t *zio;
1047 	list_t *list;
1048 	list_t *newlist = NULL;
1049 	dbuf_dirty_record_t *dr;
1050 
1051 	dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1052 
1053 	ASSERT(dmu_tx_is_syncing(tx));
1054 	/* XXX the write_done callback should really give us the tx... */
1055 	os->os_synctx = tx;
1056 
1057 	if (os->os_dsl_dataset == NULL) {
1058 		/*
1059 		 * This is the MOS.  If we have upgraded,
1060 		 * spa_max_replication() could change, so reset
1061 		 * os_copies here.
1062 		 */
1063 		os->os_copies = spa_max_replication(os->os_spa);
1064 	}
1065 
1066 	/*
1067 	 * Create the root block IO
1068 	 */
1069 	SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1070 	    os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1071 	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1072 	arc_release(os->os_phys_buf, &os->os_phys_buf);
1073 
1074 	dmu_write_policy(os, NULL, 0, 0, &zp);
1075 
1076 	zio = arc_write(pio, os->os_spa, tx->tx_txg,
1077 	    os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1078 	    DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1079 	    NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1080 	    ZIO_FLAG_MUSTSUCCEED, &zb);
1081 
1082 	/*
1083 	 * Sync special dnodes - the parent IO for the sync is the root block
1084 	 */
1085 	DMU_META_DNODE(os)->dn_zio = zio;
1086 	dnode_sync(DMU_META_DNODE(os), tx);
1087 
1088 	os->os_phys->os_flags = os->os_flags;
1089 
1090 	if (DMU_USERUSED_DNODE(os) &&
1091 	    DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1092 		DMU_USERUSED_DNODE(os)->dn_zio = zio;
1093 		dnode_sync(DMU_USERUSED_DNODE(os), tx);
1094 		DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1095 		dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1096 	}
1097 
1098 	txgoff = tx->tx_txg & TXG_MASK;
1099 
1100 	if (dmu_objset_userused_enabled(os)) {
1101 		newlist = &os->os_synced_dnodes;
1102 		/*
1103 		 * We must create the list here because it uses the
1104 		 * dn_dirty_link[] of this txg.
1105 		 */
1106 		list_create(newlist, sizeof (dnode_t),
1107 		    offsetof(dnode_t, dn_dirty_link[txgoff]));
1108 	}
1109 
1110 	dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1111 	dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1112 
1113 	list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1114 	while (dr = list_head(list)) {
1115 		ASSERT0(dr->dr_dbuf->db_level);
1116 		list_remove(list, dr);
1117 		if (dr->dr_zio)
1118 			zio_nowait(dr->dr_zio);
1119 	}
1120 	/*
1121 	 * Free intent log blocks up to this tx.
1122 	 */
1123 	zil_sync(os->os_zil, tx);
1124 	os->os_phys->os_zil_header = os->os_zil_header;
1125 	zio_nowait(zio);
1126 }
1127 
1128 boolean_t
1129 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1130 {
1131 	return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1132 	    !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1133 }
1134 
1135 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1136 
1137 void
1138 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1139 {
1140 	used_cbs[ost] = cb;
1141 }
1142 
1143 boolean_t
1144 dmu_objset_userused_enabled(objset_t *os)
1145 {
1146 	return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1147 	    used_cbs[os->os_phys->os_type] != NULL &&
1148 	    DMU_USERUSED_DNODE(os) != NULL);
1149 }
1150 
1151 static void
1152 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1153     uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1154 {
1155 	if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1156 		int64_t delta = DNODE_SIZE + used;
1157 		if (subtract)
1158 			delta = -delta;
1159 		VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1160 		    user, delta, tx));
1161 		VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1162 		    group, delta, tx));
1163 	}
1164 }
1165 
1166 void
1167 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1168 {
1169 	dnode_t *dn;
1170 	list_t *list = &os->os_synced_dnodes;
1171 
1172 	ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1173 
1174 	while (dn = list_head(list)) {
1175 		int flags;
1176 		ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1177 		ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1178 		    dn->dn_phys->dn_flags &
1179 		    DNODE_FLAG_USERUSED_ACCOUNTED);
1180 
1181 		/* Allocate the user/groupused objects if necessary. */
1182 		if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1183 			VERIFY(0 == zap_create_claim(os,
1184 			    DMU_USERUSED_OBJECT,
1185 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1186 			VERIFY(0 == zap_create_claim(os,
1187 			    DMU_GROUPUSED_OBJECT,
1188 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1189 		}
1190 
1191 		/*
1192 		 * We intentionally modify the zap object even if the
1193 		 * net delta is zero.  Otherwise
1194 		 * the block of the zap obj could be shared between
1195 		 * datasets but need to be different between them after
1196 		 * a bprewrite.
