xref: /titanic_52/usr/src/uts/common/fs/zfs/dmu_objset.c (revision b49b27dcb66b2c7f4a23f7bc158e2dde5cd79030)
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 void
521 dmu_objset_disown(objset_t *os, void *tag)
522 {
523 	dsl_dataset_disown(os->os_dsl_dataset, tag);
524 }
525 
526 void
527 dmu_objset_evict_dbufs(objset_t *os)
528 {
529 	dnode_t *dn;
530 
531 	mutex_enter(&os->os_lock);
532 
533 	/* process the mdn last, since the other dnodes have holds on it */
534 	list_remove(&os->os_dnodes, DMU_META_DNODE(os));
535 	list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
536 
537 	/*
538 	 * Find the first dnode with holds.  We have to do this dance
539 	 * because dnode_add_ref() only works if you already have a
540 	 * hold.  If there are no holds then it has no dbufs so OK to
541 	 * skip.
542 	 */
543 	for (dn = list_head(&os->os_dnodes);
544 	    dn && !dnode_add_ref(dn, FTAG);
545 	    dn = list_next(&os->os_dnodes, dn))
546 		continue;
547 
548 	while (dn) {
549 		dnode_t *next_dn = dn;
550 
551 		do {
552 			next_dn = list_next(&os->os_dnodes, next_dn);
553 		} while (next_dn && !dnode_add_ref(next_dn, FTAG));
554 
555 		mutex_exit(&os->os_lock);
556 		dnode_evict_dbufs(dn);
557 		dnode_rele(dn, FTAG);
558 		mutex_enter(&os->os_lock);
559 		dn = next_dn;
560 	}
561 	mutex_exit(&os->os_lock);
562 }
563 
564 void
565 dmu_objset_evict(objset_t *os)
566 {
567 	dsl_dataset_t *ds = os->os_dsl_dataset;
568 
569 	for (int t = 0; t < TXG_SIZE; t++)
570 		ASSERT(!dmu_objset_is_dirty(os, t));
571 
572 	if (ds) {
573 		if (!dsl_dataset_is_snapshot(ds)) {
574 			VERIFY0(dsl_prop_unregister(ds,
575 			    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
576 			    checksum_changed_cb, os));
577 			VERIFY0(dsl_prop_unregister(ds,
578 			    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
579 			    compression_changed_cb, os));
580 			VERIFY0(dsl_prop_unregister(ds,
581 			    zfs_prop_to_name(ZFS_PROP_COPIES),
582 			    copies_changed_cb, os));
583 			VERIFY0(dsl_prop_unregister(ds,
584 			    zfs_prop_to_name(ZFS_PROP_DEDUP),
585 			    dedup_changed_cb, os));
586 			VERIFY0(dsl_prop_unregister(ds,
587 			    zfs_prop_to_name(ZFS_PROP_LOGBIAS),
588 			    logbias_changed_cb, os));
589 			VERIFY0(dsl_prop_unregister(ds,
590 			    zfs_prop_to_name(ZFS_PROP_SYNC),
591 			    sync_changed_cb, os));
592 		}
593 		VERIFY0(dsl_prop_unregister(ds,
594 		    zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
595 		    primary_cache_changed_cb, os));
596 		VERIFY0(dsl_prop_unregister(ds,
597 		    zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
598 		    secondary_cache_changed_cb, os));
599 	}
600 
601 	if (os->os_sa)
602 		sa_tear_down(os);
603 
604 	dmu_objset_evict_dbufs(os);
605 
606 	dnode_special_close(&os->os_meta_dnode);
607 	if (DMU_USERUSED_DNODE(os)) {
608 		dnode_special_close(&os->os_userused_dnode);
609 		dnode_special_close(&os->os_groupused_dnode);
610 	}
611 	zil_free(os->os_zil);
612 
613 	ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
614 
615 	VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
616 
617 	/*
618 	 * This is a barrier to prevent the objset from going away in
619 	 * dnode_move() until we can safely ensure that the objset is still in
620 	 * use. We consider the objset valid before the barrier and invalid
621 	 * after the barrier.
622 	 */
623 	rw_enter(&os_lock, RW_READER);
624 	rw_exit(&os_lock);
625 
626 	mutex_destroy(&os->os_lock);
627 	mutex_destroy(&os->os_obj_lock);
628 	mutex_destroy(&os->os_user_ptr_lock);
629 	kmem_free(os, sizeof (objset_t));
630 }
631 
632 timestruc_t
633 dmu_objset_snap_cmtime(objset_t *os)
634 {
635 	return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
636 }
637 
638 /* called from dsl for meta-objset */
639 objset_t *
640 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
641     dmu_objset_type_t type, dmu_tx_t *tx)
642 {
643 	objset_t *os;
644 	dnode_t *mdn;
645 
646 	ASSERT(dmu_tx_is_syncing(tx));
647 
648 	if (ds != NULL)
649 		VERIFY0(dmu_objset_from_ds(ds, &os));
650 	else
651 		VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
652 
653 	mdn = DMU_META_DNODE(os);
654 
655 	dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
656 	    DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
657 
658 	/*
659 	 * We don't want to have to increase the meta-dnode's nlevels
660 	 * later, because then we could do it in quescing context while
661 	 * we are also accessing it in open context.
