xref: /illumos-gate/usr/src/uts/common/fs/zfs/dmu_objset.c (revision 69962b5647e4a8b9b14998733b765925381b727e)
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) 2013 by Delphix. All rights reserved.
24  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25  */
26 
27 /* Portions Copyright 2010 Robert Milkowski */
28 
29 #include <sys/cred.h>
30 #include <sys/zfs_context.h>
31 #include <sys/dmu_objset.h>
32 #include <sys/dsl_dir.h>
33 #include <sys/dsl_dataset.h>
34 #include <sys/dsl_prop.h>
35 #include <sys/dsl_pool.h>
36 #include <sys/dsl_synctask.h>
37 #include <sys/dsl_deleg.h>
38 #include <sys/dnode.h>
39 #include <sys/dbuf.h>
40 #include <sys/zvol.h>
41 #include <sys/dmu_tx.h>
42 #include <sys/zap.h>
43 #include <sys/zil.h>
44 #include <sys/dmu_impl.h>
45 #include <sys/zfs_ioctl.h>
46 #include <sys/sa.h>
47 #include <sys/zfs_onexit.h>
48 #include <sys/dsl_destroy.h>
49 
50 /*
51  * Needed to close a window in dnode_move() that allows the objset to be freed
52  * before it can be safely accessed.
53  */
54 krwlock_t os_lock;
55 
56 void
57 dmu_objset_init(void)
58 {
59 	rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
60 }
61 
62 void
63 dmu_objset_fini(void)
64 {
65 	rw_destroy(&os_lock);
66 }
67 
68 spa_t *
69 dmu_objset_spa(objset_t *os)
70 {
71 	return (os->os_spa);
72 }
73 
74 zilog_t *
75 dmu_objset_zil(objset_t *os)
76 {
77 	return (os->os_zil);
78 }
79 
80 dsl_pool_t *
81 dmu_objset_pool(objset_t *os)
82 {
83 	dsl_dataset_t *ds;
84 
85 	if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
86 		return (ds->ds_dir->dd_pool);
87 	else
88 		return (spa_get_dsl(os->os_spa));
89 }
90 
91 dsl_dataset_t *
92 dmu_objset_ds(objset_t *os)
93 {
94 	return (os->os_dsl_dataset);
95 }
96 
97 dmu_objset_type_t
98 dmu_objset_type(objset_t *os)
99 {
100 	return (os->os_phys->os_type);
101 }
102 
103 void
104 dmu_objset_name(objset_t *os, char *buf)
105 {
106 	dsl_dataset_name(os->os_dsl_dataset, buf);
107 }
108 
109 uint64_t
110 dmu_objset_id(objset_t *os)
111 {
112 	dsl_dataset_t *ds = os->os_dsl_dataset;
113 
114 	return (ds ? ds->ds_object : 0);
115 }
116 
117 uint64_t
118 dmu_objset_syncprop(objset_t *os)
119 {
120 	return (os->os_sync);
121 }
122 
123 uint64_t
124 dmu_objset_logbias(objset_t *os)
125 {
126 	return (os->os_logbias);
127 }
128 
129 static void
130 checksum_changed_cb(void *arg, uint64_t newval)
131 {
132 	objset_t *os = arg;
133 
134 	/*
135 	 * Inheritance should have been done by now.
136 	 */
137 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
138 
139 	os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
140 }
141 
142 static void
143 compression_changed_cb(void *arg, uint64_t newval)
144 {
145 	objset_t *os = arg;
146 
147 	/*
148 	 * Inheritance and range checking should have been done by now.
149 	 */
150 	ASSERT(newval != ZIO_COMPRESS_INHERIT);
151 
152 	os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
153 }
154 
155 static void
156 copies_changed_cb(void *arg, uint64_t newval)
157 {
158 	objset_t *os = arg;
159 
160 	/*
161 	 * Inheritance and range checking should have been done by now.
162 	 */
163 	ASSERT(newval > 0);
164 	ASSERT(newval <= spa_max_replication(os->os_spa));
165 
166 	os->os_copies = newval;
167 }
168 
169 static void
170 dedup_changed_cb(void *arg, uint64_t newval)
171 {
172 	objset_t *os = arg;
173 	spa_t *spa = os->os_spa;
174 	enum zio_checksum checksum;
175 
176 	/*
177 	 * Inheritance should have been done by now.
178 	 */
179 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
180 
181 	checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
182 
183 	os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
184 	os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
185 }
186 
187 static void
188 primary_cache_changed_cb(void *arg, uint64_t newval)
189 {
190 	objset_t *os = arg;
191 
192 	/*
193 	 * Inheritance and range checking should have been done by now.
194 	 */
195 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
196 	    newval == ZFS_CACHE_METADATA);
197 
198 	os->os_primary_cache = newval;
199 }
200 
201 static void
202 secondary_cache_changed_cb(void *arg, uint64_t newval)
203 {
204 	objset_t *os = arg;
205 
206 	/*
207 	 * Inheritance and range checking should have been done by now.
208 	 */
209 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
210 	    newval == ZFS_CACHE_METADATA);
211 
212 	os->os_secondary_cache = newval;
213 }
214 
215 static void
216 sync_changed_cb(void *arg, uint64_t newval)
217 {
218 	objset_t *os = arg;
219 
220 	/*
221 	 * Inheritance and range checking should have been done by now.