1197 		 */
1198 
1199 		flags = dn->dn_id_flags;
1200 		ASSERT(flags);
1201 		if (flags & DN_ID_OLD_EXIST)  {
1202 			do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1203 			    dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1204 		}
1205 		if (flags & DN_ID_NEW_EXIST) {
1206 			do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1207 			    dn->dn_phys->dn_flags,  dn->dn_newuid,
1208 			    dn->dn_newgid, B_FALSE, tx);
1209 		}
1210 
1211 		mutex_enter(&dn->dn_mtx);
1212 		dn->dn_oldused = 0;
1213 		dn->dn_oldflags = 0;
1214 		if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1215 			dn->dn_olduid = dn->dn_newuid;
1216 			dn->dn_oldgid = dn->dn_newgid;
1217 			dn->dn_id_flags |= DN_ID_OLD_EXIST;
1218 			if (dn->dn_bonuslen == 0)
1219 				dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1220 			else
1221 				dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1222 		}
1223 		dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1224 		mutex_exit(&dn->dn_mtx);
1225 
1226 		list_remove(list, dn);
1227 		dnode_rele(dn, list);
1228 	}
1229 }
1230 
1231 /*
1232  * Returns a pointer to data to find uid/gid from
1233  *
1234  * If a dirty record for transaction group that is syncing can't
1235  * be found then NULL is returned.  In the NULL case it is assumed
1236  * the uid/gid aren't changing.
1237  */
1238 static void *
1239 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1240 {
1241 	dbuf_dirty_record_t *dr, **drp;
1242 	void *data;
1243 
1244 	if (db->db_dirtycnt == 0)
1245 		return (db->db.db_data);  /* Nothing is changing */
1246 
1247 	for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1248 		if (dr->dr_txg == tx->tx_txg)
1249 			break;
1250 
1251 	if (dr == NULL) {
1252 		data = NULL;
1253 	} else {
1254 		dnode_t *dn;
1255 
1256 		DB_DNODE_ENTER(dr->dr_dbuf);
1257 		dn = DB_DNODE(dr->dr_dbuf);
1258 
1259 		if (dn->dn_bonuslen == 0 &&
1260 		    dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1261 			data = dr->dt.dl.dr_data->b_data;
1262 		else
1263 			data = dr->dt.dl.dr_data;
1264 
1265 		DB_DNODE_EXIT(dr->dr_dbuf);
1266 	}
1267 
1268 	return (data);
1269 }
1270 
1271 void
1272 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1273 {
1274 	objset_t *os = dn->dn_objset;
1275 	void *data = NULL;
1276 	dmu_buf_impl_t *db = NULL;
1277 	uint64_t *user = NULL;
1278 	uint64_t *group = NULL;
1279 	int flags = dn->dn_id_flags;
1280 	int error;
1281 	boolean_t have_spill = B_FALSE;
1282 
1283 	if (!dmu_objset_userused_enabled(dn->dn_objset))
1284 		return;
1285 
1286 	if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1287 	    DN_ID_CHKED_SPILL)))
1288 		return;
1289 
1290 	if (before && dn->dn_bonuslen != 0)
1291 		data = DN_BONUS(dn->dn_phys);
1292 	else if (!before && dn->dn_bonuslen != 0) {
1293 		if (dn->dn_bonus) {
1294 			db = dn->dn_bonus;
1295 			mutex_enter(&db->db_mtx);
1296 			data = dmu_objset_userquota_find_data(db, tx);
1297 		} else {
1298 			data = DN_BONUS(dn->dn_phys);
1299 		}
1300 	} else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1301 			int rf = 0;
1302 
1303 			if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1304 				rf |= DB_RF_HAVESTRUCT;
1305 			error = dmu_spill_hold_by_dnode(dn,
1306 			    rf | DB_RF_MUST_SUCCEED,
1307 			    FTAG, (dmu_buf_t **)&db);
1308 			ASSERT(error == 0);
1309 			mutex_enter(&db->db_mtx);
1310 			data = (before) ? db->db.db_data :
1311 			    dmu_objset_userquota_find_data(db, tx);
1312 			have_spill = B_TRUE;
1313 	} else {
1314 		mutex_enter(&dn->dn_mtx);
1315 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1316 		mutex_exit(&dn->dn_mtx);
1317 		return;
1318 	}
1319 
1320 	if (before) {
1321 		ASSERT(data);
1322 		user = &dn->dn_olduid;
1323 		group = &dn->dn_oldgid;
1324 	} else if (data) {
1325 		user = &dn->dn_newuid;
1326 		group = &dn->dn_newgid;
1327 	}
1328 
1329 	/*
1330 	 * Must always call the callback in case the object
1331 	 * type has changed and that type isn't an object type to track
1332 	 */
1333 	error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1334 	    user, group);
1335 
1336 	/*
1337 	 * Preserve existing uid/gid when the callback can't determine
1338 	 * what the new uid/gid are and the callback returned EEXIST.