662 	 *
663 	 * This precaution is not necessary for the MOS (ds == NULL),
664 	 * because the MOS is only updated in syncing context.
665 	 * This is most fortunate: the MOS is the only objset that
666 	 * needs to be synced multiple times as spa_sync() iterates
667 	 * to convergence, so minimizing its dn_nlevels matters.
668 	 */
669 	if (ds != NULL) {
670 		int levels = 1;
671 
672 		/*
673 		 * Determine the number of levels necessary for the meta-dnode
674 		 * to contain DN_MAX_OBJECT dnodes.
675 		 */
676 		while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
677 		    (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
678 		    DN_MAX_OBJECT * sizeof (dnode_phys_t))
679 			levels++;
680 
681 		mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
682 		    mdn->dn_nlevels = levels;
683 	}
684 
685 	ASSERT(type != DMU_OST_NONE);
686 	ASSERT(type != DMU_OST_ANY);
687 	ASSERT(type < DMU_OST_NUMTYPES);
688 	os->os_phys->os_type = type;
689 	if (dmu_objset_userused_enabled(os)) {
690 		os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
691 		os->os_flags = os->os_phys->os_flags;
692 	}
693 
694 	dsl_dataset_dirty(ds, tx);
695 
696 	return (os);
697 }
698 
699 typedef struct dmu_objset_create_arg {
700 	const char *doca_name;
701 	cred_t *doca_cred;
702 	void (*doca_userfunc)(objset_t *os, void *arg,
703 	    cred_t *cr, dmu_tx_t *tx);
704 	void *doca_userarg;
705 	dmu_objset_type_t doca_type;
706 	uint64_t doca_flags;
707 } dmu_objset_create_arg_t;
708 
709 /*ARGSUSED*/
710 static int
711 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
712 {
713 	dmu_objset_create_arg_t *doca = arg;
714 	dsl_pool_t *dp = dmu_tx_pool(tx);
715 	dsl_dir_t *pdd;
716 	const char *tail;
717 	int error;
718 
719 	if (strchr(doca->doca_name, '@') != NULL)
720 		return (SET_ERROR(EINVAL));
721 
722 	error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
723 	if (error != 0)
724 		return (error);
725 	if (tail == NULL) {
726 		dsl_dir_rele(pdd, FTAG);
727 		return (SET_ERROR(EEXIST));
728 	}
729 	dsl_dir_rele(pdd, FTAG);
730 
731 	return (0);
732 }
733 
734 static void
735 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
736 {
737 	dmu_objset_create_arg_t *doca = arg;
738 	dsl_pool_t *dp = dmu_tx_pool(tx);
739 	dsl_dir_t *pdd;
740 	const char *tail;
741 	dsl_dataset_t *ds;
742 	uint64_t obj;
743 	blkptr_t *bp;
744 	objset_t *os;
745 
746 	VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
747 
748 	obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
749 	    doca->doca_cred, tx);
750 
751 	VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
752 	bp = dsl_dataset_get_blkptr(ds);
753 	os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
754 	    ds, bp, doca->doca_type, tx);
755 
756 	if (doca->doca_userfunc != NULL) {
757 		doca->doca_userfunc(os, doca->doca_userarg,
758 		    doca->doca_cred, tx);
759 	}
760 
761 	spa_history_log_internal_ds(ds, "create", tx, "");
762 	dsl_dataset_rele(ds, FTAG);
763 	dsl_dir_rele(pdd, FTAG);
764 }
765 
766 int
767 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
768     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
769 {
770 	dmu_objset_create_arg_t doca;
771 
772 	doca.doca_name = name;
773 	doca.doca_cred = CRED();
774 	doca.doca_flags = flags;
775 	doca.doca_userfunc = func;
776 	doca.doca_userarg = arg;
777 	doca.doca_type = type;
778 
779 	return (dsl_sync_task(name,
780 	    dmu_objset_create_check, dmu_objset_create_sync, &doca, 5));
781 }
782 
783 typedef struct dmu_objset_clone_arg {
784 	const char *doca_clone;
785 	const char *doca_origin;
786 	cred_t *doca_cred;
787 } dmu_objset_clone_arg_t;
788 
789 /*ARGSUSED*/
790 static int
791 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
792 {
793 	dmu_objset_clone_arg_t *doca = arg;
794 	dsl_dir_t *pdd;
795 	const char *tail;
796 	int error;
797 	dsl_dataset_t *origin;
798 	dsl_pool_t *dp = dmu_tx_pool(tx);
799 
800 	if (strchr(doca->doca_clone, '@') != NULL)
801 		return (SET_ERROR(EINVAL));
802 
803 	error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
804 	if (error != 0)
805 		return (error);
806 	if (tail == NULL) {
807 		dsl_dir_rele(pdd, FTAG);
808 		return (SET_ERROR(EEXIST));
809 	}
810 	/* You can't clone across pools. */
811 	if (pdd->dd_pool != dp) {
812 		dsl_dir_rele(pdd, FTAG);
813 		return (SET_ERROR(EXDEV));
814 	}
815 	dsl_dir_rele(pdd, FTAG);
816 
817 	error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
818 	if (error != 0)
819 		return (error);
820 
821 	/* You can't clone across pools. */
