]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/dmu_objset.c
c19ebf424953ee7e236b1223e88b99a74de27a20
[mirror_zfs.git] / module / zfs / dmu_objset.c
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 https://opensource.org/licenses/CDDL-1.0.
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 * Copyright (c) 2022 Hewlett Packard Enterprise Development LP.
36 */
37
38 /* Portions Copyright 2010 Robert Milkowski */
39
40 #include <sys/cred.h>
41 #include <sys/zfs_context.h>
42 #include <sys/dmu_objset.h>
43 #include <sys/dsl_dir.h>
44 #include <sys/dsl_dataset.h>
45 #include <sys/dsl_prop.h>
46 #include <sys/dsl_pool.h>
47 #include <sys/dsl_synctask.h>
48 #include <sys/dsl_deleg.h>
49 #include <sys/dnode.h>
50 #include <sys/dbuf.h>
51 #include <sys/zvol.h>
52 #include <sys/dmu_tx.h>
53 #include <sys/zap.h>
54 #include <sys/zil.h>
55 #include <sys/dmu_impl.h>
56 #include <sys/zfs_ioctl.h>
57 #include <sys/sa.h>
58 #include <sys/zfs_onexit.h>
59 #include <sys/dsl_destroy.h>
60 #include <sys/vdev.h>
61 #include <sys/zfeature.h>
62 #include <sys/policy.h>
63 #include <sys/spa_impl.h>
64 #include <sys/dmu_recv.h>
65 #include <sys/zfs_project.h>
66 #include "zfs_namecheck.h"
67 #include <sys/vdev_impl.h>
68 #include <sys/arc.h>
69
70 /*
71 * Needed to close a window in dnode_move() that allows the objset to be freed
72 * before it can be safely accessed.
73 */
74 krwlock_t os_lock;
75
76 /*
77 * Tunable to overwrite the maximum number of threads for the parallelization
78 * of dmu_objset_find_dp, needed to speed up the import of pools with many
79 * datasets.
80 * Default is 4 times the number of leaf vdevs.
81 */
82 static const int dmu_find_threads = 0;
83
84 /*
85 * Backfill lower metadnode objects after this many have been freed.
86 * Backfilling negatively impacts object creation rates, so only do it
87 * if there are enough holes to fill.
88 */
89 static const int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT;
90
91 static const char *upgrade_tag = "upgrade_tag";
92
93 static void dmu_objset_find_dp_cb(void *arg);
94
95 static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb);
96 static void dmu_objset_upgrade_stop(objset_t *os);
97
98 void
99 dmu_objset_init(void)
100 {
101 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
102 }
103
104 void
105 dmu_objset_fini(void)
106 {
107 rw_destroy(&os_lock);
108 }
109
110 spa_t *
111 dmu_objset_spa(objset_t *os)
112 {
113 return (os->os_spa);
114 }
115
116 zilog_t *
117 dmu_objset_zil(objset_t *os)
118 {
119 return (os->os_zil);
120 }
121
122 dsl_pool_t *
123 dmu_objset_pool(objset_t *os)
124 {
125 dsl_dataset_t *ds;
126
127 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
128 return (ds->ds_dir->dd_pool);
129 else
130 return (spa_get_dsl(os->os_spa));
131 }
132
133 dsl_dataset_t *
134 dmu_objset_ds(objset_t *os)
135 {
136 return (os->os_dsl_dataset);
137 }
138
139 dmu_objset_type_t
140 dmu_objset_type(objset_t *os)
141 {
142 return (os->os_phys->os_type);
143 }
144
145 void
146 dmu_objset_name(objset_t *os, char *buf)
147 {
148 dsl_dataset_name(os->os_dsl_dataset, buf);
149 }
150
151 uint64_t
152 dmu_objset_id(objset_t *os)
153 {
154 dsl_dataset_t *ds = os->os_dsl_dataset;
155
156 return (ds ? ds->ds_object : 0);
157 }
158
159 uint64_t
160 dmu_objset_dnodesize(objset_t *os)
161 {
162 return (os->os_dnodesize);
163 }
164
165 zfs_sync_type_t
166 dmu_objset_syncprop(objset_t *os)
167 {
168 return (os->os_sync);
169 }
170
171 zfs_logbias_op_t
172 dmu_objset_logbias(objset_t *os)
173 {
174 return (os->os_logbias);
175 }
176
177 static void
178 checksum_changed_cb(void *arg, uint64_t newval)
179 {
180 objset_t *os = arg;
181
182 /*
183 * Inheritance should have been done by now.
184 */
185 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
186
187 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
188 }
189
190 static void
191 compression_changed_cb(void *arg, uint64_t newval)
192 {
193 objset_t *os = arg;
194
195 /*
196 * Inheritance and range checking should have been done by now.
197 */
198 ASSERT(newval != ZIO_COMPRESS_INHERIT);
199
200 os->os_compress = zio_compress_select(os->os_spa,
201 ZIO_COMPRESS_ALGO(newval), ZIO_COMPRESS_ON);
202 os->os_complevel = zio_complevel_select(os->os_spa, os->os_compress,
203 ZIO_COMPRESS_LEVEL(newval), ZIO_COMPLEVEL_DEFAULT);
204 }
205
206 static void
207 copies_changed_cb(void *arg, uint64_t newval)
208 {
209 objset_t *os = arg;
210
211 /*
212 * Inheritance and range checking should have been done by now.
213 */
214 ASSERT(newval > 0);
215 ASSERT(newval <= spa_max_replication(os->os_spa));
216
217 os->os_copies = newval;
218 }
219
220 static void
221 dedup_changed_cb(void *arg, uint64_t newval)
222 {
223 objset_t *os = arg;
224 spa_t *spa = os->os_spa;
225 enum zio_checksum checksum;
226
227 /*
228 * Inheritance should have been done by now.
229 */
230 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
231
232 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
233
234 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
235 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
236 }
237
238 static void
239 primary_cache_changed_cb(void *arg, uint64_t newval)
240 {
241 objset_t *os = arg;
242
243 /*
244 * Inheritance and range checking should have been done by now.
