xref: /titanic_44/usr/src/uts/common/fs/zfs/dmu_objset.c (revision 4c6ae46166796718de10c1839ffe0569daa48250)
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 by Delphix. All rights reserved.
24  */
25 
26 /* Portions Copyright 2010 Robert Milkowski */
27 
28 #include <sys/cred.h>
29 #include <sys/zfs_context.h>
30 #include <sys/dmu_objset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_dataset.h>
33 #include <sys/dsl_prop.h>
34 #include <sys/dsl_pool.h>
35 #include <sys/dsl_synctask.h>
36 #include <sys/dsl_deleg.h>
37 #include <sys/dnode.h>
38 #include <sys/dbuf.h>
39 #include <sys/zvol.h>
40 #include <sys/dmu_tx.h>
41 #include <sys/zap.h>
42 #include <sys/zil.h>
43 #include <sys/dmu_impl.h>
44 #include <sys/zfs_ioctl.h>
45 #include <sys/sa.h>
46 #include <sys/zfs_onexit.h>
47 
48 /*
49  * Needed to close a window in dnode_move() that allows the objset to be freed
50  * before it can be safely accessed.
51  */
52 krwlock_t os_lock;
53 
54 void
55 dmu_objset_init(void)
56 {
57 	rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
58 }
59 
60 void
61 dmu_objset_fini(void)
62 {
63 	rw_destroy(&os_lock);
64 }
65 
66 spa_t *
67 dmu_objset_spa(objset_t *os)
68 {
69 	return (os->os_spa);
70 }
71 
72 zilog_t *
73 dmu_objset_zil(objset_t *os)
74 {
75 	return (os->os_zil);
76 }
77 
78 dsl_pool_t *
79 dmu_objset_pool(objset_t *os)
80 {
81 	dsl_dataset_t *ds;
82 
83 	if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
84 		return (ds->ds_dir->dd_pool);
85 	else
86 		return (spa_get_dsl(os->os_spa));
87 }
88 
89 dsl_dataset_t *
90 dmu_objset_ds(objset_t *os)
91 {
92 	return (os->os_dsl_dataset);
93 }
94 
95 dmu_objset_type_t
96 dmu_objset_type(objset_t *os)
97 {
98 	return (os->os_phys->os_type);
99 }
100 
101 void
102 dmu_objset_name(objset_t *os, char *buf)
103 {
104 	dsl_dataset_name(os->os_dsl_dataset, buf);
105 }
106 
107 uint64_t
108 dmu_objset_id(objset_t *os)
109 {
110 	dsl_dataset_t *ds = os->os_dsl_dataset;
111 
112 	return (ds ? ds->ds_object : 0);
113 }
114 
115 uint64_t
116 dmu_objset_syncprop(objset_t *os)
117 {
118 	return (os->os_sync);
119 }
120 
121 uint64_t
122 dmu_objset_logbias(objset_t *os)
123 {
124 	return (os->os_logbias);
125 }
126 
127 static void
128 checksum_changed_cb(void *arg, uint64_t newval)
129 {
130 	objset_t *os = arg;
131 
132 	/*
133 	 * Inheritance should have been done by now.
134 	 */
135 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
136 
137 	os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
138 }
139 
140 static void
141 compression_changed_cb(void *arg, uint64_t newval)
142 {
143 	objset_t *os = arg;
144 
145 	/*
146 	 * Inheritance and range checking should have been done by now.
147 	 */
148 	ASSERT(newval != ZIO_COMPRESS_INHERIT);
149 
150 	os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
151 }
152 
153 static void
154 copies_changed_cb(void *arg, uint64_t newval)
155 {
156 	objset_t *os = arg;
157 
158 	/*
159 	 * Inheritance and range checking should have been done by now.
160 	 */
161 	ASSERT(newval > 0);
162 	ASSERT(newval <= spa_max_replication(os->os_spa));
163 
164 	os->os_copies = newval;
165 }
166 
167 static void
168 dedup_changed_cb(void *arg, uint64_t newval)
169 {
170 	objset_t *os = arg;
171 	spa_t *spa = os->os_spa;
172 	enum zio_checksum checksum;
173 
174 	/*
175 	 * Inheritance should have been done by now.
176 	 */
177 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
178 
179 	checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
180 
181 	os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
182 	os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
183 }
184 
185 static void
186 primary_cache_changed_cb(void *arg, uint64_t newval)
187 {
188 	objset_t *os = arg;
189 
190 	/*
191 	 * Inheritance and range checking should have been done by now.
192 	 */
193 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
194 	    newval == ZFS_CACHE_METADATA);
195 
196 	os->os_primary_cache = newval;
197 }
198 
199 static void
200 secondary_cache_changed_cb(void *arg, uint64_t newval)
201 {
202 	objset_t *os = arg;
203 
204 	/*
205 	 * Inheritance and range checking should have been done by now.
206 	 */
207 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
208 	    newval == ZFS_CACHE_METADATA);
209 
210 	os->os_secondary_cache = newval;
211 }
212 
213 static void
214 sync_changed_cb(void *arg, uint64_t newval)
215 {
216 	objset_t *os = arg;
217 
218 	/*
219 	 * Inheritance and range checking should have been done by now.
220 	 */
221 	ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
222 	    newval == ZFS_SYNC_DISABLED);
223 
224 	os->os_sync = newval;
225 	if (os->os_zil)
226 		zil_set_sync(os->os_zil, newval);
227 }
228 
229 static void
230 logbias_changed_cb(void *arg, uint64_t newval)
231 {
232 	objset_t *os = arg;
233 
234 	ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
235 	    newval == ZFS_LOGBIAS_THROUGHPUT);
236 	os->os_logbias = newval;
237 	if (os->os_zil)
238 		zil_set_logbias(os->os_zil, newval);
239 }
240 
241 void
242 dmu_objset_byteswap(void *buf, size_t size)
243 {
244 	objset_phys_t *osp = buf;
245 
246 	ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
247 	dnode_byteswap(&osp->os_meta_dnode);
248 	byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
249 	osp->os_type = BSWAP_64(osp->os_type);
250 	osp->os_flags = BSWAP_64(osp->os_flags);
251 	if (size == sizeof (objset_phys_t)) {
252 		dnode_byteswap(&osp->os_userused_dnode);
253 		dnode_byteswap(&osp->os_groupused_dnode);
254 	}
255 }
256 
257 int
258 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
259     objset_t **osp)
260 {
261 	objset_t *os;
262 	int i, err;
263 
264 	ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
265 
266 	os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
267 	os->os_dsl_dataset = ds;
268 	os->os_spa = spa;
269 	os->os_rootbp = bp;
270 	if (!BP_IS_HOLE(os->os_rootbp)) {
271 		uint32_t aflags = ARC_WAIT;
272 		zbookmark_t zb;
273 		SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
274 		    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
275 
276 		if (DMU_OS_IS_L2CACHEABLE(os))
277 			aflags |= ARC_L2CACHE;
278 
279 		dprintf_bp(os->os_rootbp, "reading %s", "");
280 		/*
281 		 * XXX when bprewrite scrub can change the bp,
282 		 * and this is called from dmu_objset_open_ds_os, the bp
283 		 * could change, and we'll need a lock.