222 	 */
223 	ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
224 	    newval == ZFS_SYNC_DISABLED);
225 
226 	os->os_sync = newval;
227 	if (os->os_zil)
228 		zil_set_sync(os->os_zil, newval);
229 }
230 
231 static void
232 logbias_changed_cb(void *arg, uint64_t newval)
233 {
234 	objset_t *os = arg;
235 
236 	ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
237 	    newval == ZFS_LOGBIAS_THROUGHPUT);
238 	os->os_logbias = newval;
239 	if (os->os_zil)
240 		zil_set_logbias(os->os_zil, newval);
241 }
242 
243 void
244 dmu_objset_byteswap(void *buf, size_t size)
245 {
246 	objset_phys_t *osp = buf;
247 
248 	ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
249 	dnode_byteswap(&osp->os_meta_dnode);
250 	byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
251 	osp->os_type = BSWAP_64(osp->os_type);
252 	osp->os_flags = BSWAP_64(osp->os_flags);
253 	if (size == sizeof (objset_phys_t)) {
254 		dnode_byteswap(&osp->os_userused_dnode);
255 		dnode_byteswap(&osp->os_groupused_dnode);
256 	}
257 }
258 
259 int
260 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
261     objset_t **osp)
262 {
263 	objset_t *os;
264 	int i, err;
265 
266 	ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
267 
268 	os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
269 	os->os_dsl_dataset = ds;
270 	os->os_spa = spa;
271 	os->os_rootbp = bp;
272 	if (!BP_IS_HOLE(os->os_rootbp)) {
273 		uint32_t aflags = ARC_WAIT;
274 		zbookmark_t zb;
275 		SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
276 		    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
277 
278 		if (DMU_OS_IS_L2CACHEABLE(os))
279 			aflags |= ARC_L2CACHE;
280 		if (DMU_OS_IS_L2COMPRESSIBLE(os))
281 			aflags |= ARC_L2COMPRESS;
282 
283 		dprintf_bp(os->os_rootbp, "reading %s", "");
284 		err = arc_read(NULL, spa, os->os_rootbp,
285 		    arc_getbuf_func, &os->os_phys_buf,
286 		    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
287 		if (err != 0) {
288 			kmem_free(os, sizeof (objset_t));
289 			/* convert checksum errors into IO errors */
290 			if (err == ECKSUM)
291 				err = SET_ERROR(EIO);
292 			return (err);
293 		}
294 
295 		/* Increase the blocksize if we are permitted. */
296 		if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
297 		    arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
298 			arc_buf_t *buf = arc_buf_alloc(spa,
299 			    sizeof (objset_phys_t), &os->os_phys_buf,
300 			    ARC_BUFC_METADATA);
301 			bzero(buf->b_data, sizeof (objset_phys_t));
302 			bcopy(os->os_phys_buf->b_data, buf->b_data,
303 			    arc_buf_size(os->os_phys_buf));
304 			(void) arc_buf_remove_ref(os->os_phys_buf,
305 			    &os->os_phys_buf);
306 			os->os_phys_buf = buf;
307 		}
308 
309 		os->os_phys = os->os_phys_buf->b_data;
310 		os->os_flags = os->os_phys->os_flags;
311 	} else {
312 		int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
313 		    sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
314 		os->os_phys_buf = arc_buf_alloc(spa, size,
315 		    &os->os_phys_buf, ARC_BUFC_METADATA);
316 		os->os_phys = os->os_phys_buf->b_data;
317 		bzero(os->os_phys, size);
318 	}
319 
320 	/*
321 	 * Note: the changed_cb will be called once before the register
322 	 * func returns, thus changing the checksum/compression from the
323 	 * default (fletcher2/off).  Snapshots don't need to know about
324 	 * checksum/compression/copies.
325 	 */
326 	if (ds) {
327 		err = dsl_prop_register(ds,
328 		    zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
329 		    primary_cache_changed_cb, os);
330 		if (err == 0) {
331 			err = dsl_prop_register(ds,
332 			    zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
333 			    secondary_cache_changed_cb, os);
334 		}
335 		if (!dsl_dataset_is_snapshot(ds)) {
336 			if (err == 0) {
337 				err = dsl_prop_register(ds,
338 				    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
339 				    checksum_changed_cb, os);
340 			}
341 			if (err == 0) {
342 				err = dsl_prop_register(ds,
343 				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
344 				    compression_changed_cb, os);
345 			}
346 			if (err == 0) {
347 				err = dsl_prop_register(ds,
348 				    zfs_prop_to_name(ZFS_PROP_COPIES),
349 				    copies_changed_cb, os);
350 			}
351 			if (err == 0) {
352 				err = dsl_prop_register(ds,
353 				    zfs_prop_to_name(ZFS_PROP_DEDUP),
354 				    dedup_changed_cb, os);
355 			}
356 			if (err == 0) {
357 				err = dsl_prop_register(ds,
358 				    zfs_prop_to_name(ZFS_PROP_LOGBIAS),
359 				    logbias_changed_cb, os);
360 			}
361 			if (err == 0) {
362 				err = dsl_prop_register(ds,
363 				    zfs_prop_to_name(ZFS_PROP_SYNC),
364 				    sync_changed_cb, os);
365 			}
366 		}
367 		if (err != 0) {
368 			VERIFY(arc_buf_remove_ref(os->os_phys_buf,
369 			    &os->os_phys_buf));
370 			kmem_free(os, sizeof (objset_t));
371 			return (err);
372 		}
373 	} else if (ds == NULL) {
374 		/* It's the meta-objset. */
375 		os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
376 		os->os_compress = ZIO_COMPRESS_LZJB;
377 		os->os_copies = spa_max_replication(spa);
378 		os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
379 		os->os_dedup_verify = 0;
380 		os->os_logbias = 0;
381 		os->os_sync = 0;
382 		os->os_primary_cache = ZFS_CACHE_ALL;
383 		os->os_secondary_cache = ZFS_CACHE_ALL;
384 	}
385 
386 	if (ds == NULL || !dsl_dataset_is_snapshot(ds))
387 		os->os_zil_header = os->os_phys->os_zil_header;
388 	os->os_zil = zil_alloc(os, &os->os_zil_header);
389 
390 	for (i = 0; i < TXG_SIZE; i++) {
391 		list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
392 		    offsetof(dnode_t, dn_dirty_link[i]));
393 		list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
394 		    offsetof(dnode_t, dn_dirty_link[i]));
395 	}
396 	list_create(&os->os_dnodes, sizeof (dnode_t),
397 	    offsetof(dnode_t, dn_link));
398 	list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
399 	    offsetof(dmu_buf_impl_t, db_link));
400 
401 	mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
402 	mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
403 	mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
404 
405 	DMU_META_DNODE(os) = dnode_special_open(os,
406 	    &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
407 	    &os->os_meta_dnode);
408 	if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
409 		DMU_USERUSED_DNODE(os) = dnode_special_open(os,
410 		    &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
411 		    &os->os_userused_dnode);
412 		DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
413 		    &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
414 		    &os->os_groupused_dnode);
415 	}
416 
417 	/*
418 	 * We should be the only thread trying to do this because we
419 	 * have ds_opening_lock
420 	 */
421 	if (ds) {
422 		mutex_enter(&ds->ds_lock);
423 		ASSERT(ds->ds_objset == NULL);
424 		ds->ds_objset = os;
425 		mutex_exit(&ds->ds_lock);
426 	}
427 
428 	*osp = os;
429 	return (0);
430 }
431 
432 int
433 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
434 {
435 	int err = 0;
436 
437 	mutex_enter(&ds->ds_opening_lock);
438 	*osp = ds->ds_objset;
439 	if (*osp == NULL) {
440 		err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
441 		    ds, dsl_dataset_get_blkptr(ds), osp);
442 	}
443 	mutex_exit(&ds->ds_opening_lock);
444 	return (err);
445 }
446 
447 /*
448  * Holds the pool while the objset is held.  Therefore only one objset
449  * can be held at a time.
450  */
451 int
452 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
453 {
454 	dsl_pool_t *dp;
455 	dsl_dataset_t *ds;
456 	int err;
457 
458 	err = dsl_pool_hold(name, tag, &dp);
459 	if (err != 0)
460 		return (err);
461 	err = dsl_dataset_hold(dp, name, tag, &ds);
462 	if (err != 0) {
463 		dsl_pool_rele(dp, tag);
464 		return (err);
465 	}
466 
467 	err = dmu_objset_from_ds(ds, osp);
468 	if (err != 0) {
469 		dsl_dataset_rele(ds, tag);
470 		dsl_pool_rele(dp, tag);
471 	}
472 
473 	return (err);
474 }
475 
476 /*
477  * dsl_pool must not be held when this is called.
478  * Upon successful return, there will be a longhold on the dataset,
479  * and the dsl_pool will not be held.
