xref: /freebsd/sys/contrib/openzfs/module/zfs/dmu_objset.c (revision dc318a4ffabcbfa23bb56a33403aad36e6de30af)
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 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2012, 2020 by Delphix. All rights reserved.
25  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
26  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
27  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28  * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
29  * Copyright (c) 2016 Actifio, Inc. All rights reserved.
30  * Copyright 2017 Nexenta Systems, Inc.
31  * Copyright (c) 2017 Open-E, Inc. All Rights Reserved.
32  * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
33  * Copyright (c) 2019, Klara Inc.
34  * Copyright (c) 2019, Allan Jude
35  */
36 
37 /* Portions Copyright 2010 Robert Milkowski */
38 
39 #include <sys/cred.h>
40 #include <sys/zfs_context.h>
41 #include <sys/dmu_objset.h>
42 #include <sys/dsl_dir.h>
43 #include <sys/dsl_dataset.h>
44 #include <sys/dsl_prop.h>
45 #include <sys/dsl_pool.h>
46 #include <sys/dsl_synctask.h>
47 #include <sys/dsl_deleg.h>
48 #include <sys/dnode.h>
49 #include <sys/dbuf.h>
50 #include <sys/zvol.h>
51 #include <sys/dmu_tx.h>
52 #include <sys/zap.h>
53 #include <sys/zil.h>
54 #include <sys/dmu_impl.h>
55 #include <sys/zfs_ioctl.h>
56 #include <sys/sa.h>
57 #include <sys/zfs_onexit.h>
58 #include <sys/dsl_destroy.h>
59 #include <sys/vdev.h>
60 #include <sys/zfeature.h>
61 #include <sys/policy.h>
62 #include <sys/spa_impl.h>
63 #include <sys/dmu_recv.h>
64 #include <sys/zfs_project.h>
65 #include "zfs_namecheck.h"
66 
67 /*
68  * Needed to close a window in dnode_move() that allows the objset to be freed
69  * before it can be safely accessed.
70  */
71 krwlock_t os_lock;
72 
73 /*
74  * Tunable to overwrite the maximum number of threads for the parallelization
75  * of dmu_objset_find_dp, needed to speed up the import of pools with many
76  * datasets.
77  * Default is 4 times the number of leaf vdevs.
78  */
79 int dmu_find_threads = 0;
80 
81 /*
82  * Backfill lower metadnode objects after this many have been freed.
83  * Backfilling negatively impacts object creation rates, so only do it
84  * if there are enough holes to fill.
85  */
86 int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT;
87 
88 static char *upgrade_tag = "upgrade_tag";
89 
90 static void dmu_objset_find_dp_cb(void *arg);
91 
92 static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb);
93 static void dmu_objset_upgrade_stop(objset_t *os);
94 
95 void
96 dmu_objset_init(void)
97 {
98 	rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
99 }
100 
101 void
102 dmu_objset_fini(void)
103 {
104 	rw_destroy(&os_lock);
105 }
106 
107 spa_t *
108 dmu_objset_spa(objset_t *os)
109 {
110 	return (os->os_spa);
111 }
112 
113 zilog_t *
114 dmu_objset_zil(objset_t *os)
115 {
116 	return (os->os_zil);
117 }
118 
119 dsl_pool_t *
120 dmu_objset_pool(objset_t *os)
121 {
122 	dsl_dataset_t *ds;
123 
124 	if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
125 		return (ds->ds_dir->dd_pool);
126 	else
127 		return (spa_get_dsl(os->os_spa));
128 }
129 
130 dsl_dataset_t *
131 dmu_objset_ds(objset_t *os)
132 {
133 	return (os->os_dsl_dataset);
134 }
135 
136 dmu_objset_type_t
137 dmu_objset_type(objset_t *os)
138 {
139 	return (os->os_phys->os_type);
140 }
141 
142 void
143 dmu_objset_name(objset_t *os, char *buf)
144 {
145 	dsl_dataset_name(os->os_dsl_dataset, buf);
146 }
147 
148 uint64_t
149 dmu_objset_id(objset_t *os)
150 {
151 	dsl_dataset_t *ds = os->os_dsl_dataset;
152 
153 	return (ds ? ds->ds_object : 0);
154 }
155 
156 uint64_t
157 dmu_objset_dnodesize(objset_t *os)
158 {
159 	return (os->os_dnodesize);
160 }
161 
162 zfs_sync_type_t
163 dmu_objset_syncprop(objset_t *os)
164 {
165 	return (os->os_sync);
166 }
167 
168 zfs_logbias_op_t
169 dmu_objset_logbias(objset_t *os)
170 {
171 	return (os->os_logbias);
172 }
173 
174 static void
175 checksum_changed_cb(void *arg, uint64_t newval)
176 {
177 	objset_t *os = arg;
178 
179 	/*
180 	 * Inheritance should have been done by now.
181 	 */
182 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
183 
184 	os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
185 }
186 
187 static void
188 compression_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 != ZIO_COMPRESS_INHERIT);
196 
197 	os->os_compress = zio_compress_select(os->os_spa,
198 	    ZIO_COMPRESS_ALGO(newval), ZIO_COMPRESS_ON);
199 	os->os_complevel = zio_complevel_select(os->os_spa, os->os_compress,
200 	    ZIO_COMPRESS_LEVEL(newval), ZIO_COMPLEVEL_DEFAULT);
201 }
202 
203 static void
204 copies_changed_cb(void *arg, uint64_t newval)
205 {
206 	objset_t *os = arg;
207 
208 	/*
209 	 * Inheritance and range checking should have been done by now.
210 	 */
211 	ASSERT(newval > 0);
212 	ASSERT(newval <= spa_max_replication(os->os_spa));
213 
214 	os->os_copies = newval;
215 }
216 
217 static void
218 dedup_changed_cb(void *arg, uint64_t newval)
219 {
220 	objset_t *os = arg;
221 	spa_t *spa = os->os_spa;
222 	enum zio_checksum checksum;
223 
224 	/*
225 	 * Inheritance should have been done by now.
226 	 */
227 	ASSERT(newval != ZIO_CHECKSUM_INHERIT);
228 
229 	checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
230 
231 	os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
232 	os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
233 }
234 
235 static void
236 primary_cache_changed_cb(void *arg, uint64_t newval)
237 {
238 	objset_t *os = arg;
239 
240 	/*
241 	 * Inheritance and range checking should have been done by now.
242 	 */
243 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
244 	    newval == ZFS_CACHE_METADATA);
245 
246 	os->os_primary_cache = newval;
247 }
248 
249 static void
250 secondary_cache_changed_cb(void *arg, uint64_t newval)
251 {
252 	objset_t *os = arg;
253 
254 	/*
255 	 * Inheritance and range checking should have been done by now.
256 	 */
257 	ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
258 	    newval == ZFS_CACHE_METADATA);
259 
260 	os->os_secondary_cache = newval;
261 }
262 
263 static void
264 sync_changed_cb(void *arg, uint64_t newval)
265 {
266 	objset_t *os = arg;
267 
268 	/*
269 	 * Inheritance and range checking should have been done by now.
270 	 */
271 	ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
272 	    newval == ZFS_SYNC_DISABLED);
273 
274 	os->os_sync = newval;
275 	if (os->os_zil)
276 		zil_set_sync(os->os_zil, newval);
277 }
278 
279 static void
280 redundant_metadata_changed_cb(void *arg, uint64_t newval)
281 {
282 	objset_t *os = arg;
283 
284 	/*
285 	 * Inheritance and range checking should have been done by now.
286 	 */
287 	ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
288 	    newval == ZFS_REDUNDANT_METADATA_MOST);
289 
290 	os->os_redundant_metadata = newval;
291 }
292 
293 static void
294 dnodesize_changed_cb(void *arg, uint64_t newval)
295 {
296 	objset_t *os = arg;
297 
298 	switch (newval) {
299 	case ZFS_DNSIZE_LEGACY:
300 		os->os_dnodesize = DNODE_MIN_SIZE;
301 		break;
302 	case ZFS_DNSIZE_AUTO:
303 		/*
304 		 * Choose a dnode size that will work well for most
305 		 * workloads if the user specified "auto". Future code
306 		 * improvements could dynamically select a dnode size
307 		 * based on observed workload patterns.
308 		 */
309 		os->os_dnodesize = DNODE_MIN_SIZE * 2;
310 		break;
311 	case ZFS_DNSIZE_1K:
312 	case ZFS_DNSIZE_2K:
313 	case ZFS_DNSIZE_4K:
314 	case ZFS_DNSIZE_8K:
315 	case ZFS_DNSIZE_16K:
316 		os->os_dnodesize = newval;
317 		break;
318 	}
319 }
320 
321 static void
322 smallblk_changed_cb(void *arg, uint64_t newval)
323 {
324 	objset_t *os = arg;
325 
326 	/*
327 	 * Inheritance and range checking should have been done by now.
328 	 */
329 	ASSERT(newval <= SPA_MAXBLOCKSIZE);
330 	ASSERT(ISP2(newval));
331 
332 	os->os_zpl_special_smallblock = newval;
333 }
334 
335 static void
336 logbias_changed_cb(void *arg, uint64_t newval)
337 {
338 	objset_t *os = arg;
339 
340 	ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
341 	    newval == ZFS_LOGBIAS_THROUGHPUT);
342 	os->os_logbias = newval;
343 	if (os->os_zil)
344 		zil_set_logbias(os->os_zil, newval);
345 }
346 
347 static void
348 recordsize_changed_cb(void *arg, uint64_t newval)
349 {
350 	objset_t *os = arg;
351 
352 	os->os_recordsize = newval;
353 }
354 
355 void
356 dmu_objset_byteswap(void *buf, size_t size)
357 {
358 	objset_phys_t *osp = buf;
359 
360 	ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 ||
361 	    size == sizeof (objset_phys_t));
362 	dnode_byteswap(&osp->os_meta_dnode);
363 	byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
364 	osp->os_type = BSWAP_64(osp->os_type);
365 	osp->os_flags = BSWAP_64(osp->os_flags);
366 	if (size >= OBJSET_PHYS_SIZE_V2) {
367 		dnode_byteswap(&osp->os_userused_dnode);
368 		dnode_byteswap(&osp->os_groupused_dnode);
369 		if (size >= sizeof (objset_phys_t))
370 			dnode_byteswap(&osp->os_projectused_dnode);
371 	}
372 }
373 
374 /*
375  * The hash is a CRC-based hash of the objset_t pointer and the object number.
376  */
377 static uint64_t
378 dnode_hash(const objset_t *os, uint64_t obj)
379 {
380 	uintptr_t osv = (uintptr_t)os;
381 	uint64_t crc = -1ULL;
382 
383 	ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
384 	/*
385 	 * The low 6 bits of the pointer don't have much entropy, because
386 	 * the objset_t is larger than 2^6 bytes long.
387 	 */
388 	crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF];
389 	crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF];
390 	crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF];
391 	crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF];
392 
393 	crc ^= (osv>>14) ^ (obj>>24);
394 
395 	return (crc);
396 }
397 
398 static unsigned int
399 dnode_multilist_index_func(multilist_t *ml, void *obj)
400 {
401 	dnode_t *dn = obj;
402 	return (dnode_hash(dn->dn_objset, dn->dn_object) %
403 	    multilist_get_num_sublists(ml));
404 }
405 
406 /*
407  * Instantiates the objset_t in-memory structure corresponding to the
408  * objset_phys_t that's pointed to by the specified blkptr_t.
409  */
410 int
411 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
412     objset_t **osp)
413 {
414 	objset_t *os;
415 	int i, err;
416 
417 	ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
418 	ASSERT(!BP_IS_REDACTED(bp));
419 
420 	/*
421 	 * We need the pool config lock to get properties.
422 	 */
423 	ASSERT(ds == NULL || dsl_pool_config_held(ds->ds_dir->dd_pool));
424 
425 	/*
426 	 * The $ORIGIN dataset (if it exists) doesn't have an associated
427 	 * objset, so there's no reason to open it. The $ORIGIN dataset
428 	 * will not exist on pools older than SPA_VERSION_ORIGIN.