284 		 */
285 		err = dsl_read_nolock(NULL, spa, os->os_rootbp,
286 		    arc_getbuf_func, &os->os_phys_buf,
287 		    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
288 		if (err) {
289 			kmem_free(os, sizeof (objset_t));
290 			/* convert checksum errors into IO errors */
291 			if (err == ECKSUM)
292 				err = EIO;
293 			return (err);
294 		}
295 
296 		/* Increase the blocksize if we are permitted. */
297 		if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
298 		    arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
299 			arc_buf_t *buf = arc_buf_alloc(spa,
300 			    sizeof (objset_phys_t), &os->os_phys_buf,
301 			    ARC_BUFC_METADATA);
302 			bzero(buf->b_data, sizeof (objset_phys_t));
303 			bcopy(os->os_phys_buf->b_data, buf->b_data,
304 			    arc_buf_size(os->os_phys_buf));
305 			(void) arc_buf_remove_ref(os->os_phys_buf,
306 			    &os->os_phys_buf);
307 			os->os_phys_buf = buf;
308 		}
309 
310 		os->os_phys = os->os_phys_buf->b_data;
311 		os->os_flags = os->os_phys->os_flags;
312 	} else {
313 		int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
314 		    sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
315 		os->os_phys_buf = arc_buf_alloc(spa, size,
316 		    &os->os_phys_buf, ARC_BUFC_METADATA);
317 		os->os_phys = os->os_phys_buf->b_data;
318 		bzero(os->os_phys, size);
319 	}
320 
321 	/*
322 	 * Note: the changed_cb will be called once before the register
323 	 * func returns, thus changing the checksum/compression from the
324 	 * default (fletcher2/off).  Snapshots don't need to know about
325 	 * checksum/compression/copies.
326 	 */
327 	if (ds) {
328 		err = dsl_prop_register(ds, "primarycache",
329 		    primary_cache_changed_cb, os);
330 		if (err == 0)
331 			err = dsl_prop_register(ds, "secondarycache",
332 			    secondary_cache_changed_cb, os);
333 		if (!dsl_dataset_is_snapshot(ds)) {
334 			if (err == 0)
335 				err = dsl_prop_register(ds, "checksum",
336 				    checksum_changed_cb, os);
337 			if (err == 0)
338 				err = dsl_prop_register(ds, "compression",
339 				    compression_changed_cb, os);
340 			if (err == 0)
341 				err = dsl_prop_register(ds, "copies",
342 				    copies_changed_cb, os);
343 			if (err == 0)
344 				err = dsl_prop_register(ds, "dedup",
345 				    dedup_changed_cb, os);
346 			if (err == 0)
347 				err = dsl_prop_register(ds, "logbias",
348 				    logbias_changed_cb, os);
349 			if (err == 0)
350 				err = dsl_prop_register(ds, "sync",
351 				    sync_changed_cb, os);
352 		}
353 		if (err) {
354 			VERIFY(arc_buf_remove_ref(os->os_phys_buf,
355 			    &os->os_phys_buf) == 1);
356 			kmem_free(os, sizeof (objset_t));
357 			return (err);
358 		}
359 	} else if (ds == NULL) {
360 		/* It's the meta-objset. */
361 		os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
362 		os->os_compress = ZIO_COMPRESS_LZJB;
363 		os->os_copies = spa_max_replication(spa);
364 		os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
365 		os->os_dedup_verify = 0;
366 		os->os_logbias = 0;
367 		os->os_sync = 0;
368 		os->os_primary_cache = ZFS_CACHE_ALL;
369 		os->os_secondary_cache = ZFS_CACHE_ALL;
370 	}
371 
372 	if (ds == NULL || !dsl_dataset_is_snapshot(ds))
373 		os->os_zil_header = os->os_phys->os_zil_header;
374 	os->os_zil = zil_alloc(os, &os->os_zil_header);
375 
376 	for (i = 0; i < TXG_SIZE; i++) {
377 		list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
378 		    offsetof(dnode_t, dn_dirty_link[i]));
379 		list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
380 		    offsetof(dnode_t, dn_dirty_link[i]));
381 	}
382 	list_create(&os->os_dnodes, sizeof (dnode_t),
383 	    offsetof(dnode_t, dn_link));
384 	list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
385 	    offsetof(dmu_buf_impl_t, db_link));
386 
387 	mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
388 	mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
389 	mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
390 
391 	DMU_META_DNODE(os) = dnode_special_open(os,
392 	    &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
393 	    &os->os_meta_dnode);
394 	if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
395 		DMU_USERUSED_DNODE(os) = dnode_special_open(os,
396 		    &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
397 		    &os->os_userused_dnode);
398 		DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
399 		    &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
400 		    &os->os_groupused_dnode);
401 	}
402 
403 	/*
404 	 * We should be the only thread trying to do this because we
405 	 * have ds_opening_lock
406 	 */
407 	if (ds) {
408 		mutex_enter(&ds->ds_lock);
409 		ASSERT(ds->ds_objset == NULL);
410 		ds->ds_objset = os;
411 		mutex_exit(&ds->ds_lock);
412 	}
413 
414 	*osp = os;
415 	return (0);
416 }
417 
418 int
419 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
420 {
421 	int err = 0;
422 
423 	mutex_enter(&ds->ds_opening_lock);
424 	*osp = ds->ds_objset;
425 	if (*osp == NULL) {
426 		err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
427 		    ds, dsl_dataset_get_blkptr(ds), osp);
428 	}
429 	mutex_exit(&ds->ds_opening_lock);
430 	return (err);
431 }
432 
433 /* called from zpl */
434 int
435 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
436 {
437 	dsl_dataset_t *ds;
438 	int err;
439 
440 	err = dsl_dataset_hold(name, tag, &ds);
441 	if (err)
442 		return (err);
443 
444 	err = dmu_objset_from_ds(ds, osp);
445 	if (err)
446 		dsl_dataset_rele(ds, tag);
447 
448 	return (err);
449 }
450 
451 /* called from zpl */
452 int
453 dmu_objset_own(const char *name, dmu_objset_type_t type,
454     boolean_t readonly, void *tag, objset_t **osp)
455 {
456 	dsl_dataset_t *ds;
457 	int err;
458 
459 	err = dsl_dataset_own(name, B_FALSE, tag, &ds);
460 	if (err)
461 		return (err);
462 
463 	err = dmu_objset_from_ds(ds, osp);
464 	if (err) {
465 		dsl_dataset_disown(ds, tag);
466 	} else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
467 		dmu_objset_disown(*osp, tag);
468 		return (EINVAL);
469 	} else if (!readonly && dsl_dataset_is_snapshot(ds)) {
470 		dmu_objset_disown(*osp, tag);
471 		return (EROFS);
472 	}
473 	return (err);
474 }
475 
476 void
477 dmu_objset_rele(objset_t *os, void *tag)
478 {
479 	dsl_dataset_rele(os->os_dsl_dataset, tag);
480 }
481 
482 void
483 dmu_objset_disown(objset_t *os, void *tag)
484 {
485 	dsl_dataset_disown(os->os_dsl_dataset, tag);
486 }
487 
488 int
489 dmu_objset_evict_dbufs(objset_t *os)
490 {
491 	dnode_t *dn;
492 
493 	mutex_enter(&os->os_lock);
494 
495 	/* process the mdn last, since the other dnodes have holds on it */
496 	list_remove(&os->os_dnodes, DMU_META_DNODE(os));
497 	list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
498 
499 	/*
500 	 * Find the first dnode with holds.  We have to do this dance
501 	 * because dnode_add_ref() only works if you already have a
502 	 * hold.  If there are no holds then it has no dbufs so OK to
503 	 * skip.