822 	if (origin->ds_dir->dd_pool != dp) {
823 		dsl_dataset_rele(origin, FTAG);
824 		return (SET_ERROR(EXDEV));
825 	}
826 
827 	/* You can only clone snapshots, not the head datasets. */
828 	if (!dsl_dataset_is_snapshot(origin)) {
829 		dsl_dataset_rele(origin, FTAG);
830 		return (SET_ERROR(EINVAL));
831 	}
832 	dsl_dataset_rele(origin, FTAG);
833 
834 	return (0);
835 }
836 
837 static void
838 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
839 {
840 	dmu_objset_clone_arg_t *doca = arg;
841 	dsl_pool_t *dp = dmu_tx_pool(tx);
842 	dsl_dir_t *pdd;
843 	const char *tail;
844 	dsl_dataset_t *origin, *ds;
845 	uint64_t obj;
846 	char namebuf[MAXNAMELEN];
847 
848 	VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
849 	VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
850 
851 	obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
852 	    doca->doca_cred, tx);
853 
854 	VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
855 	dsl_dataset_name(origin, namebuf);
856 	spa_history_log_internal_ds(ds, "clone", tx,
857 	    "origin=%s (%llu)", namebuf, origin->ds_object);
858 	dsl_dataset_rele(ds, FTAG);
859 	dsl_dataset_rele(origin, FTAG);
860 	dsl_dir_rele(pdd, FTAG);
861 }
862 
863 int
864 dmu_objset_clone(const char *clone, const char *origin)
865 {
866 	dmu_objset_clone_arg_t doca;
867 
868 	doca.doca_clone = clone;
869 	doca.doca_origin = origin;
870 	doca.doca_cred = CRED();
871 
872 	return (dsl_sync_task(clone,
873 	    dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 5));
874 }
875 
876 int
877 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
878 {
879 	int err;
880 	char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
881 	nvlist_t *snaps = fnvlist_alloc();
882 
883 	fnvlist_add_boolean(snaps, longsnap);
884 	strfree(longsnap);
885 	err = dsl_dataset_snapshot(snaps, NULL, NULL);
886 	fnvlist_free(snaps);
887 	return (err);
888 }
889 
890 static void
891 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
892 {
893 	dnode_t *dn;
894 
895 	while (dn = list_head(list)) {
896 		ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
897 		ASSERT(dn->dn_dbuf->db_data_pending);
898 		/*
899 		 * Initialize dn_zio outside dnode_sync() because the
900 		 * meta-dnode needs to set it ouside dnode_sync().
901 		 */
902 		dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
903 		ASSERT(dn->dn_zio);
904 
905 		ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
906 		list_remove(list, dn);
907 
908 		if (newlist) {
909 			(void) dnode_add_ref(dn, newlist);
910 			list_insert_tail(newlist, dn);
911 		}
912 
913 		dnode_sync(dn, tx);
914 	}
915 }
916 
917 /* ARGSUSED */
918 static void
919 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
920 {
921 	blkptr_t *bp = zio->io_bp;
922 	objset_t *os = arg;
923 	dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
924 
925 	ASSERT3P(bp, ==, os->os_rootbp);
926 	ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
927 	ASSERT0(BP_GET_LEVEL(bp));
928 
929 	/*
930 	 * Update rootbp fill count: it should be the number of objects
931 	 * allocated in the object set (not counting the "special"
932 	 * objects that are stored in the objset_phys_t -- the meta
933 	 * dnode and user/group accounting objects).
934 	 */
935 	bp->blk_fill = 0;
936 	for (int i = 0; i < dnp->dn_nblkptr; i++)
937 		bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
938 }
939 
940 /* ARGSUSED */
941 static void
942 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
943 {
944 	blkptr_t *bp = zio->io_bp;
945 	blkptr_t *bp_orig = &zio->io_bp_orig;
946 	objset_t *os = arg;
947 
948 	if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
949 		ASSERT(BP_EQUAL(bp, bp_orig));
950 	} else {
951 		dsl_dataset_t *ds = os->os_dsl_dataset;
952 		dmu_tx_t *tx = os->os_synctx;
953 
954 		(void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
955 		dsl_dataset_block_born(ds, bp, tx);
956 	}
957 }
958 
959 /* called from dsl */
960 void
961 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
962 {
963 	int txgoff;
964 	zbookmark_t zb;
965 	zio_prop_t zp;
966 	zio_t *zio;
967 	list_t *list;
968 	list_t *newlist = NULL;
969 	dbuf_dirty_record_t *dr;
970 
971 	dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
972 
973 	ASSERT(dmu_tx_is_syncing(tx));
974 	/* XXX the write_done callback should really give us the tx... */
975 	os->os_synctx = tx;
976 
977 	if (os->os_dsl_dataset == NULL) {
978 		/*
979 		 * This is the MOS.  If we have upgraded,
980 		 * spa_max_replication() could change, so reset
981 		 * os_copies here.