480  */
481 int
482 dmu_objset_own(const char *name, dmu_objset_type_t type,
483     boolean_t readonly, 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, FTAG, &dp);
490 	if (err != 0)
491 		return (err);
492 	err = dsl_dataset_own(dp, name, tag, &ds);
493 	if (err != 0) {
494 		dsl_pool_rele(dp, FTAG);
495 		return (err);
496 	}
497 
498 	err = dmu_objset_from_ds(ds, osp);
499 	dsl_pool_rele(dp, FTAG);
500 	if (err != 0) {
501 		dsl_dataset_disown(ds, tag);
502 	} else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
503 		dsl_dataset_disown(ds, tag);
504 		return (SET_ERROR(EINVAL));
505 	} else if (!readonly && dsl_dataset_is_snapshot(ds)) {
506 		dsl_dataset_disown(ds, tag);
507 		return (SET_ERROR(EROFS));
508 	}
509 	return (err);
510 }
511 
512 void
513 dmu_objset_rele(objset_t *os, void *tag)
514 {
515 	dsl_pool_t *dp = dmu_objset_pool(os);
516 	dsl_dataset_rele(os->os_dsl_dataset, tag);
517 	dsl_pool_rele(dp, tag);
518 }
519 
520 /*
521  * When we are called, os MUST refer to an objset associated with a dataset
522  * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
523  * == tag.  We will then release and reacquire ownership of the dataset while
524  * holding the pool config_rwlock to avoid intervening namespace or ownership
525  * changes may occur.
526  *
527  * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
528  * release the hold on its dataset and acquire a new one on the dataset of the
529  * same name so that it can be partially torn down and reconstructed.
530  */
531 void
532 dmu_objset_refresh_ownership(objset_t *os, void *tag)
533 {
534 	dsl_pool_t *dp;
535 	dsl_dataset_t *ds, *newds;
536 	char name[MAXNAMELEN];
537 
538 	ds = os->os_dsl_dataset;
539 	VERIFY3P(ds, !=, NULL);
540 	VERIFY3P(ds->ds_owner, ==, tag);
541 	VERIFY(dsl_dataset_long_held(ds));
542 
543 	dsl_dataset_name(ds, name);
544 	dp = dmu_objset_pool(os);
545 	dsl_pool_config_enter(dp, FTAG);
546 	dmu_objset_disown(os, tag);
547 	VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
548 	VERIFY3P(newds, ==, os->os_dsl_dataset);
549 	dsl_pool_config_exit(dp, FTAG);
550 }
551 
552 void
553 dmu_objset_disown(objset_t *os, void *tag)
554 {
555 	dsl_dataset_disown(os->os_dsl_dataset, tag);
556 }
557 
558 void
559 dmu_objset_evict_dbufs(objset_t *os)
560 {
561 	dnode_t *dn;
562 
563 	mutex_enter(&os->os_lock);
564 
565 	/* process the mdn last, since the other dnodes have holds on it */
566 	list_remove(&os->os_dnodes, DMU_META_DNODE(os));
567 	list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
568 
569 	/*
570 	 * Find the first dnode with holds.  We have to do this dance
571 	 * because dnode_add_ref() only works if you already have a
572 	 * hold.  If there are no holds then it has no dbufs so OK to
573 	 * skip.
574 	 */
575 	for (dn = list_head(&os->os_dnodes);
576 	    dn && !dnode_add_ref(dn, FTAG);
577 	    dn = list_next(&os->os_dnodes, dn))
578 		continue;
579 
580 	while (dn) {
581 		dnode_t *next_dn = dn;
582 
583 		do {
584 			next_dn = list_next(&os->os_dnodes, next_dn);
585 		} while (next_dn && !dnode_add_ref(next_dn, FTAG));
586 
587 		mutex_exit(&os->os_lock);
588 		dnode_evict_dbufs(dn);
589 		dnode_rele(dn, FTAG);
590 		mutex_enter(&os->os_lock);
591 		dn = next_dn;
592 	}
593 	mutex_exit(&os->os_lock);
594 }
595 
596 void
597 dmu_objset_evict(objset_t *os)
598 {
599 	dsl_dataset_t *ds = os->os_dsl_dataset;
600 
601 	for (int t = 0; t < TXG_SIZE; t++)
602 		ASSERT(!dmu_objset_is_dirty(os, t));
603 
604 	if (ds) {
605 		if (!dsl_dataset_is_snapshot(ds)) {
606 			VERIFY0(dsl_prop_unregister(ds,
607 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
608 			    checksum_changed_cb, os));
609 			VERIFY0(dsl_prop_unregister(ds,
610 			    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
611 			    compression_changed_cb, os));
612 			VERIFY0(dsl_prop_unregister(ds,
613 			    zfs_prop_to_name(ZFS_PROP_COPIES),
614 			    copies_changed_cb, os));
615 			VERIFY0(dsl_prop_unregister(ds,
616 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
617 			    dedup_changed_cb, os));
618 			VERIFY0(dsl_prop_unregister(ds,
619 			    zfs_prop_to_name(ZFS_PROP_LOGBIAS),
620 			    logbias_changed_cb, os));
621 			VERIFY0(dsl_prop_unregister(ds,
622 			    zfs_prop_to_name(ZFS_PROP_SYNC),
623 			    sync_changed_cb, os));
624 		}
625 		VERIFY0(dsl_prop_unregister(ds,
626 		    zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
627 		    primary_cache_changed_cb, os));
628 		VERIFY0(dsl_prop_unregister(ds,
629 		    zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
630 		    secondary_cache_changed_cb, os));
631 	}
632 
633 	if (os->os_sa)
634 		sa_tear_down(os);
635 
636 	dmu_objset_evict_dbufs(os);
637 
638 	dnode_special_close(&os->os_meta_dnode);
639 	if (DMU_USERUSED_DNODE(os)) {
640 		dnode_special_close(&os->os_userused_dnode);
641 		dnode_special_close(&os->os_groupused_dnode);
642 	}
643 	zil_free(os->os_zil);
644 
645 	ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
646 
647 	VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
648 
649 	/*
650 	 * This is a barrier to prevent the objset from going away in
651 	 * dnode_move() until we can safely ensure that the objset is still in
652 	 * use. We consider the objset valid before the barrier and invalid
653 	 * after the barrier.
654 	 */
655 	rw_enter(&os_lock, RW_READER);
656 	rw_exit(&os_lock);
657 
658 	mutex_destroy(&os->os_lock);
659 	mutex_destroy(&os->os_obj_lock);
660 	mutex_destroy(&os->os_user_ptr_lock);
661 	kmem_free(os, sizeof (objset_t));
662 }
663 
664 timestruc_t
665 dmu_objset_snap_cmtime(objset_t *os)
666 {
667 	return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
668 }
669 
670 /* called from dsl for meta-objset */
671 objset_t *
672 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
673     dmu_objset_type_t type, dmu_tx_t *tx)
674 {
675 	objset_t *os;
676 	dnode_t *mdn;
677 
678 	ASSERT(dmu_tx_is_syncing(tx));
679 
680 	if (ds != NULL)
681 		VERIFY0(dmu_objset_from_ds(ds, &os));
682 	else
683 		VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
684 
685 	mdn = DMU_META_DNODE(os);
686 
687 	dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
688 	    DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
689 
690 	/*
691 	 * We don't want to have to increase the meta-dnode's nlevels
692 	 * later, because then we could do it in quescing context while
693 	 * we are also accessing it in open context.