504 	 */
505 	for (dn = list_head(&os->os_dnodes);
506 	    dn && !dnode_add_ref(dn, FTAG);
507 	    dn = list_next(&os->os_dnodes, dn))
508 		continue;
509 
510 	while (dn) {
511 		dnode_t *next_dn = dn;
512 
513 		do {
514 			next_dn = list_next(&os->os_dnodes, next_dn);
515 		} while (next_dn && !dnode_add_ref(next_dn, FTAG));
516 
517 		mutex_exit(&os->os_lock);
518 		dnode_evict_dbufs(dn);
519 		dnode_rele(dn, FTAG);
520 		mutex_enter(&os->os_lock);
521 		dn = next_dn;
522 	}
523 	dn = list_head(&os->os_dnodes);
524 	mutex_exit(&os->os_lock);
525 	return (dn != DMU_META_DNODE(os));
526 }
527 
528 void
529 dmu_objset_evict(objset_t *os)
530 {
531 	dsl_dataset_t *ds = os->os_dsl_dataset;
532 
533 	for (int t = 0; t < TXG_SIZE; t++)
534 		ASSERT(!dmu_objset_is_dirty(os, t));
535 
536 	if (ds) {
537 		if (!dsl_dataset_is_snapshot(ds)) {
538 			VERIFY(0 == dsl_prop_unregister(ds, "checksum",
539 			    checksum_changed_cb, os));
540 			VERIFY(0 == dsl_prop_unregister(ds, "compression",
541 			    compression_changed_cb, os));
542 			VERIFY(0 == dsl_prop_unregister(ds, "copies",
543 			    copies_changed_cb, os));
544 			VERIFY(0 == dsl_prop_unregister(ds, "dedup",
545 			    dedup_changed_cb, os));
546 			VERIFY(0 == dsl_prop_unregister(ds, "logbias",
547 			    logbias_changed_cb, os));
548 			VERIFY(0 == dsl_prop_unregister(ds, "sync",
549 			    sync_changed_cb, os));
550 		}
551 		VERIFY(0 == dsl_prop_unregister(ds, "primarycache",
552 		    primary_cache_changed_cb, os));
553 		VERIFY(0 == dsl_prop_unregister(ds, "secondarycache",
554 		    secondary_cache_changed_cb, os));
555 	}
556 
557 	if (os->os_sa)
558 		sa_tear_down(os);
559 
560 	/*
561 	 * We should need only a single pass over the dnode list, since
562 	 * nothing can be added to the list at this point.
563 	 */
564 	(void) dmu_objset_evict_dbufs(os);
565 
566 	dnode_special_close(&os->os_meta_dnode);
567 	if (DMU_USERUSED_DNODE(os)) {
568 		dnode_special_close(&os->os_userused_dnode);
569 		dnode_special_close(&os->os_groupused_dnode);
570 	}
571 	zil_free(os->os_zil);
572 
573 	ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
574 
575 	VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf) == 1);
576 
577 	/*
578 	 * This is a barrier to prevent the objset from going away in
579 	 * dnode_move() until we can safely ensure that the objset is still in
580 	 * use. We consider the objset valid before the barrier and invalid
581 	 * after the barrier.
582 	 */
583 	rw_enter(&os_lock, RW_READER);
584 	rw_exit(&os_lock);
585 
586 	mutex_destroy(&os->os_lock);
587 	mutex_destroy(&os->os_obj_lock);
588 	mutex_destroy(&os->os_user_ptr_lock);
589 	kmem_free(os, sizeof (objset_t));
590 }
591 
592 timestruc_t
593 dmu_objset_snap_cmtime(objset_t *os)
594 {
595 	return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
596 }
597 
598 /* called from dsl for meta-objset */
599 objset_t *
600 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
601     dmu_objset_type_t type, dmu_tx_t *tx)
602 {
603 	objset_t *os;
604 	dnode_t *mdn;
605 
606 	ASSERT(dmu_tx_is_syncing(tx));
607 	if (ds != NULL)
608 		VERIFY(0 == dmu_objset_from_ds(ds, &os));
609 	else
610 		VERIFY(0 == dmu_objset_open_impl(spa, NULL, bp, &os));
611 
612 	mdn = DMU_META_DNODE(os);
613 
614 	dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
615 	    DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
616 
617 	/*
618 	 * We don't want to have to increase the meta-dnode's nlevels
619 	 * later, because then we could do it in quescing context while
620 	 * we are also accessing it in open context.
621 	 *
622 	 * This precaution is not necessary for the MOS (ds == NULL),
623 	 * because the MOS is only updated in syncing context.
624 	 * This is most fortunate: the MOS is the only objset that
625 	 * needs to be synced multiple times as spa_sync() iterates
626 	 * to convergence, so minimizing its dn_nlevels matters.
627 	 */
628 	if (ds != NULL) {
629 		int levels = 1;
630 
631 		/*
632 		 * Determine the number of levels necessary for the meta-dnode
633 		 * to contain DN_MAX_OBJECT dnodes.
634 		 */
635 		while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
636 		    (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
637 		    DN_MAX_OBJECT * sizeof (dnode_phys_t))
638 			levels++;
639 
640 		mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
641 		    mdn->dn_nlevels = levels;
642 	}
643 
644 	ASSERT(type != DMU_OST_NONE);
645 	ASSERT(type != DMU_OST_ANY);
646 	ASSERT(type < DMU_OST_NUMTYPES);
647 	os->os_phys->os_type = type;
648 	if (dmu_objset_userused_enabled(os)) {
649 		os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
650 		os->os_flags = os->os_phys->os_flags;
651 	}
652 
653 	dsl_dataset_dirty(ds, tx);
654 
655 	return (os);
656 }
657 
658 struct oscarg {
659 	void (*userfunc)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx);
660 	void *userarg;
661 	dsl_dataset_t *clone_origin;
662 	const char *lastname;
663 	dmu_objset_type_t type;
664 	uint64_t flags;
665 	cred_t *cr;
666 };
667 
668 /*ARGSUSED*/
669 static int
670 dmu_objset_create_check(void *arg1, void *arg2, dmu_tx_t *tx)
671 {
672 	dsl_dir_t *dd = arg1;
673 	struct oscarg *oa = arg2;
674 	objset_t *mos = dd->dd_pool->dp_meta_objset;
675 	int err;
676 	uint64_t ddobj;
677 
678 	err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
679 	    oa->lastname, sizeof (uint64_t), 1, &ddobj);
680 	if (err != ENOENT)
681 		return (err ? err : EEXIST);
682 
683 	if (oa->clone_origin != NULL) {
684 		/* You can't clone across pools. */
685 		if (oa->clone_origin->ds_dir->dd_pool != dd->dd_pool)
686 			return (EXDEV);
687 
688 		/* You can only clone snapshots, not the head datasets. */
689 		if (!dsl_dataset_is_snapshot(oa->clone_origin))
690 			return (EINVAL);
691 	}
692 
693 	return (0);
694 }
695 
696 static void
697 dmu_objset_create_sync(void *arg1, void *arg2, dmu_tx_t *tx)
698 {
699 	dsl_dir_t *dd = arg1;
700 	spa_t *spa = dd->dd_pool->dp_spa;
701 	struct oscarg *oa = arg2;
702 	uint64_t obj;