429 	 */
430 	if (ds != NULL && spa_get_dsl(spa) != NULL &&
431 	    spa_get_dsl(spa)->dp_origin_snap != NULL) {
432 		ASSERT3P(ds->ds_dir, !=,
433 		    spa_get_dsl(spa)->dp_origin_snap->ds_dir);
434 	}
435 
436 	os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
437 	os->os_dsl_dataset = ds;
438 	os->os_spa = spa;
439 	os->os_rootbp = bp;
440 	if (!BP_IS_HOLE(os->os_rootbp)) {
441 		arc_flags_t aflags = ARC_FLAG_WAIT;
442 		zbookmark_phys_t zb;
443 		int size;
444 		enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
445 		SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
446 		    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
447 
448 		if (DMU_OS_IS_L2CACHEABLE(os))
449 			aflags |= ARC_FLAG_L2CACHE;
450 
451 		if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) {
452 			ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF);
453 			ASSERT(BP_IS_AUTHENTICATED(bp));
454 			zio_flags |= ZIO_FLAG_RAW;
455 		}
456 
457 		dprintf_bp(os->os_rootbp, "reading %s", "");
458 		err = arc_read(NULL, spa, os->os_rootbp,
459 		    arc_getbuf_func, &os->os_phys_buf,
460 		    ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
461 		if (err != 0) {
462 			kmem_free(os, sizeof (objset_t));
463 			/* convert checksum errors into IO errors */
464 			if (err == ECKSUM)
465 				err = SET_ERROR(EIO);
466 			return (err);
467 		}
468 
469 		if (spa_version(spa) < SPA_VERSION_USERSPACE)
470 			size = OBJSET_PHYS_SIZE_V1;
471 		else if (!spa_feature_is_enabled(spa,
472 		    SPA_FEATURE_PROJECT_QUOTA))
473 			size = OBJSET_PHYS_SIZE_V2;
474 		else
475 			size = sizeof (objset_phys_t);
476 
477 		/* Increase the blocksize if we are permitted. */
478 		if (arc_buf_size(os->os_phys_buf) < size) {
479 			arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf,
480 			    ARC_BUFC_METADATA, size);
481 			bzero(buf->b_data, size);
482 			bcopy(os->os_phys_buf->b_data, buf->b_data,
483 			    arc_buf_size(os->os_phys_buf));
484 			arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
485 			os->os_phys_buf = buf;
486 		}
487 
488 		os->os_phys = os->os_phys_buf->b_data;
489 		os->os_flags = os->os_phys->os_flags;
490 	} else {
491 		int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
492 		    sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1;
493 		os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf,
494 		    ARC_BUFC_METADATA, size);
495 		os->os_phys = os->os_phys_buf->b_data;
496 		bzero(os->os_phys, size);
497 	}
498 	/*
499 	 * These properties will be filled in by the logic in zfs_get_zplprop()
500 	 * when they are queried for the first time.
501 	 */
502 	os->os_version = OBJSET_PROP_UNINITIALIZED;
503 	os->os_normalization = OBJSET_PROP_UNINITIALIZED;
504 	os->os_utf8only = OBJSET_PROP_UNINITIALIZED;
505 	os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED;
506 
507 	/*
508 	 * Note: the changed_cb will be called once before the register
509 	 * func returns, thus changing the checksum/compression from the
510 	 * default (fletcher2/off).  Snapshots don't need to know about
511 	 * checksum/compression/copies.
512 	 */
513 	if (ds != NULL) {
514 		os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0);
515 
516 		err = dsl_prop_register(ds,
517 		    zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
518 		    primary_cache_changed_cb, os);
519 		if (err == 0) {
520 			err = dsl_prop_register(ds,
521 			    zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
522 			    secondary_cache_changed_cb, os);
523 		}
524 		if (!ds->ds_is_snapshot) {
525 			if (err == 0) {
526 				err = dsl_prop_register(ds,
527 				    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
528 				    checksum_changed_cb, os);
529 			}
530 			if (err == 0) {
531 				err = dsl_prop_register(ds,
532 				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
533 				    compression_changed_cb, os);
534 			}
535 			if (err == 0) {
536 				err = dsl_prop_register(ds,
537 				    zfs_prop_to_name(ZFS_PROP_COPIES),
538 				    copies_changed_cb, os);
539 			}
540 			if (err == 0) {
541 				err = dsl_prop_register(ds,
542 				    zfs_prop_to_name(ZFS_PROP_DEDUP),
543 				    dedup_changed_cb, os);
544 			}
545 			if (err == 0) {
546 				err = dsl_prop_register(ds,
547 				    zfs_prop_to_name(ZFS_PROP_LOGBIAS),
548 				    logbias_changed_cb, os);
549 			}
550 			if (err == 0) {
551 				err = dsl_prop_register(ds,
552 				    zfs_prop_to_name(ZFS_PROP_SYNC),
553 				    sync_changed_cb, os);
554 			}
555 			if (err == 0) {
556 				err = dsl_prop_register(ds,
557 				    zfs_prop_to_name(
558 				    ZFS_PROP_REDUNDANT_METADATA),
559 				    redundant_metadata_changed_cb, os);
560 			}
561 			if (err == 0) {
562 				err = dsl_prop_register(ds,
563 				    zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
564 				    recordsize_changed_cb, os);
565 			}
566 			if (err == 0) {
567 				err = dsl_prop_register(ds,
568 				    zfs_prop_to_name(ZFS_PROP_DNODESIZE),
569 				    dnodesize_changed_cb, os);
570 			}
571 			if (err == 0) {
572 				err = dsl_prop_register(ds,
573 				    zfs_prop_to_name(
574 				    ZFS_PROP_SPECIAL_SMALL_BLOCKS),
575 				    smallblk_changed_cb, os);
576 			}
577 		}
578 		if (err != 0) {
579 			arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
580 			kmem_free(os, sizeof (objset_t));
581 			return (err);
582 		}
583 	} else {
584 		/* It's the meta-objset. */
585 		os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
586 		os->os_compress = ZIO_COMPRESS_ON;
587 		os->os_complevel = ZIO_COMPLEVEL_DEFAULT;
588 		os->os_encrypted = B_FALSE;
589 		os->os_copies = spa_max_replication(spa);
590 		os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
591 		os->os_dedup_verify = B_FALSE;
592 		os->os_logbias = ZFS_LOGBIAS_LATENCY;
593 		os->os_sync = ZFS_SYNC_STANDARD;
594 		os->os_primary_cache = ZFS_CACHE_ALL;
595 		os->os_secondary_cache = ZFS_CACHE_ALL;
596 		os->os_dnodesize = DNODE_MIN_SIZE;
597 	}
598 
599 	if (ds == NULL || !ds->ds_is_snapshot)
600 		os->os_zil_header = os->os_phys->os_zil_header;
601 	os->os_zil = zil_alloc(os, &os->os_zil_header);
602 
603 	for (i = 0; i < TXG_SIZE; i++) {
604 		os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t),
605 		    offsetof(dnode_t, dn_dirty_link[i]),
606 		    dnode_multilist_index_func);
607 	}
608 	list_create(&os->os_dnodes, sizeof (dnode_t),
609 	    offsetof(dnode_t, dn_link));
610 	list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
611 	    offsetof(dmu_buf_impl_t, db_link));
612 
613 	list_link_init(&os->os_evicting_node);
614 
615 	mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
616 	mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL);
617 	mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
618 	mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
619 	os->os_obj_next_percpu_len = boot_ncpus;
620 	os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len *
621 	    sizeof (os->os_obj_next_percpu[0]), KM_SLEEP);
622 
623 	dnode_special_open(os, &os->os_phys->os_meta_dnode,
624 	    DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
625 	if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) {
626 		dnode_special_open(os, &os->os_phys->os_userused_dnode,
627 		    DMU_USERUSED_OBJECT, &os->os_userused_dnode);
628 		dnode_special_open(os, &os->os_phys->os_groupused_dnode,
629 		    DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
630 		if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf))
631 			dnode_special_open(os,
632 			    &os->os_phys->os_projectused_dnode,
633 			    DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode);
634 	}
635 
636 	mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL);
637 
638 	*osp = os;
639 	return (0);
640 }
641 
642 int
643 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
644 {
645 	int err = 0;
646 
647 	/*
648 	 * We need the pool_config lock to manipulate the dsl_dataset_t.
649 	 * Even if the dataset is long-held, we need the pool_config lock
650 	 * to open the objset, as it needs to get properties.
651 	 */
652 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
653 
654 	mutex_enter(&ds->ds_opening_lock);
655 	if (ds->ds_objset == NULL) {
656 		objset_t *os;
657 		rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
658 		err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
659 		    ds, dsl_dataset_get_blkptr(ds), &os);
660 		rrw_exit(&ds->ds_bp_rwlock, FTAG);
661 
662 		if (err == 0) {
663 			mutex_enter(&ds->ds_lock);
664 			ASSERT(ds->ds_objset == NULL);
665 			ds->ds_objset = os;
666 			mutex_exit(&ds->ds_lock);
667 		}
668 	}
669 	*osp = ds->ds_objset;
670 	mutex_exit(&ds->ds_opening_lock);
671 	return (err);
672 }
673 
674 /*
675  * Holds the pool while the objset is held.  Therefore only one objset
676  * can be held at a time.
677  */
678 int
679 dmu_objset_hold_flags(const char *name, boolean_t decrypt, void *tag,
680     objset_t **osp)
681 {
682 	dsl_pool_t *dp;
683 	dsl_dataset_t *ds;
684 	int err;
685 	ds_hold_flags_t flags;
686 
687 	flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
688 	err = dsl_pool_hold(name, tag, &dp);
689 	if (err != 0)
690 		return (err);
691 	err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds);
692 	if (err != 0) {
693 		dsl_pool_rele(dp, tag);
694 		return (err);
695 	}
696 
697 	err = dmu_objset_from_ds(ds, osp);
698 	if (err != 0) {
699 		dsl_dataset_rele(ds, tag);
700 		dsl_pool_rele(dp, tag);
701 	}
702 
703 	return (err);
704 }
705 
706 int
707 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
708 {
709 	return (dmu_objset_hold_flags(name, B_FALSE, tag, osp));
710 }
711 
712 static int
713 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
714     boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
715 {
716 	int err;
717 
718 	err = dmu_objset_from_ds(ds, osp);
719 	if (err != 0) {
720 		return (err);
721 	} else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
722 		return (SET_ERROR(EINVAL));
723 	} else if (!readonly && dsl_dataset_is_snapshot(ds)) {
724 		return (SET_ERROR(EROFS));
725 	} else if (!readonly && decrypt &&
726 	    dsl_dir_incompatible_encryption_version(ds->ds_dir)) {
727 		return (SET_ERROR(EROFS));
728 	}
729 
730 	/* if we are decrypting, we can now check MACs in os->os_phys_buf */
731 	if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) {
732 		zbookmark_phys_t zb;
733 
734 		SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT,
735 		    ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
736 		err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa,
737 		    &zb, B_FALSE);
738 		if (err != 0)
739 			return (err);
740 
741 		ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf));
742 	}
743 
744 	return (0);
745 }
746 
747 /*
748  * dsl_pool must not be held when this is called.
749  * Upon successful return, there will be a longhold on the dataset,
750  * and the dsl_pool will not be held.
751  */
752 int
753 dmu_objset_own(const char *name, dmu_objset_type_t type,
754     boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
755 {
756 	dsl_pool_t *dp;
757 	dsl_dataset_t *ds;
758 	int err;
759 	ds_hold_flags_t flags;
760 
761 	flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
762 	err = dsl_pool_hold(name, FTAG, &dp);
763 	if (err != 0)
764 		return (err);
765 	err = dsl_dataset_own(dp, name, flags, tag, &ds);
766 	if (err != 0) {
767 		dsl_pool_rele(dp, FTAG);
768 		return (err);
769 	}
770 	err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp);
771 	if (err != 0) {
772 		dsl_dataset_disown(ds, flags, tag);
773 		dsl_pool_rele(dp, FTAG);
774 		return (err);
775 	}
776 
777 	/*
778 	 * User accounting requires the dataset to be decrypted and rw.
779 	 * We also don't begin user accounting during claiming to help
780 	 * speed up pool import times and to keep this txg reserved
781 	 * completely for recovery work.
782 	 */
783 	if (!readonly && !dp->dp_spa->spa_claiming &&
784 	    (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) {
785 		if (dmu_objset_userobjspace_upgradable(*osp) ||
786 		    dmu_objset_projectquota_upgradable(*osp)) {
787 			dmu_objset_id_quota_upgrade(*osp);
788 		} else if (dmu_objset_userused_enabled(*osp)) {
789 			dmu_objset_userspace_upgrade(*osp);
790 		}
791 	}
792 
793 	dsl_pool_rele(dp, FTAG);
794 	return (0);
795 }
796 
797 int
798 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
799     boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
800 {
801 	dsl_dataset_t *ds;
802 	int err;
803 	ds_hold_flags_t flags;
804 
805 	flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
806 	err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds);
807 	if (err != 0)
808 		return (err);
809 
810 	err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp);
811 	if (err != 0) {
812 		dsl_dataset_disown(ds, flags, tag);
813 		return (err);
814 	}
815 
816 	return (0);
817 }
818 
819 void
820 dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, void *tag)
821 {
822 	ds_hold_flags_t flags;
823 	dsl_pool_t *dp = dmu_objset_pool(os);
824 
825 	flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
826 	dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag);
827 	dsl_pool_rele(dp, tag);
828 }
829 
830 void
831 dmu_objset_rele(objset_t *os, void *tag)
832 {
833 	dmu_objset_rele_flags(os, B_FALSE, tag);
834 }
835 
836 /*
837  * When we are called, os MUST refer to an objset associated with a dataset
838  * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
839  * == tag.  We will then release and reacquire ownership of the dataset while
840  * holding the pool config_rwlock to avoid intervening namespace or ownership
841  * changes may occur.
842  *
843  * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
844  * release the hold on its dataset and acquire a new one on the dataset of the
845  * same name so that it can be partially torn down and reconstructed.
