xref: /illumos-gate/usr/src/uts/common/fs/zfs/dmu_objset.c (revision d2058105c61ec61df3a2dd3f839fed8c3fe7bfd6)
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  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27  * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
28  * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
29  */
30 
31 /* Portions Copyright 2010 Robert Milkowski */
32 
33 #include <sys/cred.h>
34 #include <sys/zfs_context.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dsl_dir.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/dsl_prop.h>
39 #include <sys/dsl_pool.h>
40 #include <sys/dsl_synctask.h>
41 #include <sys/dsl_deleg.h>
42 #include <sys/dnode.h>
43 #include <sys/dbuf.h>
44 #include <sys/zvol.h>
45 #include <sys/dmu_tx.h>
46 #include <sys/zap.h>
47 #include <sys/zil.h>
48 #include <sys/dmu_impl.h>
49 #include <sys/zfs_ioctl.h>
50 #include <sys/sa.h>
51 #include <sys/zfs_onexit.h>
52 #include <sys/dsl_destroy.h>
53 #include <sys/vdev.h>
54 
55 /*
56  * Needed to close a window in dnode_move() that allows the objset to be freed
57  * before it can be safely accessed.
58  */
59 krwlock_t os_lock;
60 
61 /*
62  * Tunable to overwrite the maximum number of threads for the parallization
63  * of dmu_objset_find_dp, needed to speed up the import of pools with many
64  * datasets.
65  * Default is 4 times the number of leaf vdevs.
66  */
67 int dmu_find_threads = 0;
68 
69 static void dmu_objset_find_dp_cb(void *arg);
70 
71 void
72 dmu_objset_init(void)
73 {
74 	rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
75 }
76 
77 void
78 dmu_objset_fini(void)
79 {
80 	rw_destroy(&os_lock);
81 }
82 
83 spa_t *
84 dmu_objset_spa(objset_t *os)
85 {
86 	return (os->os_spa);
87 }
88 
89 zilog_t *
90 dmu_objset_zil(objset_t *os)
91 {
92 	return (os->os_zil);
93 }
94 
95 dsl_pool_t *
96 dmu_objset_pool(objset_t *os)
97 {
98 	dsl_dataset_t *ds;
99 
100 	if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
101 		return (ds->ds_dir->dd_pool);
102 	else
103 		return (spa_get_dsl(os->os_spa));
104 }
105 
106 dsl_dataset_t *
107 dmu_objset_ds(objset_t *os)
108 {
109 	return (os->os_dsl_dataset);
110 }
111 
112 dmu_objset_type_t
113 dmu_objset_type(objset_t *os)
114 {
115 	return (os->os_phys->os_type);
116 }
117 
118 void
119 dmu_objset_name(objset_t *os, char *buf)
120 {
121 	dsl_dataset_name(os->os_dsl_dataset, buf);
122 }
123 
124 uint64_t
125 dmu_objset_id(objset_t *os)
126 {
127 	dsl_dataset_t *ds = os->os_dsl_dataset;
128 
129 	return (ds ? ds->ds_object : 0);
130 }
131 
132 zfs_sync_type_t
133 dmu_objset_syncprop(objset_t *os)
134 {
135 	return (os->os_sync);
136 }
137 
138 zfs_logbias_op_t
139 dmu_objset_logbias(objset_t *os)
140 {
141 	return (os->os_logbias);
142 }
143 
144 static void
145 checksum_changed_cb(void *arg, uint64_t newval)
146 {
147 	objset_t *os = arg;
148 
149 	/*
150 	 * Inheritance should have been done by now.
151 	 */
152 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
153 
154 	os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
155 }
156 
157 static void
158 compression_changed_cb(void *arg, uint64_t newval)
159 {
160 	objset_t *os = arg;
161 
162 	/*
163 	 * Inheritance and range checking should have been done by now.
164 	 */
165 	ASSERT(newval != ZIO_COMPRESS_INHERIT);
166 
167 	os->os_compress = zio_compress_select(os->os_spa, newval,
168 	    ZIO_COMPRESS_ON);
169 }
170 
171 static void
172 copies_changed_cb(void *arg, uint64_t newval)
173 {
174 	objset_t *os = arg;
175 
176 	/*
177 	 * Inheritance and range checking should have been done by now.
178 	 */
179 	ASSERT(newval > 0);
180 	ASSERT(newval <= spa_max_replication(os->os_spa));
181 
182 	os->os_copies = newval;
183 }
184 
185 static void
186 dedup_changed_cb(void *arg, uint64_t newval)
187 {
188 	objset_t *os = arg;
189 	spa_t *spa = os->os_spa;
190 	enum zio_checksum checksum;
191 
192 	/*
193 	 * Inheritance should have been done by now.
194 	 */
195 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
196 
197 	checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
198 
199 	os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
200 	os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
201 }
202 
203 static void
204 primary_cache_changed_cb(void *arg, uint64_t newval)
205 {
206 	objset_t *os = arg;
207 
208 	/*
209 	 * Inheritance and range checking should have been done by now.
210 	 */
211 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
212 	    newval == ZFS_CACHE_METADATA);
213 
214 	os->os_primary_cache = newval;
215 }
216 
217 static void
218 secondary_cache_changed_cb(void *arg, uint64_t newval)
219 {
220 	objset_t *os = arg;
221 
222 	/*
223 	 * Inheritance and range checking should have been done by now.
224 	 */
225 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
226 	    newval == ZFS_CACHE_METADATA);
227 
228 	os->os_secondary_cache = newval;
229 }
230 
231 static void
232 sync_changed_cb(void *arg, uint64_t newval)
233 {
234 	objset_t *os = arg;
235 
236 	/*
237 	 * Inheritance and range checking should have been done by now.
238 	 */
239 	ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
240 	    newval == ZFS_SYNC_DISABLED);
241 
242 	os->os_sync = newval;
243 	if (os->os_zil)
244 		zil_set_sync(os->os_zil, newval);
245 }
246 
247 static void
248 redundant_metadata_changed_cb(void *arg, uint64_t newval)
249 {
250 	objset_t *os = arg;
251 
252 	/*
253 	 * Inheritance and range checking should have been done by now.