982 		 */
983 		os->os_copies = spa_max_replication(os->os_spa);
984 	}
985 
986 	/*
987 	 * Create the root block IO
988 	 */
989 	SET_BOOKMARK(&zb, os->os_dsl_dataset ?
990 	    os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
991 	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
992 	arc_release(os->os_phys_buf, &os->os_phys_buf);
993 
994 	dmu_write_policy(os, NULL, 0, 0, &zp);
995 
996 	zio = arc_write(pio, os->os_spa, tx->tx_txg,
997 	    os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
998 	    DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
999 	    dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1000 	    ZIO_FLAG_MUSTSUCCEED, &zb);
1001 
1002 	/*
1003 	 * Sync special dnodes - the parent IO for the sync is the root block
1004 	 */
1005 	DMU_META_DNODE(os)->dn_zio = zio;
1006 	dnode_sync(DMU_META_DNODE(os), tx);
1007 
1008 	os->os_phys->os_flags = os->os_flags;
1009 
1010 	if (DMU_USERUSED_DNODE(os) &&
1011 	    DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1012 		DMU_USERUSED_DNODE(os)->dn_zio = zio;
1013 		dnode_sync(DMU_USERUSED_DNODE(os), tx);
1014 		DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1015 		dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1016 	}
1017 
1018 	txgoff = tx->tx_txg & TXG_MASK;
1019 
1020 	if (dmu_objset_userused_enabled(os)) {
1021 		newlist = &os->os_synced_dnodes;
1022 		/*
1023 		 * We must create the list here because it uses the
1024 		 * dn_dirty_link[] of this txg.
1025 		 */
1026 		list_create(newlist, sizeof (dnode_t),
1027 		    offsetof(dnode_t, dn_dirty_link[txgoff]));
1028 	}
1029 
1030 	dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1031 	dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1032 
1033 	list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1034 	while (dr = list_head(list)) {
1035 		ASSERT0(dr->dr_dbuf->db_level);
1036 		list_remove(list, dr);
1037 		if (dr->dr_zio)
1038 			zio_nowait(dr->dr_zio);
1039 	}
1040 	/*
1041 	 * Free intent log blocks up to this tx.
1042 	 */
1043 	zil_sync(os->os_zil, tx);
1044 	os->os_phys->os_zil_header = os->os_zil_header;
1045 	zio_nowait(zio);
1046 }
1047 
1048 boolean_t
1049 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1050 {
1051 	return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1052 	    !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1053 }
1054 
1055 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1056 
1057 void
1058 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1059 {
1060 	used_cbs[ost] = cb;
1061 }
1062 
1063 boolean_t
1064 dmu_objset_userused_enabled(objset_t *os)
1065 {
1066 	return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1067 	    used_cbs[os->os_phys->os_type] != NULL &&
1068 	    DMU_USERUSED_DNODE(os) != NULL);
1069 }
1070 
1071 static void
1072 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1073     uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1074 {
1075 	if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1076 		int64_t delta = DNODE_SIZE + used;
1077 		if (subtract)
1078 			delta = -delta;
1079 		VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1080 		    user, delta, tx));
1081 		VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1082 		    group, delta, tx));
1083 	}
1084 }
1085 
1086 void
1087 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1088 {
1089 	dnode_t *dn;
1090 	list_t *list = &os->os_synced_dnodes;
1091 
1092 	ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1093 
1094 	while (dn = list_head(list)) {
1095 		int flags;
1096 		ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1097 		ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1098 		    dn->dn_phys->dn_flags &
1099 		    DNODE_FLAG_USERUSED_ACCOUNTED);
1100 
1101 		/* Allocate the user/groupused objects if necessary. */
1102 		if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1103 			VERIFY(0 == zap_create_claim(os,
1104 			    DMU_USERUSED_OBJECT,
1105 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1106 			VERIFY(0 == zap_create_claim(os,
1107 			    DMU_GROUPUSED_OBJECT,
1108 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1109 		}
1110 
1111 		/*
1112 		 * We intentionally modify the zap object even if the
1113 		 * net delta is zero.  Otherwise
1114 		 * the block of the zap obj could be shared between
1115 		 * datasets but need to be different between them after
1116 		 * a bprewrite.
1117 		 */
1118 
1119 		flags = dn->dn_id_flags;
1120 		ASSERT(flags);
1121 		if (flags & DN_ID_OLD_EXIST)  {
1122 			do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1123 			    dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1124 		}
1125 		if (flags & DN_ID_NEW_EXIST) {
1126 			do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1127 			    dn->dn_phys->dn_flags,  dn->dn_newuid,
1128 			    dn->dn_newgid, B_FALSE, tx);
1129 		}
1130 
1131 		mutex_enter(&dn->dn_mtx);
1132 		dn->dn_oldused = 0;
1133 		dn->dn_oldflags = 0;
1134 		if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1135 			dn->dn_olduid = dn->dn_newuid;
1136 			dn->dn_oldgid = dn->dn_newgid;
1137 			dn->dn_id_flags |= DN_ID_OLD_EXIST;
1138 			if (dn->dn_bonuslen == 0)
1139 				dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1140 			else
1141 				dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1142 		}
1143 		dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1144 		mutex_exit(&dn->dn_mtx);
1145 
1146 		list_remove(list, dn);
1147 		dnode_rele(dn, list);
1148 	}
1149 }
1150 
1151 /*
1152  * Returns a pointer to data to find uid/gid from
1153  *
1154  * If a dirty record for transaction group that is syncing can't
1155  * be found then NULL is returned.  In the NULL case it is assumed
1156  * the uid/gid aren't changing.