846  */
847 void
848 dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds,
849     boolean_t decrypt, void *tag)
850 {
851 	dsl_pool_t *dp;
852 	char name[ZFS_MAX_DATASET_NAME_LEN];
853 	ds_hold_flags_t flags;
854 
855 	flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
856 	VERIFY3P(ds, !=, NULL);
857 	VERIFY3P(ds->ds_owner, ==, tag);
858 	VERIFY(dsl_dataset_long_held(ds));
859 
860 	dsl_dataset_name(ds, name);
861 	dp = ds->ds_dir->dd_pool;
862 	dsl_pool_config_enter(dp, FTAG);
863 	dsl_dataset_disown(ds, flags, tag);
864 	VERIFY0(dsl_dataset_own(dp, name, flags, tag, newds));
865 	dsl_pool_config_exit(dp, FTAG);
866 }
867 
868 void
869 dmu_objset_disown(objset_t *os, boolean_t decrypt, void *tag)
870 {
871 	ds_hold_flags_t flags;
872 
873 	flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
874 	/*
875 	 * Stop upgrading thread
876 	 */
877 	dmu_objset_upgrade_stop(os);
878 	dsl_dataset_disown(os->os_dsl_dataset, flags, tag);
879 }
880 
881 void
882 dmu_objset_evict_dbufs(objset_t *os)
883 {
884 	dnode_t *dn_marker;
885 	dnode_t *dn;
886 
887 	dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP);
888 
889 	mutex_enter(&os->os_lock);
890 	dn = list_head(&os->os_dnodes);
891 	while (dn != NULL) {
892 		/*
893 		 * Skip dnodes without holds.  We have to do this dance
894 		 * because dnode_add_ref() only works if there is already a
895 		 * hold.  If the dnode has no holds, then it has no dbufs.
896 		 */
897 		if (dnode_add_ref(dn, FTAG)) {
898 			list_insert_after(&os->os_dnodes, dn, dn_marker);
899 			mutex_exit(&os->os_lock);
900 
901 			dnode_evict_dbufs(dn);
902 			dnode_rele(dn, FTAG);
903 
904 			mutex_enter(&os->os_lock);
905 			dn = list_next(&os->os_dnodes, dn_marker);
906 			list_remove(&os->os_dnodes, dn_marker);
907 		} else {
908 			dn = list_next(&os->os_dnodes, dn);
909 		}
910 	}
911 	mutex_exit(&os->os_lock);
912 
913 	kmem_free(dn_marker, sizeof (dnode_t));
914 
915 	if (DMU_USERUSED_DNODE(os) != NULL) {
916 		if (DMU_PROJECTUSED_DNODE(os) != NULL)
917 			dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os));
918 		dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
919 		dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
920 	}
921 	dnode_evict_dbufs(DMU_META_DNODE(os));
922 }
923 
924 /*
925  * Objset eviction processing is split into into two pieces.
926  * The first marks the objset as evicting, evicts any dbufs that
927  * have a refcount of zero, and then queues up the objset for the
928  * second phase of eviction.  Once os->os_dnodes has been cleared by
929  * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
930  * The second phase closes the special dnodes, dequeues the objset from
931  * the list of those undergoing eviction, and finally frees the objset.
932  *
933  * NOTE: Due to asynchronous eviction processing (invocation of
934  *       dnode_buf_pageout()), it is possible for the meta dnode for the
935  *       objset to have no holds even though os->os_dnodes is not empty.
936  */
937 void
938 dmu_objset_evict(objset_t *os)
939 {
940 	dsl_dataset_t *ds = os->os_dsl_dataset;
941 
942 	for (int t = 0; t < TXG_SIZE; t++)
943 		ASSERT(!dmu_objset_is_dirty(os, t));
944 
945 	if (ds)
946 		dsl_prop_unregister_all(ds, os);
947 
948 	if (os->os_sa)
949 		sa_tear_down(os);
950 
951 	dmu_objset_evict_dbufs(os);
952 
953 	mutex_enter(&os->os_lock);
954 	spa_evicting_os_register(os->os_spa, os);
955 	if (list_is_empty(&os->os_dnodes)) {
956 		mutex_exit(&os->os_lock);
957 		dmu_objset_evict_done(os);
958 	} else {
959 		mutex_exit(&os->os_lock);
960 	}
961 
962 
963 }
964 
965 void
966 dmu_objset_evict_done(objset_t *os)
967 {
968 	ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
969 
970 	dnode_special_close(&os->os_meta_dnode);
971 	if (DMU_USERUSED_DNODE(os)) {
972 		if (DMU_PROJECTUSED_DNODE(os))
973 			dnode_special_close(&os->os_projectused_dnode);
974 		dnode_special_close(&os->os_userused_dnode);
975 		dnode_special_close(&os->os_groupused_dnode);
976 	}
977 	zil_free(os->os_zil);
978 
979 	arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
980 
981 	/*
982 	 * This is a barrier to prevent the objset from going away in
983 	 * dnode_move() until we can safely ensure that the objset is still in
984 	 * use. We consider the objset valid before the barrier and invalid
985 	 * after the barrier.
986 	 */
987 	rw_enter(&os_lock, RW_READER);
988 	rw_exit(&os_lock);
989 
990 	kmem_free(os->os_obj_next_percpu,
991 	    os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0]));
992 
993 	mutex_destroy(&os->os_lock);
994 	mutex_destroy(&os->os_userused_lock);
995 	mutex_destroy(&os->os_obj_lock);
996 	mutex_destroy(&os->os_user_ptr_lock);
997 	mutex_destroy(&os->os_upgrade_lock);
998 	for (int i = 0; i < TXG_SIZE; i++) {
999 		multilist_destroy(os->os_dirty_dnodes[i]);
1000 	}
1001 	spa_evicting_os_deregister(os->os_spa, os);
1002 	kmem_free(os, sizeof (objset_t));
1003 }
1004 
1005 inode_timespec_t
1006 dmu_objset_snap_cmtime(objset_t *os)
1007 {
1008 	return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
1009 }
1010 
1011 objset_t *
1012 dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
1013     dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx)
1014 {
1015 	objset_t *os;
1016 	dnode_t *mdn;
1017 
1018 	ASSERT(dmu_tx_is_syncing(tx));
1019 
1020 	if (blksz == 0)
1021 		blksz = DNODE_BLOCK_SIZE;
1022 	if (ibs == 0)
1023 		ibs = DN_MAX_INDBLKSHIFT;
1024 
1025 	if (ds != NULL)
1026 		VERIFY0(dmu_objset_from_ds(ds, &os));
1027 	else
1028 		VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
1029 
1030 	mdn = DMU_META_DNODE(os);
1031 
1032 	dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0,
1033 	    DNODE_MIN_SLOTS, tx);
1034 
1035 	/*
1036 	 * We don't want to have to increase the meta-dnode's nlevels
1037 	 * later, because then we could do it in quiescing context while
1038 	 * we are also accessing it in open context.
1039 	 *
1040 	 * This precaution is not necessary for the MOS (ds == NULL),
1041 	 * because the MOS is only updated in syncing context.
1042 	 * This is most fortunate: the MOS is the only objset that
1043 	 * needs to be synced multiple times as spa_sync() iterates
1044 	 * to convergence, so minimizing its dn_nlevels matters.
1045 	 */
1046 	if (ds != NULL) {
1047 		if (levels == 0) {
1048 			levels = 1;
1049 
1050 			/*
1051 			 * Determine the number of levels necessary for the
1052 			 * meta-dnode to contain DN_MAX_OBJECT dnodes.  Note
1053 			 * that in order to ensure that we do not overflow
1054 			 * 64 bits, there has to be a nlevels that gives us a
1055 			 * number of blocks > DN_MAX_OBJECT but < 2^64.
1056 			 * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)
1057 			 * (10) must be less than (64 - log2(DN_MAX_OBJECT))
1058 			 * (16).
1059 			 */
1060 			while ((uint64_t)mdn->dn_nblkptr <<
1061 			    (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) *
1062 			    (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
1063 			    DN_MAX_OBJECT)
1064 				levels++;
1065 		}
1066 
1067 		mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
1068 		    mdn->dn_nlevels = levels;
1069 	}
1070 
1071 	ASSERT(type != DMU_OST_NONE);
1072 	ASSERT(type != DMU_OST_ANY);
1073 	ASSERT(type < DMU_OST_NUMTYPES);
1074 	os->os_phys->os_type = type;
1075 
1076 	/*
1077 	 * Enable user accounting if it is enabled and this is not an
1078 	 * encrypted receive.
1079 	 */
1080 	if (dmu_objset_userused_enabled(os) &&
1081 	    (!os->os_encrypted || !dmu_objset_is_receiving(os))) {
1082 		os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1083 		if (dmu_objset_userobjused_enabled(os)) {
1084 			ds->ds_feature_activation[
1085 			    SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE;
1086 			os->os_phys->os_flags |=
1087 			    OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE;
1088 		}
1089 		if (dmu_objset_projectquota_enabled(os)) {
1090 			ds->ds_feature_activation[
1091 			    SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE;
1092 			os->os_phys->os_flags |=
1093 			    OBJSET_FLAG_PROJECTQUOTA_COMPLETE;
1094 		}
1095 		os->os_flags = os->os_phys->os_flags;
1096 	}
1097 
1098 	dsl_dataset_dirty(ds, tx);
1099 
1100 	return (os);
1101 }
1102 
1103 /* called from dsl for meta-objset */
1104 objset_t *
1105 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
1106     dmu_objset_type_t type, dmu_tx_t *tx)
1107 {
1108 	return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx));
1109 }
1110 
1111 typedef struct dmu_objset_create_arg {
1112 	const char *doca_name;
1113 	cred_t *doca_cred;
1114 	proc_t *doca_proc;
1115 	void (*doca_userfunc)(objset_t *os, void *arg,
1116 	    cred_t *cr, dmu_tx_t *tx);
1117 	void *doca_userarg;
1118 	dmu_objset_type_t doca_type;
1119 	uint64_t doca_flags;
1120 	dsl_crypto_params_t *doca_dcp;
1121 } dmu_objset_create_arg_t;
1122 
1123 /*ARGSUSED*/
1124 static int
1125 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
1126 {
1127 	dmu_objset_create_arg_t *doca = arg;
1128 	dsl_pool_t *dp = dmu_tx_pool(tx);
1129 	dsl_dir_t *pdd;
1130 	dsl_dataset_t *parentds;
1131 	objset_t *parentos;
1132 	const char *tail;
1133 	int error;
1134 
1135 	if (strchr(doca->doca_name, '@') != NULL)
1136 		return (SET_ERROR(EINVAL));
1137 
1138 	if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN)
1139 		return (SET_ERROR(ENAMETOOLONG));
1140 
1141 	if (dataset_nestcheck(doca->doca_name) != 0)
1142 		return (SET_ERROR(ENAMETOOLONG));
1143 
1144 	error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
1145 	if (error != 0)
1146 		return (error);
1147 	if (tail == NULL) {
1148 		dsl_dir_rele(pdd, FTAG);
1149 		return (SET_ERROR(EEXIST));
1150 	}
1151 
1152 	error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL);
1153 	if (error != 0) {
1154 		dsl_dir_rele(pdd, FTAG);
1155 		return (error);
1156 	}
1157 
1158 	error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
1159 	    doca->doca_cred, doca->doca_proc);
1160 	if (error != 0) {
1161 		dsl_dir_rele(pdd, FTAG);
1162 		return (error);
1163 	}
1164 
1165 	/* can't create below anything but filesystems (eg. no ZVOLs) */
1166 	error = dsl_dataset_hold_obj(pdd->dd_pool,
1167 	    dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds);
1168 	if (error != 0) {
1169 		dsl_dir_rele(pdd, FTAG);
1170 		return (error);
1171 	}
1172 	error = dmu_objset_from_ds(parentds, &parentos);
1173 	if (error != 0) {
1174 		dsl_dataset_rele(parentds, FTAG);
1175 		dsl_dir_rele(pdd, FTAG);
1176 		return (error);
1177 	}
1178 	if (dmu_objset_type(parentos) != DMU_OST_ZFS) {
1179 		dsl_dataset_rele(parentds, FTAG);
1180 		dsl_dir_rele(pdd, FTAG);
1181 		return (SET_ERROR(ZFS_ERR_WRONG_PARENT));
1182 	}
1183 	dsl_dataset_rele(parentds, FTAG);
1184 	dsl_dir_rele(pdd, FTAG);
1185 
1186 	return (error);
1187 }
1188 
1189 static void
1190 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
1191 {
1192 	dmu_objset_create_arg_t *doca = arg;
1193 	dsl_pool_t *dp = dmu_tx_pool(tx);
1194 	spa_t *spa = dp->dp_spa;
1195 	dsl_dir_t *pdd;
1196 	const char *tail;
1197 	dsl_dataset_t *ds;
1198 	uint64_t obj;
1199 	blkptr_t *bp;
1200 	objset_t *os;
1201 	zio_t *rzio;
1202 
1203 	VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
1204 
1205 	obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
1206 	    doca->doca_cred, doca->doca_dcp, tx);
1207 
1208 	VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj,
1209 	    DS_HOLD_FLAG_DECRYPT, FTAG, &ds));
1210 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1211 	bp = dsl_dataset_get_blkptr(ds);
1212 	os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx);
1213 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
1214 
1215 	if (doca->doca_userfunc != NULL) {
1216 		doca->doca_userfunc(os, doca->doca_userarg,
1217 		    doca->doca_cred, tx);
1218 	}
1219 
1220 	/*
1221 	 * The doca_userfunc() may write out some data that needs to be
1222 	 * encrypted if the dataset is encrypted (specifically the root
1223 	 * directory).  This data must be written out before the encryption
1224 	 * key mapping is removed by dsl_dataset_rele_flags().  Force the
1225 	 * I/O to occur immediately by invoking the relevant sections of
1226 	 * dsl_pool_sync().