254 	 */
255 	ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
256 	    newval == ZFS_REDUNDANT_METADATA_MOST);
257 
258 	os->os_redundant_metadata = newval;
259 }
260 
261 static void
262 logbias_changed_cb(void *arg, uint64_t newval)
263 {
264 	objset_t *os = arg;
265 
266 	ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
267 	    newval == ZFS_LOGBIAS_THROUGHPUT);
268 	os->os_logbias = newval;
269 	if (os->os_zil)
270 		zil_set_logbias(os->os_zil, newval);
271 }
272 
273 static void
274 recordsize_changed_cb(void *arg, uint64_t newval)
275 {
276 	objset_t *os = arg;
277 
278 	os->os_recordsize = newval;
279 }
280 
281 void
282 dmu_objset_byteswap(void *buf, size_t size)
283 {
284 	objset_phys_t *osp = buf;
285 
286 	ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
287 	dnode_byteswap(&osp->os_meta_dnode);
288 	byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
289 	osp->os_type = BSWAP_64(osp->os_type);
290 	osp->os_flags = BSWAP_64(osp->os_flags);
291 	if (size == sizeof (objset_phys_t)) {
292 		dnode_byteswap(&osp->os_userused_dnode);
293 		dnode_byteswap(&osp->os_groupused_dnode);
294 	}
295 }
296 
297 int
298 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
299     objset_t **osp)
300 {
301 	objset_t *os;
302 	int i, err;
303 
304 	ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
305 
306 	os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
307 	os->os_dsl_dataset = ds;
308 	os->os_spa = spa;
309 	os->os_rootbp = bp;
310 	if (!BP_IS_HOLE(os->os_rootbp)) {
311 		arc_flags_t aflags = ARC_FLAG_WAIT;
312 		zbookmark_phys_t zb;
313 		SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
314 		    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
315 
316 		if (DMU_OS_IS_L2CACHEABLE(os))
317 			aflags |= ARC_FLAG_L2CACHE;
318 		if (DMU_OS_IS_L2COMPRESSIBLE(os))
319 			aflags |= ARC_FLAG_L2COMPRESS;
320 
321 		dprintf_bp(os->os_rootbp, "reading %s", "");
322 		err = arc_read(NULL, spa, os->os_rootbp,
323 		    arc_getbuf_func, &os->os_phys_buf,
324 		    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
325 		if (err != 0) {
326 			kmem_free(os, sizeof (objset_t));
327 			/* convert checksum errors into IO errors */
328 			if (err == ECKSUM)
329 				err = SET_ERROR(EIO);
330 			return (err);
331 		}
332 
333 		/* Increase the blocksize if we are permitted. */
334 		if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
335 		    arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
336 			arc_buf_t *buf = arc_buf_alloc(spa,
337 			    sizeof (objset_phys_t), &os->os_phys_buf,
338 			    ARC_BUFC_METADATA);
339 			bzero(buf->b_data, sizeof (objset_phys_t));
340 			bcopy(os->os_phys_buf->b_data, buf->b_data,
341 			    arc_buf_size(os->os_phys_buf));
342 			(void) arc_buf_remove_ref(os->os_phys_buf,
343 			    &os->os_phys_buf);
344 			os->os_phys_buf = buf;
345 		}
346 
347 		os->os_phys = os->os_phys_buf->b_data;
348 		os->os_flags = os->os_phys->os_flags;
349 	} else {
350 		int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
351 		    sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
352 		os->os_phys_buf = arc_buf_alloc(spa, size,
353 		    &os->os_phys_buf, ARC_BUFC_METADATA);
354 		os->os_phys = os->os_phys_buf->b_data;
355 		bzero(os->os_phys, size);
356 	}
357 
358 	/*
359 	 * Note: the changed_cb will be called once before the register
360 	 * func returns, thus changing the checksum/compression from the
361 	 * default (fletcher2/off).  Snapshots don't need to know about
362 	 * checksum/compression/copies.
363 	 */
364 	if (ds != NULL) {
365 		boolean_t needlock = B_FALSE;
366 
367 		/*
368 		 * Note: it's valid to open the objset if the dataset is
369 		 * long-held, in which case the pool_config lock will not
370 		 * be held.
371 		 */
372 		if (!dsl_pool_config_held(dmu_objset_pool(os))) {
373 			needlock = B_TRUE;
374 			dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
375 		}
376 		err = dsl_prop_register(ds,
377 		    zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
378 		    primary_cache_changed_cb, os);
379 		if (err == 0) {
380 			err = dsl_prop_register(ds,
381 			    zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
382 			    secondary_cache_changed_cb, os);
383 		}
384 		if (!ds->ds_is_snapshot) {
385 			if (err == 0) {
386 				err = dsl_prop_register(ds,
387 				    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
388 				    checksum_changed_cb, os);
389 			}
390 			if (err == 0) {
391 				err = dsl_prop_register(ds,
392 				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
393 				    compression_changed_cb, os);
394 			}
395 			if (err == 0) {
396 				err = dsl_prop_register(ds,
397 				    zfs_prop_to_name(ZFS_PROP_COPIES),
398 				    copies_changed_cb, os);
399 			}
400 			if (err == 0) {
401 				err = dsl_prop_register(ds,
402 				    zfs_prop_to_name(ZFS_PROP_DEDUP),
403 				    dedup_changed_cb, os);
404 			}
405 			if (err == 0) {
406 				err = dsl_prop_register(ds,
407 				    zfs_prop_to_name(ZFS_PROP_LOGBIAS),
408 				    logbias_changed_cb, os);
409 			}
410 			if (err == 0) {
411 				err = dsl_prop_register(ds,
412 				    zfs_prop_to_name(ZFS_PROP_SYNC),
413 				    sync_changed_cb, os);
414 			}
415 			if (err == 0) {
416 				err = dsl_prop_register(ds,
417 				    zfs_prop_to_name(
418 				    ZFS_PROP_REDUNDANT_METADATA),
419 				    redundant_metadata_changed_cb, os);
420 			}
421 			if (err == 0) {
422 				err = dsl_prop_register(ds,
423 				    zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
424 				    recordsize_changed_cb, os);
425 			}
426 		}
427 		if (needlock)
428 			dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
429 		if (err != 0) {
430 			VERIFY(arc_buf_remove_ref(os->os_phys_buf,
431 			    &os->os_phys_buf));
432 			kmem_free(os, sizeof (objset_t));
433 			return (err);
434 		}
435 	} else {
436 		/* It's the meta-objset. */
437 		os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
438 		os->os_compress = ZIO_COMPRESS_ON;
439 		os->os_copies = spa_max_replication(spa);
440 		os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
441 		os->os_dedup_verify = B_FALSE;
442 		os->os_logbias = ZFS_LOGBIAS_LATENCY;
443 		os->os_sync = ZFS_SYNC_STANDARD;
444 		os->os_primary_cache = ZFS_CACHE_ALL;
445 		os->os_secondary_cache = ZFS_CACHE_ALL;
446 	}
447 
448 	if (ds == NULL || !ds->ds_is_snapshot)
449 		os->os_zil_header = os->os_phys->os_zil_header;
450 	os->os_zil = zil_alloc(os, &os->os_zil_header);
451 
452 	for (i = 0; i < TXG_SIZE; i++) {
453 		list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
454 		    offsetof(dnode_t, dn_dirty_link[i]));
455 		list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
456 		    offsetof(dnode_t, dn_dirty_link[i]));
457 	}
458 	list_create(&os->os_dnodes, sizeof (dnode_t),
459 	    offsetof(dnode_t, dn_link));
460 	list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
461 	    offsetof(dmu_buf_impl_t, db_link));
462 
463 	mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
464 	mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
465 	mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
466 
467 	dnode_special_open(os, &os->os_phys->os_meta_dnode,
468 	    DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
469 	if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
470 		dnode_special_open(os, &os->os_phys->os_userused_dnode,
471 		    DMU_USERUSED_OBJECT, &os->os_userused_dnode);
472 		dnode_special_open(os, &os->os_phys->os_groupused_dnode,
473 		    DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
474 	}
475 
476 	*osp = os;
477 	return (0);
478 }
479 
480 int
481 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
482 {
483 	int err = 0;
484 
485 	/*
486 	 * We shouldn't be doing anything with dsl_dataset_t's unless the
487 	 * pool_config lock is held, or the dataset is long-held.
488 	 */
489 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) ||
490 	    dsl_dataset_long_held(ds));
491 
492 	mutex_enter(&ds->ds_opening_lock);
493 	if (ds->ds_objset == NULL) {
494 		objset_t *os;
495 		err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
496 		    ds, dsl_dataset_get_blkptr(ds), &os);
497 
498 		if (err == 0) {
499 			mutex_enter(&ds->ds_lock);
500 			ASSERT(ds->ds_objset == NULL);
501 			ds->ds_objset = os;
502 			mutex_exit(&ds->ds_lock);
503 		}
504 	}
505 	*osp = ds->ds_objset;
506 	mutex_exit(&ds->ds_opening_lock);
507 	return (err);
508 }
509 
510 /*
511  * Holds the pool while the objset is held.  Therefore only one objset
512  * can be held at a time.
513  */
514 int
515 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
516 {
517 	dsl_pool_t *dp;
518 	dsl_dataset_t *ds;
519 	int err;
520 
521 	err = dsl_pool_hold(name, tag, &dp);
522 	if (err != 0)
523 		return (err);
524 	err = dsl_dataset_hold(dp, name, tag, &ds);
525 	if (err != 0) {
526 		dsl_pool_rele(dp, tag);
527 		return (err);
528 	}
529 
530 	err = dmu_objset_from_ds(ds, osp);
531 	if (err != 0) {
532 		dsl_dataset_rele(ds, tag);
533 		dsl_pool_rele(dp, tag);
534 	}
535 
536 	return (err);
537 }
538 
539 static int
540 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
541     boolean_t readonly, void *tag, objset_t **osp)
542 {
543 	int err;
544 
545 	err = dmu_objset_from_ds(ds, osp);
546 	if (err != 0) {
547 		dsl_dataset_disown(ds, tag);
548 	} else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
549 		dsl_dataset_disown(ds, tag);
550 		return (SET_ERROR(EINVAL));
551 	} else if (!readonly && dsl_dataset_is_snapshot(ds)) {
552 		dsl_dataset_disown(ds, tag);
553 		return (SET_ERROR(EROFS));
554 	}
555 	return (err);
556 }
557 
558 /*
559  * dsl_pool must not be held when this is called.