1157  */
1158 static void *
1159 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1160 {
1161 	dbuf_dirty_record_t *dr, **drp;
1162 	void *data;
1163 
1164 	if (db->db_dirtycnt == 0)
1165 		return (db->db.db_data);  /* Nothing is changing */
1166 
1167 	for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1168 		if (dr->dr_txg == tx->tx_txg)
1169 			break;
1170 
1171 	if (dr == NULL) {
1172 		data = NULL;
1173 	} else {
1174 		dnode_t *dn;
1175 
1176 		DB_DNODE_ENTER(dr->dr_dbuf);
1177 		dn = DB_DNODE(dr->dr_dbuf);
1178 
1179 		if (dn->dn_bonuslen == 0 &&
1180 		    dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1181 			data = dr->dt.dl.dr_data->b_data;
1182 		else
1183 			data = dr->dt.dl.dr_data;
1184 
1185 		DB_DNODE_EXIT(dr->dr_dbuf);
1186 	}
1187 
1188 	return (data);
1189 }
1190 
1191 void
1192 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1193 {
1194 	objset_t *os = dn->dn_objset;
1195 	void *data = NULL;
1196 	dmu_buf_impl_t *db = NULL;
1197 	uint64_t *user = NULL;
1198 	uint64_t *group = NULL;
1199 	int flags = dn->dn_id_flags;
1200 	int error;
1201 	boolean_t have_spill = B_FALSE;
1202 
1203 	if (!dmu_objset_userused_enabled(dn->dn_objset))
1204 		return;
1205 
1206 	if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1207 	    DN_ID_CHKED_SPILL)))
1208 		return;
1209 
1210 	if (before && dn->dn_bonuslen != 0)
1211 		data = DN_BONUS(dn->dn_phys);
1212 	else if (!before && dn->dn_bonuslen != 0) {
1213 		if (dn->dn_bonus) {
1214 			db = dn->dn_bonus;
1215 			mutex_enter(&db->db_mtx);
1216 			data = dmu_objset_userquota_find_data(db, tx);
1217 		} else {
1218 			data = DN_BONUS(dn->dn_phys);
1219 		}
1220 	} else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1221 			int rf = 0;
1222 
1223 			if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1224 				rf |= DB_RF_HAVESTRUCT;
1225 			error = dmu_spill_hold_by_dnode(dn,
1226 			    rf | DB_RF_MUST_SUCCEED,
1227 			    FTAG, (dmu_buf_t **)&db);
1228 			ASSERT(error == 0);
1229 			mutex_enter(&db->db_mtx);
1230 			data = (before) ? db->db.db_data :
1231 			    dmu_objset_userquota_find_data(db, tx);
1232 			have_spill = B_TRUE;
1233 	} else {
1234 		mutex_enter(&dn->dn_mtx);
1235 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1236 		mutex_exit(&dn->dn_mtx);
1237 		return;
1238 	}
1239 
1240 	if (before) {
1241 		ASSERT(data);
1242 		user = &dn->dn_olduid;
1243 		group = &dn->dn_oldgid;
1244 	} else if (data) {
1245 		user = &dn->dn_newuid;
1246 		group = &dn->dn_newgid;
1247 	}
1248 
1249 	/*
1250 	 * Must always call the callback in case the object
1251 	 * type has changed and that type isn't an object type to track
1252 	 */
1253 	error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1254 	    user, group);
1255 
1256 	/*
1257 	 * Preserve existing uid/gid when the callback can't determine
1258 	 * what the new uid/gid are and the callback returned EEXIST.
1259 	 * The EEXIST error tells us to just use the existing uid/gid.
1260 	 * If we don't know what the old values are then just assign
1261 	 * them to 0, since that is a new file  being created.