245 */
246 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
247 newval == ZFS_CACHE_METADATA);
248
249 os->os_primary_cache = newval;
250 }
251
252 static void
253 secondary_cache_changed_cb(void *arg, uint64_t newval)
254 {
255 objset_t *os = arg;
256
257 /*
258 * Inheritance and range checking should have been done by now.
259 */
260 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
261 newval == ZFS_CACHE_METADATA);
262
263 os->os_secondary_cache = newval;
264 }
265
266 static void
267 sync_changed_cb(void *arg, uint64_t newval)
268 {
269 objset_t *os = arg;
270
271 /*
272 * Inheritance and range checking should have been done by now.
273 */
274 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
275 newval == ZFS_SYNC_DISABLED);
276
277 os->os_sync = newval;
278 if (os->os_zil)
279 zil_set_sync(os->os_zil, newval);
280 }
281
282 static void
283 redundant_metadata_changed_cb(void *arg, uint64_t newval)
284 {
285 objset_t *os = arg;
286
287 /*
288 * Inheritance and range checking should have been done by now.
289 */
290 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
291 newval == ZFS_REDUNDANT_METADATA_MOST ||
292 newval == ZFS_REDUNDANT_METADATA_SOME ||
293 newval == ZFS_REDUNDANT_METADATA_NONE);
294
295 os->os_redundant_metadata = newval;
296 }
297
298 static void
299 dnodesize_changed_cb(void *arg, uint64_t newval)
300 {
301 objset_t *os = arg;
302
303 switch (newval) {
304 case ZFS_DNSIZE_LEGACY:
305 os->os_dnodesize = DNODE_MIN_SIZE;
306 break;
307 case ZFS_DNSIZE_AUTO:
308 /*
309 * Choose a dnode size that will work well for most
310 * workloads if the user specified "auto". Future code
311 * improvements could dynamically select a dnode size
312 * based on observed workload patterns.
313 */
314 os->os_dnodesize = DNODE_MIN_SIZE * 2;
315 break;
316 case ZFS_DNSIZE_1K:
317 case ZFS_DNSIZE_2K:
318 case ZFS_DNSIZE_4K:
319 case ZFS_DNSIZE_8K:
320 case ZFS_DNSIZE_16K:
321 os->os_dnodesize = newval;
322 break;
323 }
324 }
325
326 static void
327 smallblk_changed_cb(void *arg, uint64_t newval)
328 {
329 objset_t *os = arg;
330
331 /*
332 * Inheritance and range checking should have been done by now.
333 */
334 ASSERT(newval <= SPA_MAXBLOCKSIZE);
335 ASSERT(ISP2(newval));
336
337 os->os_zpl_special_smallblock = newval;
338 }
339
340 static void
341 logbias_changed_cb(void *arg, uint64_t newval)
342 {
343 objset_t *os = arg;
344
345 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
346 newval == ZFS_LOGBIAS_THROUGHPUT);
347 os->os_logbias = newval;
348 if (os->os_zil)
349 zil_set_logbias(os->os_zil, newval);
350 }
351
352 static void
353 recordsize_changed_cb(void *arg, uint64_t newval)
354 {
355 objset_t *os = arg;
356
357 os->os_recordsize = newval;
358 }
359
360 void
361 dmu_objset_byteswap(void *buf, size_t size)
362 {
363 objset_phys_t *osp = buf;
364
365 ASSERT(size == OBJSET_PHYS_SIZE_V1 || size == OBJSET_PHYS_SIZE_V2 ||
366 size == sizeof (objset_phys_t));
367 dnode_byteswap(&osp->os_meta_dnode);
368 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
369 osp->os_type = BSWAP_64(osp->os_type);
370 osp->os_flags = BSWAP_64(osp->os_flags);
371 if (size >= OBJSET_PHYS_SIZE_V2) {
372 dnode_byteswap(&osp->os_userused_dnode);
373 dnode_byteswap(&osp->os_groupused_dnode);
374 if (size >= sizeof (objset_phys_t))
375 dnode_byteswap(&osp->os_projectused_dnode);
376 }
377 }
378
379 /*
380 * The hash is a CRC-based hash of the objset_t pointer and the object number.
381 */
382 static uint64_t
383 dnode_hash(const objset_t *os, uint64_t obj)
384 {
385 uintptr_t osv = (uintptr_t)os;
386 uint64_t crc = -1ULL;
387
388 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
389 /*
390 * The low 6 bits of the pointer don't have much entropy, because
391 * the objset_t is larger than 2^6 bytes long.
392 */
393 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF];
394 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF];
395 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF];
396 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF];
397
398 crc ^= (osv>>14) ^ (obj>>24);
399
400 return (crc);
401 }
402
403 static unsigned int
404 dnode_multilist_index_func(multilist_t *ml, void *obj)
405 {
406 dnode_t *dn = obj;
407
408 /*
409 * The low order bits of the hash value are thought to be
410 * distributed evenly. Otherwise, in the case that the multilist
411 * has a power of two number of sublists, each sublists' usage
412 * would not be evenly distributed. In this context full 64bit
413 * division would be a waste of time, so limit it to 32 bits.
414 */
415 return ((unsigned int)dnode_hash(dn->dn_objset, dn->dn_object) %
416 multilist_get_num_sublists(ml));
417 }
418
419 static inline boolean_t
420 dmu_os_is_l2cacheable(objset_t *os)
421 {
422 if (os->os_secondary_cache == ZFS_CACHE_ALL ||
423 os->os_secondary_cache == ZFS_CACHE_METADATA) {
424 if (l2arc_exclude_special == 0)
425 return (B_TRUE);
426
427 blkptr_t *bp = os->os_rootbp;
428 if (bp == NULL || BP_IS_HOLE(bp))
429 return (B_FALSE);
430 uint64_t vdev = DVA_GET_VDEV(bp->blk_dva);
431 vdev_t *rvd = os->os_spa->spa_root_vdev;
432 vdev_t *vd = NULL;
433
434 if (vdev < rvd->vdev_children)
435 vd = rvd->vdev_child[vdev];
436
437 if (vd == NULL)
438 return (B_TRUE);
439
440 if (vd->vdev_alloc_bias != VDEV_BIAS_SPECIAL &&
441 vd->vdev_alloc_bias != VDEV_BIAS_DEDUP)
442 return (B_TRUE);
443 }
444 return (B_FALSE);
445 }
446
447 /*
448 * Instantiates the objset_t in-memory structure corresponding to the
449 * objset_phys_t that's pointed to by the specified blkptr_t.