703 	dsl_dataset_t *ds;
704 	blkptr_t *bp;
705 
706 	ASSERT(dmu_tx_is_syncing(tx));
707 
708 	obj = dsl_dataset_create_sync(dd, oa->lastname,
709 	    oa->clone_origin, oa->flags, oa->cr, tx);
710 
711 	VERIFY3U(0, ==, dsl_dataset_hold_obj(dd->dd_pool, obj, FTAG, &ds));
712 	bp = dsl_dataset_get_blkptr(ds);
713 	if (BP_IS_HOLE(bp)) {
714 		objset_t *os =
715 		    dmu_objset_create_impl(spa, ds, bp, oa->type, tx);
716 
717 		if (oa->userfunc)
718 			oa->userfunc(os, oa->userarg, oa->cr, tx);
719 	}
720 
721 	if (oa->clone_origin == NULL) {
722 		spa_history_log_internal_ds(ds, "create", tx, "");
723 	} else {
724 		char namebuf[MAXNAMELEN];
725 		dsl_dataset_name(oa->clone_origin, namebuf);
726 		spa_history_log_internal_ds(ds, "clone", tx,
727 		    "origin=%s (%llu)", namebuf, oa->clone_origin->ds_object);
728 	}
729 	dsl_dataset_rele(ds, FTAG);
730 }
731 
732 int
733 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
734     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
735 {
736 	dsl_dir_t *pdd;
737 	const char *tail;
738 	int err = 0;
739 	struct oscarg oa = { 0 };
740 
741 	ASSERT(strchr(name, '@') == NULL);
742 	err = dsl_dir_open(name, FTAG, &pdd, &tail);
743 	if (err)
744 		return (err);
745 	if (tail == NULL) {
746 		dsl_dir_close(pdd, FTAG);
747 		return (EEXIST);
748 	}
749 
750 	oa.userfunc = func;
751 	oa.userarg = arg;
752 	oa.lastname = tail;
753 	oa.type = type;
754 	oa.flags = flags;
755 	oa.cr = CRED();
756 
757 	err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
758 	    dmu_objset_create_sync, pdd, &oa, 5);
759 	dsl_dir_close(pdd, FTAG);
760 	return (err);
761 }
762 
763 int
764 dmu_objset_clone(const char *name, dsl_dataset_t *clone_origin, uint64_t flags)
765 {
766 	dsl_dir_t *pdd;
767 	const char *tail;
768 	int err = 0;
769 	struct oscarg oa = { 0 };
770 
771 	ASSERT(strchr(name, '@') == NULL);
772 	err = dsl_dir_open(name, FTAG, &pdd, &tail);
773 	if (err)
774 		return (err);
775 	if (tail == NULL) {
776 		dsl_dir_close(pdd, FTAG);
777 		return (EEXIST);
778 	}
779 
780 	oa.lastname = tail;
781 	oa.clone_origin = clone_origin;
782 	oa.flags = flags;
783 	oa.cr = CRED();
784 
785 	err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
786 	    dmu_objset_create_sync, pdd, &oa, 5);
787 	dsl_dir_close(pdd, FTAG);
788 	return (err);
789 }
790 
791 int
792 dmu_objset_destroy(const char *name, boolean_t defer)
793 {
794 	dsl_dataset_t *ds;
795 	int error;
796 
797 	error = dsl_dataset_own(name, B_TRUE, FTAG, &ds);
798 	if (error == 0) {
799 		error = dsl_dataset_destroy(ds, FTAG, defer);
800 		/* dsl_dataset_destroy() closes the ds. */
801 	}
802 
803 	return (error);
804 }
805 
806 typedef struct snapallarg {
807 	dsl_sync_task_group_t *saa_dstg;
808 	boolean_t saa_needsuspend;
809 	nvlist_t *saa_props;
810 
811 	/* the following are used only if 'temporary' is set: */
812 	boolean_t saa_temporary;
813 	const char *saa_htag;
814 	struct dsl_ds_holdarg *saa_ha;
815 	dsl_dataset_t *saa_newds;
816 } snapallarg_t;
817 
818 typedef struct snaponearg {
819 	const char *soa_longname; /* long snap name */
820 	const char *soa_snapname; /* short snap name */
821 	snapallarg_t *soa_saa;
822 } snaponearg_t;
823 
824 static int
825 snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
826 {
827 	objset_t *os = arg1;
828 	snaponearg_t *soa = arg2;
829 	snapallarg_t *saa = soa->soa_saa;
830 	int error;
831 
832 	/* The props have already been checked by zfs_check_userprops(). */
833 
834 	error = dsl_dataset_snapshot_check(os->os_dsl_dataset,
835 	    soa->soa_snapname, tx);
836 	if (error)
837 		return (error);
838 
839 	if (saa->saa_temporary) {
840 		/*
841 		 * Ideally we would just call
842 		 * dsl_dataset_user_hold_check() and
843 		 * dsl_dataset_destroy_check() here.  However the
844 		 * dataset we want to hold and destroy is the snapshot
845 		 * that we just confirmed we can create, but it won't
846 		 * exist until after these checks are run.  Do any
847 		 * checks we can here and if more checks are added to
848 		 * those routines in the future, similar checks may be
849 		 * necessary here.
850 		 */
851 		if (spa_version(os->os_spa) < SPA_VERSION_USERREFS)
852 			return (ENOTSUP);
853 		/*
854 		 * Not checking number of tags because the tag will be
855 		 * unique, as it will be the only tag.
856 		 */
857 		if (strlen(saa->saa_htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
858 			return (E2BIG);
859 
860 		saa->saa_ha = kmem_alloc(sizeof (struct dsl_ds_holdarg),
861 		    KM_SLEEP);
862 		saa->saa_ha->temphold = B_TRUE;
863 		saa->saa_ha->htag = saa->saa_htag;
864 	}
865 	return (error);
866 }
867 
868 static void
869 snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
870 {
871 	objset_t *os = arg1;
872 	dsl_dataset_t *ds = os->os_dsl_dataset;
873 	snaponearg_t *soa = arg2;
874 	snapallarg_t *saa = soa->soa_saa;
875 
876 	dsl_dataset_snapshot_sync(ds, soa->soa_snapname, tx);
877 
878 	if (saa->saa_props != NULL) {
879 		dsl_props_arg_t pa;
880 		pa.pa_props = saa->saa_props;
881 		pa.pa_source = ZPROP_SRC_LOCAL;
882 		dsl_props_set_sync(ds->ds_prev, &pa, tx);
883 	}
884 
885 	if (saa->saa_temporary) {
886 		struct dsl_ds_destroyarg da;
887 
888 		dsl_dataset_user_hold_sync(ds->ds_prev, saa->saa_ha, tx);
889 		kmem_free(saa->saa_ha, sizeof (struct dsl_ds_holdarg));
890 		saa->saa_ha = NULL;
891 		saa->saa_newds = ds->ds_prev;
892 
893 		da.ds = ds->ds_prev;
894 		da.defer = B_TRUE;
895 		dsl_dataset_destroy_sync(&da, FTAG, tx);
896 	}
897 }
898 
899 static int
900 snapshot_one_impl(const char *snapname, void *arg)
901 {
902 	char fsname[MAXPATHLEN];
903 	snapallarg_t *saa = arg;
904 	snaponearg_t *soa;
905 	objset_t *os;
906 	int err;
907 
908 	(void) strlcpy(fsname, snapname, sizeof (fsname));
909 	strchr(fsname, '@')[0] = '\0';
910 
911 	err = dmu_objset_hold(fsname, saa, &os);
912 	if (err != 0)
913 		return (err);
914 
915 	/*
916 	 * If the objset is in an inconsistent state (eg, in the process
917 	 * of being destroyed), don't snapshot it.