1262 	 */
1263 	if (!before && data == NULL && error == EEXIST) {
1264 		if (flags & DN_ID_OLD_EXIST) {
1265 			dn->dn_newuid = dn->dn_olduid;
1266 			dn->dn_newgid = dn->dn_oldgid;
1267 		} else {
1268 			dn->dn_newuid = 0;
1269 			dn->dn_newgid = 0;
1270 		}
1271 		error = 0;
1272 	}
1273 
1274 	if (db)
1275 		mutex_exit(&db->db_mtx);
1276 
1277 	mutex_enter(&dn->dn_mtx);
1278 	if (error == 0 && before)
1279 		dn->dn_id_flags |= DN_ID_OLD_EXIST;
1280 	if (error == 0 && !before)
1281 		dn->dn_id_flags |= DN_ID_NEW_EXIST;
1282 
1283 	if (have_spill) {
1284 		dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1285 	} else {
1286 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1287 	}
1288 	mutex_exit(&dn->dn_mtx);
1289 	if (have_spill)
1290 		dmu_buf_rele((dmu_buf_t *)db, FTAG);
1291 }
1292 
1293 boolean_t
1294 dmu_objset_userspace_present(objset_t *os)
1295 {
1296 	return (os->os_phys->os_flags &
1297 	    OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1298 }
1299 
1300 int
1301 dmu_objset_userspace_upgrade(objset_t *os)
1302 {
1303 	uint64_t obj;
1304 	int err = 0;
1305 
1306 	if (dmu_objset_userspace_present(os))
1307 		return (0);
1308 	if (!dmu_objset_userused_enabled(os))
1309 		return (SET_ERROR(ENOTSUP));
1310 	if (dmu_objset_is_snapshot(os))
1311 		return (SET_ERROR(EINVAL));
1312 
1313 	/*
1314 	 * We simply need to mark every object dirty, so that it will be
1315 	 * synced out and now accounted.  If this is called
1316 	 * concurrently, or if we already did some work before crashing,
1317 	 * that's fine, since we track each object's accounted state
1318 	 * independently.
1319 	 */
1320 
1321 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1322 		dmu_tx_t *tx;
1323 		dmu_buf_t *db;
1324 		int objerr;
1325 
1326 		if (issig(JUSTLOOKING) && issig(FORREAL))
1327 			return (SET_ERROR(EINTR));
1328 
1329 		objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1330 		if (objerr != 0)
1331 			continue;
1332 		tx = dmu_tx_create(os);
1333 		dmu_tx_hold_bonus(tx, obj);
1334 		objerr = dmu_tx_assign(tx, TXG_WAIT);
1335 		if (objerr != 0) {
1336 			dmu_tx_abort(tx);
1337 			continue;
1338 		}
1339 		dmu_buf_will_dirty(db, tx);
1340 		dmu_buf_rele(db, FTAG);
1341 		dmu_tx_commit(tx);
1342 	}
1343 
1344 	os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1345 	txg_wait_synced(dmu_objset_pool(os), 0);
1346 	return (0);
1347 }
1348 
1349 void
1350 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1351     uint64_t *usedobjsp, uint64_t *availobjsp)
1352 {
1353 	dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1354 	    usedobjsp, availobjsp);
1355 }
1356 
1357 uint64_t
1358 dmu_objset_fsid_guid(objset_t *os)
1359 {
1360 	return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1361 }
1362 
1363 void
1364 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1365 {
1366 	stat->dds_type = os->os_phys->os_type;
1367 	if (os->os_dsl_dataset)
1368 		dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1369 }
1370 
1371 void
1372 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1373 {
1374 	ASSERT(os->os_dsl_dataset ||
1375 	    os->os_phys->os_type == DMU_OST_META);
1376 
1377 	if (os->os_dsl_dataset != NULL)
1378 		dsl_dataset_stats(os->os_dsl_dataset, nv);
1379 
1380 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1381 	    os->os_phys->os_type);
1382 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1383 	    dmu_objset_userspace_present(os));
1384 }
1385 
1386 int
1387 dmu_objset_is_snapshot(objset_t *os)
1388 {
1389 	if (os->os_dsl_dataset != NULL)
1390 		return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1391 	else
1392 		return (B_FALSE);
1393 }
1394 
1395 int
1396 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1397     boolean_t *conflict)
1398 {
1399 	dsl_dataset_t *ds = os->os_dsl_dataset;
1400 	uint64_t ignored;
1401 
1402 	if (ds->ds_phys->ds_snapnames_zapobj == 0)
1403 		return (SET_ERROR(ENOENT));
1404 
1405 	return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1406 	    ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1407 	    real, maxlen, conflict));
1408 }
1409 
1410 int
1411 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1412     uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1413 {
1414 	dsl_dataset_t *ds = os->os_dsl_dataset;
1415 	zap_cursor_t cursor;
1416 	zap_attribute_t attr;
1417 
1418 	ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1419 
1420 	if (ds->ds_phys->ds_snapnames_zapobj == 0)
1421 		return (SET_ERROR(ENOENT));
1422 
1423 	zap_cursor_init_serialized(&cursor,
1424 	    ds->ds_dir->dd_pool->dp_meta_objset,
1425 	    ds->ds_phys->ds_snapnames_zapobj, *offp);
1426 