1227 	 */
1228 	if (os->os_encrypted) {
1229 		dsl_dataset_t *tmpds = NULL;
1230 		boolean_t need_sync_done = B_FALSE;
1231 
1232 		mutex_enter(&ds->ds_lock);
1233 		ds->ds_owner = FTAG;
1234 		mutex_exit(&ds->ds_lock);
1235 
1236 		rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1237 		tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds,
1238 		    tx->tx_txg);
1239 		if (tmpds != NULL) {
1240 			dsl_dataset_sync(ds, rzio, tx);
1241 			need_sync_done = B_TRUE;
1242 		}
1243 		VERIFY0(zio_wait(rzio));
1244 
1245 		dmu_objset_sync_done(os, tx);
1246 		taskq_wait(dp->dp_sync_taskq);
1247 		if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1248 			ASSERT3P(ds->ds_key_mapping, !=, NULL);
1249 			key_mapping_rele(spa, ds->ds_key_mapping, ds);
1250 		}
1251 
1252 		rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1253 		tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds,
1254 		    tx->tx_txg);
1255 		if (tmpds != NULL) {
1256 			dmu_buf_rele(ds->ds_dbuf, ds);
1257 			dsl_dataset_sync(ds, rzio, tx);
1258 		}
1259 		VERIFY0(zio_wait(rzio));
1260 
1261 		if (need_sync_done) {
1262 			ASSERT3P(ds->ds_key_mapping, !=, NULL);
1263 			key_mapping_rele(spa, ds->ds_key_mapping, ds);
1264 			dsl_dataset_sync_done(ds, tx);
1265 		}
1266 
1267 		mutex_enter(&ds->ds_lock);
1268 		ds->ds_owner = NULL;
1269 		mutex_exit(&ds->ds_lock);
1270 	}
1271 
1272 	spa_history_log_internal_ds(ds, "create", tx, " ");
1273 
1274 	dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG);
1275 	dsl_dir_rele(pdd, FTAG);
1276 }
1277 
1278 int
1279 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
1280     dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg)
1281 {
1282 	dmu_objset_create_arg_t doca;
1283 	dsl_crypto_params_t tmp_dcp = { 0 };
1284 
1285 	doca.doca_name = name;
1286 	doca.doca_cred = CRED();
1287 	doca.doca_proc = curproc;
1288 	doca.doca_flags = flags;
1289 	doca.doca_userfunc = func;
1290 	doca.doca_userarg = arg;
1291 	doca.doca_type = type;
1292 
1293 	/*
1294 	 * Some callers (mostly for testing) do not provide a dcp on their
1295 	 * own but various code inside the sync task will require it to be
1296 	 * allocated. Rather than adding NULL checks throughout this code
1297 	 * or adding dummy dcp's to all of the callers we simply create a
1298 	 * dummy one here and use that. This zero dcp will have the same
1299 	 * effect as asking for inheritance of all encryption params.
1300 	 */
1301 	doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp;
1302 
1303 	int rv = dsl_sync_task(name,
1304 	    dmu_objset_create_check, dmu_objset_create_sync, &doca,
1305 	    6, ZFS_SPACE_CHECK_NORMAL);
1306 
1307 	if (rv == 0)
1308 		zvol_create_minor(name);
1309 	return (rv);
1310 }
1311 
1312 typedef struct dmu_objset_clone_arg {
1313 	const char *doca_clone;
1314 	const char *doca_origin;
1315 	cred_t *doca_cred;
1316 	proc_t *doca_proc;
1317 } dmu_objset_clone_arg_t;
1318 
1319 /*ARGSUSED*/
1320 static int
1321 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
1322 {
1323 	dmu_objset_clone_arg_t *doca = arg;
1324 	dsl_dir_t *pdd;
1325 	const char *tail;
1326 	int error;
1327 	dsl_dataset_t *origin;
1328 	dsl_pool_t *dp = dmu_tx_pool(tx);
1329 
1330 	if (strchr(doca->doca_clone, '@') != NULL)
1331 		return (SET_ERROR(EINVAL));
1332 
1333 	if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN)
1334 		return (SET_ERROR(ENAMETOOLONG));
1335 
1336 	error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
1337 	if (error != 0)
1338 		return (error);
1339 	if (tail == NULL) {
1340 		dsl_dir_rele(pdd, FTAG);
1341 		return (SET_ERROR(EEXIST));
1342 	}
1343 
1344 	error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
1345 	    doca->doca_cred, doca->doca_proc);
1346 	if (error != 0) {
1347 		dsl_dir_rele(pdd, FTAG);
1348 		return (SET_ERROR(EDQUOT));
1349 	}
1350 
1351 	error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
1352 	if (error != 0) {
1353 		dsl_dir_rele(pdd, FTAG);
1354 		return (error);
1355 	}
1356 
1357 	/* You can only clone snapshots, not the head datasets. */
1358 	if (!origin->ds_is_snapshot) {
1359 		dsl_dataset_rele(origin, FTAG);
1360 		dsl_dir_rele(pdd, FTAG);
1361 		return (SET_ERROR(EINVAL));
1362 	}
1363 
1364 	dsl_dataset_rele(origin, FTAG);
1365 	dsl_dir_rele(pdd, FTAG);
1366 
1367 	return (0);
1368 }
1369 
1370 static void
1371 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
1372 {
1373 	dmu_objset_clone_arg_t *doca = arg;
1374 	dsl_pool_t *dp = dmu_tx_pool(tx);
1375 	dsl_dir_t *pdd;
1376 	const char *tail;
1377 	dsl_dataset_t *origin, *ds;
1378 	uint64_t obj;
1379 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1380 
1381 	VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
1382 	VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
1383 
1384 	obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
1385 	    doca->doca_cred, NULL, tx);
1386 
1387 	VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
1388 	dsl_dataset_name(origin, namebuf);
1389 	spa_history_log_internal_ds(ds, "clone", tx,
1390 	    "origin=%s (%llu)", namebuf, (u_longlong_t)origin->ds_object);
1391 	dsl_dataset_rele(ds, FTAG);
1392 	dsl_dataset_rele(origin, FTAG);
1393 	dsl_dir_rele(pdd, FTAG);
1394 }
1395 
1396 int
1397 dmu_objset_clone(const char *clone, const char *origin)
1398 {
1399 	dmu_objset_clone_arg_t doca;
1400 
1401 	doca.doca_clone = clone;
1402 	doca.doca_origin = origin;
1403 	doca.doca_cred = CRED();
1404 	doca.doca_proc = curproc;
1405 
1406 	int rv = dsl_sync_task(clone,
1407 	    dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
1408 	    6, ZFS_SPACE_CHECK_NORMAL);
1409 
1410 	if (rv == 0)
1411 		zvol_create_minor(clone);
1412 
1413 	return (rv);
1414 }
1415 
1416 int
1417 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1418 {
1419 	int err;
1420 	char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1421 	nvlist_t *snaps = fnvlist_alloc();
1422 
1423 	fnvlist_add_boolean(snaps, longsnap);
1424 	kmem_strfree(longsnap);
1425 	err = dsl_dataset_snapshot(snaps, NULL, NULL);
1426 	fnvlist_free(snaps);
1427 	return (err);
1428 }
1429 
1430 static void
1431 dmu_objset_upgrade_task_cb(void *data)
1432 {
1433 	objset_t *os = data;
1434 
1435 	mutex_enter(&os->os_upgrade_lock);
1436 	os->os_upgrade_status = EINTR;
1437 	if (!os->os_upgrade_exit) {
1438 		int status;
1439 
1440 		mutex_exit(&os->os_upgrade_lock);
1441 
1442 		status = os->os_upgrade_cb(os);
1443 
1444 		mutex_enter(&os->os_upgrade_lock);
1445 
1446 		os->os_upgrade_status = status;
1447 	}
1448 	os->os_upgrade_exit = B_TRUE;
1449 	os->os_upgrade_id = 0;
1450 	mutex_exit(&os->os_upgrade_lock);
1451 	dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
1452 }
1453 
1454 static void
1455 dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb)
1456 {
1457 	if (os->os_upgrade_id != 0)
1458 		return;
1459 
1460 	ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1461 	dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag);
1462 
1463 	mutex_enter(&os->os_upgrade_lock);
1464 	if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) {
1465 		os->os_upgrade_exit = B_FALSE;
1466 		os->os_upgrade_cb = cb;
1467 		os->os_upgrade_id = taskq_dispatch(
1468 		    os->os_spa->spa_upgrade_taskq,
1469 		    dmu_objset_upgrade_task_cb, os, TQ_SLEEP);
1470 		if (os->os_upgrade_id == TASKQID_INVALID) {
1471 			dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
1472 			os->os_upgrade_status = ENOMEM;
1473 		}
1474 	} else {
1475 		dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
1476 	}
1477 	mutex_exit(&os->os_upgrade_lock);
1478 }
1479 
1480 static void
1481 dmu_objset_upgrade_stop(objset_t *os)
1482 {
1483 	mutex_enter(&os->os_upgrade_lock);
1484 	os->os_upgrade_exit = B_TRUE;
1485 	if (os->os_upgrade_id != 0) {
1486 		taskqid_t id = os->os_upgrade_id;
1487 
1488 		os->os_upgrade_id = 0;
1489 		mutex_exit(&os->os_upgrade_lock);
1490 
1491 		if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) {
1492 			dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
1493 		}
1494 		txg_wait_synced(os->os_spa->spa_dsl_pool, 0);
1495 	} else {
1496 		mutex_exit(&os->os_upgrade_lock);
1497 	}
1498 }
1499 
1500 static void
1501 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx)
1502 {
1503 	dnode_t *dn;
1504 
1505 	while ((dn = multilist_sublist_head(list)) != NULL) {
1506 		ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1507 		ASSERT(dn->dn_dbuf->db_data_pending);
1508 		/*
1509 		 * Initialize dn_zio outside dnode_sync() because the
1510 		 * meta-dnode needs to set it outside dnode_sync().
1511 		 */
1512 		dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1513 		ASSERT(dn->dn_zio);
1514 
1515 		ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1516 		multilist_sublist_remove(list, dn);
1517 
1518 		/*
1519 		 * See the comment above dnode_rele_task() for an explanation
1520 		 * of why this dnode hold is always needed (even when not
1521 		 * doing user accounting).
1522 		 */
1523 		multilist_t *newlist = dn->dn_objset->os_synced_dnodes;
1524 		(void) dnode_add_ref(dn, newlist);
1525 		multilist_insert(newlist, dn);
1526 
1527 		dnode_sync(dn, tx);
1528 	}
1529 }
1530 
1531 /* ARGSUSED */
1532 static void
1533 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1534 {
1535 	blkptr_t *bp = zio->io_bp;
1536 	objset_t *os = arg;
1537 	dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1538 	uint64_t fill = 0;
1539 
1540 	ASSERT(!BP_IS_EMBEDDED(bp));
1541 	ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1542 	ASSERT0(BP_GET_LEVEL(bp));
1543 
1544 	/*
1545 	 * Update rootbp fill count: it should be the number of objects
1546 	 * allocated in the object set (not counting the "special"
1547 	 * objects that are stored in the objset_phys_t -- the meta
1548 	 * dnode and user/group/project accounting objects).
1549 	 */
1550 	for (int i = 0; i < dnp->dn_nblkptr; i++)
1551 		fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1552 
1553 	BP_SET_FILL(bp, fill);
1554 
1555 	if (os->os_dsl_dataset != NULL)
1556 		rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG);
1557 	*os->os_rootbp = *bp;
1558 	if (os->os_dsl_dataset != NULL)
1559 		rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG);
1560 }
1561 
1562 /* ARGSUSED */
1563 static void
1564 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1565 {
1566 	blkptr_t *bp = zio->io_bp;
1567 	blkptr_t *bp_orig = &zio->io_bp_orig;
1568 	objset_t *os = arg;
1569 
1570 	if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1571 		ASSERT(BP_EQUAL(bp, bp_orig));
1572 	} else {
1573 		dsl_dataset_t *ds = os->os_dsl_dataset;
1574 		dmu_tx_t *tx = os->os_synctx;
1575 
1576 		(void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1577 		dsl_dataset_block_born(ds, bp, tx);
1578 	}
1579 	kmem_free(bp, sizeof (*bp));
1580 }
1581 
1582 typedef struct sync_dnodes_arg {
1583 	multilist_t *sda_list;
1584 	int sda_sublist_idx;
1585 	multilist_t *sda_newlist;
1586 	dmu_tx_t *sda_tx;
1587 } sync_dnodes_arg_t;
1588 
1589 static void
1590 sync_dnodes_task(void *arg)
1591 {
1592 	sync_dnodes_arg_t *sda = arg;
1593 
1594 	multilist_sublist_t *ms =
1595 	    multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx);
1596 
1597 	dmu_objset_sync_dnodes(ms, sda->sda_tx);
1598 
1599 	multilist_sublist_unlock(ms);
1600 
1601 	kmem_free(sda, sizeof (*sda));
1602 }
1603 
1604 
1605 /* called from dsl */
1606 void
1607 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1608 {
1609 	int txgoff;
1610 	zbookmark_phys_t zb;
1611 	zio_prop_t zp;
1612 	zio_t *zio;
1613 	list_t *list;
1614 	dbuf_dirty_record_t *dr;
1615 	int num_sublists;
1616 	multilist_t *ml;
1617 	blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP);
1618 	*blkptr_copy = *os->os_rootbp;
1619 
1620 	dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1621 
1622 	ASSERT(dmu_tx_is_syncing(tx));
1623 	/* XXX the write_done callback should really give us the tx... */
1624 	os->os_synctx = tx;
1625 
1626 	if (os->os_dsl_dataset == NULL) {
1627 		/*
1628 		 * This is the MOS.  If we have upgraded,
1629 		 * spa_max_replication() could change, so reset
1630 		 * os_copies here.