450 */
451 int
452 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
453 objset_t **osp)
454 {
455 objset_t *os;
456 int i, err;
457
458 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
459 ASSERT(!BP_IS_REDACTED(bp));
460
461 /*
462 * We need the pool config lock to get properties.
463 */
464 ASSERT(ds == NULL || dsl_pool_config_held(ds->ds_dir->dd_pool));
465
466 /*
467 * The $ORIGIN dataset (if it exists) doesn't have an associated
468 * objset, so there's no reason to open it. The $ORIGIN dataset
469 * will not exist on pools older than SPA_VERSION_ORIGIN.
470 */
471 if (ds != NULL && spa_get_dsl(spa) != NULL &&
472 spa_get_dsl(spa)->dp_origin_snap != NULL) {
473 ASSERT3P(ds->ds_dir, !=,
474 spa_get_dsl(spa)->dp_origin_snap->ds_dir);
475 }
476
477 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
478 os->os_dsl_dataset = ds;
479 os->os_spa = spa;
480 os->os_rootbp = bp;
481 if (!BP_IS_HOLE(os->os_rootbp)) {
482 arc_flags_t aflags = ARC_FLAG_WAIT;
483 zbookmark_phys_t zb;
484 int size;
485 zio_flag_t zio_flags = ZIO_FLAG_CANFAIL;
486 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
487 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
488
489 if (dmu_os_is_l2cacheable(os))
490 aflags |= ARC_FLAG_L2CACHE;
491
492 if (ds != NULL && ds->ds_dir->dd_crypto_obj != 0) {
493 ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF);
494 ASSERT(BP_IS_AUTHENTICATED(bp));
495 zio_flags |= ZIO_FLAG_RAW;
496 }
497
498 dprintf_bp(os->os_rootbp, "reading %s", "");
499 err = arc_read(NULL, spa, os->os_rootbp,
500 arc_getbuf_func, &os->os_phys_buf,
501 ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
502 if (err != 0) {
503 kmem_free(os, sizeof (objset_t));
504 /* convert checksum errors into IO errors */
505 if (err == ECKSUM)
506 err = SET_ERROR(EIO);
507 return (err);
508 }
509
510 if (spa_version(spa) < SPA_VERSION_USERSPACE)
511 size = OBJSET_PHYS_SIZE_V1;
512 else if (!spa_feature_is_enabled(spa,
513 SPA_FEATURE_PROJECT_QUOTA))
514 size = OBJSET_PHYS_SIZE_V2;
515 else
516 size = sizeof (objset_phys_t);
517
518 /* Increase the blocksize if we are permitted. */
519 if (arc_buf_size(os->os_phys_buf) < size) {
520 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf,
521 ARC_BUFC_METADATA, size);
522 memset(buf->b_data, 0, size);
523 memcpy(buf->b_data, os->os_phys_buf->b_data,
524 arc_buf_size(os->os_phys_buf));
525 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
526 os->os_phys_buf = buf;
527 }
528
529 os->os_phys = os->os_phys_buf->b_data;
530 os->os_flags = os->os_phys->os_flags;
531 } else {
532 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
533 sizeof (objset_phys_t) : OBJSET_PHYS_SIZE_V1;
534 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf,
535 ARC_BUFC_METADATA, size);
536 os->os_phys = os->os_phys_buf->b_data;
537 memset(os->os_phys, 0, size);
538 }
539 /*
540 * These properties will be filled in by the logic in zfs_get_zplprop()
541 * when they are queried for the first time.
542 */
543 os->os_version = OBJSET_PROP_UNINITIALIZED;
544 os->os_normalization = OBJSET_PROP_UNINITIALIZED;
545 os->os_utf8only = OBJSET_PROP_UNINITIALIZED;
546 os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED;
547
548 /*
549 * Note: the changed_cb will be called once before the register
550 * func returns, thus changing the checksum/compression from the
551 * default (fletcher2/off). Snapshots don't need to know about
552 * checksum/compression/copies.
553 */
554 if (ds != NULL) {
555 os->os_encrypted = (ds->ds_dir->dd_crypto_obj != 0);
556
557 err = dsl_prop_register(ds,
558 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
559 primary_cache_changed_cb, os);
560 if (err == 0) {
561 err = dsl_prop_register(ds,
562 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
563 secondary_cache_changed_cb, os);
564 }
565 if (!ds->ds_is_snapshot) {
566 if (err == 0) {
567 err = dsl_prop_register(ds,
568 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
569 checksum_changed_cb, os);
570 }
571 if (err == 0) {
572 err = dsl_prop_register(ds,
573 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
574 compression_changed_cb, os);
575 }
576 if (err == 0) {
577 err = dsl_prop_register(ds,
578 zfs_prop_to_name(ZFS_PROP_COPIES),
579 copies_changed_cb, os);
580 }
581 if (err == 0) {
582 err = dsl_prop_register(ds,
583 zfs_prop_to_name(ZFS_PROP_DEDUP),
584 dedup_changed_cb, os);
585 }
586 if (err == 0) {
587 err = dsl_prop_register(ds,
588 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
589 logbias_changed_cb, os);
590 }
591 if (err == 0) {
592 err = dsl_prop_register(ds,
593 zfs_prop_to_name(ZFS_PROP_SYNC),
594 sync_changed_cb, os);
595 }
596 if (err == 0) {
597 err = dsl_prop_register(ds,
598 zfs_prop_to_name(
599 ZFS_PROP_REDUNDANT_METADATA),
600 redundant_metadata_changed_cb, os);
601 }
602 if (err == 0) {
603 err = dsl_prop_register(ds,
604 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
605 recordsize_changed_cb, os);
606 }
607 if (err == 0) {
608 err = dsl_prop_register(ds,
609 zfs_prop_to_name(ZFS_PROP_DNODESIZE),
610 dnodesize_changed_cb, os);
611 }
612 if (err == 0) {
613 err = dsl_prop_register(ds,
614 zfs_prop_to_name(
615 ZFS_PROP_SPECIAL_SMALL_BLOCKS),
616 smallblk_changed_cb, os);
617 }
618 }
619 if (err != 0) {
620 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
621 kmem_free(os, sizeof (objset_t));
622 return (err);
623 }
624 } else {
625 /* It's the meta-objset. */
626 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
627 os->os_compress = ZIO_COMPRESS_ON;
628 os->os_complevel = ZIO_COMPLEVEL_DEFAULT;
629 os->os_encrypted = B_FALSE;
630 os->os_copies = spa_max_replication(spa);
631 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