1427 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1428 		zap_cursor_fini(&cursor);
1429 		return (SET_ERROR(ENOENT));
1430 	}
1431 
1432 	if (strlen(attr.za_name) + 1 > namelen) {
1433 		zap_cursor_fini(&cursor);
1434 		return (SET_ERROR(ENAMETOOLONG));
1435 	}
1436 
1437 	(void) strcpy(name, attr.za_name);
1438 	if (idp)
1439 		*idp = attr.za_first_integer;
1440 	if (case_conflict)
1441 		*case_conflict = attr.za_normalization_conflict;
1442 	zap_cursor_advance(&cursor);
1443 	*offp = zap_cursor_serialize(&cursor);
1444 	zap_cursor_fini(&cursor);
1445 
1446 	return (0);
1447 }
1448 
1449 int
1450 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1451     uint64_t *idp, uint64_t *offp)
1452 {
1453 	dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1454 	zap_cursor_t cursor;
1455 	zap_attribute_t attr;
1456 
1457 	/* there is no next dir on a snapshot! */
1458 	if (os->os_dsl_dataset->ds_object !=
1459 	    dd->dd_phys->dd_head_dataset_obj)
1460 		return (SET_ERROR(ENOENT));
1461 
1462 	zap_cursor_init_serialized(&cursor,
1463 	    dd->dd_pool->dp_meta_objset,
1464 	    dd->dd_phys->dd_child_dir_zapobj, *offp);
1465 
1466 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1467 		zap_cursor_fini(&cursor);
1468 		return (SET_ERROR(ENOENT));
1469 	}
1470 
1471 	if (strlen(attr.za_name) + 1 > namelen) {
1472 		zap_cursor_fini(&cursor);
1473 		return (SET_ERROR(ENAMETOOLONG));
1474 	}
1475 
1476 	(void) strcpy(name, attr.za_name);
1477 	if (idp)
1478 		*idp = attr.za_first_integer;
1479 	zap_cursor_advance(&cursor);
1480 	*offp = zap_cursor_serialize(&cursor);
1481 	zap_cursor_fini(&cursor);
1482 
1483 	return (0);
1484 }
1485 
1486 /*
1487  * Find objsets under and including ddobj, call func(ds) on each.
1488  */
1489 int
1490 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1491     int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1492 {
1493 	dsl_dir_t *dd;
1494 	dsl_dataset_t *ds;
1495 	zap_cursor_t zc;
1496 	zap_attribute_t *attr;
1497 	uint64_t thisobj;
1498 	int err;
1499 
1500 	ASSERT(dsl_pool_config_held(dp));
1501 
1502 	err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1503 	if (err != 0)
1504 		return (err);
1505 
1506 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1507 	if (dd->dd_myname[0] == '$') {
1508 		dsl_dir_rele(dd, FTAG);
1509 		return (0);
1510 	}
1511 
1512 	thisobj = dd->dd_phys->dd_head_dataset_obj;
1513 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1514 
1515 	/*
1516 	 * Iterate over all children.
1517 	 */
1518 	if (flags & DS_FIND_CHILDREN) {
1519 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
1520 		    dd->dd_phys->dd_child_dir_zapobj);
1521 		    zap_cursor_retrieve(&zc, attr) == 0;
1522 		    (void) zap_cursor_advance(&zc)) {
1523 			ASSERT3U(attr->za_integer_length, ==,
1524 			    sizeof (uint64_t));
1525 			ASSERT3U(attr->za_num_integers, ==, 1);
1526 
1527 			err = dmu_objset_find_dp(dp, attr->za_first_integer,
1528 			    func, arg, flags);
1529 			if (err != 0)
1530 				break;
1531 		}
1532 		zap_cursor_fini(&zc);
1533 
1534 		if (err != 0) {
1535 			dsl_dir_rele(dd, FTAG);
1536 			kmem_free(attr, sizeof (zap_attribute_t));
1537 			return (err);
1538 		}
1539 	}
1540 
1541 	/*
1542 	 * Iterate over all snapshots.
1543 	 */
1544 	if (flags & DS_FIND_SNAPSHOTS) {
1545 		dsl_dataset_t *ds;
1546 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1547 
1548 		if (err == 0) {
1549 			uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1550 			dsl_dataset_rele(ds, FTAG);
1551 
1552 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
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 = dsl_dataset_hold_obj(dp,
1560 				    attr->za_first_integer, FTAG, &ds);
1561 				if (err != 0)
1562 					break;
1563 				err = func(dp, ds, arg);
1564 				dsl_dataset_rele(ds, FTAG);
1565 				if (err != 0)
1566 					break;
1567 			}
1568 			zap_cursor_fini(&zc);
1569 		}
1570 	}
1571 
1572 	dsl_dir_rele(dd, FTAG);
1573 	kmem_free(attr, sizeof (zap_attribute_t));
1574 
1575 	if (err != 0)
1576 		return (err);
1577 
1578 	/*
1579 	 * Apply to self.
1580 	 */
1581 	err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1582 	if (err != 0)
1583 		return (err);
1584 	err = func(dp, ds, arg);
1585 	dsl_dataset_rele(ds, FTAG);
1586 	return (err);
1587 }
1588 
1589 /*
1590  * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1591  * The dp_config_rwlock must not be held when this is called, and it
1592  * will not be held when the callback is called.
1593  * Therefore this function should only be used when the pool is not changing
1594  * (e.g. in syncing context), or the callback can deal with the possible races.