560  * Upon successful return, there will be a longhold on the dataset,
561  * and the dsl_pool will not be held.
562  */
563 int
564 dmu_objset_own(const char *name, dmu_objset_type_t type,
565     boolean_t readonly, void *tag, objset_t **osp)
566 {
567 	dsl_pool_t *dp;
568 	dsl_dataset_t *ds;
569 	int err;
570 
571 	err = dsl_pool_hold(name, FTAG, &dp);
572 	if (err != 0)
573 		return (err);
574 	err = dsl_dataset_own(dp, name, tag, &ds);
575 	if (err != 0) {
576 		dsl_pool_rele(dp, FTAG);
577 		return (err);
578 	}
579 	err = dmu_objset_own_impl(ds, type, readonly, tag, osp);
580 	dsl_pool_rele(dp, FTAG);
581 
582 	return (err);
583 }
584 
585 int
586 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
587     boolean_t readonly, void *tag, objset_t **osp)
588 {
589 	dsl_dataset_t *ds;
590 	int err;
591 
592 	err = dsl_dataset_own_obj(dp, obj, tag, &ds);
593 	if (err != 0)
594 		return (err);
595 
596 	return (dmu_objset_own_impl(ds, type, readonly, tag, osp));
597 }
598 
599 void
600 dmu_objset_rele(objset_t *os, void *tag)
601 {
602 	dsl_pool_t *dp = dmu_objset_pool(os);
603 	dsl_dataset_rele(os->os_dsl_dataset, tag);
604 	dsl_pool_rele(dp, tag);
605 }
606 
607 /*
608  * When we are called, os MUST refer to an objset associated with a dataset
609  * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
610  * == tag.  We will then release and reacquire ownership of the dataset while
611  * holding the pool config_rwlock to avoid intervening namespace or ownership
612  * changes may occur.
613  *
614  * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
615  * release the hold on its dataset and acquire a new one on the dataset of the
616  * same name so that it can be partially torn down and reconstructed.
617  */
618 void
619 dmu_objset_refresh_ownership(objset_t *os, void *tag)
620 {
621 	dsl_pool_t *dp;
622 	dsl_dataset_t *ds, *newds;
623 	char name[MAXNAMELEN];
624 
625 	ds = os->os_dsl_dataset;
626 	VERIFY3P(ds, !=, NULL);
627 	VERIFY3P(ds->ds_owner, ==, tag);
628 	VERIFY(dsl_dataset_long_held(ds));
629 
630 	dsl_dataset_name(ds, name);
631 	dp = dmu_objset_pool(os);
632 	dsl_pool_config_enter(dp, FTAG);
633 	dmu_objset_disown(os, tag);
634 	VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
635 	VERIFY3P(newds, ==, os->os_dsl_dataset);
636 	dsl_pool_config_exit(dp, FTAG);
637 }
638 
639 void
640 dmu_objset_disown(objset_t *os, void *tag)
641 {
642 	dsl_dataset_disown(os->os_dsl_dataset, tag);
643 }
644 
645 void
646 dmu_objset_evict_dbufs(objset_t *os)
647 {
648 	dnode_t dn_marker;
649 	dnode_t *dn;
650 
651 	mutex_enter(&os->os_lock);
652 	dn = list_head(&os->os_dnodes);
653 	while (dn != NULL) {
654 		/*
655 		 * Skip dnodes without holds.  We have to do this dance
656 		 * because dnode_add_ref() only works if there is already a
657 		 * hold.  If the dnode has no holds, then it has no dbufs.
658 		 */
659 		if (dnode_add_ref(dn, FTAG)) {
660 			list_insert_after(&os->os_dnodes, dn, &dn_marker);
661 			mutex_exit(&os->os_lock);
662 
663 			dnode_evict_dbufs(dn);
664 			dnode_rele(dn, FTAG);
665 
666 			mutex_enter(&os->os_lock);
667 			dn = list_next(&os->os_dnodes, &dn_marker);
668 			list_remove(&os->os_dnodes, &dn_marker);
669 		} else {
670 			dn = list_next(&os->os_dnodes, dn);
671 		}
672 	}
673 	mutex_exit(&os->os_lock);
674 
675 	if (DMU_USERUSED_DNODE(os) != NULL) {
676 		dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
677 		dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
678 	}
679 	dnode_evict_dbufs(DMU_META_DNODE(os));
680 }
681 
682 /*
683  * Objset eviction processing is split into into two pieces.
684  * The first marks the objset as evicting, evicts any dbufs that
685  * have a refcount of zero, and then queues up the objset for the
686  * second phase of eviction.  Once os->os_dnodes has been cleared by
687  * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
688  * The second phase closes the special dnodes, dequeues the objset from
689  * the list of those undergoing eviction, and finally frees the objset.
690  *
691  * NOTE: Due to asynchronous eviction processing (invocation of
692  *       dnode_buf_pageout()), it is possible for the meta dnode for the
693  *       objset to have no holds even though os->os_dnodes is not empty.
694  */
695 void
696 dmu_objset_evict(objset_t *os)
697 {
698 	dsl_dataset_t *ds = os->os_dsl_dataset;
699 
700 	for (int t = 0; t < TXG_SIZE; t++)
701 		ASSERT(!dmu_objset_is_dirty(os, t));
702 
703 	if (ds)
704 		dsl_prop_unregister_all(ds, os);
705 
706 	if (os->os_sa)
707 		sa_tear_down(os);
708 
709 	dmu_objset_evict_dbufs(os);
710 
711 	mutex_enter(&os->os_lock);
712 	spa_evicting_os_register(os->os_spa, os);
713 	if (list_is_empty(&os->os_dnodes)) {
714 		mutex_exit(&os->os_lock);
715 		dmu_objset_evict_done(os);
716 	} else {
717 		mutex_exit(&os->os_lock);
718 	}
719 }
720 
721 void
722 dmu_objset_evict_done(objset_t *os)
723 {
724 	ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
725 
726 	dnode_special_close(&os->os_meta_dnode);
727 	if (DMU_USERUSED_DNODE(os)) {
728 		dnode_special_close(&os->os_userused_dnode);
729 		dnode_special_close(&os->os_groupused_dnode);
730 	}
731 	zil_free(os->os_zil);
732 
733 	VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
734 
735 	/*
736 	 * This is a barrier to prevent the objset from going away in
737 	 * dnode_move() until we can safely ensure that the objset is still in
738 	 * use. We consider the objset valid before the barrier and invalid
739 	 * after the barrier.
740 	 */
741 	rw_enter(&os_lock, RW_READER);
742 	rw_exit(&os_lock);
743 
744 	mutex_destroy(&os->os_lock);
745 	mutex_destroy(&os->os_obj_lock);
746 	mutex_destroy(&os->os_user_ptr_lock);
747 	spa_evicting_os_deregister(os->os_spa, os);
748 	kmem_free(os, sizeof (objset_t));
749 }
750 
751 timestruc_t
752 dmu_objset_snap_cmtime(objset_t *os)
753 {
754 	return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
755 }
756 
757 /* called from dsl for meta-objset */
758 objset_t *
759 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
760     dmu_objset_type_t type, dmu_tx_t *tx)
761 {
762 	objset_t *os;
763 	dnode_t *mdn;
764 
765 	ASSERT(dmu_tx_is_syncing(tx));
766 
767 	if (ds != NULL)
768 		VERIFY0(dmu_objset_from_ds(ds, &os));
769 	else
770 		VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
771 
772 	mdn = DMU_META_DNODE(os);
773 
774 	dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
775 	    DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
776 
777 	/*
778 	 * We don't want to have to increase the meta-dnode's nlevels
779 	 * later, because then we could do it in quescing context while
780 	 * we are also accessing it in open context.