632 os->os_dedup_verify = B_FALSE;
633 os->os_logbias = ZFS_LOGBIAS_LATENCY;
634 os->os_sync = ZFS_SYNC_STANDARD;
635 os->os_primary_cache = ZFS_CACHE_ALL;
636 os->os_secondary_cache = ZFS_CACHE_ALL;
637 os->os_dnodesize = DNODE_MIN_SIZE;
638 }
639
640 if (ds == NULL || !ds->ds_is_snapshot)
641 os->os_zil_header = os->os_phys->os_zil_header;
642 os->os_zil = zil_alloc(os, &os->os_zil_header);
643
644 for (i = 0; i < TXG_SIZE; i++) {
645 multilist_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
646 offsetof(dnode_t, dn_dirty_link[i]),
647 dnode_multilist_index_func);
648 }
649 list_create(&os->os_dnodes, sizeof (dnode_t),
650 offsetof(dnode_t, dn_link));
651 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
652 offsetof(dmu_buf_impl_t, db_link));
653
654 list_link_init(&os->os_evicting_node);
655
656 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
657 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL);
658 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
659 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
660 os->os_obj_next_percpu_len = boot_ncpus;
661 os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len *
662 sizeof (os->os_obj_next_percpu[0]), KM_SLEEP);
663
664 dnode_special_open(os, &os->os_phys->os_meta_dnode,
665 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
666 if (OBJSET_BUF_HAS_USERUSED(os->os_phys_buf)) {
667 dnode_special_open(os, &os->os_phys->os_userused_dnode,
668 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
669 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
670 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
671 if (OBJSET_BUF_HAS_PROJECTUSED(os->os_phys_buf))
672 dnode_special_open(os,
673 &os->os_phys->os_projectused_dnode,
674 DMU_PROJECTUSED_OBJECT, &os->os_projectused_dnode);
675 }
676
677 mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL);
678
679 *osp = os;
680 return (0);
681 }
682
683 int
684 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
685 {
686 int err = 0;
687
688 /*
689 * We need the pool_config lock to manipulate the dsl_dataset_t.
690 * Even if the dataset is long-held, we need the pool_config lock
691 * to open the objset, as it needs to get properties.
692 */
693 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
694
695 mutex_enter(&ds->ds_opening_lock);
696 if (ds->ds_objset == NULL) {
697 objset_t *os;
698 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
699 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
700 ds, dsl_dataset_get_blkptr(ds), &os);
701 rrw_exit(&ds->ds_bp_rwlock, FTAG);
702
703 if (err == 0) {
704 mutex_enter(&ds->ds_lock);
705 ASSERT(ds->ds_objset == NULL);
706 ds->ds_objset = os;
707 mutex_exit(&ds->ds_lock);
708 }
709 }
710 *osp = ds->ds_objset;
711 mutex_exit(&ds->ds_opening_lock);
712 return (err);
713 }
714
715 /*
716 * Holds the pool while the objset is held. Therefore only one objset
717 * can be held at a time.
718 */
719 int
720 dmu_objset_hold_flags(const char *name, boolean_t decrypt, const void *tag,
721 objset_t **osp)
722 {
723 dsl_pool_t *dp;
724 dsl_dataset_t *ds;
725 int err;
726 ds_hold_flags_t flags;
727
728 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
729 err = dsl_pool_hold(name, tag, &dp);
730 if (err != 0)
731 return (err);
732 err = dsl_dataset_hold_flags(dp, name, flags, tag, &ds);
733 if (err != 0) {
734 dsl_pool_rele(dp, tag);
735 return (err);
736 }
737
738 err = dmu_objset_from_ds(ds, osp);
739 if (err != 0) {
740 dsl_dataset_rele(ds, tag);
741 dsl_pool_rele(dp, tag);
742 }
743
744 return (err);
745 }
746
747 int
748 dmu_objset_hold(const char *name, const void *tag, objset_t **osp)
749 {
750 return (dmu_objset_hold_flags(name, B_FALSE, tag, osp));
751 }
752
753 static int
754 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
755 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp)
756 {
757 (void) tag;
758
759 int err = dmu_objset_from_ds(ds, osp);
760 if (err != 0) {
761 return (err);
762 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
763 return (SET_ERROR(EINVAL));
764 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
765 return (SET_ERROR(EROFS));
766 } else if (!readonly && decrypt &&
767 dsl_dir_incompatible_encryption_version(ds->ds_dir)) {
768 return (SET_ERROR(EROFS));
769 }
770
771 /* if we are decrypting, we can now check MACs in os->os_phys_buf */
772 if (decrypt && arc_is_unauthenticated((*osp)->os_phys_buf)) {
773 zbookmark_phys_t zb;
774
775 SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT,
776 ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
777 err = arc_untransform((*osp)->os_phys_buf, (*osp)->os_spa,
778 &zb, B_FALSE);
779 if (err != 0)
780 return (err);
781
782 ASSERT0(arc_is_unauthenticated((*osp)->os_phys_buf));
783 }
784
785 return (0);
786 }
787
788 /*
789 * dsl_pool must not be held when this is called.
790 * Upon successful return, there will be a longhold on the dataset,
791 * and the dsl_pool will not be held.
792 */
793 int
794 dmu_objset_own(const char *name, dmu_objset_type_t type,
795 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp)
796 {
797 dsl_pool_t *dp;
798 dsl_dataset_t *ds;
799 int err;
800 ds_hold_flags_t flags;
801
802 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
803 err = dsl_pool_hold(name, FTAG, &dp);
804 if (err != 0)
805 return (err);
806 err = dsl_dataset_own(dp, name, flags, tag, &ds);
807 if (err != 0) {
808 dsl_pool_rele(dp, FTAG);
809 return (err);
810 }
811 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp);
812 if (err != 0) {
813 dsl_dataset_disown(ds, flags, tag);
814 dsl_pool_rele(dp, FTAG);
815 return (err);
816 }
817
818 /*
819 * User accounting requires the dataset to be decrypted and rw.