1595  */
1596 static int
1597 dmu_objset_find_impl(spa_t *spa, const char *name,
1598     int func(const char *, void *), void *arg, int flags)
1599 {
1600 	dsl_dir_t *dd;
1601 	dsl_pool_t *dp = spa_get_dsl(spa);
1602 	dsl_dataset_t *ds;
1603 	zap_cursor_t zc;
1604 	zap_attribute_t *attr;
1605 	char *child;
1606 	uint64_t thisobj;
1607 	int err;
1608 
1609 	dsl_pool_config_enter(dp, FTAG);
1610 
1611 	err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1612 	if (err != 0) {
1613 		dsl_pool_config_exit(dp, FTAG);
1614 		return (err);
1615 	}
1616 
1617 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1618 	if (dd->dd_myname[0] == '$') {
1619 		dsl_dir_rele(dd, FTAG);
1620 		dsl_pool_config_exit(dp, FTAG);
1621 		return (0);
1622 	}
1623 
1624 	thisobj = dd->dd_phys->dd_head_dataset_obj;
1625 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1626 
1627 	/*
1628 	 * Iterate over all children.
1629 	 */
1630 	if (flags & DS_FIND_CHILDREN) {
1631 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
1632 		    dd->dd_phys->dd_child_dir_zapobj);
1633 		    zap_cursor_retrieve(&zc, attr) == 0;
1634 		    (void) zap_cursor_advance(&zc)) {
1635 			ASSERT3U(attr->za_integer_length, ==,
1636 			    sizeof (uint64_t));
1637 			ASSERT3U(attr->za_num_integers, ==, 1);
1638 
1639 			child = kmem_asprintf("%s/%s", name, attr->za_name);
1640 			dsl_pool_config_exit(dp, FTAG);
1641 			err = dmu_objset_find_impl(spa, child,
1642 			    func, arg, flags);
1643 			dsl_pool_config_enter(dp, FTAG);
1644 			strfree(child);
1645 			if (err != 0)
1646 				break;
1647 		}
1648 		zap_cursor_fini(&zc);
1649 
1650 		if (err != 0) {
1651 			dsl_dir_rele(dd, FTAG);
1652 			dsl_pool_config_exit(dp, FTAG);
1653 			kmem_free(attr, sizeof (zap_attribute_t));
1654 			return (err);
1655 		}
1656 	}
1657 
1658 	/*
1659 	 * Iterate over all snapshots.
1660 	 */
1661 	if (flags & DS_FIND_SNAPSHOTS) {
1662 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1663 
1664 		if (err == 0) {
1665 			uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1666 			dsl_dataset_rele(ds, FTAG);
1667 
1668 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1669 			    zap_cursor_retrieve(&zc, attr) == 0;
1670 			    (void) zap_cursor_advance(&zc)) {
1671 				ASSERT3U(attr->za_integer_length, ==,
1672 				    sizeof (uint64_t));
1673 				ASSERT3U(attr->za_num_integers, ==, 1);
1674 
1675 				child = kmem_asprintf("%s@%s",
1676 				    name, attr->za_name);
1677 				dsl_pool_config_exit(dp, FTAG);
1678 				err = func(child, arg);
1679 				dsl_pool_config_enter(dp, FTAG);
1680 				strfree(child);
1681 				if (err != 0)
1682 					break;
1683 			}
1684 			zap_cursor_fini(&zc);
1685 		}
1686 	}
1687 
1688 	dsl_dir_rele(dd, FTAG);
1689 	kmem_free(attr, sizeof (zap_attribute_t));
1690 	dsl_pool_config_exit(dp, FTAG);
1691 
1692 	if (err != 0)
1693 		return (err);
1694 
1695 	/* Apply to self. */
1696 	return (func(name, arg));
1697 }
1698 
1699 /*
1700  * See comment above dmu_objset_find_impl().
1701  */
1702 int
1703 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1704     int flags)
1705 {
1706 	spa_t *spa;
1707 	int error;
1708 
1709 	error = spa_open(name, &spa, FTAG);
1710 	if (error != 0)
1711 		return (error);
1712 	error = dmu_objset_find_impl(spa, name, func, arg, flags);
1713 	spa_close(spa, FTAG);
1714 	return (error);
1715 }
1716 
1717 void
1718 dmu_objset_set_user(objset_t *os, void *user_ptr)
1719 {
1720 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1721 	os->os_user_ptr = user_ptr;
1722 }
1723 
1724 void *
1725 dmu_objset_get_user(objset_t *os)
1726 {
1727 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1728 	return (os->os_user_ptr);
1729 }
1730 
1731 /*
1732  * Determine name of filesystem, given name of snapshot.
1733  * buf must be at least MAXNAMELEN bytes
1734  */
1735 int
1736 dmu_fsname(const char *snapname, char *buf)
1737 {
1738 	char *atp = strchr(snapname, '@');
1739 	if (atp == NULL)
1740 		return (SET_ERROR(EINVAL));
1741 	if (atp - snapname >= MAXNAMELEN)
1742 		return (SET_ERROR(ENAMETOOLONG));
1743 	(void) strlcpy(buf, snapname, atp - snapname + 1);
1744 	return (0);
1745 }
1746