694 	 *
695 	 * This precaution is not necessary for the MOS (ds == NULL),
696 	 * because the MOS is only updated in syncing context.
697 	 * This is most fortunate: the MOS is the only objset that
698 	 * needs to be synced multiple times as spa_sync() iterates
699 	 * to convergence, so minimizing its dn_nlevels matters.
700 	 */
701 	if (ds != NULL) {
702 		int levels = 1;
703 
704 		/*
705 		 * Determine the number of levels necessary for the meta-dnode
706 		 * to contain DN_MAX_OBJECT dnodes.
707 		 */
708 		while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
709 		    (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
710 		    DN_MAX_OBJECT * sizeof (dnode_phys_t))
711 			levels++;
712 
713 		mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
714 		    mdn->dn_nlevels = levels;
715 	}
716 
717 	ASSERT(type != DMU_OST_NONE);
718 	ASSERT(type != DMU_OST_ANY);
719 	ASSERT(type < DMU_OST_NUMTYPES);
720 	os->os_phys->os_type = type;
721 	if (dmu_objset_userused_enabled(os)) {
722 		os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
723 		os->os_flags = os->os_phys->os_flags;
724 	}
725 
726 	dsl_dataset_dirty(ds, tx);
727 
728 	return (os);
729 }
730 
731 typedef struct dmu_objset_create_arg {
732 	const char *doca_name;
733 	cred_t *doca_cred;
734 	void (*doca_userfunc)(objset_t *os, void *arg,
735 	    cred_t *cr, dmu_tx_t *tx);
736 	void *doca_userarg;
737 	dmu_objset_type_t doca_type;
738 	uint64_t doca_flags;
739 } dmu_objset_create_arg_t;
740 
741 /*ARGSUSED*/
742 static int
743 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
744 {
745 	dmu_objset_create_arg_t *doca = arg;
746 	dsl_pool_t *dp = dmu_tx_pool(tx);
747 	dsl_dir_t *pdd;
748 	const char *tail;
749 	int error;
750 
751 	if (strchr(doca->doca_name, '@') != NULL)
752 		return (SET_ERROR(EINVAL));
753 
754 	error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
755 	if (error != 0)
756 		return (error);
757 	if (tail == NULL) {
758 		dsl_dir_rele(pdd, FTAG);
759 		return (SET_ERROR(EEXIST));
760 	}
761 	dsl_dir_rele(pdd, FTAG);
762 
763 	return (0);
764 }
765 
766 static void
767 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
768 {
769 	dmu_objset_create_arg_t *doca = arg;
770 	dsl_pool_t *dp = dmu_tx_pool(tx);
771 	dsl_dir_t *pdd;
772 	const char *tail;
773 	dsl_dataset_t *ds;
774 	uint64_t obj;
775 	blkptr_t *bp;
776 	objset_t *os;
777 
778 	VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
779 
780 	obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
781 	    doca->doca_cred, tx);
782 
783 	VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
784 	bp = dsl_dataset_get_blkptr(ds);
785 	os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
786 	    ds, bp, doca->doca_type, tx);
787 
788 	if (doca->doca_userfunc != NULL) {
789 		doca->doca_userfunc(os, doca->doca_userarg,
790 		    doca->doca_cred, tx);
791 	}
792 
793 	spa_history_log_internal_ds(ds, "create", tx, "");
794 	dsl_dataset_rele(ds, FTAG);
795 	dsl_dir_rele(pdd, FTAG);
796 }
797 
798 int
799 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
800     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
801 {
802 	dmu_objset_create_arg_t doca;
803 
804 	doca.doca_name = name;
805 	doca.doca_cred = CRED();
806 	doca.doca_flags = flags;
807 	doca.doca_userfunc = func;
808 	doca.doca_userarg = arg;
809 	doca.doca_type = type;
810 
811 	return (dsl_sync_task(name,
812 	    dmu_objset_create_check, dmu_objset_create_sync, &doca, 5));
813 }
814 
815 typedef struct dmu_objset_clone_arg {
816 	const char *doca_clone;
817 	const char *doca_origin;
818 	cred_t *doca_cred;
819 } dmu_objset_clone_arg_t;
820 
821 /*ARGSUSED*/
822 static int
823 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
824 {
825 	dmu_objset_clone_arg_t *doca = arg;
826 	dsl_dir_t *pdd;
827 	const char *tail;
828 	int error;
829 	dsl_dataset_t *origin;
830 	dsl_pool_t *dp = dmu_tx_pool(tx);
831 
832 	if (strchr(doca->doca_clone, '@') != NULL)
833 		return (SET_ERROR(EINVAL));
834 
835 	error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
836 	if (error != 0)
837 		return (error);
838 	if (tail == NULL) {
839 		dsl_dir_rele(pdd, FTAG);
840 		return (SET_ERROR(EEXIST));
841 	}
842 	/* You can't clone across pools. */
843 	if (pdd->dd_pool != dp) {
844 		dsl_dir_rele(pdd, FTAG);
845 		return (SET_ERROR(EXDEV));
846 	}
847 	dsl_dir_rele(pdd, FTAG);
848 
849 	error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
850 	if (error != 0)
851 		return (error);
852 
853 	/* You can't clone across pools. */
854 	if (origin->ds_dir->dd_pool != dp) {
855 		dsl_dataset_rele(origin, FTAG);
856 		return (SET_ERROR(EXDEV));
857 	}
858 
859 	/* You can only clone snapshots, not the head datasets. */
860 	if (!dsl_dataset_is_snapshot(origin)) {
861 		dsl_dataset_rele(origin, FTAG);
862 		return (SET_ERROR(EINVAL));
863 	}
864 	dsl_dataset_rele(origin, FTAG);
865 
866 	return (0);
867 }
868 
869 static void
870 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
871 {
872 	dmu_objset_clone_arg_t *doca = arg;
873 	dsl_pool_t *dp = dmu_tx_pool(tx);
874 	dsl_dir_t *pdd;
875 	const char *tail;
876 	dsl_dataset_t *origin, *ds;
877 	uint64_t obj;
878 	char namebuf[MAXNAMELEN];
879 
880 	VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
881 	VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
882 
883 	obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
884 	    doca->doca_cred, tx);
885 
886 	VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
887 	dsl_dataset_name(origin, namebuf);
888 	spa_history_log_internal_ds(ds, "clone", tx,
889 	    "origin=%s (%llu)", namebuf, origin->ds_object);
890 	dsl_dataset_rele(ds, FTAG);
891 	dsl_dataset_rele(origin, FTAG);
892 	dsl_dir_rele(pdd, FTAG);
893 }
894 
895 int
896 dmu_objset_clone(const char *clone, const char *origin)
897 {
898 	dmu_objset_clone_arg_t doca;
899 
900 	doca.doca_clone = clone;
901 	doca.doca_origin = origin;
902 	doca.doca_cred = CRED();
903 
904 	return (dsl_sync_task(clone,
905 	    dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 5));
906 }
907 
908 int
909 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
910 {
911 	int err;
912 	char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
913 	nvlist_t *snaps = fnvlist_alloc();
914 
915 	fnvlist_add_boolean(snaps, longsnap);
916 	strfree(longsnap);
917 	err = dsl_dataset_snapshot(snaps, NULL, NULL);
918 	fnvlist_free(snaps);
919 	return (err);
920 }
921 
922 static void
923 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
924 {
925 	dnode_t *dn;
926 
927 	while (dn = list_head(list)) {
928 		ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
929 		ASSERT(dn->dn_dbuf->db_data_pending);
930 		/*
931 		 * Initialize dn_zio outside dnode_sync() because the
932 		 * meta-dnode needs to set it ouside dnode_sync().