1631 		 */
1632 		os->os_copies = spa_max_replication(os->os_spa);
1633 	}
1634 
1635 	/*
1636 	 * Create the root block IO
1637 	 */
1638 	SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1639 	    os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1640 	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1641 	arc_release(os->os_phys_buf, &os->os_phys_buf);
1642 
1643 	dmu_write_policy(os, NULL, 0, 0, &zp);
1644 
1645 	/*
1646 	 * If we are either claiming the ZIL or doing a raw receive, write
1647 	 * out the os_phys_buf raw. Neither of these actions will effect the
1648 	 * MAC at this point.
1649 	 */
1650 	if (os->os_raw_receive ||
1651 	    os->os_next_write_raw[tx->tx_txg & TXG_MASK]) {
1652 		ASSERT(os->os_encrypted);
1653 		arc_convert_to_raw(os->os_phys_buf,
1654 		    os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER,
1655 		    DMU_OT_OBJSET, NULL, NULL, NULL);
1656 	}
1657 
1658 	zio = arc_write(pio, os->os_spa, tx->tx_txg,
1659 	    blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1660 	    &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done,
1661 	    os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1662 
1663 	/*
1664 	 * Sync special dnodes - the parent IO for the sync is the root block
1665 	 */
1666 	DMU_META_DNODE(os)->dn_zio = zio;
1667 	dnode_sync(DMU_META_DNODE(os), tx);
1668 
1669 	os->os_phys->os_flags = os->os_flags;
1670 
1671 	if (DMU_USERUSED_DNODE(os) &&
1672 	    DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1673 		DMU_USERUSED_DNODE(os)->dn_zio = zio;
1674 		dnode_sync(DMU_USERUSED_DNODE(os), tx);
1675 		DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1676 		dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1677 	}
1678 
1679 	if (DMU_PROJECTUSED_DNODE(os) &&
1680 	    DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1681 		DMU_PROJECTUSED_DNODE(os)->dn_zio = zio;
1682 		dnode_sync(DMU_PROJECTUSED_DNODE(os), tx);
1683 	}
1684 
1685 	txgoff = tx->tx_txg & TXG_MASK;
1686 
1687 	/*
1688 	 * We must create the list here because it uses the
1689 	 * dn_dirty_link[] of this txg.  But it may already
1690 	 * exist because we call dsl_dataset_sync() twice per txg.
1691 	 */
1692 	if (os->os_synced_dnodes == NULL) {
1693 		os->os_synced_dnodes =
1694 		    multilist_create(sizeof (dnode_t),
1695 		    offsetof(dnode_t, dn_dirty_link[txgoff]),
1696 		    dnode_multilist_index_func);
1697 	} else {
1698 		ASSERT3U(os->os_synced_dnodes->ml_offset, ==,
1699 		    offsetof(dnode_t, dn_dirty_link[txgoff]));
1700 	}
1701 
1702 	ml = os->os_dirty_dnodes[txgoff];
1703 	num_sublists = multilist_get_num_sublists(ml);
1704 	for (int i = 0; i < num_sublists; i++) {
1705 		if (multilist_sublist_is_empty_idx(ml, i))
1706 			continue;
1707 		sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP);
1708 		sda->sda_list = ml;
1709 		sda->sda_sublist_idx = i;
1710 		sda->sda_tx = tx;
1711 		(void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1712 		    sync_dnodes_task, sda, 0);
1713 		/* callback frees sda */
1714 	}
1715 	taskq_wait(dmu_objset_pool(os)->dp_sync_taskq);
1716 
1717 	list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1718 	while ((dr = list_head(list)) != NULL) {
1719 		ASSERT0(dr->dr_dbuf->db_level);
1720 		list_remove(list, dr);
1721 		zio_nowait(dr->dr_zio);
1722 	}
1723 
1724 	/* Enable dnode backfill if enough objects have been freed. */
1725 	if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) {
1726 		os->os_rescan_dnodes = B_TRUE;
1727 		os->os_freed_dnodes = 0;
1728 	}
1729 
1730 	/*
1731 	 * Free intent log blocks up to this tx.
1732 	 */
1733 	zil_sync(os->os_zil, tx);
1734 	os->os_phys->os_zil_header = os->os_zil_header;
1735 	zio_nowait(zio);
1736 }
1737 
1738 boolean_t
1739 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1740 {
1741 	return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK]));
1742 }
1743 
1744 static file_info_cb_t *file_cbs[DMU_OST_NUMTYPES];
1745 
1746 void
1747 dmu_objset_register_type(dmu_objset_type_t ost, file_info_cb_t *cb)
1748 {
1749 	file_cbs[ost] = cb;
1750 }
1751 
1752 int
1753 dmu_get_file_info(objset_t *os, dmu_object_type_t bonustype, const void *data,
1754     zfs_file_info_t *zfi)
1755 {
1756 	file_info_cb_t *cb = file_cbs[os->os_phys->os_type];
1757 	if (cb == NULL)
1758 		return (EINVAL);
1759 	return (cb(bonustype, data, zfi));
1760 }
1761 
1762 boolean_t
1763 dmu_objset_userused_enabled(objset_t *os)
1764 {
1765 	return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1766 	    file_cbs[os->os_phys->os_type] != NULL &&
1767 	    DMU_USERUSED_DNODE(os) != NULL);
1768 }
1769 
1770 boolean_t
1771 dmu_objset_userobjused_enabled(objset_t *os)
1772 {
1773 	return (dmu_objset_userused_enabled(os) &&
1774 	    spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING));
1775 }
1776 
1777 boolean_t
1778 dmu_objset_projectquota_enabled(objset_t *os)
1779 {
1780 	return (file_cbs[os->os_phys->os_type] != NULL &&
1781 	    DMU_PROJECTUSED_DNODE(os) != NULL &&
1782 	    spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA));
1783 }
1784 
1785 typedef struct userquota_node {
1786 	/* must be in the first filed, see userquota_update_cache() */
1787 	char		uqn_id[20 + DMU_OBJACCT_PREFIX_LEN];
1788 	int64_t		uqn_delta;
1789 	avl_node_t	uqn_node;
1790 } userquota_node_t;
1791 
1792 typedef struct userquota_cache {
1793 	avl_tree_t uqc_user_deltas;
1794 	avl_tree_t uqc_group_deltas;
1795 	avl_tree_t uqc_project_deltas;
1796 } userquota_cache_t;
1797 
1798 static int
1799 userquota_compare(const void *l, const void *r)
1800 {
1801 	const userquota_node_t *luqn = l;
1802 	const userquota_node_t *ruqn = r;
1803 	int rv;
1804 
1805 	/*
1806 	 * NB: can only access uqn_id because userquota_update_cache() doesn't
1807 	 * pass in an entire userquota_node_t.
1808 	 */
1809 	rv = strcmp(luqn->uqn_id, ruqn->uqn_id);
1810 
1811 	return (TREE_ISIGN(rv));
1812 }
1813 
1814 static void
1815 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx)
1816 {
1817 	void *cookie;
1818 	userquota_node_t *uqn;
1819 
1820 	ASSERT(dmu_tx_is_syncing(tx));
1821 
1822 	cookie = NULL;
1823 	while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas,
1824 	    &cookie)) != NULL) {
1825 		/*
1826 		 * os_userused_lock protects against concurrent calls to
1827 		 * zap_increment_int().  It's needed because zap_increment_int()
1828 		 * is not thread-safe (i.e. not atomic).
1829 		 */
1830 		mutex_enter(&os->os_userused_lock);
1831 		VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT,
1832 		    uqn->uqn_id, uqn->uqn_delta, tx));
1833 		mutex_exit(&os->os_userused_lock);
1834 		kmem_free(uqn, sizeof (*uqn));
1835 	}
1836 	avl_destroy(&cache->uqc_user_deltas);
1837 
1838 	cookie = NULL;
1839 	while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas,
1840 	    &cookie)) != NULL) {
1841 		mutex_enter(&os->os_userused_lock);
1842 		VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT,
1843 		    uqn->uqn_id, uqn->uqn_delta, tx));
1844 		mutex_exit(&os->os_userused_lock);
1845 		kmem_free(uqn, sizeof (*uqn));
1846 	}
1847 	avl_destroy(&cache->uqc_group_deltas);
1848 
1849 	if (dmu_objset_projectquota_enabled(os)) {
1850 		cookie = NULL;
1851 		while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas,
1852 		    &cookie)) != NULL) {
1853 			mutex_enter(&os->os_userused_lock);
1854 			VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT,
1855 			    uqn->uqn_id, uqn->uqn_delta, tx));
1856 			mutex_exit(&os->os_userused_lock);
1857 			kmem_free(uqn, sizeof (*uqn));
1858 		}
1859 		avl_destroy(&cache->uqc_project_deltas);
1860 	}
1861 }
1862 
1863 static void
1864 userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta)
1865 {
1866 	userquota_node_t *uqn;
1867 	avl_index_t idx;
1868 
1869 	ASSERT(strlen(id) < sizeof (uqn->uqn_id));
1870 	/*
1871 	 * Use id directly for searching because uqn_id is the first field of
1872 	 * userquota_node_t and fields after uqn_id won't be accessed in
1873 	 * avl_find().
1874 	 */
1875 	uqn = avl_find(avl, (const void *)id, &idx);
1876 	if (uqn == NULL) {
1877 		uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP);
1878 		strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id));
1879 		avl_insert(avl, uqn, idx);
1880 	}
1881 	uqn->uqn_delta += delta;
1882 }
1883 
1884 static void
1885 do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used,
1886     uint64_t flags, uint64_t user, uint64_t group, uint64_t project,
1887     boolean_t subtract)
1888 {
1889 	if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) {
1890 		int64_t delta = DNODE_MIN_SIZE + used;
1891 		char name[20];
1892 
1893 		if (subtract)
1894 			delta = -delta;
1895 
1896 		(void) snprintf(name, sizeof (name), "%llx", (longlong_t)user);
1897 		userquota_update_cache(&cache->uqc_user_deltas, name, delta);
1898 
1899 		(void) snprintf(name, sizeof (name), "%llx", (longlong_t)group);
1900 		userquota_update_cache(&cache->uqc_group_deltas, name, delta);
1901 
1902 		if (dmu_objset_projectquota_enabled(os)) {
1903 			(void) snprintf(name, sizeof (name), "%llx",
1904 			    (longlong_t)project);
1905 			userquota_update_cache(&cache->uqc_project_deltas,
1906 			    name, delta);
1907 		}
1908 	}
1909 }
1910 
1911 static void
1912 do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags,
1913     uint64_t user, uint64_t group, uint64_t project, boolean_t subtract)
1914 {
1915 	if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) {
1916 		char name[20 + DMU_OBJACCT_PREFIX_LEN];
1917 		int delta = subtract ? -1 : 1;
1918 
1919 		(void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx",
1920 		    (longlong_t)user);
1921 		userquota_update_cache(&cache->uqc_user_deltas, name, delta);
1922 
1923 		(void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx",
1924 		    (longlong_t)group);
1925 		userquota_update_cache(&cache->uqc_group_deltas, name, delta);
1926 
1927 		if (dmu_objset_projectquota_enabled(os)) {
1928 			(void) snprintf(name, sizeof (name),
1929 			    DMU_OBJACCT_PREFIX "%llx", (longlong_t)project);
1930 			userquota_update_cache(&cache->uqc_project_deltas,
1931 			    name, delta);
1932 		}
1933 	}
1934 }
1935 
1936 typedef struct userquota_updates_arg {
1937 	objset_t *uua_os;
1938 	int uua_sublist_idx;
1939 	dmu_tx_t *uua_tx;
1940 } userquota_updates_arg_t;
1941 
1942 static void
1943 userquota_updates_task(void *arg)
1944 {
1945 	userquota_updates_arg_t *uua = arg;
1946 	objset_t *os = uua->uua_os;
1947 	dmu_tx_t *tx = uua->uua_tx;
1948 	dnode_t *dn;
1949 	userquota_cache_t cache = { { 0 } };
1950 
1951 	multilist_sublist_t *list =
1952 	    multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx);
1953 
1954 	ASSERT(multilist_sublist_head(list) == NULL ||
1955 	    dmu_objset_userused_enabled(os));
1956 	avl_create(&cache.uqc_user_deltas, userquota_compare,
1957 	    sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1958 	avl_create(&cache.uqc_group_deltas, userquota_compare,
1959 	    sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1960 	if (dmu_objset_projectquota_enabled(os))
1961 		avl_create(&cache.uqc_project_deltas, userquota_compare,
1962 		    sizeof (userquota_node_t), offsetof(userquota_node_t,
1963 		    uqn_node));
1964 
1965 	while ((dn = multilist_sublist_head(list)) != NULL) {
1966 		int flags;
1967 		ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1968 		ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1969 		    dn->dn_phys->dn_flags &
1970 		    DNODE_FLAG_USERUSED_ACCOUNTED);
1971 
1972 		flags = dn->dn_id_flags;
1973 		ASSERT(flags);
1974 		if (flags & DN_ID_OLD_EXIST)  {
1975 			do_userquota_update(os, &cache, dn->dn_oldused,
1976 			    dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid,
1977 			    dn->dn_oldprojid, B_TRUE);
1978 			do_userobjquota_update(os, &cache, dn->dn_oldflags,
1979 			    dn->dn_olduid, dn->dn_oldgid,
1980 			    dn->dn_oldprojid, B_TRUE);
1981 		}
1982 		if (flags & DN_ID_NEW_EXIST) {
1983 			do_userquota_update(os, &cache,
1984 			    DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags,
1985 			    dn->dn_newuid, dn->dn_newgid,
1986 			    dn->dn_newprojid, B_FALSE);
1987 			do_userobjquota_update(os, &cache,
1988 			    dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid,
1989 			    dn->dn_newprojid, B_FALSE);
1990 		}
1991 
1992 		mutex_enter(&dn->dn_mtx);
1993 		dn->dn_oldused = 0;
1994 		dn->dn_oldflags = 0;
1995 		if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1996 			dn->dn_olduid = dn->dn_newuid;
1997 			dn->dn_oldgid = dn->dn_newgid;
1998 			dn->dn_oldprojid = dn->dn_newprojid;
1999 			dn->dn_id_flags |= DN_ID_OLD_EXIST;
2000 			if (dn->dn_bonuslen == 0)
2001 				dn->dn_id_flags |= DN_ID_CHKED_SPILL;
2002 			else
2003 				dn->dn_id_flags |= DN_ID_CHKED_BONUS;
2004 		}
2005 		dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
2006 		mutex_exit(&dn->dn_mtx);
2007 
2008 		multilist_sublist_remove(list, dn);
2009 		dnode_rele(dn, os->os_synced_dnodes);
2010 	}
2011 	do_userquota_cacheflush(os, &cache, tx);
2012 	multilist_sublist_unlock(list);
2013 	kmem_free(uua, sizeof (*uua));
2014 }
2015 
2016 /*
2017  * Release dnode holds from dmu_objset_sync_dnodes().  When the dnode is being
2018  * synced (i.e. we have issued the zio's for blocks in the dnode), it can't be
2019  * evicted because the block containing the dnode can't be evicted until it is
2020  * written out.  However, this hold is necessary to prevent the dnode_t from
2021  * being moved (via dnode_move()) while it's still referenced by
2022  * dbuf_dirty_record_t:dr_dnode.  And dr_dnode is needed for
2023  * dirty_lightweight_leaf-type dirty records.