781 	 *
782 	 * This precaution is not necessary for the MOS (ds == NULL),
783 	 * because the MOS is only updated in syncing context.
784 	 * This is most fortunate: the MOS is the only objset that
785 	 * needs to be synced multiple times as spa_sync() iterates
786 	 * to convergence, so minimizing its dn_nlevels matters.
787 	 */
788 	if (ds != NULL) {
789 		int levels = 1;
790 
791 		/*
792 		 * Determine the number of levels necessary for the meta-dnode
793 		 * to contain DN_MAX_OBJECT dnodes.
794 		 */
795 		while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
796 		    (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
797 		    DN_MAX_OBJECT * sizeof (dnode_phys_t))
798 			levels++;
799 
800 		mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
801 		    mdn->dn_nlevels = levels;
802 	}
803 
804 	ASSERT(type != DMU_OST_NONE);
805 	ASSERT(type != DMU_OST_ANY);
806 	ASSERT(type < DMU_OST_NUMTYPES);
807 	os->os_phys->os_type = type;
808 	if (dmu_objset_userused_enabled(os)) {
809 		os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
810 		os->os_flags = os->os_phys->os_flags;
811 	}
812 
813 	dsl_dataset_dirty(ds, tx);
814 
815 	return (os);
816 }
817 
818 typedef struct dmu_objset_create_arg {
819 	const char *doca_name;
820 	cred_t *doca_cred;
821 	void (*doca_userfunc)(objset_t *os, void *arg,
822 	    cred_t *cr, dmu_tx_t *tx);
823 	void *doca_userarg;
824 	dmu_objset_type_t doca_type;
825 	uint64_t doca_flags;
826 } dmu_objset_create_arg_t;
827 
828 /*ARGSUSED*/
829 static int
830 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
831 {
832 	dmu_objset_create_arg_t *doca = arg;
833 	dsl_pool_t *dp = dmu_tx_pool(tx);
834 	dsl_dir_t *pdd;
835 	const char *tail;
836 	int error;
837 
838 	if (strchr(doca->doca_name, '@') != NULL)
839 		return (SET_ERROR(EINVAL));
840 
841 	error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
842 	if (error != 0)
843 		return (error);
844 	if (tail == NULL) {
845 		dsl_dir_rele(pdd, FTAG);
846 		return (SET_ERROR(EEXIST));
847 	}
848 	error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
849 	    doca->doca_cred);
850 	dsl_dir_rele(pdd, FTAG);
851 
852 	return (error);
853 }
854 
855 static void
856 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
857 {
858 	dmu_objset_create_arg_t *doca = arg;
859 	dsl_pool_t *dp = dmu_tx_pool(tx);
860 	dsl_dir_t *pdd;
861 	const char *tail;
862 	dsl_dataset_t *ds;
863 	uint64_t obj;
864 	blkptr_t *bp;
865 	objset_t *os;
866 
867 	VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
868 
869 	obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
870 	    doca->doca_cred, tx);
871 
872 	VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
873 	bp = dsl_dataset_get_blkptr(ds);
874 	os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
875 	    ds, bp, doca->doca_type, tx);
876 
877 	if (doca->doca_userfunc != NULL) {
878 		doca->doca_userfunc(os, doca->doca_userarg,
879 		    doca->doca_cred, tx);
880 	}
881 
882 	spa_history_log_internal_ds(ds, "create", tx, "");
883 	dsl_dataset_rele(ds, FTAG);
884 	dsl_dir_rele(pdd, FTAG);
885 }
886 
887 int
888 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
889     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
890 {
891 	dmu_objset_create_arg_t doca;
892 
893 	doca.doca_name = name;
894 	doca.doca_cred = CRED();
895 	doca.doca_flags = flags;
896 	doca.doca_userfunc = func;
897 	doca.doca_userarg = arg;
898 	doca.doca_type = type;
899 
900 	return (dsl_sync_task(name,
901 	    dmu_objset_create_check, dmu_objset_create_sync, &doca,
902 	    5, ZFS_SPACE_CHECK_NORMAL));
903 }
904 
905 typedef struct dmu_objset_clone_arg {
906 	const char *doca_clone;
907 	const char *doca_origin;
908 	cred_t *doca_cred;
909 } dmu_objset_clone_arg_t;
910 
911 /*ARGSUSED*/
912 static int
913 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
914 {
915 	dmu_objset_clone_arg_t *doca = arg;
916 	dsl_dir_t *pdd;
917 	const char *tail;
918 	int error;
919 	dsl_dataset_t *origin;
920 	dsl_pool_t *dp = dmu_tx_pool(tx);
921 
922 	if (strchr(doca->doca_clone, '@') != NULL)
923 		return (SET_ERROR(EINVAL));
924 
925 	error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
926 	if (error != 0)
927 		return (error);
928 	if (tail == NULL) {
929 		dsl_dir_rele(pdd, FTAG);
930 		return (SET_ERROR(EEXIST));
931 	}
932 
933 	error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
934 	    doca->doca_cred);
935 	if (error != 0) {
936 		dsl_dir_rele(pdd, FTAG);
937 		return (SET_ERROR(EDQUOT));
938 	}
939 	dsl_dir_rele(pdd, FTAG);
940 
941 	error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
942 	if (error != 0)
943 		return (error);
944 
945 	/* You can only clone snapshots, not the head datasets. */
946 	if (!origin->ds_is_snapshot) {
947 		dsl_dataset_rele(origin, FTAG);
948 		return (SET_ERROR(EINVAL));
949 	}
950 	dsl_dataset_rele(origin, FTAG);
951 
952 	return (0);
953 }
954 
955 static void
956 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
957 {
958 	dmu_objset_clone_arg_t *doca = arg;
959 	dsl_pool_t *dp = dmu_tx_pool(tx);
960 	dsl_dir_t *pdd;
961 	const char *tail;
962 	dsl_dataset_t *origin, *ds;
963 	uint64_t obj;
964 	char namebuf[MAXNAMELEN];
965 
966 	VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
967 	VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
968 
969 	obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
970 	    doca->doca_cred, tx);
971 
972 	VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
973 	dsl_dataset_name(origin, namebuf);
974 	spa_history_log_internal_ds(ds, "clone", tx,
975 	    "origin=%s (%llu)", namebuf, origin->ds_object);
976 	dsl_dataset_rele(ds, FTAG);
977 	dsl_dataset_rele(origin, FTAG);
978 	dsl_dir_rele(pdd, FTAG);
979 }
980 
981 int
982 dmu_objset_clone(const char *clone, const char *origin)
983 {
984 	dmu_objset_clone_arg_t doca;
985 
986 	doca.doca_clone = clone;
987 	doca.doca_origin = origin;
988 	doca.doca_cred = CRED();
989 
990 	return (dsl_sync_task(clone,
991 	    dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
992 	    5, ZFS_SPACE_CHECK_NORMAL));
993 }
994 
995 int
996 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
997 {
998 	int err;
999 	char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1000 	nvlist_t *snaps = fnvlist_alloc();
1001 
1002 	fnvlist_add_boolean(snaps, longsnap);
1003 	strfree(longsnap);
1004 	err = dsl_dataset_snapshot(snaps, NULL, NULL);
1005 	fnvlist_free(snaps);
1006 	return (err);
1007 }
1008 
1009 static void
1010 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
1011 {
1012 	dnode_t *dn;
1013 
1014 	while (dn = list_head(list)) {
1015 		ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1016 		ASSERT(dn->dn_dbuf->db_data_pending);
1017 		/*
1018 		 * Initialize dn_zio outside dnode_sync() because the
1019 		 * meta-dnode needs to set it ouside dnode_sync().