933 		 */
934 		dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
935 		ASSERT(dn->dn_zio);
936 
937 		ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
938 		list_remove(list, dn);
939 
940 		if (newlist) {
941 			(void) dnode_add_ref(dn, newlist);
942 			list_insert_tail(newlist, dn);
943 		}
944 
945 		dnode_sync(dn, tx);
946 	}
947 }
948 
949 /* ARGSUSED */
950 static void
951 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
952 {
953 	blkptr_t *bp = zio->io_bp;
954 	objset_t *os = arg;
955 	dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
956 
957 	ASSERT3P(bp, ==, os->os_rootbp);
958 	ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
959 	ASSERT0(BP_GET_LEVEL(bp));
960 
961 	/*
962 	 * Update rootbp fill count: it should be the number of objects
963 	 * allocated in the object set (not counting the "special"
964 	 * objects that are stored in the objset_phys_t -- the meta
965 	 * dnode and user/group accounting objects).
966 	 */
967 	bp->blk_fill = 0;
968 	for (int i = 0; i < dnp->dn_nblkptr; i++)
969 		bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
970 }
971 
972 /* ARGSUSED */
973 static void
974 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
975 {
976 	blkptr_t *bp = zio->io_bp;
977 	blkptr_t *bp_orig = &zio->io_bp_orig;
978 	objset_t *os = arg;
979 
980 	if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
981 		ASSERT(BP_EQUAL(bp, bp_orig));
982 	} else {
983 		dsl_dataset_t *ds = os->os_dsl_dataset;
984 		dmu_tx_t *tx = os->os_synctx;
985 
986 		(void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
987 		dsl_dataset_block_born(ds, bp, tx);
988 	}
989 }
990 
991 /* called from dsl */
992 void
993 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
994 {
995 	int txgoff;
996 	zbookmark_t zb;
997 	zio_prop_t zp;
998 	zio_t *zio;
999 	list_t *list;
1000 	list_t *newlist = NULL;
1001 	dbuf_dirty_record_t *dr;
1002 
1003 	dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1004 
1005 	ASSERT(dmu_tx_is_syncing(tx));
1006 	/* XXX the write_done callback should really give us the tx... */
1007 	os->os_synctx = tx;
1008 
1009 	if (os->os_dsl_dataset == NULL) {
1010 		/*
1011 		 * This is the MOS.  If we have upgraded,
1012 		 * spa_max_replication() could change, so reset
1013 		 * os_copies here.
1014 		 */
1015 		os->os_copies = spa_max_replication(os->os_spa);
1016 	}
1017 
1018 	/*
1019 	 * Create the root block IO
1020 	 */
1021 	SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1022 	    os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1023 	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1024 	arc_release(os->os_phys_buf, &os->os_phys_buf);
1025 
1026 	dmu_write_policy(os, NULL, 0, 0, &zp);
1027 
1028 	zio = arc_write(pio, os->os_spa, tx->tx_txg,
1029 	    os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1030 	    DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1031 	    NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1032 	    ZIO_FLAG_MUSTSUCCEED, &zb);
1033 
1034 	/*
1035 	 * Sync special dnodes - the parent IO for the sync is the root block
1036 	 */
1037 	DMU_META_DNODE(os)->dn_zio = zio;
1038 	dnode_sync(DMU_META_DNODE(os), tx);
1039 
1040 	os->os_phys->os_flags = os->os_flags;
1041 
1042 	if (DMU_USERUSED_DNODE(os) &&
1043 	    DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1044 		DMU_USERUSED_DNODE(os)->dn_zio = zio;
1045 		dnode_sync(DMU_USERUSED_DNODE(os), tx);
1046 		DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1047 		dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1048 	}
1049 
1050 	txgoff = tx->tx_txg & TXG_MASK;
1051 
1052 	if (dmu_objset_userused_enabled(os)) {
1053 		newlist = &os->os_synced_dnodes;
1054 		/*
1055 		 * We must create the list here because it uses the
1056 		 * dn_dirty_link[] of this txg.
1057 		 */
1058 		list_create(newlist, sizeof (dnode_t),
1059 		    offsetof(dnode_t, dn_dirty_link[txgoff]));
1060 	}
1061 
1062 	dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1063 	dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1064 
1065 	list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1066 	while (dr = list_head(list)) {
1067 		ASSERT0(dr->dr_dbuf->db_level);
1068 		list_remove(list, dr);
1069 		if (dr->dr_zio)
1070 			zio_nowait(dr->dr_zio);
1071 	}
1072 	/*
1073 	 * Free intent log blocks up to this tx.
1074 	 */
1075 	zil_sync(os->os_zil, tx);
1076 	os->os_phys->os_zil_header = os->os_zil_header;
1077 	zio_nowait(zio);
1078 }
1079 
1080 boolean_t
1081 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1082 {
1083 	return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1084 	    !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1085 }
1086 
1087 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1088 
1089 void
1090 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1091 {
1092 	used_cbs[ost] = cb;
1093 }
1094 
1095 boolean_t
1096 dmu_objset_userused_enabled(objset_t *os)
1097 {
1098 	return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1099 	    used_cbs[os->os_phys->os_type] != NULL &&
1100 	    DMU_USERUSED_DNODE(os) != NULL);
1101 }
1102 
1103 static void
1104 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1105     uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1106 {
1107 	if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1108 		int64_t delta = DNODE_SIZE + used;
1109 		if (subtract)
1110 			delta = -delta;
1111 		VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1112 		    user, delta, tx));
1113 		VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1114 		    group, delta, tx));
1115 	}
1116 }
1117 
1118 void
1119 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1120 {
1121 	dnode_t *dn;
1122 	list_t *list = &os->os_synced_dnodes;
1123 
1124 	ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1125 
1126 	while (dn = list_head(list)) {
1127 		int flags;
1128 		ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1129 		ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1130 		    dn->dn_phys->dn_flags &
1131 		    DNODE_FLAG_USERUSED_ACCOUNTED);
1132 
1133 		/* Allocate the user/groupused objects if necessary. */
1134 		if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1135 			VERIFY(0 == zap_create_claim(os,
1136 			    DMU_USERUSED_OBJECT,
1137 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1138 			VERIFY(0 == zap_create_claim(os,
1139 			    DMU_GROUPUSED_OBJECT,
1140 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1141 		}
1142 
1143 		/*
1144 		 * We intentionally modify the zap object even if the
1145 		 * net delta is zero.  Otherwise
1146 		 * the block of the zap obj could be shared between
1147 		 * datasets but need to be different between them after
1148 		 * a bprewrite.