820 * We also don't begin user accounting during claiming to help
821 * speed up pool import times and to keep this txg reserved
822 * completely for recovery work.
823 */
824 if (!readonly && !dp->dp_spa->spa_claiming &&
825 (ds->ds_dir->dd_crypto_obj == 0 || decrypt)) {
826 if (dmu_objset_userobjspace_upgradable(*osp) ||
827 dmu_objset_projectquota_upgradable(*osp)) {
828 dmu_objset_id_quota_upgrade(*osp);
829 } else if (dmu_objset_userused_enabled(*osp)) {
830 dmu_objset_userspace_upgrade(*osp);
831 }
832 }
833
834 dsl_pool_rele(dp, FTAG);
835 return (0);
836 }
837
838 int
839 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
840 boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp)
841 {
842 dsl_dataset_t *ds;
843 int err;
844 ds_hold_flags_t flags;
845
846 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
847 err = dsl_dataset_own_obj(dp, obj, flags, tag, &ds);
848 if (err != 0)
849 return (err);
850
851 err = dmu_objset_own_impl(ds, type, readonly, decrypt, tag, osp);
852 if (err != 0) {
853 dsl_dataset_disown(ds, flags, tag);
854 return (err);
855 }
856
857 return (0);
858 }
859
860 void
861 dmu_objset_rele_flags(objset_t *os, boolean_t decrypt, const void *tag)
862 {
863 ds_hold_flags_t flags;
864 dsl_pool_t *dp = dmu_objset_pool(os);
865
866 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
867 dsl_dataset_rele_flags(os->os_dsl_dataset, flags, tag);
868 dsl_pool_rele(dp, tag);
869 }
870
871 void
872 dmu_objset_rele(objset_t *os, const void *tag)
873 {
874 dmu_objset_rele_flags(os, B_FALSE, tag);
875 }
876
877 /*
878 * When we are called, os MUST refer to an objset associated with a dataset
879 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
880 * == tag. We will then release and reacquire ownership of the dataset while
881 * holding the pool config_rwlock to avoid intervening namespace or ownership
882 * changes may occur.
883 *
884 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
885 * release the hold on its dataset and acquire a new one on the dataset of the
886 * same name so that it can be partially torn down and reconstructed.
887 */
888 void
889 dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds,
890 boolean_t decrypt, const void *tag)
891 {
892 dsl_pool_t *dp;
893 char name[ZFS_MAX_DATASET_NAME_LEN];
894 ds_hold_flags_t flags;
895
896 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
897 VERIFY3P(ds, !=, NULL);
898 VERIFY3P(ds->ds_owner, ==, tag);
899 VERIFY(dsl_dataset_long_held(ds));
900
901 dsl_dataset_name(ds, name);
902 dp = ds->ds_dir->dd_pool;
903 dsl_pool_config_enter(dp, FTAG);
904 dsl_dataset_disown(ds, flags, tag);
905 VERIFY0(dsl_dataset_own(dp, name, flags, tag, newds));
906 dsl_pool_config_exit(dp, FTAG);
907 }
908
909 void
910 dmu_objset_disown(objset_t *os, boolean_t decrypt, const void *tag)
911 {
912 ds_hold_flags_t flags;
913
914 flags = (decrypt) ? DS_HOLD_FLAG_DECRYPT : DS_HOLD_FLAG_NONE;
915 /*
916 * Stop upgrading thread
917 */
918 dmu_objset_upgrade_stop(os);
919 dsl_dataset_disown(os->os_dsl_dataset, flags, tag);
920 }
921
922 void
923 dmu_objset_evict_dbufs(objset_t *os)
924 {
925 dnode_t *dn_marker;
926 dnode_t *dn;
927
928 dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP);
929
930 mutex_enter(&os->os_lock);
931 dn = list_head(&os->os_dnodes);
932 while (dn != NULL) {
933 /*
934 * Skip dnodes without holds. We have to do this dance
935 * because dnode_add_ref() only works if there is already a
936 * hold. If the dnode has no holds, then it has no dbufs.
937 */
938 if (dnode_add_ref(dn, FTAG)) {
939 list_insert_after(&os->os_dnodes, dn, dn_marker);
940 mutex_exit(&os->os_lock);
941
942 dnode_evict_dbufs(dn);
943 dnode_rele(dn, FTAG);
944
945 mutex_enter(&os->os_lock);
946 dn = list_next(&os->os_dnodes, dn_marker);
947 list_remove(&os->os_dnodes, dn_marker);
948 } else {
949 dn = list_next(&os->os_dnodes, dn);
950 }
951 }
952 mutex_exit(&os->os_lock);
953
954 kmem_free(dn_marker, sizeof (dnode_t));
955
956 if (DMU_USERUSED_DNODE(os) != NULL) {
957 if (DMU_PROJECTUSED_DNODE(os) != NULL)
958 dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os));
959 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
960 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
961 }
962 dnode_evict_dbufs(DMU_META_DNODE(os));
963 }
964
965 /*
966 * Objset eviction processing is split into into two pieces.
967 * The first marks the objset as evicting, evicts any dbufs that
968 * have a refcount of zero, and then queues up the objset for the
969 * second phase of eviction. Once os->os_dnodes has been cleared by
970 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
971 * The second phase closes the special dnodes, dequeues the objset from
972 * the list of those undergoing eviction, and finally frees the objset.
973 *
974 * NOTE: Due to asynchronous eviction processing (invocation of
975 * dnode_buf_pageout()), it is possible for the meta dnode for the
976 * objset to have no holds even though os->os_dnodes is not empty.