1020 		 */
1021 		dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1022 		ASSERT(dn->dn_zio);
1023 
1024 		ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1025 		list_remove(list, dn);
1026 
1027 		if (newlist) {
1028 			(void) dnode_add_ref(dn, newlist);
1029 			list_insert_tail(newlist, dn);
1030 		}
1031 
1032 		dnode_sync(dn, tx);
1033 	}
1034 }
1035 
1036 /* ARGSUSED */
1037 static void
1038 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1039 {
1040 	blkptr_t *bp = zio->io_bp;
1041 	objset_t *os = arg;
1042 	dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1043 
1044 	ASSERT(!BP_IS_EMBEDDED(bp));
1045 	ASSERT3P(bp, ==, os->os_rootbp);
1046 	ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1047 	ASSERT0(BP_GET_LEVEL(bp));
1048 
1049 	/*
1050 	 * Update rootbp fill count: it should be the number of objects
1051 	 * allocated in the object set (not counting the "special"
1052 	 * objects that are stored in the objset_phys_t -- the meta
1053 	 * dnode and user/group accounting objects).
1054 	 */
1055 	bp->blk_fill = 0;
1056 	for (int i = 0; i < dnp->dn_nblkptr; i++)
1057 		bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1058 }
1059 
1060 /* ARGSUSED */
1061 static void
1062 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1063 {
1064 	blkptr_t *bp = zio->io_bp;
1065 	blkptr_t *bp_orig = &zio->io_bp_orig;
1066 	objset_t *os = arg;
1067 
1068 	if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1069 		ASSERT(BP_EQUAL(bp, bp_orig));
1070 	} else {
1071 		dsl_dataset_t *ds = os->os_dsl_dataset;
1072 		dmu_tx_t *tx = os->os_synctx;
1073 
1074 		(void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1075 		dsl_dataset_block_born(ds, bp, tx);
1076 	}
1077 }
1078 
1079 /* called from dsl */
1080 void
1081 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1082 {
1083 	int txgoff;
1084 	zbookmark_phys_t zb;
1085 	zio_prop_t zp;
1086 	zio_t *zio;
1087 	list_t *list;
1088 	list_t *newlist = NULL;
1089 	dbuf_dirty_record_t *dr;
1090 
1091 	dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1092 
1093 	ASSERT(dmu_tx_is_syncing(tx));
1094 	/* XXX the write_done callback should really give us the tx... */
1095 	os->os_synctx = tx;
1096 
1097 	if (os->os_dsl_dataset == NULL) {
1098 		/*
1099 		 * This is the MOS.  If we have upgraded,
1100 		 * spa_max_replication() could change, so reset
1101 		 * os_copies here.
1102 		 */
1103 		os->os_copies = spa_max_replication(os->os_spa);
1104 	}
1105 
1106 	/*
1107 	 * Create the root block IO
1108 	 */
1109 	SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1110 	    os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1111 	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1112 	arc_release(os->os_phys_buf, &os->os_phys_buf);
1113 
1114 	dmu_write_policy(os, NULL, 0, 0, &zp);
1115 
1116 	zio = arc_write(pio, os->os_spa, tx->tx_txg,
1117 	    os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1118 	    DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1119 	    NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1120 	    ZIO_FLAG_MUSTSUCCEED, &zb);
1121 
1122 	/*
1123 	 * Sync special dnodes - the parent IO for the sync is the root block
1124 	 */
1125 	DMU_META_DNODE(os)->dn_zio = zio;
1126 	dnode_sync(DMU_META_DNODE(os), tx);
1127 
1128 	os->os_phys->os_flags = os->os_flags;
1129 
1130 	if (DMU_USERUSED_DNODE(os) &&
1131 	    DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1132 		DMU_USERUSED_DNODE(os)->dn_zio = zio;
1133 		dnode_sync(DMU_USERUSED_DNODE(os), tx);
1134 		DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1135 		dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1136 	}
1137 
1138 	txgoff = tx->tx_txg & TXG_MASK;
1139 
1140 	if (dmu_objset_userused_enabled(os)) {
1141 		newlist = &os->os_synced_dnodes;
1142 		/*
1143 		 * We must create the list here because it uses the
1144 		 * dn_dirty_link[] of this txg.
1145 		 */
1146 		list_create(newlist, sizeof (dnode_t),
1147 		    offsetof(dnode_t, dn_dirty_link[txgoff]));
1148 	}
1149 
1150 	dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1151 	dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1152 
1153 	list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1154 	while (dr = list_head(list)) {
1155 		ASSERT0(dr->dr_dbuf->db_level);
1156 		list_remove(list, dr);
1157 		if (dr->dr_zio)
1158 			zio_nowait(dr->dr_zio);
1159 	}
1160 	/*
1161 	 * Free intent log blocks up to this tx.
1162 	 */
1163 	zil_sync(os->os_zil, tx);
1164 	os->os_phys->os_zil_header = os->os_zil_header;
1165 	zio_nowait(zio);
1166 }
1167 
1168 boolean_t
1169 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1170 {
1171 	return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1172 	    !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1173 }
1174 
1175 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1176 
1177 void
1178 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1179 {
1180 	used_cbs[ost] = cb;
1181 }
1182 
1183 boolean_t
1184 dmu_objset_userused_enabled(objset_t *os)
1185 {
1186 	return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1187 	    used_cbs[os->os_phys->os_type] != NULL &&
1188 	    DMU_USERUSED_DNODE(os) != NULL);
1189 }
1190 
1191 static void
1192 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1193     uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1194 {
1195 	if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1196 		int64_t delta = DNODE_SIZE + used;
1197 		if (subtract)
1198 			delta = -delta;
1199 		VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1200 		    user, delta, tx));
1201 		VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1202 		    group, delta, tx));
1203 	}
1204 }
1205 
1206 void
1207 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1208 {
1209 	dnode_t *dn;
1210 	list_t *list = &os->os_synced_dnodes;
1211 
1212 	ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1213 
1214 	while (dn = list_head(list)) {
1215 		int flags;
1216 		ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1217 		ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1218 		    dn->dn_phys->dn_flags &
1219 		    DNODE_FLAG_USERUSED_ACCOUNTED);
1220 
1221 		/* Allocate the user/groupused objects if necessary. */
1222 		if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1223 			VERIFY(0 == zap_create_claim(os,
1224 			    DMU_USERUSED_OBJECT,
1225 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1226 			VERIFY(0 == zap_create_claim(os,
1227 			    DMU_GROUPUSED_OBJECT,
1228 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1229 		}
1230 
1231 		/*
1232 		 * We intentionally modify the zap object even if the
1233 		 * net delta is zero.  Otherwise
1234 		 * the block of the zap obj could be shared between
1235 		 * datasets but need to be different between them after
1236 		 * a bprewrite.
1237 		 */
1238 
1239 		flags = dn->dn_id_flags;
1240 		ASSERT(flags);
1241 		if (flags & DN_ID_OLD_EXIST)  {
1242 			do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1243 			    dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1244 		}
1245 		if (flags & DN_ID_NEW_EXIST) {
1246 			do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1247 			    dn->dn_phys->dn_flags,  dn->dn_newuid,
1248 			    dn->dn_newgid, B_FALSE, tx);
1249 		}
1250 
1251 		mutex_enter(&dn->dn_mtx);
1252 		dn->dn_oldused = 0;
1253 		dn->dn_oldflags = 0;
1254 		if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1255 			dn->dn_olduid = dn->dn_newuid;
1256 			dn->dn_oldgid = dn->dn_newgid;
1257 			dn->dn_id_flags |= DN_ID_OLD_EXIST;
1258 			if (dn->dn_bonuslen == 0)
1259 				dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1260 			else
1261 				dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1262 		}
1263 		dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1264 		mutex_exit(&dn->dn_mtx);
1265 
1266 		list_remove(list, dn);
1267 		dnode_rele(dn, list);
1268 	}
1269 }
1270 
1271 /*
1272  * Returns a pointer to data to find uid/gid from
1273  *
1274  * If a dirty record for transaction group that is syncing can't
1275  * be found then NULL is returned.  In the NULL case it is assumed
1276  * the uid/gid aren't changing.