2024  *
2025  * If we are doing user-object accounting, the dnode_rele() happens from
2026  * userquota_updates_task() instead.
2027  */
2028 static void
2029 dnode_rele_task(void *arg)
2030 {
2031 	userquota_updates_arg_t *uua = arg;
2032 	objset_t *os = uua->uua_os;
2033 
2034 	multilist_sublist_t *list =
2035 	    multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx);
2036 
2037 	dnode_t *dn;
2038 	while ((dn = multilist_sublist_head(list)) != NULL) {
2039 		multilist_sublist_remove(list, dn);
2040 		dnode_rele(dn, os->os_synced_dnodes);
2041 	}
2042 	multilist_sublist_unlock(list);
2043 	kmem_free(uua, sizeof (*uua));
2044 }
2045 
2046 /*
2047  * Return TRUE if userquota updates are needed.
2048  */
2049 static boolean_t
2050 dmu_objset_do_userquota_updates_prep(objset_t *os, dmu_tx_t *tx)
2051 {
2052 	if (!dmu_objset_userused_enabled(os))
2053 		return (B_FALSE);
2054 
2055 	/*
2056 	 * If this is a raw receive just return and handle accounting
2057 	 * later when we have the keys loaded. We also don't do user
2058 	 * accounting during claiming since the datasets are not owned
2059 	 * for the duration of claiming and this txg should only be
2060 	 * used for recovery.
2061 	 */
2062 	if (os->os_encrypted && dmu_objset_is_receiving(os))
2063 		return (B_FALSE);
2064 
2065 	if (tx->tx_txg <= os->os_spa->spa_claim_max_txg)
2066 		return (B_FALSE);
2067 
2068 	/* Allocate the user/group/project used objects if necessary. */
2069 	if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
2070 		VERIFY0(zap_create_claim(os,
2071 		    DMU_USERUSED_OBJECT,
2072 		    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
2073 		VERIFY0(zap_create_claim(os,
2074 		    DMU_GROUPUSED_OBJECT,
2075 		    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
2076 	}
2077 
2078 	if (dmu_objset_projectquota_enabled(os) &&
2079 	    DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
2080 		VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT,
2081 		    DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
2082 	}
2083 	return (B_TRUE);
2084 }
2085 
2086 /*
2087  * Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and
2088  * also release the holds on the dnodes from dmu_objset_sync_dnodes().
2089  * The caller must taskq_wait(dp_sync_taskq).
2090  */
2091 void
2092 dmu_objset_sync_done(objset_t *os, dmu_tx_t *tx)
2093 {
2094 	boolean_t need_userquota = dmu_objset_do_userquota_updates_prep(os, tx);
2095 
2096 	int num_sublists = multilist_get_num_sublists(os->os_synced_dnodes);
2097 	for (int i = 0; i < num_sublists; i++) {
2098 		userquota_updates_arg_t *uua =
2099 		    kmem_alloc(sizeof (*uua), KM_SLEEP);
2100 		uua->uua_os = os;
2101 		uua->uua_sublist_idx = i;
2102 		uua->uua_tx = tx;
2103 
2104 		/*
2105 		 * If we don't need to update userquotas, use
2106 		 * dnode_rele_task() to call dnode_rele()
2107 		 */
2108 		(void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
2109 		    need_userquota ? userquota_updates_task : dnode_rele_task,
2110 		    uua, 0);
2111 		/* callback frees uua */
2112 	}
2113 }
2114 
2115 
2116 /*
2117  * Returns a pointer to data to find uid/gid from
2118  *
2119  * If a dirty record for transaction group that is syncing can't
2120  * be found then NULL is returned.  In the NULL case it is assumed
2121  * the uid/gid aren't changing.
2122  */
2123 static void *
2124 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
2125 {
2126 	dbuf_dirty_record_t *dr;
2127 	void *data;
2128 
2129 	if (db->db_dirtycnt == 0)
2130 		return (db->db.db_data);  /* Nothing is changing */
2131 
2132 	dr = dbuf_find_dirty_eq(db, tx->tx_txg);
2133 
2134 	if (dr == NULL) {
2135 		data = NULL;
2136 	} else {
2137 		if (dr->dr_dnode->dn_bonuslen == 0 &&
2138 		    dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
2139 			data = dr->dt.dl.dr_data->b_data;
2140 		else
2141 			data = dr->dt.dl.dr_data;
2142 	}
2143 
2144 	return (data);
2145 }
2146 
2147 void
2148 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
2149 {
2150 	objset_t *os = dn->dn_objset;
2151 	void *data = NULL;
2152 	dmu_buf_impl_t *db = NULL;
2153 	int flags = dn->dn_id_flags;
2154 	int error;
2155 	boolean_t have_spill = B_FALSE;
2156 
2157 	if (!dmu_objset_userused_enabled(dn->dn_objset))
2158 		return;
2159 
2160 	/*
2161 	 * Raw receives introduce a problem with user accounting. Raw
2162 	 * receives cannot update the user accounting info because the
2163 	 * user ids and the sizes are encrypted. To guarantee that we
2164 	 * never end up with bad user accounting, we simply disable it
2165 	 * during raw receives. We also disable this for normal receives
2166 	 * so that an incremental raw receive may be done on top of an
2167 	 * existing non-raw receive.
2168 	 */
2169 	if (os->os_encrypted && dmu_objset_is_receiving(os))
2170 		return;
2171 
2172 	if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
2173 	    DN_ID_CHKED_SPILL)))
2174 		return;
2175 
2176 	if (before && dn->dn_bonuslen != 0)
2177 		data = DN_BONUS(dn->dn_phys);
2178 	else if (!before && dn->dn_bonuslen != 0) {
2179 		if (dn->dn_bonus) {
2180 			db = dn->dn_bonus;
2181 			mutex_enter(&db->db_mtx);
2182 			data = dmu_objset_userquota_find_data(db, tx);
2183 		} else {
2184 			data = DN_BONUS(dn->dn_phys);
2185 		}
2186 	} else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
2187 			int rf = 0;
2188 
2189 			if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
2190 				rf |= DB_RF_HAVESTRUCT;
2191 			error = dmu_spill_hold_by_dnode(dn,
2192 			    rf | DB_RF_MUST_SUCCEED,
2193 			    FTAG, (dmu_buf_t **)&db);
2194 			ASSERT(error == 0);
2195 			mutex_enter(&db->db_mtx);
2196 			data = (before) ? db->db.db_data :
2197 			    dmu_objset_userquota_find_data(db, tx);
2198 			have_spill = B_TRUE;
2199 	} else {
2200 		mutex_enter(&dn->dn_mtx);
2201 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
2202 		mutex_exit(&dn->dn_mtx);
2203 		return;
2204 	}
2205 
2206 	/*
2207 	 * Must always call the callback in case the object
2208 	 * type has changed and that type isn't an object type to track
2209 	 */
2210 	zfs_file_info_t zfi;
2211 	error = file_cbs[os->os_phys->os_type](dn->dn_bonustype, data, &zfi);
2212 
2213 	if (before) {
2214 		ASSERT(data);
2215 		dn->dn_olduid = zfi.zfi_user;
2216 		dn->dn_oldgid = zfi.zfi_group;
2217 		dn->dn_oldprojid = zfi.zfi_project;
2218 	} else if (data) {
2219 		dn->dn_newuid = zfi.zfi_user;
2220 		dn->dn_newgid = zfi.zfi_group;
2221 		dn->dn_newprojid = zfi.zfi_project;
2222 	}
2223 
2224 	/*
2225 	 * Preserve existing uid/gid when the callback can't determine
2226 	 * what the new uid/gid are and the callback returned EEXIST.
2227 	 * The EEXIST error tells us to just use the existing uid/gid.
2228 	 * If we don't know what the old values are then just assign
2229 	 * them to 0, since that is a new file  being created.
2230 	 */
2231 	if (!before && data == NULL && error == EEXIST) {
2232 		if (flags & DN_ID_OLD_EXIST) {
2233 			dn->dn_newuid = dn->dn_olduid;
2234 			dn->dn_newgid = dn->dn_oldgid;
2235 			dn->dn_newprojid = dn->dn_oldprojid;
2236 		} else {
2237 			dn->dn_newuid = 0;
2238 			dn->dn_newgid = 0;
2239 			dn->dn_newprojid = ZFS_DEFAULT_PROJID;
2240 		}
2241 		error = 0;
2242 	}
2243 
2244 	if (db)
2245 		mutex_exit(&db->db_mtx);
2246 
2247 	mutex_enter(&dn->dn_mtx);
2248 	if (error == 0 && before)
2249 		dn->dn_id_flags |= DN_ID_OLD_EXIST;
2250 	if (error == 0 && !before)
2251 		dn->dn_id_flags |= DN_ID_NEW_EXIST;
2252 
2253 	if (have_spill) {
2254 		dn->dn_id_flags |= DN_ID_CHKED_SPILL;
2255 	} else {
2256 		dn->dn_id_flags |= DN_ID_CHKED_BONUS;
2257 	}
2258 	mutex_exit(&dn->dn_mtx);
2259 	if (have_spill)
2260 		dmu_buf_rele((dmu_buf_t *)db, FTAG);
2261 }
2262 
2263 boolean_t
2264 dmu_objset_userspace_present(objset_t *os)
2265 {
2266 	return (os->os_phys->os_flags &
2267 	    OBJSET_FLAG_USERACCOUNTING_COMPLETE);
2268 }
2269 
2270 boolean_t
2271 dmu_objset_userobjspace_present(objset_t *os)
2272 {
2273 	return (os->os_phys->os_flags &
2274 	    OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE);
2275 }
2276 
2277 boolean_t
2278 dmu_objset_projectquota_present(objset_t *os)
2279 {
2280 	return (os->os_phys->os_flags &
2281 	    OBJSET_FLAG_PROJECTQUOTA_COMPLETE);
2282 }
2283 
2284 static int
2285 dmu_objset_space_upgrade(objset_t *os)
2286 {
2287 	uint64_t obj;
2288 	int err = 0;
2289 
2290 	/*
2291 	 * We simply need to mark every object dirty, so that it will be
2292 	 * synced out and now accounted.  If this is called
2293 	 * concurrently, or if we already did some work before crashing,
2294 	 * that's fine, since we track each object's accounted state
2295 	 * independently.