977 */
978 void
979 dmu_objset_evict(objset_t *os)
980 {
981 dsl_dataset_t *ds = os->os_dsl_dataset;
982
983 for (int t = 0; t < TXG_SIZE; t++)
984 ASSERT(!dmu_objset_is_dirty(os, t));
985
986 if (ds)
987 dsl_prop_unregister_all(ds, os);
988
989 if (os->os_sa)
990 sa_tear_down(os);
991
992 dmu_objset_evict_dbufs(os);
993
994 mutex_enter(&os->os_lock);
995 spa_evicting_os_register(os->os_spa, os);
996 if (list_is_empty(&os->os_dnodes)) {
997 mutex_exit(&os->os_lock);
998 dmu_objset_evict_done(os);
999 } else {
1000 mutex_exit(&os->os_lock);
1001 }
1002
1003
1004 }
1005
1006 void
1007 dmu_objset_evict_done(objset_t *os)
1008 {
1009 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
1010
1011 dnode_special_close(&os->os_meta_dnode);
1012 if (DMU_USERUSED_DNODE(os)) {
1013 if (DMU_PROJECTUSED_DNODE(os))
1014 dnode_special_close(&os->os_projectused_dnode);
1015 dnode_special_close(&os->os_userused_dnode);
1016 dnode_special_close(&os->os_groupused_dnode);
1017 }
1018 zil_free(os->os_zil);
1019
1020 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
1021
1022 /*
1023 * This is a barrier to prevent the objset from going away in
1024 * dnode_move() until we can safely ensure that the objset is still in
1025 * use. We consider the objset valid before the barrier and invalid
1026 * after the barrier.
1027 */
1028 rw_enter(&os_lock, RW_READER);
1029 rw_exit(&os_lock);
1030
1031 kmem_free(os->os_obj_next_percpu,
1032 os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0]));
1033
1034 mutex_destroy(&os->os_lock);
1035 mutex_destroy(&os->os_userused_lock);
1036 mutex_destroy(&os->os_obj_lock);
1037 mutex_destroy(&os->os_user_ptr_lock);
1038 mutex_destroy(&os->os_upgrade_lock);
1039 for (int i = 0; i < TXG_SIZE; i++)
1040 multilist_destroy(&os->os_dirty_dnodes[i]);
1041 spa_evicting_os_deregister(os->os_spa, os);
1042 kmem_free(os, sizeof (objset_t));
1043 }
1044
1045 inode_timespec_t
1046 dmu_objset_snap_cmtime(objset_t *os)
1047 {
1048 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
1049 }
1050
1051 objset_t *
1052 dmu_objset_create_impl_dnstats(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
1053 dmu_objset_type_t type, int levels, int blksz, int ibs, dmu_tx_t *tx)
1054 {
1055 objset_t *os;
1056 dnode_t *mdn;
1057
1058 ASSERT(dmu_tx_is_syncing(tx));
1059
1060 if (blksz == 0)
1061 blksz = DNODE_BLOCK_SIZE;
1062 if (ibs == 0)
1063 ibs = DN_MAX_INDBLKSHIFT;
1064
1065 if (ds != NULL)
1066 VERIFY0(dmu_objset_from_ds(ds, &os));
1067 else
1068 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
1069
1070 mdn = DMU_META_DNODE(os);
1071
1072 dnode_allocate(mdn, DMU_OT_DNODE, blksz, ibs, DMU_OT_NONE, 0,
1073 DNODE_MIN_SLOTS, tx);
1074
1075 /*
1076 * We don't want to have to increase the meta-dnode's nlevels
1077 * later, because then we could do it in quiescing context while
1078 * we are also accessing it in open context.
1079 *
1080 * This precaution is not necessary for the MOS (ds == NULL),
1081 * because the MOS is only updated in syncing context.
1082 * This is most fortunate: the MOS is the only objset that
1083 * needs to be synced multiple times as spa_sync() iterates
1084 * to convergence, so minimizing its dn_nlevels matters.
1085 */
1086 if (ds != NULL) {
1087 if (levels == 0) {
1088 levels = 1;
1089
1090 /*
1091 * Determine the number of levels necessary for the
1092 * meta-dnode to contain DN_MAX_OBJECT dnodes. Note
1093 * that in order to ensure that we do not overflow
1094 * 64 bits, there has to be a nlevels that gives us a
1095 * number of blocks > DN_MAX_OBJECT but < 2^64.
1096 * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)
1097 * (10) must be less than (64 - log2(DN_MAX_OBJECT))
1098 * (16).
1099 */
1100 while ((uint64_t)mdn->dn_nblkptr <<
1101 (mdn->dn_datablkshift - DNODE_SHIFT + (levels - 1) *
1102 (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
1103 DN_MAX_OBJECT)
1104 levels++;
1105 }
1106
1107 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
1108 mdn->dn_nlevels = levels;
1109 }
1110
1111 ASSERT(type != DMU_OST_NONE);
1112 ASSERT(type != DMU_OST_ANY);
1113 ASSERT(type < DMU_OST_NUMTYPES);
1114 os->os_phys->os_type = type;
1115
1116 /*
1117 * Enable user accounting if it is enabled and this is not an
1118 * encrypted receive.