918 	 */
919 	if (os->os_dsl_dataset->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) {
920 		dmu_objset_rele(os, saa);
921 		return (EBUSY);
922 	}
923 
924 	if (saa->saa_needsuspend) {
925 		err = zil_suspend(dmu_objset_zil(os));
926 		if (err) {
927 			dmu_objset_rele(os, saa);
928 			return (err);
929 		}
930 	}
931 
932 	soa = kmem_zalloc(sizeof (*soa), KM_SLEEP);
933 	soa->soa_saa = saa;
934 	soa->soa_longname = snapname;
935 	soa->soa_snapname = strchr(snapname, '@') + 1;
936 
937 	dsl_sync_task_create(saa->saa_dstg, snapshot_check, snapshot_sync,
938 	    os, soa, 3);
939 
940 	return (0);
941 }
942 
943 /*
944  * The snapshots must all be in the same pool.
945  */
946 int
947 dmu_objset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
948 {
949 	dsl_sync_task_t *dst;
950 	snapallarg_t saa = { 0 };
951 	spa_t *spa;
952 	int rv = 0;
953 	int err;
954 	nvpair_t *pair;
955 
956 	pair = nvlist_next_nvpair(snaps, NULL);
957 	if (pair == NULL)
958 		return (0);
959 
960 	err = spa_open(nvpair_name(pair), &spa, FTAG);
961 	if (err)
962 		return (err);
963 	saa.saa_dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
964 	saa.saa_props = props;
965 	saa.saa_needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
966 
967 	for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
968 	    pair = nvlist_next_nvpair(snaps, pair)) {
969 		err = snapshot_one_impl(nvpair_name(pair), &saa);
970 		if (err != 0) {
971 			if (errors != NULL) {
972 				fnvlist_add_int32(errors,
973 				    nvpair_name(pair), err);
974 			}
975 			rv = err;
976 		}
977 	}
978 
979 	/*
980 	 * If any call to snapshot_one_impl() failed, don't execute the
981 	 * sync task.  The error handling code below will clean up the
982 	 * snaponearg_t from any successful calls to
983 	 * snapshot_one_impl().
984 	 */
985 	if (rv == 0)
986 		err = dsl_sync_task_group_wait(saa.saa_dstg);
987 	if (err != 0)
988 		rv = err;
989 
990 	for (dst = list_head(&saa.saa_dstg->dstg_tasks); dst;
991 	    dst = list_next(&saa.saa_dstg->dstg_tasks, dst)) {
992 		objset_t *os = dst->dst_arg1;
993 		snaponearg_t *soa = dst->dst_arg2;
994 		if (dst->dst_err != 0) {
995 			if (errors != NULL) {
996 				fnvlist_add_int32(errors,
997 				    soa->soa_longname, dst->dst_err);
998 			}
999 			rv = dst->dst_err;
1000 		}
1001 
1002 		if (saa.saa_needsuspend)
1003 			zil_resume(dmu_objset_zil(os));
1004 		dmu_objset_rele(os, &saa);
1005 		kmem_free(soa, sizeof (*soa));
1006 	}
1007 
1008 	dsl_sync_task_group_destroy(saa.saa_dstg);
1009 	spa_close(spa, FTAG);
1010 	return (rv);
1011 }
1012 
1013 int
1014 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1015 {
1016 	int err;
1017 	char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1018 	nvlist_t *snaps = fnvlist_alloc();
1019 
1020 	fnvlist_add_boolean(snaps, longsnap);
1021 	err = dmu_objset_snapshot(snaps, NULL, NULL);
1022 	fnvlist_free(snaps);
1023 	strfree(longsnap);
1024 	return (err);
1025 }
1026 
1027 int
1028 dmu_objset_snapshot_tmp(const char *snapname, const char *tag, int cleanup_fd)
1029 {
1030 	dsl_sync_task_t *dst;
1031 	snapallarg_t saa = { 0 };
1032 	spa_t *spa;
1033 	minor_t minor;
1034 	int err;
1035 
1036 	err = spa_open(snapname, &spa, FTAG);
1037 	if (err)
1038 		return (err);
1039 	saa.saa_dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
1040 	saa.saa_htag = tag;
1041 	saa.saa_needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1042 	saa.saa_temporary = B_TRUE;
1043 
1044 	if (cleanup_fd < 0) {
1045 		spa_close(spa, FTAG);
1046 		return (EINVAL);
1047 	}
1048 	if ((err = zfs_onexit_fd_hold(cleanup_fd, &minor)) != 0) {
1049 		spa_close(spa, FTAG);
1050 		return (err);
1051 	}
1052 
1053 	err = snapshot_one_impl(snapname, &saa);
1054 
1055 	if (err == 0)
1056 		err = dsl_sync_task_group_wait(saa.saa_dstg);
1057 
1058 	for (dst = list_head(&saa.saa_dstg->dstg_tasks); dst;
1059 	    dst = list_next(&saa.saa_dstg->dstg_tasks, dst)) {
1060 		objset_t *os = dst->dst_arg1;
1061 		dsl_register_onexit_hold_cleanup(saa.saa_newds, tag, minor);
1062 		if (saa.saa_needsuspend)
1063 			zil_resume(dmu_objset_zil(os));
1064 		dmu_objset_rele(os, &saa);
1065 	}
1066 
1067 	zfs_onexit_fd_rele(cleanup_fd);
1068 	dsl_sync_task_group_destroy(saa.saa_dstg);
1069 	spa_close(spa, FTAG);
1070 	return (err);
1071 }
1072 
1073 
1074 static void
1075 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
1076 {
1077 	dnode_t *dn;
1078 
1079 	while (dn = list_head(list)) {
1080 		ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1081 		ASSERT(dn->dn_dbuf->db_data_pending);
1082 		/*
1083 		 * Initialize dn_zio outside dnode_sync() because the
1084 		 * meta-dnode needs to set it ouside dnode_sync().
1085 		 */
1086 		dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1087 		ASSERT(dn->dn_zio);
1088 
1089 		ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1090 		list_remove(list, dn);
1091 
1092 		if (newlist) {
1093 			(void) dnode_add_ref(dn, newlist);
1094 			list_insert_tail(newlist, dn);
1095 		}
1096 
1097 		dnode_sync(dn, tx);
1098 	}
1099 }
1100 
1101 /* ARGSUSED */
1102 static void
1103 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1104 {
1105 	blkptr_t *bp = zio->io_bp;
1106 	objset_t *os = arg;
1107 	dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1108 
1109 	ASSERT(bp == os->os_rootbp);
1110 	ASSERT(BP_GET_TYPE(bp) == DMU_OT_OBJSET);
1111 	ASSERT(BP_GET_LEVEL(bp) == 0);
1112 
1113 	/*
1114 	 * Update rootbp fill count: it should be the number of objects
1115 	 * allocated in the object set (not counting the "special"
1116 	 * objects that are stored in the objset_phys_t -- the meta
1117 	 * dnode and user/group accounting objects).
1118 	 */
1119 	bp->blk_fill = 0;
1120 	for (int i = 0; i < dnp->dn_nblkptr; i++)
1121 		bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
1122 }
1123 
1124 /* ARGSUSED */
1125 static void
1126 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1127 {
1128 	blkptr_t *bp = zio->io_bp;
1129 	blkptr_t *bp_orig = &zio->io_bp_orig;
1130 	objset_t *os = arg;
1131 
1132 	if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1133 		ASSERT(BP_EQUAL(bp, bp_orig));
1134 	} else {
1135 		dsl_dataset_t *ds = os->os_dsl_dataset;
1136 		dmu_tx_t *tx = os->os_synctx;
1137 
1138 		(void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1139 		dsl_dataset_block_born(ds, bp, tx);
1140 	}
1141 }
1142 
1143 /* called from dsl */
1144 void
1145 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1146 {
1147 	int txgoff;
1148 	zbookmark_t zb;
1149 	zio_prop_t zp;
1150 	zio_t *zio;
1151 	list_t *list;
1152 	list_t *newlist = NULL;
1153 	dbuf_dirty_record_t *dr;
1154 
1155 	dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1156 
1157 	ASSERT(dmu_tx_is_syncing(tx));
1158 	/* XXX the write_done callback should really give us the tx... */
1159 	os->os_synctx = tx;
1160 
1161 	if (os->os_dsl_dataset == NULL) {
1162 		/*
1163 		 * This is the MOS.  If we have upgraded,
1164 		 * spa_max_replication() could change, so reset
1165 		 * os_copies here.