2296 	 */
2297 
2298 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
2299 		dmu_tx_t *tx;
2300 		dmu_buf_t *db;
2301 		int objerr;
2302 
2303 		mutex_enter(&os->os_upgrade_lock);
2304 		if (os->os_upgrade_exit)
2305 			err = SET_ERROR(EINTR);
2306 		mutex_exit(&os->os_upgrade_lock);
2307 		if (err != 0)
2308 			return (err);
2309 
2310 		if (issig(JUSTLOOKING) && issig(FORREAL))
2311 			return (SET_ERROR(EINTR));
2312 
2313 		objerr = dmu_bonus_hold(os, obj, FTAG, &db);
2314 		if (objerr != 0)
2315 			continue;
2316 		tx = dmu_tx_create(os);
2317 		dmu_tx_hold_bonus(tx, obj);
2318 		objerr = dmu_tx_assign(tx, TXG_WAIT);
2319 		if (objerr != 0) {
2320 			dmu_buf_rele(db, FTAG);
2321 			dmu_tx_abort(tx);
2322 			continue;
2323 		}
2324 		dmu_buf_will_dirty(db, tx);
2325 		dmu_buf_rele(db, FTAG);
2326 		dmu_tx_commit(tx);
2327 	}
2328 	return (0);
2329 }
2330 
2331 static int
2332 dmu_objset_userspace_upgrade_cb(objset_t *os)
2333 {
2334 	int err = 0;
2335 
2336 	if (dmu_objset_userspace_present(os))
2337 		return (0);
2338 	if (dmu_objset_is_snapshot(os))
2339 		return (SET_ERROR(EINVAL));
2340 	if (!dmu_objset_userused_enabled(os))
2341 		return (SET_ERROR(ENOTSUP));
2342 
2343 	err = dmu_objset_space_upgrade(os);
2344 	if (err)
2345 		return (err);
2346 
2347 	os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
2348 	txg_wait_synced(dmu_objset_pool(os), 0);
2349 	return (0);
2350 }
2351 
2352 void
2353 dmu_objset_userspace_upgrade(objset_t *os)
2354 {
2355 	dmu_objset_upgrade(os, dmu_objset_userspace_upgrade_cb);
2356 }
2357 
2358 static int
2359 dmu_objset_id_quota_upgrade_cb(objset_t *os)
2360 {
2361 	int err = 0;
2362 
2363 	if (dmu_objset_userobjspace_present(os) &&
2364 	    dmu_objset_projectquota_present(os))
2365 		return (0);
2366 	if (dmu_objset_is_snapshot(os))
2367 		return (SET_ERROR(EINVAL));
2368 	if (!dmu_objset_userused_enabled(os))
2369 		return (SET_ERROR(ENOTSUP));
2370 	if (!dmu_objset_projectquota_enabled(os) &&
2371 	    dmu_objset_userobjspace_present(os))
2372 		return (SET_ERROR(ENOTSUP));
2373 
2374 	if (dmu_objset_userobjused_enabled(os))
2375 		dmu_objset_ds(os)->ds_feature_activation[
2376 		    SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE;
2377 	if (dmu_objset_projectquota_enabled(os))
2378 		dmu_objset_ds(os)->ds_feature_activation[
2379 		    SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE;
2380 
2381 	err = dmu_objset_space_upgrade(os);
2382 	if (err)
2383 		return (err);
2384 
2385 	os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
2386 	if (dmu_objset_userobjused_enabled(os))
2387 		os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE;
2388 	if (dmu_objset_projectquota_enabled(os))
2389 		os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE;
2390 
2391 	txg_wait_synced(dmu_objset_pool(os), 0);
2392 	return (0);
2393 }
2394 
2395 void
2396 dmu_objset_id_quota_upgrade(objset_t *os)
2397 {
2398 	dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb);
2399 }
2400 
2401 boolean_t
2402 dmu_objset_userobjspace_upgradable(objset_t *os)
2403 {
2404 	return (dmu_objset_type(os) == DMU_OST_ZFS &&
2405 	    !dmu_objset_is_snapshot(os) &&
2406 	    dmu_objset_userobjused_enabled(os) &&
2407 	    !dmu_objset_userobjspace_present(os) &&
2408 	    spa_writeable(dmu_objset_spa(os)));
2409 }
2410 
2411 boolean_t
2412 dmu_objset_projectquota_upgradable(objset_t *os)
2413 {
2414 	return (dmu_objset_type(os) == DMU_OST_ZFS &&
2415 	    !dmu_objset_is_snapshot(os) &&
2416 	    dmu_objset_projectquota_enabled(os) &&
2417 	    !dmu_objset_projectquota_present(os) &&
2418 	    spa_writeable(dmu_objset_spa(os)));
2419 }
2420 
2421 void
2422 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
2423     uint64_t *usedobjsp, uint64_t *availobjsp)
2424 {
2425 	dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
2426 	    usedobjsp, availobjsp);
2427 }
2428 
2429 uint64_t
2430 dmu_objset_fsid_guid(objset_t *os)
2431 {
2432 	return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
2433 }
2434 
2435 void
2436 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
2437 {
2438 	stat->dds_type = os->os_phys->os_type;
2439 	if (os->os_dsl_dataset)
2440 		dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
2441 }
2442 
2443 void
2444 dmu_objset_stats(objset_t *os, nvlist_t *nv)
2445 {
2446 	ASSERT(os->os_dsl_dataset ||
2447 	    os->os_phys->os_type == DMU_OST_META);
2448 
2449 	if (os->os_dsl_dataset != NULL)
2450 		dsl_dataset_stats(os->os_dsl_dataset, nv);
2451 
2452 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
2453 	    os->os_phys->os_type);
2454 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
2455 	    dmu_objset_userspace_present(os));
2456 }
2457 
2458 int
2459 dmu_objset_is_snapshot(objset_t *os)
2460 {
2461 	if (os->os_dsl_dataset != NULL)
2462 		return (os->os_dsl_dataset->ds_is_snapshot);
2463 	else
2464 		return (B_FALSE);
2465 }
2466 
2467 int
2468 dmu_snapshot_realname(objset_t *os, const char *name, char *real, int maxlen,
2469     boolean_t *conflict)
2470 {
2471 	dsl_dataset_t *ds = os->os_dsl_dataset;
2472 	uint64_t ignored;
2473 
2474 	if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
2475 		return (SET_ERROR(ENOENT));
2476 
2477 	return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
2478 	    dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
2479 	    MT_NORMALIZE, real, maxlen, conflict));
2480 }
2481 
2482 int
2483 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
2484     uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
2485 {
2486 	dsl_dataset_t *ds = os->os_dsl_dataset;
2487 	zap_cursor_t cursor;
2488 	zap_attribute_t attr;
2489 
2490 	ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
2491 
2492 	if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
2493 		return (SET_ERROR(ENOENT));
2494 
2495 	zap_cursor_init_serialized(&cursor,
2496 	    ds->ds_dir->dd_pool->dp_meta_objset,
2497 	    dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
2498 
2499 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
2500 		zap_cursor_fini(&cursor);
2501 		return (SET_ERROR(ENOENT));
2502 	}
2503 
2504 	if (strlen(attr.za_name) + 1 > namelen) {
2505 		zap_cursor_fini(&cursor);
2506 		return (SET_ERROR(ENAMETOOLONG));
2507 	}
2508 
2509 	(void) strlcpy(name, attr.za_name, namelen);
2510 	if (idp)
2511 		*idp = attr.za_first_integer;
2512 	if (case_conflict)
2513 		*case_conflict = attr.za_normalization_conflict;
2514 	zap_cursor_advance(&cursor);
2515 	*offp = zap_cursor_serialize(&cursor);
2516 	zap_cursor_fini(&cursor);
2517 
2518 	return (0);
2519 }
2520 
2521 int
2522 dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value)
2523 {
2524 	return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value));
2525 }
2526 
2527 int
2528 dmu_dir_list_next(objset_t *os, int namelen, char *name,
2529     uint64_t *idp, uint64_t *offp)
2530 {
2531 	dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
2532 	zap_cursor_t cursor;
2533 	zap_attribute_t attr;
2534 
2535 	/* there is no next dir on a snapshot! */
2536 	if (os->os_dsl_dataset->ds_object !=
2537 	    dsl_dir_phys(dd)->dd_head_dataset_obj)
2538 		return (SET_ERROR(ENOENT));
2539 
2540 	zap_cursor_init_serialized(&cursor,
2541 	    dd->dd_pool->dp_meta_objset,
2542 	    dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
2543 
2544 	if (zap_cursor_retrieve(&cursor, &attr) != 0) {
2545 		zap_cursor_fini(&cursor);
2546 		return (SET_ERROR(ENOENT));
2547 	}
2548 
2549 	if (strlen(attr.za_name) + 1 > namelen) {
2550 		zap_cursor_fini(&cursor);
2551 		return (SET_ERROR(ENAMETOOLONG));
2552 	}
2553 
2554 	(void) strlcpy(name, attr.za_name, namelen);
2555 	if (idp)
2556 		*idp = attr.za_first_integer;
2557 	zap_cursor_advance(&cursor);
2558 	*offp = zap_cursor_serialize(&cursor);
2559 	zap_cursor_fini(&cursor);
2560 
2561 	return (0);
2562 }
2563 
2564 typedef struct dmu_objset_find_ctx {
2565 	taskq_t		*dc_tq;
2566 	dsl_pool_t	*dc_dp;
2567 	uint64_t	dc_ddobj;
2568 	char		*dc_ddname; /* last component of ddobj's name */
2569 	int		(*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
2570 	void		*dc_arg;
2571 	int		dc_flags;
2572 	kmutex_t	*dc_error_lock;
2573 	int		*dc_error;
2574 } dmu_objset_find_ctx_t;
2575 
2576 static void
2577 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
2578 {
2579 	dsl_pool_t *dp = dcp->dc_dp;
2580 	dsl_dir_t *dd;
2581 	dsl_dataset_t *ds;
2582 	zap_cursor_t zc;
2583 	zap_attribute_t *attr;
2584 	uint64_t thisobj;
2585 	int err = 0;
2586 
2587 	/* don't process if there already was an error */
2588 	if (*dcp->dc_error != 0)
2589 		goto out;
2590 
2591 	/*
2592 	 * Note: passing the name (dc_ddname) here is optional, but it
2593 	 * improves performance because we don't need to call
2594 	 * zap_value_search() to determine the name.
2595 	 */
2596 	err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd);
2597 	if (err != 0)
2598 		goto out;
2599 
2600 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2601 	if (dd->dd_myname[0] == '$') {
2602 		dsl_dir_rele(dd, FTAG);
2603 		goto out;
2604 	}
2605 
2606 	thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2607 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2608 
2609 	/*
2610 	 * Iterate over all children.
2611 	 */
2612 	if (dcp->dc_flags & DS_FIND_CHILDREN) {
2613 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
2614 		    dsl_dir_phys(dd)->dd_child_dir_zapobj);
2615 		    zap_cursor_retrieve(&zc, attr) == 0;
2616 		    (void) zap_cursor_advance(&zc)) {
2617 			ASSERT3U(attr->za_integer_length, ==,
2618 			    sizeof (uint64_t));
2619 			ASSERT3U(attr->za_num_integers, ==, 1);
2620 
2621 			dmu_objset_find_ctx_t *child_dcp =
2622 			    kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
2623 			*child_dcp = *dcp;
2624 			child_dcp->dc_ddobj = attr->za_first_integer;
2625 			child_dcp->dc_ddname = spa_strdup(attr->za_name);
2626 			if (dcp->dc_tq != NULL)
2627 				(void) taskq_dispatch(dcp->dc_tq,
2628 				    dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
2629 			else
2630 				dmu_objset_find_dp_impl(child_dcp);
2631 		}
2632 		zap_cursor_fini(&zc);
2633 	}
2634 
2635 	/*
2636 	 * Iterate over all snapshots.
2637 	 */
2638 	if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
2639 		dsl_dataset_t *ds;
2640 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2641 
2642 		if (err == 0) {
2643 			uint64_t snapobj;
2644 
2645 			snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2646 			dsl_dataset_rele(ds, FTAG);
2647 
2648 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2649 			    zap_cursor_retrieve(&zc, attr) == 0;
2650 			    (void) zap_cursor_advance(&zc)) {
2651 				ASSERT3U(attr->za_integer_length, ==,
2652 				    sizeof (uint64_t));
2653 				ASSERT3U(attr->za_num_integers, ==, 1);
2654 
2655 				err = dsl_dataset_hold_obj(dp,
2656 				    attr->za_first_integer, FTAG, &ds);
2657 				if (err != 0)
2658 					break;
2659 				err = dcp->dc_func(dp, ds, dcp->dc_arg);
2660 				dsl_dataset_rele(ds, FTAG);
2661 				if (err != 0)
2662 					break;
2663 			}
2664 			zap_cursor_fini(&zc);
2665 		}
2666 	}
2667 
2668 	kmem_free(attr, sizeof (zap_attribute_t));
2669 
2670 	if (err != 0) {
2671 		dsl_dir_rele(dd, FTAG);
2672 		goto out;
2673 	}
2674 
2675 	/*
2676 	 * Apply to self.
2677 	 */
2678 	err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2679 
2680 	/*
2681 	 * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2682 	 * that the dir will remain cached, and we won't have to re-instantiate
2683 	 * it (which could be expensive due to finding its name via
2684 	 * zap_value_search()).
2685 	 */
2686 	dsl_dir_rele(dd, FTAG);
2687 	if (err != 0)
2688 		goto out;
2689 	err = dcp->dc_func(dp, ds, dcp->dc_arg);
2690 	dsl_dataset_rele(ds, FTAG);
2691 
2692 out:
2693 	if (err != 0) {
2694 		mutex_enter(dcp->dc_error_lock);
2695 		/* only keep first error */
2696 		if (*dcp->dc_error == 0)
2697 			*dcp->dc_error = err;
2698 		mutex_exit(dcp->dc_error_lock);
2699 	}
2700 
2701 	if (dcp->dc_ddname != NULL)
2702 		spa_strfree(dcp->dc_ddname);
2703 	kmem_free(dcp, sizeof (*dcp));
2704 }
2705 
2706 static void
2707 dmu_objset_find_dp_cb(void *arg)
2708 {
2709 	dmu_objset_find_ctx_t *dcp = arg;
2710 	dsl_pool_t *dp = dcp->dc_dp;
2711 
2712 	/*
2713 	 * We need to get a pool_config_lock here, as there are several
2714 	 * assert(pool_config_held) down the stack. Getting a lock via
2715 	 * dsl_pool_config_enter is risky, as it might be stalled by a
2716 	 * pending writer. This would deadlock, as the write lock can
2717 	 * only be granted when our parent thread gives up the lock.