1119 */
1120 if (dmu_objset_userused_enabled(os) &&
1121 (!os->os_encrypted || !dmu_objset_is_receiving(os))) {
1122 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1123 if (dmu_objset_userobjused_enabled(os)) {
1124 ASSERT3P(ds, !=, NULL);
1125 ds->ds_feature_activation[
1126 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE;
1127 os->os_phys->os_flags |=
1128 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE;
1129 }
1130 if (dmu_objset_projectquota_enabled(os)) {
1131 ASSERT3P(ds, !=, NULL);
1132 ds->ds_feature_activation[
1133 SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE;
1134 os->os_phys->os_flags |=
1135 OBJSET_FLAG_PROJECTQUOTA_COMPLETE;
1136 }
1137 os->os_flags = os->os_phys->os_flags;
1138 }
1139
1140 dsl_dataset_dirty(ds, tx);
1141
1142 return (os);
1143 }
1144
1145 /* called from dsl for meta-objset */
1146 objset_t *
1147 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
1148 dmu_objset_type_t type, dmu_tx_t *tx)
1149 {
1150 return (dmu_objset_create_impl_dnstats(spa, ds, bp, type, 0, 0, 0, tx));
1151 }
1152
1153 typedef struct dmu_objset_create_arg {
1154 const char *doca_name;
1155 cred_t *doca_cred;
1156 proc_t *doca_proc;
1157 void (*doca_userfunc)(objset_t *os, void *arg,
1158 cred_t *cr, dmu_tx_t *tx);
1159 void *doca_userarg;
1160 dmu_objset_type_t doca_type;
1161 uint64_t doca_flags;
1162 dsl_crypto_params_t *doca_dcp;
1163 } dmu_objset_create_arg_t;
1164
1165 static int
1166 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
1167 {
1168 dmu_objset_create_arg_t *doca = arg;
1169 dsl_pool_t *dp = dmu_tx_pool(tx);
1170 dsl_dir_t *pdd;
1171 dsl_dataset_t *parentds;
1172 objset_t *parentos;
1173 const char *tail;
1174 int error;
1175
1176 if (strchr(doca->doca_name, '@') != NULL)
1177 return (SET_ERROR(EINVAL));
1178
1179 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN)
1180 return (SET_ERROR(ENAMETOOLONG));
1181
1182 if (dataset_nestcheck(doca->doca_name) != 0)
1183 return (SET_ERROR(ENAMETOOLONG));
1184
1185 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
1186 if (error != 0)
1187 return (error);
1188 if (tail == NULL) {
1189 dsl_dir_rele(pdd, FTAG);
1190 return (SET_ERROR(EEXIST));
1191 }
1192
1193 error = dmu_objset_create_crypt_check(pdd, doca->doca_dcp, NULL);
1194 if (error != 0) {
1195 dsl_dir_rele(pdd, FTAG);
1196 return (error);
1197 }
1198
1199 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
1200 doca->doca_cred, doca->doca_proc);
1201 if (error != 0) {
1202 dsl_dir_rele(pdd, FTAG);
1203 return (error);
1204 }
1205
1206 /* can't create below anything but filesystems (eg. no ZVOLs) */
1207 error = dsl_dataset_hold_obj(pdd->dd_pool,
1208 dsl_dir_phys(pdd)->dd_head_dataset_obj, FTAG, &parentds);
1209 if (error != 0) {
1210 dsl_dir_rele(pdd, FTAG);
1211 return (error);
1212 }
1213 error = dmu_objset_from_ds(parentds, &parentos);
1214 if (error != 0) {
1215 dsl_dataset_rele(parentds, FTAG);
1216 dsl_dir_rele(pdd, FTAG);
1217 return (error);
1218 }
1219 if (dmu_objset_type(parentos) != DMU_OST_ZFS) {
1220 dsl_dataset_rele(parentds, FTAG);
1221 dsl_dir_rele(pdd, FTAG);
1222 return (SET_ERROR(ZFS_ERR_WRONG_PARENT));
1223 }
1224 dsl_dataset_rele(parentds, FTAG);
1225 dsl_dir_rele(pdd, FTAG);
1226
1227 return (error);
1228 }
1229
1230 static void
1231 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
1232 {
1233 dmu_objset_create_arg_t *doca = arg;
1234 dsl_pool_t *dp = dmu_tx_pool(tx);
1235 spa_t *spa = dp->dp_spa;
1236 dsl_dir_t *pdd;
1237 const char *tail;
1238 dsl_dataset_t *ds;
1239 uint64_t obj;
1240 blkptr_t *bp;
1241 objset_t *os;
1242 zio_t *rzio;
1243
1244 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
1245
1246 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
1247 doca->doca_cred, doca->doca_dcp, tx);
1248
1249 VERIFY0(dsl_dataset_hold_obj_flags(pdd->dd_pool, obj,
1250 DS_HOLD_FLAG_DECRYPT, FTAG, &ds));
1251 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1252 bp = dsl_dataset_get_blkptr(ds);
1253 os = dmu_objset_create_impl(spa, ds, bp, doca->doca_type, tx);
1254 rrw_exit(&ds->ds_bp_rwlock, FTAG);
1255
1256 if (doca->doca_userfunc != NULL) {
1257 doca->doca_userfunc(os, doca->doca_userarg,
1258 doca->doca_cred, tx);
1259 }
1260
1261 /*
1262 * The doca_userfunc() may write out some data that needs to be
1263 * encrypted if the dataset is encrypted (specifically the root
1264 * directory). This data must be written out before the encryption
1265 * key mapping is removed by dsl_dataset_rele_flags(). Force the
1266 * I/O to occur immediately by invoking the relevant sections of
1267 * dsl_pool_sync().
1268 */
1269 if (os->os_encrypted) {
1270 dsl_dataset_t *tmpds = NULL;
1271 boolean_t need_sync_done = B_FALSE;
1272
1273 mutex_enter(&ds->ds_lock);
1274 ds->ds_owner = FTAG;
1275 mutex_exit(&ds->ds_lock);
1276
1277 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1278 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds,
1279 tx->tx_txg);
1280 if (tmpds != NULL) {
1281 dsl_dataset_sync(ds, rzio, tx);
1282 need_sync_done = B_TRUE;
1283 }
1284 VERIFY0(zio_wait(rzio));
1285
1286 dmu_objset_sync_done(os, tx);
1287 taskq_wait(dp->dp_sync_taskq);
1288 if (txg_list_member(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1289 ASSERT3P(ds->ds_key_mapping, !=, NULL);
1290 key_mapping_rele(spa, ds->ds_key_mapping, ds);
1291 }
1292
1293 rzio = zio_root(spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1294 tmpds = txg_list_remove_this(&dp->dp_dirty_datasets, ds,
1295 tx->tx_txg);
1296 if (tmpds != NULL) {
1297 dmu_buf_rele(ds->ds_dbuf, ds);
1298 dsl_dataset_sync(ds, rzio, tx);
1299 }
1300 VERIFY0(zio_wait(rzio));
1301
1302 if (need_sync_done) {
1303 ASSERT3P(ds->ds_key_mapping, !=, NULL);
1304 key_mapping_rele(spa, ds->ds_key_mapping, ds);
1305 dsl_dataset_sync_done(ds, tx);
1306 dmu_buf_rele(ds->ds_dbuf, ds);
1307 }
1308
1309 mutex_enter(&ds->ds_lock);
1310 ds->ds_owner = NULL;
1311 mutex_exit(&ds->ds_lock);
1312 }
1313
1314 spa_history_log_internal_ds(ds, "create", tx, " ");
1315
1316 dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG);
1317 dsl_dir_rele(pdd, FTAG);
1318 }
1319
1320 int
1321 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
1322 dsl_crypto_params_t *dcp, dmu_objset_create_sync_func_t func, void *arg)
1323 {
1324 dmu_objset_create_arg_t doca;
1325 dsl_crypto_params_t tmp_dcp = { 0 };
1326
1327 doca.doca_name = name;
1328 doca.doca_cred = CRED();
1329 doca.doca_proc = curproc;
1330 doca.doca_flags = flags;
1331 doca.doca_userfunc = func;
1332 doca.doca_userarg = arg;
1333 doca.doca_type = type;
1334
1335 /*
1336 * Some callers (mostly for testing) do not provide a dcp on their
1337 * own but various code inside the sync task will require it to be
1338 * allocated. Rather than adding NULL checks throughout this code
1339 * or adding dummy dcp's to all of the callers we simply create a
1340 * dummy one here and use that. This zero dcp will have the same
1341 * effect as asking for inheritance of all encryption params.