1339 	 * The EEXIST error tells us to just use the existing uid/gid.
1340 	 * If we don't know what the old values are then just assign
1341 	 * them to 0, since that is a new file  being created.
1342 	 */
1343 	if (!before && data == NULL && error == EEXIST) {
1344 		if (flags & DN_ID_OLD_EXIST) {
1345 			dn->dn_newuid = dn->dn_olduid;
1346 			dn->dn_newgid = dn->dn_oldgid;
1347 		} else {
1348 			dn->dn_newuid = 0;
1349 			dn->dn_newgid = 0;
1350 		}
1351 		error = 0;
1352 	}
1353 
1354 	if (db)
1355 		mutex_exit(&db->db_mtx);
1356 
1357 	mutex_enter(&dn->dn_mtx);
1358 	if (error == 0 && before)
1359 		dn->dn_id_flags |= DN_ID_OLD_EXIST;
1360 	if (error == 0 && !before)
1361 		dn->dn_id_flags |= DN_ID_NEW_EXIST;
1362 
1363 	if (have_spill) {
1364 		dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1365 	} else {
1366 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1367 	}
1368 	mutex_exit(&dn->dn_mtx);
1369 	if (have_spill)
1370 		dmu_buf_rele((dmu_buf_t *)db, FTAG);
1371 }
1372 
1373 boolean_t
1374 dmu_objset_userspace_present(objset_t *os)
1375 {
1376 	return (os->os_phys->os_flags &
1377 	    OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1378 }
1379 
1380 int
1381 dmu_objset_userspace_upgrade(objset_t *os)
1382 {
1383 	uint64_t obj;
1384 	int err = 0;
1385 
1386 	if (dmu_objset_userspace_present(os))
1387 		return (0);
1388 	if (!dmu_objset_userused_enabled(os))
1389 		return (SET_ERROR(ENOTSUP));
1390 	if (dmu_objset_is_snapshot(os))
1391 		return (SET_ERROR(EINVAL));
1392 
1393 	/*
1394 	 * We simply need to mark every object dirty, so that it will be
1395 	 * synced out and now accounted.  If this is called
1396 	 * concurrently, or if we already did some work before crashing,
1397 	 * that's fine, since we track each object's accounted state
1398 	 * independently.