1166 		 */
1167 		os->os_copies = spa_max_replication(os->os_spa);
1168 	}
1169 
1170 	/*
1171 	 * Create the root block IO
1172 	 */
1173 	SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1174 	    os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1175 	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1176 	VERIFY3U(0, ==, arc_release_bp(os->os_phys_buf, &os->os_phys_buf,
1177 	    os->os_rootbp, os->os_spa, &zb));
1178 
1179 	dmu_write_policy(os, NULL, 0, 0, &zp);
1180 
1181 	zio = arc_write(pio, os->os_spa, tx->tx_txg,
1182 	    os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), &zp,
1183 	    dmu_objset_write_ready, dmu_objset_write_done, os,
1184 	    ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1185 
1186 	/*
1187 	 * Sync special dnodes - the parent IO for the sync is the root block
1188 	 */
1189 	DMU_META_DNODE(os)->dn_zio = zio;
1190 	dnode_sync(DMU_META_DNODE(os), tx);
1191 
1192 	os->os_phys->os_flags = os->os_flags;
1193 
1194 	if (DMU_USERUSED_DNODE(os) &&
1195 	    DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1196 		DMU_USERUSED_DNODE(os)->dn_zio = zio;
1197 		dnode_sync(DMU_USERUSED_DNODE(os), tx);
1198 		DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1199 		dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1200 	}
1201 
1202 	txgoff = tx->tx_txg & TXG_MASK;
1203 
1204 	if (dmu_objset_userused_enabled(os)) {
1205 		newlist = &os->os_synced_dnodes;
1206 		/*
1207 		 * We must create the list here because it uses the
1208 		 * dn_dirty_link[] of this txg.
1209 		 */
1210 		list_create(newlist, sizeof (dnode_t),
1211 		    offsetof(dnode_t, dn_dirty_link[txgoff]));
1212 	}
1213 
1214 	dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1215 	dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1216 
1217 	list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1218 	while (dr = list_head(list)) {
1219 		ASSERT(dr->dr_dbuf->db_level == 0);
1220 		list_remove(list, dr);
1221 		if (dr->dr_zio)
1222 			zio_nowait(dr->dr_zio);
1223 	}
1224 	/*
1225 	 * Free intent log blocks up to this tx.
1226 	 */
1227 	zil_sync(os->os_zil, tx);
1228 	os->os_phys->os_zil_header = os->os_zil_header;
1229 	zio_nowait(zio);
1230 }
1231 
1232 boolean_t
1233 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1234 {
1235 	return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1236 	    !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1237 }
1238 
1239 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1240 
1241 void
1242 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1243 {
1244 	used_cbs[ost] = cb;
1245 }
1246 
1247 boolean_t
1248 dmu_objset_userused_enabled(objset_t *os)
1249 {
1250 	return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1251 	    used_cbs[os->os_phys->os_type] != NULL &&
1252 	    DMU_USERUSED_DNODE(os) != NULL);
1253 }
1254 
1255 static void
1256 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1257     uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1258 {
1259 	if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1260 		int64_t delta = DNODE_SIZE + used;
1261 		if (subtract)
1262 			delta = -delta;
1263 		VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1264 		    user, delta, tx));
1265 		VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1266 		    group, delta, tx));
1267 	}
1268 }
1269 
1270 void
1271 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1272 {
1273 	dnode_t *dn;
1274 	list_t *list = &os->os_synced_dnodes;
1275 
1276 	ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1277 
1278 	while (dn = list_head(list)) {
1279 		int flags;
1280 		ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1281 		ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1282 		    dn->dn_phys->dn_flags &
1283 		    DNODE_FLAG_USERUSED_ACCOUNTED);
1284 
1285 		/* Allocate the user/groupused objects if necessary. */
1286 		if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1287 			VERIFY(0 == zap_create_claim(os,
1288 			    DMU_USERUSED_OBJECT,
1289 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1290 			VERIFY(0 == zap_create_claim(os,
1291 			    DMU_GROUPUSED_OBJECT,
1292 			    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1293 		}
1294 
1295 		/*
1296 		 * We intentionally modify the zap object even if the
1297 		 * net delta is zero.  Otherwise
1298 		 * the block of the zap obj could be shared between
1299 		 * datasets but need to be different between them after
1300 		 * a bprewrite.
1301 		 */
1302 
1303 		flags = dn->dn_id_flags;
1304 		ASSERT(flags);
1305 		if (flags & DN_ID_OLD_EXIST)  {
1306 			do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1307 			    dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1308 		}
1309 		if (flags & DN_ID_NEW_EXIST) {
1310 			do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1311 			    dn->dn_phys->dn_flags,  dn->dn_newuid,
1312 			    dn->dn_newgid, B_FALSE, tx);
1313 		}
1314 
1315 		mutex_enter(&dn->dn_mtx);
1316 		dn->dn_oldused = 0;
1317 		dn->dn_oldflags = 0;
1318 		if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1319 			dn->dn_olduid = dn->dn_newuid;
1320 			dn->dn_oldgid = dn->dn_newgid;
1321 			dn->dn_id_flags |= DN_ID_OLD_EXIST;
1322 			if (dn->dn_bonuslen == 0)
1323 				dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1324 			else
1325 				dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1326 		}
1327 		dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1328 		mutex_exit(&dn->dn_mtx);
1329 
1330 		list_remove(list, dn);
1331 		dnode_rele(dn, list);
1332 	}
1333 }
1334 
1335 /*
1336  * Returns a pointer to data to find uid/gid from
1337  *
1338  * If a dirty record for transaction group that is syncing can't
1339  * be found then NULL is returned.  In the NULL case it is assumed
1340  * the uid/gid aren't changing.