1277  */
1278 static void *
1279 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1280 {
1281 	dbuf_dirty_record_t *dr, **drp;
1282 	void *data;
1283 
1284 	if (db->db_dirtycnt == 0)
1285 		return (db->db.db_data);  /* Nothing is changing */
1286 
1287 	for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1288 		if (dr->dr_txg == tx->tx_txg)
1289 			break;
1290 
1291 	if (dr == NULL) {
1292 		data = NULL;
1293 	} else {
1294 		dnode_t *dn;
1295 
1296 		DB_DNODE_ENTER(dr->dr_dbuf);
1297 		dn = DB_DNODE(dr->dr_dbuf);
1298 
1299 		if (dn->dn_bonuslen == 0 &&
1300 		    dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1301 			data = dr->dt.dl.dr_data->b_data;
1302 		else
1303 			data = dr->dt.dl.dr_data;
1304 
1305 		DB_DNODE_EXIT(dr->dr_dbuf);
1306 	}
1307 
1308 	return (data);
1309 }
1310 
1311 void
1312 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1313 {
1314 	objset_t *os = dn->dn_objset;
1315 	void *data = NULL;
1316 	dmu_buf_impl_t *db = NULL;
1317 	uint64_t *user = NULL;
1318 	uint64_t *group = NULL;
1319 	int flags = dn->dn_id_flags;
1320 	int error;
1321 	boolean_t have_spill = B_FALSE;
1322 
1323 	if (!dmu_objset_userused_enabled(dn->dn_objset))
1324 		return;
1325 
1326 	if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1327 	    DN_ID_CHKED_SPILL)))
1328 		return;
1329 
1330 	if (before && dn->dn_bonuslen != 0)
1331 		data = DN_BONUS(dn->dn_phys);
1332 	else if (!before && dn->dn_bonuslen != 0) {
1333 		if (dn->dn_bonus) {
1334 			db = dn->dn_bonus;
1335 			mutex_enter(&db->db_mtx);
1336 			data = dmu_objset_userquota_find_data(db, tx);
1337 		} else {
1338 			data = DN_BONUS(dn->dn_phys);
1339 		}
1340 	} else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1341 			int rf = 0;
1342 
1343 			if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1344 				rf |= DB_RF_HAVESTRUCT;
1345 			error = dmu_spill_hold_by_dnode(dn,
1346 			    rf | DB_RF_MUST_SUCCEED,
1347 			    FTAG, (dmu_buf_t **)&db);
1348 			ASSERT(error == 0);
1349 			mutex_enter(&db->db_mtx);
1350 			data = (before) ? db->db.db_data :
1351 			    dmu_objset_userquota_find_data(db, tx);
1352 			have_spill = B_TRUE;
1353 	} else {
1354 		mutex_enter(&dn->dn_mtx);
1355 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1356 		mutex_exit(&dn->dn_mtx);
1357 		return;
1358 	}
1359 
1360 	if (before) {
1361 		ASSERT(data);
1362 		user = &dn->dn_olduid;
1363 		group = &dn->dn_oldgid;
1364 	} else if (data) {
1365 		user = &dn->dn_newuid;
1366 		group = &dn->dn_newgid;
1367 	}
1368 
1369 	/*
1370 	 * Must always call the callback in case the object
1371 	 * type has changed and that type isn't an object type to track
1372 	 */
1373 	error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1374 	    user, group);
1375 
1376 	/*
1377 	 * Preserve existing uid/gid when the callback can't determine
1378 	 * what the new uid/gid are and the callback returned EEXIST.
1379 	 * The EEXIST error tells us to just use the existing uid/gid.
1380 	 * If we don't know what the old values are then just assign
1381 	 * them to 0, since that is a new file  being created.
1382 	 */
1383 	if (!before && data == NULL && error == EEXIST) {
1384 		if (flags & DN_ID_OLD_EXIST) {
1385 			dn->dn_newuid = dn->dn_olduid;
1386 			dn->dn_newgid = dn->dn_oldgid;
1387 		} else {
1388 			dn->dn_newuid = 0;
1389 			dn->dn_newgid = 0;
1390 		}
1391 		error = 0;
1392 	}
1393 
1394 	if (db)
1395 		mutex_exit(&db->db_mtx);
1396 
1397 	mutex_enter(&dn->dn_mtx);
1398 	if (error == 0 && before)
1399 		dn->dn_id_flags |= DN_ID_OLD_EXIST;
1400 	if (error == 0 && !before)
1401 		dn->dn_id_flags |= DN_ID_NEW_EXIST;
1402 
1403 	if (have_spill) {
1404 		dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1405 	} else {
1406 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1407 	}
1408 	mutex_exit(&dn->dn_mtx);
1409 	if (have_spill)
1410 		dmu_buf_rele((dmu_buf_t *)db, FTAG);
1411 }
1412 
1413 boolean_t
1414 dmu_objset_userspace_present(objset_t *os)
1415 {
1416 	return (os->os_phys->os_flags &
1417 	    OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1418 }
1419 
1420 int
1421 dmu_objset_userspace_upgrade(objset_t *os)
1422 {
1423 	uint64_t obj;
1424 	int err = 0;
1425 
1426 	if (dmu_objset_userspace_present(os))
1427 		return (0);
1428 	if (!dmu_objset_userused_enabled(os))
1429 		return (SET_ERROR(ENOTSUP));
1430 	if (dmu_objset_is_snapshot(os))
1431 		return (SET_ERROR(EINVAL));
1432 
1433 	/*
1434 	 * We simply need to mark every object dirty, so that it will be
1435 	 * synced out and now accounted.  If this is called
1436 	 * concurrently, or if we already did some work before crashing,
1437 	 * that's fine, since we track each object's accounted state
1438 	 * independently.
1439 	 */
1440 
1441 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1442 		dmu_tx_t *tx;
1443 		dmu_buf_t *db;
1444 		int objerr;
1445 
1446 		if (issig(JUSTLOOKING) && issig(FORREAL))
1447 			return (SET_ERROR(EINTR));
1448 
1449 		objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1450 		if (objerr != 0)
1451 			continue;
1452 		tx = dmu_tx_create(os);
1453 		dmu_tx_hold_bonus(tx, obj);
1454 		objerr = dmu_tx_assign(tx, TXG_WAIT);
1455 		if (objerr != 0) {
1456 			dmu_tx_abort(tx);
1457 			continue;
1458 		}
1459 		dmu_buf_will_dirty(db, tx);
1460 		dmu_buf_rele(db, FTAG);
1461 		dmu_tx_commit(tx);
1462 	}
1463 
1464 	os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1465 	txg_wait_synced(dmu_objset_pool(os), 0);
1466 	return (0);
1467 }
1468 
1469 void
1470 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1471     uint64_t *usedobjsp, uint64_t *availobjsp)
1472 {
1473 	dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1474 	    usedobjsp, availobjsp);
1475 }
1476 
1477 uint64_t
1478 dmu_objset_fsid_guid(objset_t *os)
1479 {
1480 	return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1481 }
1482 
1483 void
1484 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1485 {
1486 	stat->dds_type = os->os_phys->os_type;
1487 	if (os->os_dsl_dataset)
1488 		dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1489 }
1490 
1491 void
1492 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1493 {
1494 	ASSERT(os->os_dsl_dataset ||
1495 	    os->os_phys->os_type == DMU_OST_META);
1496 
1497 	if (os->os_dsl_dataset != NULL)
1498 		dsl_dataset_stats(os->os_dsl_dataset, nv);
1499 
1500 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1501 	    os->os_phys->os_type);
1502 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1503 	    dmu_objset_userspace_present(os));
1504 }
1505 
1506 int
1507 dmu_objset_is_snapshot(objset_t *os)
1508 {
1509 	if (os->os_dsl_dataset != NULL)
1510 		return (os->os_dsl_dataset->ds_is_snapshot);
1511 	else
1512 		return (B_FALSE);
1513 }
1514 
1515 int
1516 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1517     boolean_t *conflict)
1518 {
1519 	dsl_dataset_t *ds = os->os_dsl_dataset;
1520 	uint64_t ignored;
1521 