1399 	 */
1400 
1401 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1402 		dmu_tx_t *tx;
1403 		dmu_buf_t *db;
1404 		int objerr;
1405 
1406 		if (issig(JUSTLOOKING) && issig(FORREAL))
1407 			return (SET_ERROR(EINTR));
1408 
1409 		objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1410 		if (objerr != 0)
1411 			continue;
1412 		tx = dmu_tx_create(os);
1413 		dmu_tx_hold_bonus(tx, obj);
1414 		objerr = dmu_tx_assign(tx, TXG_WAIT);
1415 		if (objerr != 0) {
1416 			dmu_tx_abort(tx);
1417 			continue;
1418 		}
1419 		dmu_buf_will_dirty(db, tx);
1420 		dmu_buf_rele(db, FTAG);
1421 		dmu_tx_commit(tx);
1422 	}
1423 
1424 	os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1425 	txg_wait_synced(dmu_objset_pool(os), 0);
1426 	return (0);
1427 }
1428 
1429 void
1430 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1431     uint64_t *usedobjsp, uint64_t *availobjsp)
1432 {
1433 	dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1434 	    usedobjsp, availobjsp);
1435 }
1436 
1437 uint64_t
1438 dmu_objset_fsid_guid(objset_t *os)
1439 {
1440 	return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1441 }
1442 
1443 void
1444 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1445 {
1446 	stat->dds_type = os->os_phys->os_type;
1447 	if (os->os_dsl_dataset)
1448 		dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1449 }
1450 
1451 void
1452 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1453 {
1454 	ASSERT(os->os_dsl_dataset ||
1455 	    os->os_phys->os_type == DMU_OST_META);
1456 
1457 	if (os->os_dsl_dataset != NULL)
1458 		dsl_dataset_stats(os->os_dsl_dataset, nv);
1459 
1460 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1461 	    os->os_phys->os_type);
1462 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1463 	    dmu_objset_userspace_present(os));
1464 }
1465 
1466 int
1467 dmu_objset_is_snapshot(objset_t *os)
1468 {
1469 	if (os->os_dsl_dataset != NULL)
1470 		return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1471 	else
1472 		return (B_FALSE);
1473 }
1474 
1475 int
1476 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1477     boolean_t *conflict)
1478 {
1479 	dsl_dataset_t *ds = os->os_dsl_dataset;
1480 	uint64_t ignored;
1481 
1482 	if (ds->ds_phys->ds_snapnames_zapobj == 0)
1483 		return (SET_ERROR(ENOENT));
1484 
1485 	return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1486 	    ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1487 	    real, maxlen, conflict));
1488 }
1489 
1490 int
1491 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1492     uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1493 {
1494 	dsl_dataset_t *ds = os->os_dsl_dataset;
1495 	zap_cursor_t cursor;
1496 	zap_attribute_t attr;
1497 
1498 	ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1499 
1500 	if (ds->ds_phys->ds_snapnames_zapobj == 0)
1501 		return (SET_ERROR(ENOENT));
1502 
1503 	zap_cursor_init_serialized(&cursor,
1504 	    ds->ds_dir->dd_pool->dp_meta_objset,
1505 	    ds->ds_phys->ds_snapnames_zapobj, *offp);
1506 
1507 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1508 		zap_cursor_fini(&cursor);
1509 		return (SET_ERROR(ENOENT));
1510 	}
1511 
1512 	if (strlen(attr.za_name) + 1 > namelen) {
1513 		zap_cursor_fini(&cursor);
1514 		return (SET_ERROR(ENAMETOOLONG));
1515 	}
1516 
1517 	(void) strcpy(name, attr.za_name);
1518 	if (idp)
1519 		*idp = attr.za_first_integer;
1520 	if (case_conflict)
1521 		*case_conflict = attr.za_normalization_conflict;
1522 	zap_cursor_advance(&cursor);
1523 	*offp = zap_cursor_serialize(&cursor);
1524 	zap_cursor_fini(&cursor);
1525 
1526 	return (0);
1527 }
1528 
1529 int
1530 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1531     uint64_t *idp, uint64_t *offp)
1532 {
1533 	dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1534 	zap_cursor_t cursor;
1535 	zap_attribute_t attr;
1536 
1537 	/* there is no next dir on a snapshot! */
1538 	if (os->os_dsl_dataset->ds_object !=
1539 	    dd->dd_phys->dd_head_dataset_obj)
1540 		return (SET_ERROR(ENOENT));
1541 
1542 	zap_cursor_init_serialized(&cursor,
1543 	    dd->dd_pool->dp_meta_objset,
1544 	    dd->dd_phys->dd_child_dir_zapobj, *offp);
1545 
1546 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1547 		zap_cursor_fini(&cursor);
1548 		return (SET_ERROR(ENOENT));
1549 	}
1550 
1551 	if (strlen(attr.za_name) + 1 > namelen) {
1552 		zap_cursor_fini(&cursor);
1553 		return (SET_ERROR(ENAMETOOLONG));
1554 	}
1555 
1556 	(void) strcpy(name, attr.za_name);
1557 	if (idp)
1558 		*idp = attr.za_first_integer;
1559 	zap_cursor_advance(&cursor);
1560 	*offp = zap_cursor_serialize(&cursor);
1561 	zap_cursor_fini(&cursor);
1562 
1563 	return (0);
1564 }
1565 
1566 /*
1567  * Find objsets under and including ddobj, call func(ds) on each.