1341  */
1342 static void *
1343 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1344 {
1345 	dbuf_dirty_record_t *dr, **drp;
1346 	void *data;
1347 
1348 	if (db->db_dirtycnt == 0)
1349 		return (db->db.db_data);  /* Nothing is changing */
1350 
1351 	for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1352 		if (dr->dr_txg == tx->tx_txg)
1353 			break;
1354 
1355 	if (dr == NULL) {
1356 		data = NULL;
1357 	} else {
1358 		dnode_t *dn;
1359 
1360 		DB_DNODE_ENTER(dr->dr_dbuf);
1361 		dn = DB_DNODE(dr->dr_dbuf);
1362 
1363 		if (dn->dn_bonuslen == 0 &&
1364 		    dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1365 			data = dr->dt.dl.dr_data->b_data;
1366 		else
1367 			data = dr->dt.dl.dr_data;
1368 
1369 		DB_DNODE_EXIT(dr->dr_dbuf);
1370 	}
1371 
1372 	return (data);
1373 }
1374 
1375 void
1376 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1377 {
1378 	objset_t *os = dn->dn_objset;
1379 	void *data = NULL;
1380 	dmu_buf_impl_t *db = NULL;
1381 	uint64_t *user, *group;
1382 	int flags = dn->dn_id_flags;
1383 	int error;
1384 	boolean_t have_spill = B_FALSE;
1385 
1386 	if (!dmu_objset_userused_enabled(dn->dn_objset))
1387 		return;
1388 
1389 	if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1390 	    DN_ID_CHKED_SPILL)))
1391 		return;
1392 
1393 	if (before && dn->dn_bonuslen != 0)
1394 		data = DN_BONUS(dn->dn_phys);
1395 	else if (!before && dn->dn_bonuslen != 0) {
1396 		if (dn->dn_bonus) {
1397 			db = dn->dn_bonus;
1398 			mutex_enter(&db->db_mtx);
1399 			data = dmu_objset_userquota_find_data(db, tx);
1400 		} else {
1401 			data = DN_BONUS(dn->dn_phys);
1402 		}
1403 	} else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1404 			int rf = 0;
1405 
1406 			if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1407 				rf |= DB_RF_HAVESTRUCT;
1408 			error = dmu_spill_hold_by_dnode(dn,
1409 			    rf | DB_RF_MUST_SUCCEED,
1410 			    FTAG, (dmu_buf_t **)&db);
1411 			ASSERT(error == 0);
1412 			mutex_enter(&db->db_mtx);
1413 			data = (before) ? db->db.db_data :
1414 			    dmu_objset_userquota_find_data(db, tx);
1415 			have_spill = B_TRUE;
1416 	} else {
1417 		mutex_enter(&dn->dn_mtx);
1418 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1419 		mutex_exit(&dn->dn_mtx);
1420 		return;
1421 	}
1422 
1423 	if (before) {
1424 		ASSERT(data);
1425 		user = &dn->dn_olduid;
1426 		group = &dn->dn_oldgid;
1427 	} else if (data) {
1428 		user = &dn->dn_newuid;
1429 		group = &dn->dn_newgid;
1430 	}
1431 
1432 	/*
1433 	 * Must always call the callback in case the object
1434 	 * type has changed and that type isn't an object type to track
1435 	 */
1436 	error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1437 	    user, group);
1438 
1439 	/*
1440 	 * Preserve existing uid/gid when the callback can't determine
1441 	 * what the new uid/gid are and the callback returned EEXIST.
1442 	 * The EEXIST error tells us to just use the existing uid/gid.
1443 	 * If we don't know what the old values are then just assign
1444 	 * them to 0, since that is a new file  being created.
1445 	 */
1446 	if (!before && data == NULL && error == EEXIST) {
1447 		if (flags & DN_ID_OLD_EXIST) {
1448 			dn->dn_newuid = dn->dn_olduid;
1449 			dn->dn_newgid = dn->dn_oldgid;
1450 		} else {
1451 			dn->dn_newuid = 0;
1452 			dn->dn_newgid = 0;
1453 		}
1454 		error = 0;
1455 	}
1456 
1457 	if (db)
1458 		mutex_exit(&db->db_mtx);
1459 
1460 	mutex_enter(&dn->dn_mtx);
1461 	if (error == 0 && before)
1462 		dn->dn_id_flags |= DN_ID_OLD_EXIST;
1463 	if (error == 0 && !before)
1464 		dn->dn_id_flags |= DN_ID_NEW_EXIST;
1465 
1466 	if (have_spill) {
1467 		dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1468 	} else {
1469 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1470 	}
1471 	mutex_exit(&dn->dn_mtx);
1472 	if (have_spill)
1473 		dmu_buf_rele((dmu_buf_t *)db, FTAG);
1474 }
1475 
1476 boolean_t
1477 dmu_objset_userspace_present(objset_t *os)
1478 {
1479 	return (os->os_phys->os_flags &
1480 	    OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1481 }
1482 
1483 int
1484 dmu_objset_userspace_upgrade(objset_t *os)
1485 {
1486 	uint64_t obj;
1487 	int err = 0;
1488 
1489 	if (dmu_objset_userspace_present(os))
1490 		return (0);
1491 	if (!dmu_objset_userused_enabled(os))
1492 		return (ENOTSUP);
1493 	if (dmu_objset_is_snapshot(os))
1494 		return (EINVAL);
1495 
1496 	/*
1497 	 * We simply need to mark every object dirty, so that it will be
1498 	 * synced out and now accounted.  If this is called
1499 	 * concurrently, or if we already did some work before crashing,
1500 	 * that's fine, since we track each object's accounted state
1501 	 * independently.
1502 	 */
1503 
1504 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1505 		dmu_tx_t *tx;
1506 		dmu_buf_t *db;
1507 		int objerr;
1508 
1509 		if (issig(JUSTLOOKING) && issig(FORREAL))
1510 			return (EINTR);
1511 
1512 		objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1513 		if (objerr)
1514 			continue;
1515 		tx = dmu_tx_create(os);
1516 		dmu_tx_hold_bonus(tx, obj);
1517 		objerr = dmu_tx_assign(tx, TXG_WAIT);
1518 		if (objerr) {
1519 			dmu_tx_abort(tx);
1520 			continue;
1521 		}
1522 		dmu_buf_will_dirty(db, tx);
1523 		dmu_buf_rele(db, FTAG);
1524 		dmu_tx_commit(tx);
1525 	}
1526 
1527 	os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1528 	txg_wait_synced(dmu_objset_pool(os), 0);
1529 	return (0);
1530 }
1531 
1532 void
1533 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1534     uint64_t *usedobjsp, uint64_t *availobjsp)
1535 {
1536 	dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1537 	    usedobjsp, availobjsp);
1538 }
1539 
1540 uint64_t
1541 dmu_objset_fsid_guid(objset_t *os)
1542 {
1543 	return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1544 }
1545 
1546 void
1547 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1548 {
1549 	stat->dds_type = os->os_phys->os_type;
1550 	if (os->os_dsl_dataset)
1551 		dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1552 }
1553 
1554 void
1555 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1556 {
1557 	ASSERT(os->os_dsl_dataset ||
1558 	    os->os_phys->os_type == DMU_OST_META);
1559 
1560 	if (os->os_dsl_dataset != NULL)
1561 		dsl_dataset_stats(os->os_dsl_dataset, nv);
1562 
1563 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1564 	    os->os_phys->os_type);
1565 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1566 	    dmu_objset_userspace_present(os));
1567 }
1568 
1569 int
1570 dmu_objset_is_snapshot(objset_t *os)
1571 {
1572 	if (os->os_dsl_dataset != NULL)
1573 		return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1574 	else
1575 		return (B_FALSE);
1576 }
1577 
1578 int
1579 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1580     boolean_t *conflict)
1581 {
1582 	dsl_dataset_t *ds = os->os_dsl_dataset;
1583 	uint64_t ignored;
1584 
1585 	if (ds->ds_phys->ds_snapnames_zapobj == 0)
1586 		return (ENOENT);
1587 
1588 	return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1589 	    ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1590 	    real, maxlen, conflict));
1591 }
1592 
1593 int
1594 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1595     uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1596 {
1597 	dsl_dataset_t *ds = os->os_dsl_dataset;
1598 	zap_cursor_t cursor;
1599 	zap_attribute_t attr;
1600 
1601 	if (ds->ds_phys->ds_snapnames_zapobj == 0)
1602 		return (ENOENT);
1603 
1604 	zap_cursor_init_serialized(&cursor,
1605 	    ds->ds_dir->dd_pool->dp_meta_objset,
1606 	    ds->ds_phys->ds_snapnames_zapobj, *offp);
1607 
1608 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1609 		zap_cursor_fini(&cursor);
1610 		return (ENOENT);
1611 	}
1612 
1613 	if (strlen(attr.za_name) + 1 > namelen) {
1614 		zap_cursor_fini(&cursor);
1615 		return (ENAMETOOLONG);
1616 	}
1617 
1618 	(void) strcpy(name, attr.za_name);
1619 	if (idp)
1620 		*idp = attr.za_first_integer;
1621 	if (case_conflict)
1622 		*case_conflict = attr.za_normalization_conflict;
1623 	zap_cursor_advance(&cursor);
1624 	*offp = zap_cursor_serialize(&cursor);
1625 	zap_cursor_fini(&cursor);
1626 
1627 	return (0);
1628 }
1629 
1630 int
1631 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1632     uint64_t *idp, uint64_t *offp)
1633 {
1634 	dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1635 	zap_cursor_t cursor;
1636 	zap_attribute_t attr;
1637 
1638 	/* there is no next dir on a snapshot! */
1639 	if (os->os_dsl_dataset->ds_object !=
1640 	    dd->dd_phys->dd_head_dataset_obj)
1641 		return (ENOENT);
1642 
1643 	zap_cursor_init_serialized(&cursor,
1644 	    dd->dd_pool->dp_meta_objset,
1645 	    dd->dd_phys->dd_child_dir_zapobj, *offp);
1646 
1647 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1648 		zap_cursor_fini(&cursor);
1649 		return (ENOENT);
1650 	}
1651 
1652 	if (strlen(attr.za_name) + 1 > namelen) {
1653 		zap_cursor_fini(&cursor);
1654 		return (ENAMETOOLONG);
1655 	}
1656 
1657 	(void) strcpy(name, attr.za_name);
1658 	if (idp)
1659 		*idp = attr.za_first_integer;
1660 	zap_cursor_advance(&cursor);
1661 	*offp = zap_cursor_serialize(&cursor);
1662 	zap_cursor_fini(&cursor);
1663 
1664 	return (0);
1665 }
1666 
1667 struct findarg {
1668 	int (*func)(const char *, void *);
1669 	void *arg;
1670 };
1671 
1672 /* ARGSUSED */
1673 static int
1674 findfunc(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
1675 {
1676 	struct findarg *fa = arg;
1677 	return (fa->func(dsname, fa->arg));
1678 }
1679 
1680 /*
1681  * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1682  * Perhaps change all callers to use dmu_objset_find_spa()?