1149 		 */
1150 
1151 		flags = dn->dn_id_flags;
1152 		ASSERT(flags);
1153 		if (flags & DN_ID_OLD_EXIST)  {
1154 			do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1155 			    dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1156 		}
1157 		if (flags & DN_ID_NEW_EXIST) {
1158 			do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1159 			    dn->dn_phys->dn_flags,  dn->dn_newuid,
1160 			    dn->dn_newgid, B_FALSE, tx);
1161 		}
1162 
1163 		mutex_enter(&dn->dn_mtx);
1164 		dn->dn_oldused = 0;
1165 		dn->dn_oldflags = 0;
1166 		if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1167 			dn->dn_olduid = dn->dn_newuid;
1168 			dn->dn_oldgid = dn->dn_newgid;
1169 			dn->dn_id_flags |= DN_ID_OLD_EXIST;
1170 			if (dn->dn_bonuslen == 0)
1171 				dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1172 			else
1173 				dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1174 		}
1175 		dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1176 		mutex_exit(&dn->dn_mtx);
1177 
1178 		list_remove(list, dn);
1179 		dnode_rele(dn, list);
1180 	}
1181 }
1182 
1183 /*
1184  * Returns a pointer to data to find uid/gid from
1185  *
1186  * If a dirty record for transaction group that is syncing can't
1187  * be found then NULL is returned.  In the NULL case it is assumed
1188  * the uid/gid aren't changing.
1189  */
1190 static void *
1191 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1192 {
1193 	dbuf_dirty_record_t *dr, **drp;
1194 	void *data;
1195 
1196 	if (db->db_dirtycnt == 0)
1197 		return (db->db.db_data);  /* Nothing is changing */
1198 
1199 	for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1200 		if (dr->dr_txg == tx->tx_txg)
1201 			break;
1202 
1203 	if (dr == NULL) {
1204 		data = NULL;
1205 	} else {
1206 		dnode_t *dn;
1207 
1208 		DB_DNODE_ENTER(dr->dr_dbuf);
1209 		dn = DB_DNODE(dr->dr_dbuf);
1210 
1211 		if (dn->dn_bonuslen == 0 &&
1212 		    dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1213 			data = dr->dt.dl.dr_data->b_data;
1214 		else
1215 			data = dr->dt.dl.dr_data;
1216 
1217 		DB_DNODE_EXIT(dr->dr_dbuf);
1218 	}
1219 
1220 	return (data);
1221 }
1222 
1223 void
1224 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1225 {
1226 	objset_t *os = dn->dn_objset;
1227 	void *data = NULL;
1228 	dmu_buf_impl_t *db = NULL;
1229 	uint64_t *user = NULL;
1230 	uint64_t *group = NULL;
1231 	int flags = dn->dn_id_flags;
1232 	int error;
1233 	boolean_t have_spill = B_FALSE;
1234 
1235 	if (!dmu_objset_userused_enabled(dn->dn_objset))
1236 		return;
1237 
1238 	if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1239 	    DN_ID_CHKED_SPILL)))
1240 		return;
1241 
1242 	if (before && dn->dn_bonuslen != 0)
1243 		data = DN_BONUS(dn->dn_phys);
1244 	else if (!before && dn->dn_bonuslen != 0) {
1245 		if (dn->dn_bonus) {
1246 			db = dn->dn_bonus;
1247 			mutex_enter(&db->db_mtx);
1248 			data = dmu_objset_userquota_find_data(db, tx);
1249 		} else {
1250 			data = DN_BONUS(dn->dn_phys);
1251 		}
1252 	} else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1253 			int rf = 0;
1254 
1255 			if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1256 				rf |= DB_RF_HAVESTRUCT;
1257 			error = dmu_spill_hold_by_dnode(dn,
1258 			    rf | DB_RF_MUST_SUCCEED,
1259 			    FTAG, (dmu_buf_t **)&db);
1260 			ASSERT(error == 0);
1261 			mutex_enter(&db->db_mtx);
1262 			data = (before) ? db->db.db_data :
1263 			    dmu_objset_userquota_find_data(db, tx);
1264 			have_spill = B_TRUE;
1265 	} else {
1266 		mutex_enter(&dn->dn_mtx);
1267 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1268 		mutex_exit(&dn->dn_mtx);
1269 		return;
1270 	}
1271 
1272 	if (before) {
1273 		ASSERT(data);
1274 		user = &dn->dn_olduid;
1275 		group = &dn->dn_oldgid;
1276 	} else if (data) {
1277 		user = &dn->dn_newuid;
1278 		group = &dn->dn_newgid;
1279 	}
1280 
1281 	/*
1282 	 * Must always call the callback in case the object
1283 	 * type has changed and that type isn't an object type to track
1284 	 */
1285 	error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1286 	    user, group);
1287 
1288 	/*
1289 	 * Preserve existing uid/gid when the callback can't determine
1290 	 * what the new uid/gid are and the callback returned EEXIST.
1291 	 * The EEXIST error tells us to just use the existing uid/gid.
1292 	 * If we don't know what the old values are then just assign
1293 	 * them to 0, since that is a new file  being created.
1294 	 */
1295 	if (!before && data == NULL && error == EEXIST) {
1296 		if (flags & DN_ID_OLD_EXIST) {
1297 			dn->dn_newuid = dn->dn_olduid;
1298 			dn->dn_newgid = dn->dn_oldgid;
1299 		} else {
1300 			dn->dn_newuid = 0;
1301 			dn->dn_newgid = 0;
1302 		}
1303 		error = 0;
1304 	}
1305 
1306 	if (db)
1307 		mutex_exit(&db->db_mtx);
1308 
1309 	mutex_enter(&dn->dn_mtx);
1310 	if (error == 0 && before)
1311 		dn->dn_id_flags |= DN_ID_OLD_EXIST;
1312 	if (error == 0 && !before)
1313 		dn->dn_id_flags |= DN_ID_NEW_EXIST;
1314 
1315 	if (have_spill) {
1316 		dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1317 	} else {
1318 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1319 	}
1320 	mutex_exit(&dn->dn_mtx);
1321 	if (have_spill)
1322 		dmu_buf_rele((dmu_buf_t *)db, FTAG);
1323 }
1324 
1325 boolean_t
1326 dmu_objset_userspace_present(objset_t *os)
1327 {
1328 	return (os->os_phys->os_flags &
1329 	    OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1330 }
1331 
1332 int
1333 dmu_objset_userspace_upgrade(objset_t *os)
1334 {
1335 	uint64_t obj;
1336 	int err = 0;
1337 
1338 	if (dmu_objset_userspace_present(os))
1339 		return (0);
1340 	if (!dmu_objset_userused_enabled(os))
1341 		return (SET_ERROR(ENOTSUP));
1342 	if (dmu_objset_is_snapshot(os))
1343 		return (SET_ERROR(EINVAL));
1344 
1345 	/*
1346 	 * We simply need to mark every object dirty, so that it will be
1347 	 * synced out and now accounted.  If this is called
1348 	 * concurrently, or if we already did some work before crashing,
1349 	 * that's fine, since we track each object's accounted state
1350 	 * independently.