1568  */
1569 int
1570 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1571     int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1572 {
1573 	dsl_dir_t *dd;
1574 	dsl_dataset_t *ds;
1575 	zap_cursor_t zc;
1576 	zap_attribute_t *attr;
1577 	uint64_t thisobj;
1578 	int err;
1579 
1580 	ASSERT(dsl_pool_config_held(dp));
1581 
1582 	err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1583 	if (err != 0)
1584 		return (err);
1585 
1586 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1587 	if (dd->dd_myname[0] == '$') {
1588 		dsl_dir_rele(dd, FTAG);
1589 		return (0);
1590 	}
1591 
1592 	thisobj = dd->dd_phys->dd_head_dataset_obj;
1593 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1594 
1595 	/*
1596 	 * Iterate over all children.
1597 	 */
1598 	if (flags & DS_FIND_CHILDREN) {
1599 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
1600 		    dd->dd_phys->dd_child_dir_zapobj);
1601 		    zap_cursor_retrieve(&zc, attr) == 0;
1602 		    (void) zap_cursor_advance(&zc)) {
1603 			ASSERT3U(attr->za_integer_length, ==,
1604 			    sizeof (uint64_t));
1605 			ASSERT3U(attr->za_num_integers, ==, 1);
1606 
1607 			err = dmu_objset_find_dp(dp, attr->za_first_integer,
1608 			    func, arg, flags);
1609 			if (err != 0)
1610 				break;
1611 		}
1612 		zap_cursor_fini(&zc);
1613 
1614 		if (err != 0) {
1615 			dsl_dir_rele(dd, FTAG);
1616 			kmem_free(attr, sizeof (zap_attribute_t));
1617 			return (err);
1618 		}
1619 	}
1620 
1621 	/*
1622 	 * Iterate over all snapshots.
1623 	 */
1624 	if (flags & DS_FIND_SNAPSHOTS) {
1625 		dsl_dataset_t *ds;
1626 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1627 
1628 		if (err == 0) {
1629 			uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1630 			dsl_dataset_rele(ds, FTAG);
1631 
1632 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1633 			    zap_cursor_retrieve(&zc, attr) == 0;
1634 			    (void) zap_cursor_advance(&zc)) {
1635 				ASSERT3U(attr->za_integer_length, ==,
1636 				    sizeof (uint64_t));
1637 				ASSERT3U(attr->za_num_integers, ==, 1);
1638 
1639 				err = dsl_dataset_hold_obj(dp,
1640 				    attr->za_first_integer, FTAG, &ds);
1641 				if (err != 0)
1642 					break;
1643 				err = func(dp, ds, arg);
1644 				dsl_dataset_rele(ds, FTAG);
1645 				if (err != 0)
1646 					break;
1647 			}
1648 			zap_cursor_fini(&zc);
1649 		}
1650 	}
1651 
1652 	dsl_dir_rele(dd, FTAG);
1653 	kmem_free(attr, sizeof (zap_attribute_t));
1654 
1655 	if (err != 0)
1656 		return (err);
1657 
1658 	/*
1659 	 * Apply to self.
1660 	 */
1661 	err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1662 	if (err != 0)
1663 		return (err);
1664 	err = func(dp, ds, arg);
1665 	dsl_dataset_rele(ds, FTAG);
1666 	return (err);
1667 }
1668 
1669 /*
1670  * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1671  * The dp_config_rwlock must not be held when this is called, and it
1672  * will not be held when the callback is called.
1673  * Therefore this function should only be used when the pool is not changing
1674  * (e.g. in syncing context), or the callback can deal with the possible races.
1675  */
1676 static int
1677 dmu_objset_find_impl(spa_t *spa, const char *name,
1678     int func(const char *, void *), void *arg, int flags)
1679 {
1680 	dsl_dir_t *dd;
1681 	dsl_pool_t *dp = spa_get_dsl(spa);
1682 	dsl_dataset_t *ds;
1683 	zap_cursor_t zc;
1684 	zap_attribute_t *attr;
1685 	char *child;
1686 	uint64_t thisobj;
1687 	int err;
1688 
1689 	dsl_pool_config_enter(dp, FTAG);
1690 
1691 	err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1692 	if (err != 0) {
1693 		dsl_pool_config_exit(dp, FTAG);
1694 		return (err);
1695 	}
1696 
1697 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1698 	if (dd->dd_myname[0] == '$') {
1699 		dsl_dir_rele(dd, FTAG);
1700 		dsl_pool_config_exit(dp, FTAG);
1701 		return (0);
1702 	}
1703 
1704 	thisobj = dd->dd_phys->dd_head_dataset_obj;
1705 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1706 
1707 	/*
1708 	 * Iterate over all children.