1683  */
1684 int
1685 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1686     int flags)
1687 {
1688 	struct findarg fa;
1689 	fa.func = func;
1690 	fa.arg = arg;
1691 	return (dmu_objset_find_spa(NULL, name, findfunc, &fa, flags));
1692 }
1693 
1694 /*
1695  * Find all objsets under name, call func on each
1696  */
1697 int
1698 dmu_objset_find_spa(spa_t *spa, const char *name,
1699     int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags)
1700 {
1701 	dsl_dir_t *dd;
1702 	dsl_pool_t *dp;
1703 	dsl_dataset_t *ds;
1704 	zap_cursor_t zc;
1705 	zap_attribute_t *attr;
1706 	char *child;
1707 	uint64_t thisobj;
1708 	int err;
1709 
1710 	if (name == NULL)
1711 		name = spa_name(spa);
1712 	err = dsl_dir_open_spa(spa, name, FTAG, &dd, NULL);
1713 	if (err)
1714 		return (err);
1715 
1716 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1717 	if (dd->dd_myname[0] == '$') {
1718 		dsl_dir_close(dd, FTAG);
1719 		return (0);
1720 	}
1721 
1722 	thisobj = dd->dd_phys->dd_head_dataset_obj;
1723 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1724 	dp = dd->dd_pool;
1725 
1726 	/*
1727 	 * Iterate over all children.
1728 	 */
1729 	if (flags & DS_FIND_CHILDREN) {
1730 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
1731 		    dd->dd_phys->dd_child_dir_zapobj);
1732 		    zap_cursor_retrieve(&zc, attr) == 0;
1733 		    (void) zap_cursor_advance(&zc)) {
1734 			ASSERT(attr->za_integer_length == sizeof (uint64_t));
1735 			ASSERT(attr->za_num_integers == 1);
1736 
1737 			child = kmem_asprintf("%s/%s", name, attr->za_name);
1738 			err = dmu_objset_find_spa(spa, child, func, arg, flags);
1739 			strfree(child);
1740 			if (err)
1741 				break;
1742 		}
1743 		zap_cursor_fini(&zc);
1744 
1745 		if (err) {
1746 			dsl_dir_close(dd, FTAG);
1747 			kmem_free(attr, sizeof (zap_attribute_t));
1748 			return (err);
1749 		}
1750 	}
1751 
1752 	/*
1753 	 * Iterate over all snapshots.
1754 	 */
1755 	if (flags & DS_FIND_SNAPSHOTS) {
1756 		if (!dsl_pool_sync_context(dp))
1757 			rw_enter(&dp->dp_config_rwlock, RW_READER);
1758 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1759 		if (!dsl_pool_sync_context(dp))
1760 			rw_exit(&dp->dp_config_rwlock);
1761 
1762 		if (err == 0) {
1763 			uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1764 			dsl_dataset_rele(ds, FTAG);
1765 
1766 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1767 			    zap_cursor_retrieve(&zc, attr) == 0;
1768 			    (void) zap_cursor_advance(&zc)) {
1769 				ASSERT(attr->za_integer_length ==
1770 				    sizeof (uint64_t));
1771 				ASSERT(attr->za_num_integers == 1);
1772 
1773 				child = kmem_asprintf("%s@%s",
1774 				    name, attr->za_name);
1775 				err = func(spa, attr->za_first_integer,
1776 				    child, arg);
1777 				strfree(child);
1778 				if (err)
1779 					break;
1780 			}
1781 			zap_cursor_fini(&zc);
1782 		}
1783 	}
1784 
1785 	dsl_dir_close(dd, FTAG);
1786 	kmem_free(attr, sizeof (zap_attribute_t));
1787 
1788 	if (err)
1789 		return (err);
1790 
1791 	/*
1792 	 * Apply to self if appropriate.
1793 	 */
1794 	err = func(spa, thisobj, name, arg);
1795 	return (err);
1796 }
1797 
1798 /* ARGSUSED */
1799 int
1800 dmu_objset_prefetch(const char *name, void *arg)
1801 {
1802 	dsl_dataset_t *ds;
1803 
1804 	if (dsl_dataset_hold(name, FTAG, &ds))
1805 		return (0);
1806 
1807 	if (!BP_IS_HOLE(&ds->ds_phys->ds_bp)) {
1808 		mutex_enter(&ds->ds_opening_lock);
1809 		if (ds->ds_objset == NULL) {
1810 			uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
1811 			zbookmark_t zb;
1812 
1813 			SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT,
1814 			    ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1815 
1816 			(void) dsl_read_nolock(NULL, dsl_dataset_get_spa(ds),
1817 			    &ds->ds_phys->ds_bp, NULL, NULL,
1818 			    ZIO_PRIORITY_ASYNC_READ,
1819 			    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
1820 			    &aflags, &zb);
1821 		}
1822 		mutex_exit(&ds->ds_opening_lock);
1823 	}
1824 
1825 	dsl_dataset_rele(ds, FTAG);
1826 	return (0);
1827 }
1828 
1829 void
1830 dmu_objset_set_user(objset_t *os, void *user_ptr)
1831 {
1832 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1833 	os->os_user_ptr = user_ptr;
1834 }
1835 
1836 void *
1837 dmu_objset_get_user(objset_t *os)
1838 {
1839 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1840 	return (os->os_user_ptr);
1841 }
1842