2718 	 * The _prio interface gives us priority over a pending writer.
2719 	 */
2720 	dsl_pool_config_enter_prio(dp, FTAG);
2721 
2722 	dmu_objset_find_dp_impl(dcp);
2723 
2724 	dsl_pool_config_exit(dp, FTAG);
2725 }
2726 
2727 /*
2728  * Find objsets under and including ddobj, call func(ds) on each.
2729  * The order for the enumeration is completely undefined.
2730  * func is called with dsl_pool_config held.
2731  */
2732 int
2733 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
2734     int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
2735 {
2736 	int error = 0;
2737 	taskq_t *tq = NULL;
2738 	int ntasks;
2739 	dmu_objset_find_ctx_t *dcp;
2740 	kmutex_t err_lock;
2741 
2742 	mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
2743 	dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
2744 	dcp->dc_tq = NULL;
2745 	dcp->dc_dp = dp;
2746 	dcp->dc_ddobj = ddobj;
2747 	dcp->dc_ddname = NULL;
2748 	dcp->dc_func = func;
2749 	dcp->dc_arg = arg;
2750 	dcp->dc_flags = flags;
2751 	dcp->dc_error_lock = &err_lock;
2752 	dcp->dc_error = &error;
2753 
2754 	if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
2755 		/*
2756 		 * In case a write lock is held we can't make use of
2757 		 * parallelism, as down the stack of the worker threads
2758 		 * the lock is asserted via dsl_pool_config_held.
2759 		 * In case of a read lock this is solved by getting a read
2760 		 * lock in each worker thread, which isn't possible in case
2761 		 * of a writer lock. So we fall back to the synchronous path
2762 		 * here.
2763 		 * In the future it might be possible to get some magic into
2764 		 * dsl_pool_config_held in a way that it returns true for
2765 		 * the worker threads so that a single lock held from this
2766 		 * thread suffices. For now, stay single threaded.
2767 		 */
2768 		dmu_objset_find_dp_impl(dcp);
2769 		mutex_destroy(&err_lock);
2770 
2771 		return (error);
2772 	}
2773 
2774 	ntasks = dmu_find_threads;
2775 	if (ntasks == 0)
2776 		ntasks = vdev_count_leaves(dp->dp_spa) * 4;
2777 	tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks,
2778 	    INT_MAX, 0);
2779 	if (tq == NULL) {
2780 		kmem_free(dcp, sizeof (*dcp));
2781 		mutex_destroy(&err_lock);
2782 
2783 		return (SET_ERROR(ENOMEM));
2784 	}
2785 	dcp->dc_tq = tq;
2786 
2787 	/* dcp will be freed by task */
2788 	(void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
2789 
2790 	/*
2791 	 * PORTING: this code relies on the property of taskq_wait to wait
2792 	 * until no more tasks are queued and no more tasks are active. As
2793 	 * we always queue new tasks from within other tasks, task_wait
2794 	 * reliably waits for the full recursion to finish, even though we
2795 	 * enqueue new tasks after taskq_wait has been called.
2796 	 * On platforms other than illumos, taskq_wait may not have this
2797 	 * property.
2798 	 */
2799 	taskq_wait(tq);
2800 	taskq_destroy(tq);
2801 	mutex_destroy(&err_lock);
2802 
2803 	return (error);
2804 }
2805 
2806 /*
2807  * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2808  * The dp_config_rwlock must not be held when this is called, and it
2809  * will not be held when the callback is called.
2810  * Therefore this function should only be used when the pool is not changing
2811  * (e.g. in syncing context), or the callback can deal with the possible races.
2812  */
2813 static int
2814 dmu_objset_find_impl(spa_t *spa, const char *name,
2815     int func(const char *, void *), void *arg, int flags)
2816 {
2817 	dsl_dir_t *dd;
2818 	dsl_pool_t *dp = spa_get_dsl(spa);
2819 	dsl_dataset_t *ds;
2820 	zap_cursor_t zc;
2821 	zap_attribute_t *attr;
2822 	char *child;
2823 	uint64_t thisobj;
2824 	int err;
2825 
2826 	dsl_pool_config_enter(dp, FTAG);
2827 
2828 	err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
2829 	if (err != 0) {
2830 		dsl_pool_config_exit(dp, FTAG);
2831 		return (err);
2832 	}
2833 
2834 	/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2835 	if (dd->dd_myname[0] == '$') {
2836 		dsl_dir_rele(dd, FTAG);
2837 		dsl_pool_config_exit(dp, FTAG);
2838 		return (0);
2839 	}
2840 
2841 	thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2842 	attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2843 
2844 	/*
2845 	 * Iterate over all children.
2846 	 */
2847 	if (flags & DS_FIND_CHILDREN) {
2848 		for (zap_cursor_init(&zc, dp->dp_meta_objset,
2849 		    dsl_dir_phys(dd)->dd_child_dir_zapobj);
2850 		    zap_cursor_retrieve(&zc, attr) == 0;
2851 		    (void) zap_cursor_advance(&zc)) {
2852 			ASSERT3U(attr->za_integer_length, ==,
2853 			    sizeof (uint64_t));
2854 			ASSERT3U(attr->za_num_integers, ==, 1);
2855 
2856 			child = kmem_asprintf("%s/%s", name, attr->za_name);
2857 			dsl_pool_config_exit(dp, FTAG);
2858 			err = dmu_objset_find_impl(spa, child,
2859 			    func, arg, flags);
2860 			dsl_pool_config_enter(dp, FTAG);
2861 			kmem_strfree(child);
2862 			if (err != 0)
2863 				break;
2864 		}
2865 		zap_cursor_fini(&zc);
2866 
2867 		if (err != 0) {
2868 			dsl_dir_rele(dd, FTAG);
2869 			dsl_pool_config_exit(dp, FTAG);
2870 			kmem_free(attr, sizeof (zap_attribute_t));
2871 			return (err);
2872 		}
2873 	}
2874 
2875 	/*
2876 	 * Iterate over all snapshots.
2877 	 */
2878 	if (flags & DS_FIND_SNAPSHOTS) {
2879 		err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2880 
2881 		if (err == 0) {
2882 			uint64_t snapobj;
2883 
2884 			snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2885 			dsl_dataset_rele(ds, FTAG);
2886 
2887 			for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2888 			    zap_cursor_retrieve(&zc, attr) == 0;
2889 			    (void) zap_cursor_advance(&zc)) {
2890 				ASSERT3U(attr->za_integer_length, ==,
2891 				    sizeof (uint64_t));
2892 				ASSERT3U(attr->za_num_integers, ==, 1);
2893 
2894 				child = kmem_asprintf("%s@%s",
2895 				    name, attr->za_name);
2896 				dsl_pool_config_exit(dp, FTAG);
2897 				err = func(child, arg);
2898 				dsl_pool_config_enter(dp, FTAG);
2899 				kmem_strfree(child);
2900 				if (err != 0)
2901 					break;
2902 			}
2903 			zap_cursor_fini(&zc);
2904 		}
2905 	}
2906 
2907 	dsl_dir_rele(dd, FTAG);
2908 	kmem_free(attr, sizeof (zap_attribute_t));
2909 	dsl_pool_config_exit(dp, FTAG);
2910 
2911 	if (err != 0)
2912 		return (err);
2913 
2914 	/* Apply to self. */
2915 	return (func(name, arg));
2916 }
2917 
2918 /*
2919  * See comment above dmu_objset_find_impl().
2920  */
2921 int
2922 dmu_objset_find(const char *name, int func(const char *, void *), void *arg,
2923     int flags)
2924 {
2925 	spa_t *spa;
2926 	int error;
2927 
2928 	error = spa_open(name, &spa, FTAG);
2929 	if (error != 0)
2930 		return (error);
2931 	error = dmu_objset_find_impl(spa, name, func, arg, flags);
2932 	spa_close(spa, FTAG);
2933 	return (error);
2934 }
2935 
2936 boolean_t
2937 dmu_objset_incompatible_encryption_version(objset_t *os)
2938 {
2939 	return (dsl_dir_incompatible_encryption_version(
2940 	    os->os_dsl_dataset->ds_dir));
2941 }
2942 
2943 void
2944 dmu_objset_set_user(objset_t *os, void *user_ptr)
2945 {
2946 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2947 	os->os_user_ptr = user_ptr;
2948 }
2949 
2950 void *
2951 dmu_objset_get_user(objset_t *os)
2952 {
2953 	ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2954 	return (os->os_user_ptr);
2955 }
2956 
2957 /*
2958  * Determine name of filesystem, given name of snapshot.
2959  * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2960  */
2961 int
2962 dmu_fsname(const char *snapname, char *buf)
2963 {
2964 	char *atp = strchr(snapname, '@');
2965 	if (atp == NULL)
2966 		return (SET_ERROR(EINVAL));
2967 	if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN)
2968 		return (SET_ERROR(ENAMETOOLONG));
2969 	(void) strlcpy(buf, snapname, atp - snapname + 1);
2970 	return (0);
2971 }
2972 
2973 /*
2974  * Call when we think we're going to write/free space in open context
2975  * to track the amount of dirty data in the open txg, which is also the
2976  * amount of memory that can not be evicted until this txg syncs.
2977  *
2978  * Note that there are two conditions where this can be called from
2979  * syncing context:
2980  *
2981  * [1] When we just created the dataset, in which case we go on with
2982  *     updating any accounting of dirty data as usual.
2983  * [2] When we are dirtying MOS data, in which case we only update the
2984  *     pool's accounting of dirty data.
2985  */
2986 void
2987 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx)
2988 {
2989 	dsl_dataset_t *ds = os->os_dsl_dataset;
2990 	int64_t aspace = spa_get_worst_case_asize(os->os_spa, space);
2991 
2992 	if (ds != NULL) {
2993 		dsl_dir_willuse_space(ds->ds_dir, aspace, tx);
2994 	}
2995 
2996 	dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx);
2997 }
2998 
2999 #if defined(_KERNEL)
3000 EXPORT_SYMBOL(dmu_objset_zil);
3001 EXPORT_SYMBOL(dmu_objset_pool);
3002 EXPORT_SYMBOL(dmu_objset_ds);
3003 EXPORT_SYMBOL(dmu_objset_type);
3004 EXPORT_SYMBOL(dmu_objset_name);
3005 EXPORT_SYMBOL(dmu_objset_hold);
3006 EXPORT_SYMBOL(dmu_objset_hold_flags);
3007 EXPORT_SYMBOL(dmu_objset_own);
3008 EXPORT_SYMBOL(dmu_objset_rele);
3009 EXPORT_SYMBOL(dmu_objset_rele_flags);
3010 EXPORT_SYMBOL(dmu_objset_disown);
3011 EXPORT_SYMBOL(dmu_objset_from_ds);
3012 EXPORT_SYMBOL(dmu_objset_create);
3013 EXPORT_SYMBOL(dmu_objset_clone);
3014 EXPORT_SYMBOL(dmu_objset_stats);
3015 EXPORT_SYMBOL(dmu_objset_fast_stat);
3016 EXPORT_SYMBOL(dmu_objset_spa);
3017 EXPORT_SYMBOL(dmu_objset_space);
3018 EXPORT_SYMBOL(dmu_objset_fsid_guid);
3019 EXPORT_SYMBOL(dmu_objset_find);
3020 EXPORT_SYMBOL(dmu_objset_byteswap);
3021 EXPORT_SYMBOL(dmu_objset_evict_dbufs);
3022 EXPORT_SYMBOL(dmu_objset_snap_cmtime);
3023 EXPORT_SYMBOL(dmu_objset_dnodesize);
3024 
3025 EXPORT_SYMBOL(dmu_objset_sync);
3026 EXPORT_SYMBOL(dmu_objset_is_dirty);
3027 EXPORT_SYMBOL(dmu_objset_create_impl_dnstats);
3028 EXPORT_SYMBOL(dmu_objset_create_impl);
3029 EXPORT_SYMBOL(dmu_objset_open_impl);
3030 EXPORT_SYMBOL(dmu_objset_evict);
3031 EXPORT_SYMBOL(dmu_objset_register_type);
3032 EXPORT_SYMBOL(dmu_objset_sync_done);
3033 EXPORT_SYMBOL(dmu_objset_userquota_get_ids);
3034 EXPORT_SYMBOL(dmu_objset_userused_enabled);
3035 EXPORT_SYMBOL(dmu_objset_userspace_upgrade);
3036 EXPORT_SYMBOL(dmu_objset_userspace_present);
3037 EXPORT_SYMBOL(dmu_objset_userobjused_enabled);
3038 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable);
3039 EXPORT_SYMBOL(dmu_objset_userobjspace_present);
3040 EXPORT_SYMBOL(dmu_objset_projectquota_enabled);
3041 EXPORT_SYMBOL(dmu_objset_projectquota_present);
3042 EXPORT_SYMBOL(dmu_objset_projectquota_upgradable);
3043 EXPORT_SYMBOL(dmu_objset_id_quota_upgrade);
3044 #endif
3045