1709 	 */
1710 	if (flags & DS_FIND_CHILDREN) {
1711 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
1712 		    dd->dd_phys->dd_child_dir_zapobj);
1713 		    zap_cursor_retrieve(&zc, attr) == 0;
1714 		    (void) zap_cursor_advance(&zc)) {
1715 			ASSERT3U(attr->za_integer_length, ==,
1716 			    sizeof (uint64_t));
1717 			ASSERT3U(attr->za_num_integers, ==, 1);
1718 
1719 			child = kmem_asprintf("%s/%s", name, attr->za_name);
1720 			dsl_pool_config_exit(dp, FTAG);
1721 			err = dmu_objset_find_impl(spa, child,
1722 			    func, arg, flags);
1723 			dsl_pool_config_enter(dp, FTAG);
1724 			strfree(child);
1725 			if (err != 0)
1726 				break;
1727 		}
1728 		zap_cursor_fini(&zc);
1729 
1730 		if (err != 0) {
1731 			dsl_dir_rele(dd, FTAG);
1732 			dsl_pool_config_exit(dp, FTAG);
1733 			kmem_free(attr, sizeof (zap_attribute_t));
1734 			return (err);
1735 		}
1736 	}
1737 
1738 	/*
1739 	 * Iterate over all snapshots.
1740 	 */
1741 	if (flags & DS_FIND_SNAPSHOTS) {
1742 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1743 
1744 		if (err == 0) {
1745 			uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1746 			dsl_dataset_rele(ds, FTAG);
1747 
1748 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1749 			    zap_cursor_retrieve(&zc, attr) == 0;
1750 			    (void) zap_cursor_advance(&zc)) {
1751 				ASSERT3U(attr->za_integer_length, ==,
1752 				    sizeof (uint64_t));
1753 				ASSERT3U(attr->za_num_integers, ==, 1);
1754 
1755 				child = kmem_asprintf("%s@%s",
1756 				    name, attr->za_name);
1757 				dsl_pool_config_exit(dp, FTAG);
1758 				err = func(child, arg);
1759 				dsl_pool_config_enter(dp, FTAG);
1760 				strfree(child);
1761 				if (err != 0)
1762 					break;
1763 			}
1764 			zap_cursor_fini(&zc);
1765 		}
1766 	}
1767 
1768 	dsl_dir_rele(dd, FTAG);
1769 	kmem_free(attr, sizeof (zap_attribute_t));
1770 	dsl_pool_config_exit(dp, FTAG);
1771 
1772 	if (err != 0)
1773 		return (err);
1774 
1775 	/* Apply to self. */
1776 	return (func(name, arg));
1777 }
1778 
1779 /*
1780  * See comment above dmu_objset_find_impl().
1781  */
1782 int
1783 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1784     int flags)
1785 {
1786 	spa_t *spa;
1787 	int error;
1788 
1789 	error = spa_open(name, &spa, FTAG);
1790 	if (error != 0)
1791 		return (error);
1792 	error = dmu_objset_find_impl(spa, name, func, arg, flags);
1793 	spa_close(spa, FTAG);
1794 	return (error);
1795 }
1796 
1797 void
1798 dmu_objset_set_user(objset_t *os, void *user_ptr)
1799 {
1800 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1801 	os->os_user_ptr = user_ptr;
1802 }
1803 
1804 void *
1805 dmu_objset_get_user(objset_t *os)
1806 {
1807 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1808 	return (os->os_user_ptr);
1809 }
1810 
1811 /*
1812  * Determine name of filesystem, given name of snapshot.
1813  * buf must be at least MAXNAMELEN bytes
1814  */
1815 int
1816 dmu_fsname(const char *snapname, char *buf)
1817 {
1818 	char *atp = strchr(snapname, '@');
1819 	if (atp == NULL)
1820 		return (SET_ERROR(EINVAL));
1821 	if (atp - snapname >= MAXNAMELEN)
1822 		return (SET_ERROR(ENAMETOOLONG));
1823 	(void) strlcpy(buf, snapname, atp - snapname + 1);
1824 	return (0);
1825 }
1826