1522 	if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1523 		return (SET_ERROR(ENOENT));
1524 
1525 	return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1526 	    dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1527 	    MT_FIRST, real, maxlen, conflict));
1528 }
1529 
1530 int
1531 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1532     uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1533 {
1534 	dsl_dataset_t *ds = os->os_dsl_dataset;
1535 	zap_cursor_t cursor;
1536 	zap_attribute_t attr;
1537 
1538 	ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1539 
1540 	if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1541 		return (SET_ERROR(ENOENT));
1542 
1543 	zap_cursor_init_serialized(&cursor,
1544 	    ds->ds_dir->dd_pool->dp_meta_objset,
1545 	    dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1546 
1547 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1548 		zap_cursor_fini(&cursor);
1549 		return (SET_ERROR(ENOENT));
1550 	}
1551 
1552 	if (strlen(attr.za_name) + 1 > namelen) {
1553 		zap_cursor_fini(&cursor);
1554 		return (SET_ERROR(ENAMETOOLONG));
1555 	}
1556 
1557 	(void) strcpy(name, attr.za_name);
1558 	if (idp)
1559 		*idp = attr.za_first_integer;
1560 	if (case_conflict)
1561 		*case_conflict = attr.za_normalization_conflict;
1562 	zap_cursor_advance(&cursor);
1563 	*offp = zap_cursor_serialize(&cursor);
1564 	zap_cursor_fini(&cursor);
1565 
1566 	return (0);
1567 }
1568 
1569 int
1570 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1571     uint64_t *idp, uint64_t *offp)
1572 {
1573 	dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1574 	zap_cursor_t cursor;
1575 	zap_attribute_t attr;
1576 
1577 	/* there is no next dir on a snapshot! */
1578 	if (os->os_dsl_dataset->ds_object !=
1579 	    dsl_dir_phys(dd)->dd_head_dataset_obj)
1580 		return (SET_ERROR(ENOENT));
1581 
1582 	zap_cursor_init_serialized(&cursor,
1583 	    dd->dd_pool->dp_meta_objset,
1584 	    dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1585 
1586 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1587 		zap_cursor_fini(&cursor);
1588 		return (SET_ERROR(ENOENT));
1589 	}
1590 
1591 	if (strlen(attr.za_name) + 1 > namelen) {
1592 		zap_cursor_fini(&cursor);
1593 		return (SET_ERROR(ENAMETOOLONG));
1594 	}
1595 
1596 	(void) strcpy(name, attr.za_name);
1597 	if (idp)
1598 		*idp = attr.za_first_integer;
1599 	zap_cursor_advance(&cursor);
1600 	*offp = zap_cursor_serialize(&cursor);
1601 	zap_cursor_fini(&cursor);
1602 
1603 	return (0);
1604 }
1605 
1606 typedef struct dmu_objset_find_ctx {
1607 	taskq_t		*dc_tq;
1608 	dsl_pool_t	*dc_dp;
1609 	uint64_t	dc_ddobj;
1610 	int		(*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
1611 	void		*dc_arg;
1612 	int		dc_flags;
1613 	kmutex_t	*dc_error_lock;
1614 	int		*dc_error;
1615 } dmu_objset_find_ctx_t;
1616 
1617 static void
1618 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
1619 {
1620 	dsl_pool_t *dp = dcp->dc_dp;
1621 	dmu_objset_find_ctx_t *child_dcp;
1622 	dsl_dir_t *dd;
1623 	dsl_dataset_t *ds;
1624 	zap_cursor_t zc;
1625 	zap_attribute_t *attr;
1626 	uint64_t thisobj;
1627 	int err = 0;
1628 
1629 	/* don't process if there already was an error */
1630 	if (*dcp->dc_error != 0)
1631 		goto out;
1632 
1633 	err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, NULL, FTAG, &dd);
1634 	if (err != 0)
1635 		goto out;
1636 
1637 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1638 	if (dd->dd_myname[0] == '$') {
1639 		dsl_dir_rele(dd, FTAG);
1640 		goto out;
1641 	}
1642 
1643 	thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1644 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1645 
1646 	/*
1647 	 * Iterate over all children.
1648 	 */
1649 	if (dcp->dc_flags & DS_FIND_CHILDREN) {
1650 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
1651 		    dsl_dir_phys(dd)->dd_child_dir_zapobj);
1652 		    zap_cursor_retrieve(&zc, attr) == 0;
1653 		    (void) zap_cursor_advance(&zc)) {
1654 			ASSERT3U(attr->za_integer_length, ==,
1655 			    sizeof (uint64_t));
1656 			ASSERT3U(attr->za_num_integers, ==, 1);
1657 
1658 			child_dcp = kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
1659 			*child_dcp = *dcp;
1660 			child_dcp->dc_ddobj = attr->za_first_integer;
1661 			if (dcp->dc_tq != NULL)
1662 				(void) taskq_dispatch(dcp->dc_tq,
1663 				    dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
1664 			else
1665 				dmu_objset_find_dp_impl(child_dcp);
1666 		}
1667 		zap_cursor_fini(&zc);
1668 	}
1669 
1670 	/*
1671 	 * Iterate over all snapshots.
1672 	 */
1673 	if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
1674 		dsl_dataset_t *ds;
1675 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1676 
1677 		if (err == 0) {
1678 			uint64_t snapobj;
1679 
1680 			snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1681 			dsl_dataset_rele(ds, FTAG);
1682 
1683 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1684 			    zap_cursor_retrieve(&zc, attr) == 0;
1685 			    (void) zap_cursor_advance(&zc)) {
1686 				ASSERT3U(attr->za_integer_length, ==,
1687 				    sizeof (uint64_t));
1688 				ASSERT3U(attr->za_num_integers, ==, 1);
1689 
1690 				err = dsl_dataset_hold_obj(dp,
1691 				    attr->za_first_integer, FTAG, &ds);
1692 				if (err != 0)
1693 					break;
1694 				err = dcp->dc_func(dp, ds, dcp->dc_arg);
1695 				dsl_dataset_rele(ds, FTAG);
1696 				if (err != 0)
1697 					break;
1698 			}
1699 			zap_cursor_fini(&zc);
1700 		}
1701 	}
1702 
1703 	dsl_dir_rele(dd, FTAG);
1704 	kmem_free(attr, sizeof (zap_attribute_t));
1705 
1706 	if (err != 0)
1707 		goto out;
1708 
1709 	/*
1710 	 * Apply to self.
1711 	 */
1712 	err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1713 	if (err != 0)
1714 		goto out;
1715 	err = dcp->dc_func(dp, ds, dcp->dc_arg);
1716 	dsl_dataset_rele(ds, FTAG);
1717 
1718 out:
1719 	if (err != 0) {
1720 		mutex_enter(dcp->dc_error_lock);
1721 		/* only keep first error */
1722 		if (*dcp->dc_error == 0)
1723 			*dcp->dc_error = err;
1724 		mutex_exit(dcp->dc_error_lock);
1725 	}
1726 
1727 	kmem_free(dcp, sizeof (*dcp));
1728 }
1729 
1730 static void
1731 dmu_objset_find_dp_cb(void *arg)
1732 {
1733 	dmu_objset_find_ctx_t *dcp = arg;
1734 	dsl_pool_t *dp = dcp->dc_dp;
1735 
1736 	/*
1737 	 * We need to get a pool_config_lock here, as there are several
1738 	 * asssert(pool_config_held) down the stack. Getting a lock via
1739 	 * dsl_pool_config_enter is risky, as it might be stalled by a
1740 	 * pending writer. This would deadlock, as the write lock can
1741 	 * only be granted when our parent thread gives up the lock.
1742 	 * The _prio interface gives us priority over a pending writer.
1743 	 */
1744 	dsl_pool_config_enter_prio(dp, FTAG);
1745 
1746 	dmu_objset_find_dp_impl(dcp);
1747 
1748 	dsl_pool_config_exit(dp, FTAG);
1749 }
1750 
1751 /*
1752  * Find objsets under and including ddobj, call func(ds) on each.
1753  * The order for the enumeration is completely undefined.