1351 	 */
1352 
1353 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1354 		dmu_tx_t *tx;
1355 		dmu_buf_t *db;
1356 		int objerr;
1357 
1358 		if (issig(JUSTLOOKING) && issig(FORREAL))
1359 			return (SET_ERROR(EINTR));
1360 
1361 		objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1362 		if (objerr != 0)
1363 			continue;
1364 		tx = dmu_tx_create(os);
1365 		dmu_tx_hold_bonus(tx, obj);
1366 		objerr = dmu_tx_assign(tx, TXG_WAIT);
1367 		if (objerr != 0) {
1368 			dmu_tx_abort(tx);
1369 			continue;
1370 		}
1371 		dmu_buf_will_dirty(db, tx);
1372 		dmu_buf_rele(db, FTAG);
1373 		dmu_tx_commit(tx);
1374 	}
1375 
1376 	os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1377 	txg_wait_synced(dmu_objset_pool(os), 0);
1378 	return (0);
1379 }
1380 
1381 void
1382 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1383     uint64_t *usedobjsp, uint64_t *availobjsp)
1384 {
1385 	dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1386 	    usedobjsp, availobjsp);
1387 }
1388 
1389 uint64_t
1390 dmu_objset_fsid_guid(objset_t *os)
1391 {
1392 	return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1393 }
1394 
1395 void
1396 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1397 {
1398 	stat->dds_type = os->os_phys->os_type;
1399 	if (os->os_dsl_dataset)
1400 		dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1401 }
1402 
1403 void
1404 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1405 {
1406 	ASSERT(os->os_dsl_dataset ||
1407 	    os->os_phys->os_type == DMU_OST_META);
1408 
1409 	if (os->os_dsl_dataset != NULL)
1410 		dsl_dataset_stats(os->os_dsl_dataset, nv);
1411 
1412 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1413 	    os->os_phys->os_type);
1414 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1415 	    dmu_objset_userspace_present(os));
1416 }
1417 
1418 int
1419 dmu_objset_is_snapshot(objset_t *os)
1420 {
1421 	if (os->os_dsl_dataset != NULL)
1422 		return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1423 	else
1424 		return (B_FALSE);
1425 }
1426 
1427 int
1428 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1429     boolean_t *conflict)
1430 {
1431 	dsl_dataset_t *ds = os->os_dsl_dataset;
1432 	uint64_t ignored;
1433 
1434 	if (ds->ds_phys->ds_snapnames_zapobj == 0)
1435 		return (SET_ERROR(ENOENT));
1436 
1437 	return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1438 	    ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1439 	    real, maxlen, conflict));
1440 }
1441 
1442 int
1443 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1444     uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1445 {
1446 	dsl_dataset_t *ds = os->os_dsl_dataset;
1447 	zap_cursor_t cursor;
1448 	zap_attribute_t attr;
1449 
1450 	ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1451 
1452 	if (ds->ds_phys->ds_snapnames_zapobj == 0)
1453 		return (SET_ERROR(ENOENT));
1454 
1455 	zap_cursor_init_serialized(&cursor,
1456 	    ds->ds_dir->dd_pool->dp_meta_objset,
1457 	    ds->ds_phys->ds_snapnames_zapobj, *offp);
1458 
1459 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1460 		zap_cursor_fini(&cursor);
1461 		return (SET_ERROR(ENOENT));
1462 	}
1463 
1464 	if (strlen(attr.za_name) + 1 > namelen) {
1465 		zap_cursor_fini(&cursor);
1466 		return (SET_ERROR(ENAMETOOLONG));
1467 	}
1468 
1469 	(void) strcpy(name, attr.za_name);
1470 	if (idp)
1471 		*idp = attr.za_first_integer;
1472 	if (case_conflict)
1473 		*case_conflict = attr.za_normalization_conflict;
1474 	zap_cursor_advance(&cursor);
1475 	*offp = zap_cursor_serialize(&cursor);
1476 	zap_cursor_fini(&cursor);
1477 
1478 	return (0);
1479 }
1480 
1481 int
1482 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1483     uint64_t *idp, uint64_t *offp)
1484 {
1485 	dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1486 	zap_cursor_t cursor;
1487 	zap_attribute_t attr;
1488 
1489 	/* there is no next dir on a snapshot! */
1490 	if (os->os_dsl_dataset->ds_object !=
1491 	    dd->dd_phys->dd_head_dataset_obj)
1492 		return (SET_ERROR(ENOENT));
1493 
1494 	zap_cursor_init_serialized(&cursor,
1495 	    dd->dd_pool->dp_meta_objset,
1496 	    dd->dd_phys->dd_child_dir_zapobj, *offp);
1497 
1498 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1499 		zap_cursor_fini(&cursor);
1500 		return (SET_ERROR(ENOENT));
1501 	}
1502 
1503 	if (strlen(attr.za_name) + 1 > namelen) {
1504 		zap_cursor_fini(&cursor);
1505 		return (SET_ERROR(ENAMETOOLONG));
1506 	}
1507 
1508 	(void) strcpy(name, attr.za_name);
1509 	if (idp)
1510 		*idp = attr.za_first_integer;
1511 	zap_cursor_advance(&cursor);
1512 	*offp = zap_cursor_serialize(&cursor);
1513 	zap_cursor_fini(&cursor);
1514 
1515 	return (0);
1516 }
1517 
1518 /*
1519  * Find objsets under and including ddobj, call func(ds) on each.
1520  */
1521 int
1522 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1523     int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1524 {
1525 	dsl_dir_t *dd;
1526 	dsl_dataset_t *ds;
1527 	zap_cursor_t zc;
1528 	zap_attribute_t *attr;
1529 	uint64_t thisobj;
1530 	int err;
1531 
1532 	ASSERT(dsl_pool_config_held(dp));
1533 
1534 	err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1535 	if (err != 0)
1536 		return (err);
1537 
1538 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1539 	if (dd->dd_myname[0] == '$') {
1540 		dsl_dir_rele(dd, FTAG);
1541 		return (0);
1542 	}
1543 
1544 	thisobj = dd->dd_phys->dd_head_dataset_obj;
1545 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1546 
1547 	/*
1548 	 * Iterate over all children.
1549 	 */
1550 	if (flags & DS_FIND_CHILDREN) {
1551 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
1552 		    dd->dd_phys->dd_child_dir_zapobj);
1553 		    zap_cursor_retrieve(&zc, attr) == 0;
1554 		    (void) zap_cursor_advance(&zc)) {
1555 			ASSERT3U(attr->za_integer_length, ==,
1556 			    sizeof (uint64_t));
1557 			ASSERT3U(attr->za_num_integers, ==, 1);
1558 
1559 			err = dmu_objset_find_dp(dp, attr->za_first_integer,
1560 			    func, arg, flags);
1561 			if (err != 0)
1562 				break;
1563 		}
1564 		zap_cursor_fini(&zc);
1565 
1566 		if (err != 0) {
1567 			dsl_dir_rele(dd, FTAG);
1568 			kmem_free(attr, sizeof (zap_attribute_t));
1569 			return (err);
1570 		}
1571 	}
1572 
1573 	/*
1574 	 * Iterate over all snapshots.