1754  * func is called with dsl_pool_config held.
1755  */
1756 int
1757 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1758     int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1759 {
1760 	int error = 0;
1761 	taskq_t *tq = NULL;
1762 	int ntasks;
1763 	dmu_objset_find_ctx_t *dcp;
1764 	kmutex_t err_lock;
1765 
1766 	mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
1767 	dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
1768 	dcp->dc_tq = NULL;
1769 	dcp->dc_dp = dp;
1770 	dcp->dc_ddobj = ddobj;
1771 	dcp->dc_func = func;
1772 	dcp->dc_arg = arg;
1773 	dcp->dc_flags = flags;
1774 	dcp->dc_error_lock = &err_lock;
1775 	dcp->dc_error = &error;
1776 
1777 	if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
1778 		/*
1779 		 * In case a write lock is held we can't make use of
1780 		 * parallelism, as down the stack of the worker threads
1781 		 * the lock is asserted via dsl_pool_config_held.
1782 		 * In case of a read lock this is solved by getting a read
1783 		 * lock in each worker thread, which isn't possible in case
1784 		 * of a writer lock. So we fall back to the synchronous path
1785 		 * here.
1786 		 * In the future it might be possible to get some magic into
1787 		 * dsl_pool_config_held in a way that it returns true for
1788 		 * the worker threads so that a single lock held from this
1789 		 * thread suffices. For now, stay single threaded.
1790 		 */
1791 		dmu_objset_find_dp_impl(dcp);
1792 
1793 		return (error);
1794 	}
1795 
1796 	ntasks = dmu_find_threads;
1797 	if (ntasks == 0)
1798 		ntasks = vdev_count_leaves(dp->dp_spa) * 4;
1799 	tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks,
1800 	    INT_MAX, 0);
1801 	if (tq == NULL) {
1802 		kmem_free(dcp, sizeof (*dcp));
1803 		return (SET_ERROR(ENOMEM));
1804 	}
1805 	dcp->dc_tq = tq;
1806 
1807 	/* dcp will be freed by task */
1808 	(void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
1809 
1810 	/*
1811 	 * PORTING: this code relies on the property of taskq_wait to wait
1812 	 * until no more tasks are queued and no more tasks are active. As
1813 	 * we always queue new tasks from within other tasks, task_wait
1814 	 * reliably waits for the full recursion to finish, even though we
1815 	 * enqueue new tasks after taskq_wait has been called.
1816 	 * On platforms other than illumos, taskq_wait may not have this
1817 	 * property.
1818 	 */
1819 	taskq_wait(tq);
1820 	taskq_destroy(tq);
1821 	mutex_destroy(&err_lock);
1822 
1823 	return (error);
1824 }
1825 
1826 /*
1827  * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1828  * The dp_config_rwlock must not be held when this is called, and it
1829  * will not be held when the callback is called.
1830  * Therefore this function should only be used when the pool is not changing
1831  * (e.g. in syncing context), or the callback can deal with the possible races.
1832  */
1833 static int
1834 dmu_objset_find_impl(spa_t *spa, const char *name,
1835     int func(const char *, void *), void *arg, int flags)
1836 {
1837 	dsl_dir_t *dd;
1838 	dsl_pool_t *dp = spa_get_dsl(spa);
1839 	dsl_dataset_t *ds;
1840 	zap_cursor_t zc;
1841 	zap_attribute_t *attr;
1842 	char *child;
1843 	uint64_t thisobj;
1844 	int err;
1845 
1846 	dsl_pool_config_enter(dp, FTAG);
1847 
1848 	err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1849 	if (err != 0) {
1850 		dsl_pool_config_exit(dp, FTAG);
1851 		return (err);
1852 	}
1853 
1854 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1855 	if (dd->dd_myname[0] == '$') {
1856 		dsl_dir_rele(dd, FTAG);
1857 		dsl_pool_config_exit(dp, FTAG);
1858 		return (0);
1859 	}
1860 
1861 	thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1862 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1863 
1864 	/*
1865 	 * Iterate over all children.
1866 	 */
1867 	if (flags & DS_FIND_CHILDREN) {
1868 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
1869 		    dsl_dir_phys(dd)->dd_child_dir_zapobj);
1870 		    zap_cursor_retrieve(&zc, attr) == 0;
1871 		    (void) zap_cursor_advance(&zc)) {
1872 			ASSERT3U(attr->za_integer_length, ==,
1873 			    sizeof (uint64_t));
1874 			ASSERT3U(attr->za_num_integers, ==, 1);
1875 
1876 			child = kmem_asprintf("%s/%s", name, attr->za_name);
1877 			dsl_pool_config_exit(dp, FTAG);
1878 			err = dmu_objset_find_impl(spa, child,
1879 			    func, arg, flags);
1880 			dsl_pool_config_enter(dp, FTAG);
1881 			strfree(child);
1882 			if (err != 0)
1883 				break;
1884 		}
1885 		zap_cursor_fini(&zc);
1886 
1887 		if (err != 0) {
1888 			dsl_dir_rele(dd, FTAG);
1889 			dsl_pool_config_exit(dp, FTAG);
1890 			kmem_free(attr, sizeof (zap_attribute_t));
1891 			return (err);
1892 		}
1893 	}
1894 
1895 	/*
1896 	 * Iterate over all snapshots.
1897 	 */
1898 	if (flags & DS_FIND_SNAPSHOTS) {
1899 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1900 
1901 		if (err == 0) {
1902 			uint64_t snapobj;
1903 
1904 			snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1905 			dsl_dataset_rele(ds, FTAG);
1906 
1907 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1908 			    zap_cursor_retrieve(&zc, attr) == 0;
1909 			    (void) zap_cursor_advance(&zc)) {
1910 				ASSERT3U(attr->za_integer_length, ==,
1911 				    sizeof (uint64_t));
1912 				ASSERT3U(attr->za_num_integers, ==, 1);
1913 
1914 				child = kmem_asprintf("%s@%s",
1915 				    name, attr->za_name);
1916 				dsl_pool_config_exit(dp, FTAG);
1917 				err = func(child, arg);
1918 				dsl_pool_config_enter(dp, FTAG);
1919 				strfree(child);
1920 				if (err != 0)
1921 					break;
1922 			}
1923 			zap_cursor_fini(&zc);
1924 		}
1925 	}
1926 
1927 	dsl_dir_rele(dd, FTAG);
1928 	kmem_free(attr, sizeof (zap_attribute_t));
1929 	dsl_pool_config_exit(dp, FTAG);
1930 
1931 	if (err != 0)
1932 		return (err);
1933 
1934 	/* Apply to self. */
1935 	return (func(name, arg));
1936 }
1937 
1938 /*
1939  * See comment above dmu_objset_find_impl().
1940  */
1941 int
1942 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1943     int flags)
1944 {
1945 	spa_t *spa;
1946 	int error;
1947 
1948 	error = spa_open(name, &spa, FTAG);
1949 	if (error != 0)
1950 		return (error);
1951 	error = dmu_objset_find_impl(spa, name, func, arg, flags);
1952 	spa_close(spa, FTAG);
1953 	return (error);
1954 }
1955 
1956 void
1957 dmu_objset_set_user(objset_t *os, void *user_ptr)
1958 {
1959 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1960 	os->os_user_ptr = user_ptr;
1961 }
1962 
1963 void *
1964 dmu_objset_get_user(objset_t *os)
1965 {
1966 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1967 	return (os->os_user_ptr);
1968 }
1969 
1970 /*
1971  * Determine name of filesystem, given name of snapshot.
1972  * buf must be at least MAXNAMELEN bytes
1973  */
1974 int
1975 dmu_fsname(const char *snapname, char *buf)
1976 {
1977 	char *atp = strchr(snapname, '@');
1978 	if (atp == NULL)
1979 		return (SET_ERROR(EINVAL));
1980 	if (atp - snapname >= MAXNAMELEN)
1981 		return (SET_ERROR(ENAMETOOLONG));
1982 	(void) strlcpy(buf, snapname, atp - snapname + 1);
1983 	return (0);
1984 }
1985