1575 	 */
1576 	if (flags & DS_FIND_SNAPSHOTS) {
1577 		dsl_dataset_t *ds;
1578 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1579 
1580 		if (err == 0) {
1581 			uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1582 			dsl_dataset_rele(ds, FTAG);
1583 
1584 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1585 			    zap_cursor_retrieve(&zc, attr) == 0;
1586 			    (void) zap_cursor_advance(&zc)) {
1587 				ASSERT3U(attr->za_integer_length, ==,
1588 				    sizeof (uint64_t));
1589 				ASSERT3U(attr->za_num_integers, ==, 1);
1590 
1591 				err = dsl_dataset_hold_obj(dp,
1592 				    attr->za_first_integer, FTAG, &ds);
1593 				if (err != 0)
1594 					break;
1595 				err = func(dp, ds, arg);
1596 				dsl_dataset_rele(ds, FTAG);
1597 				if (err != 0)
1598 					break;
1599 			}
1600 			zap_cursor_fini(&zc);
1601 		}
1602 	}
1603 
1604 	dsl_dir_rele(dd, FTAG);
1605 	kmem_free(attr, sizeof (zap_attribute_t));
1606 
1607 	if (err != 0)
1608 		return (err);
1609 
1610 	/*
1611 	 * Apply to self.
1612 	 */
1613 	err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1614 	if (err != 0)
1615 		return (err);
1616 	err = func(dp, ds, arg);
1617 	dsl_dataset_rele(ds, FTAG);
1618 	return (err);
1619 }
1620 
1621 /*
1622  * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1623  * The dp_config_rwlock must not be held when this is called, and it
1624  * will not be held when the callback is called.
1625  * Therefore this function should only be used when the pool is not changing
1626  * (e.g. in syncing context), or the callback can deal with the possible races.
1627  */
1628 static int
1629 dmu_objset_find_impl(spa_t *spa, const char *name,
1630     int func(const char *, void *), void *arg, int flags)
1631 {
1632 	dsl_dir_t *dd;
1633 	dsl_pool_t *dp = spa_get_dsl(spa);
1634 	dsl_dataset_t *ds;
1635 	zap_cursor_t zc;
1636 	zap_attribute_t *attr;
1637 	char *child;
1638 	uint64_t thisobj;
1639 	int err;
1640 
1641 	dsl_pool_config_enter(dp, FTAG);
1642 
1643 	err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1644 	if (err != 0) {
1645 		dsl_pool_config_exit(dp, FTAG);
1646 		return (err);
1647 	}
1648 
1649 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1650 	if (dd->dd_myname[0] == '$') {
1651 		dsl_dir_rele(dd, FTAG);
1652 		dsl_pool_config_exit(dp, FTAG);
1653 		return (0);
1654 	}
1655 
1656 	thisobj = dd->dd_phys->dd_head_dataset_obj;
1657 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1658 
1659 	/*
1660 	 * Iterate over all children.
1661 	 */
1662 	if (flags & DS_FIND_CHILDREN) {
1663 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
1664 		    dd->dd_phys->dd_child_dir_zapobj);
1665 		    zap_cursor_retrieve(&zc, attr) == 0;
1666 		    (void) zap_cursor_advance(&zc)) {
1667 			ASSERT3U(attr->za_integer_length, ==,
1668 			    sizeof (uint64_t));
1669 			ASSERT3U(attr->za_num_integers, ==, 1);
1670 
1671 			child = kmem_asprintf("%s/%s", name, attr->za_name);
1672 			dsl_pool_config_exit(dp, FTAG);
1673 			err = dmu_objset_find_impl(spa, child,
1674 			    func, arg, flags);
1675 			dsl_pool_config_enter(dp, FTAG);
1676 			strfree(child);
1677 			if (err != 0)
1678 				break;
1679 		}
1680 		zap_cursor_fini(&zc);
1681 
1682 		if (err != 0) {
1683 			dsl_dir_rele(dd, FTAG);
1684 			dsl_pool_config_exit(dp, FTAG);
1685 			kmem_free(attr, sizeof (zap_attribute_t));
1686 			return (err);
1687 		}
1688 	}
1689 
1690 	/*
1691 	 * Iterate over all snapshots.
1692 	 */
1693 	if (flags & DS_FIND_SNAPSHOTS) {
1694 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1695 
1696 		if (err == 0) {
1697 			uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1698 			dsl_dataset_rele(ds, FTAG);
1699 
1700 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1701 			    zap_cursor_retrieve(&zc, attr) == 0;
1702 			    (void) zap_cursor_advance(&zc)) {
1703 				ASSERT3U(attr->za_integer_length, ==,
1704 				    sizeof (uint64_t));
1705 				ASSERT3U(attr->za_num_integers, ==, 1);
1706 
1707 				child = kmem_asprintf("%s@%s",
1708 				    name, attr->za_name);
1709 				dsl_pool_config_exit(dp, FTAG);
1710 				err = func(child, arg);
1711 				dsl_pool_config_enter(dp, FTAG);
1712 				strfree(child);
1713 				if (err != 0)
1714 					break;
1715 			}
1716 			zap_cursor_fini(&zc);
1717 		}
1718 	}
1719 
1720 	dsl_dir_rele(dd, FTAG);
1721 	kmem_free(attr, sizeof (zap_attribute_t));
1722 	dsl_pool_config_exit(dp, FTAG);
1723 
1724 	if (err != 0)
1725 		return (err);
1726 
1727 	/* Apply to self. */
1728 	return (func(name, arg));
1729 }
1730 
1731 /*
1732  * See comment above dmu_objset_find_impl().
1733  */
1734 int
1735 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1736     int flags)
1737 {
1738 	spa_t *spa;
1739 	int error;
1740 
1741 	error = spa_open(name, &spa, FTAG);
1742 	if (error != 0)
1743 		return (error);
1744 	error = dmu_objset_find_impl(spa, name, func, arg, flags);
1745 	spa_close(spa, FTAG);
1746 	return (error);
1747 }
1748 
1749 void
1750 dmu_objset_set_user(objset_t *os, void *user_ptr)
1751 {
1752 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1753 	os->os_user_ptr = user_ptr;
1754 }
1755 
1756 void *
1757 dmu_objset_get_user(objset_t *os)
1758 {
1759 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1760 	return (os->os_user_ptr);
1761 }
1762 
1763 /*
1764  * Determine name of filesystem, given name of snapshot.
1765  * buf must be at least MAXNAMELEN bytes
1766  */
1767 int
1768 dmu_fsname(const char *snapname, char *buf)
1769 {
1770 	char *atp = strchr(snapname, '@');
1771 	if (atp == NULL)
1772 		return (SET_ERROR(EINVAL));
1773 	if (atp - snapname >= MAXNAMELEN)
1774 		return (SET_ERROR(ENAMETOOLONG));
1775 	(void) strlcpy(buf, snapname, atp - snapname + 1);
1776 	return (0);
1777 }
1778