1342 */
1343 doca.doca_dcp = (dcp != NULL) ? dcp : &tmp_dcp;
1344
1345 int rv = dsl_sync_task(name,
1346 dmu_objset_create_check, dmu_objset_create_sync, &doca,
1347 6, ZFS_SPACE_CHECK_NORMAL);
1348
1349 if (rv == 0)
1350 zvol_create_minor(name);
1351 return (rv);
1352 }
1353
1354 typedef struct dmu_objset_clone_arg {
1355 const char *doca_clone;
1356 const char *doca_origin;
1357 cred_t *doca_cred;
1358 proc_t *doca_proc;
1359 } dmu_objset_clone_arg_t;
1360
1361 static int
1362 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
1363 {
1364 dmu_objset_clone_arg_t *doca = arg;
1365 dsl_dir_t *pdd;
1366 const char *tail;
1367 int error;
1368 dsl_dataset_t *origin;
1369 dsl_pool_t *dp = dmu_tx_pool(tx);
1370
1371 if (strchr(doca->doca_clone, '@') != NULL)
1372 return (SET_ERROR(EINVAL));
1373
1374 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN)
1375 return (SET_ERROR(ENAMETOOLONG));
1376
1377 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
1378 if (error != 0)
1379 return (error);
1380 if (tail == NULL) {
1381 dsl_dir_rele(pdd, FTAG);
1382 return (SET_ERROR(EEXIST));
1383 }
1384
1385 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
1386 doca->doca_cred, doca->doca_proc);
1387 if (error != 0) {
1388 dsl_dir_rele(pdd, FTAG);
1389 return (SET_ERROR(EDQUOT));
1390 }
1391
1392 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
1393 if (error != 0) {
1394 dsl_dir_rele(pdd, FTAG);
1395 return (error);
1396 }
1397
1398 /* You can only clone snapshots, not the head datasets. */
1399 if (!origin->ds_is_snapshot) {
1400 dsl_dataset_rele(origin, FTAG);
1401 dsl_dir_rele(pdd, FTAG);
1402 return (SET_ERROR(EINVAL));
1403 }
1404
1405 dsl_dataset_rele(origin, FTAG);
1406 dsl_dir_rele(pdd, FTAG);
1407
1408 return (0);
1409 }
1410
1411 static void
1412 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
1413 {
1414 dmu_objset_clone_arg_t *doca = arg;
1415 dsl_pool_t *dp = dmu_tx_pool(tx);
1416 dsl_dir_t *pdd;
1417 const char *tail;
1418 dsl_dataset_t *origin, *ds;
1419 uint64_t obj;
1420 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1421
1422 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
1423 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
1424
1425 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
1426 doca->doca_cred, NULL, tx);
1427
1428 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
1429 dsl_dataset_name(origin, namebuf);
1430 spa_history_log_internal_ds(ds, "clone", tx,
1431 "origin=%s (%llu)", namebuf, (u_longlong_t)origin->ds_object);
1432 dsl_dataset_rele(ds, FTAG);
1433 dsl_dataset_rele(origin, FTAG);
1434 dsl_dir_rele(pdd, FTAG);
1435 }
1436
1437 int
1438 dmu_objset_clone(const char *clone, const char *origin)
1439 {
1440 dmu_objset_clone_arg_t doca;
1441
1442 doca.doca_clone = clone;
1443 doca.doca_origin = origin;
1444 doca.doca_cred = CRED();
1445 doca.doca_proc = curproc;
1446
1447 int rv = dsl_sync_task(clone,
1448 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
1449 6, ZFS_SPACE_CHECK_NORMAL);
1450
1451 if (rv == 0)
1452 zvol_create_minor(clone);
1453
1454 return (rv);
1455 }
1456
1457 int
1458 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1459 {
1460 int err;
1461 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1462 nvlist_t *snaps = fnvlist_alloc();
1463
1464 fnvlist_add_boolean(snaps, longsnap);
1465 kmem_strfree(longsnap);
1466 err = dsl_dataset_snapshot(snaps, NULL, NULL);
1467 fnvlist_free(snaps);
1468 return (err);
1469 }
1470
1471 static void
1472 dmu_objset_upgrade_task_cb(void *data)
1473 {
1474 objset_t *os = data;
1475
1476 mutex_enter(&os->os_upgrade_lock);
1477 os->os_upgrade_status = EINTR;
1478 if (!os->os_upgrade_exit) {
1479 int status;
1480
1481 mutex_exit(&os->os_upgrade_lock);
1482
1483 status = os->os_upgrade_cb(os);
1484
1485 mutex_enter(&os->os_upgrade_lock);
1486
1487 os->os_upgrade_status = status;
1488 }
1489 os->os_upgrade_exit = B_TRUE;
1490 os->os_upgrade_id = 0;
1491 mutex_exit(&os->os_upgrade_lock);
1492 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
1493 }
1494
1495 static void
1496 dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb)
1497 {
1498 if (os->os_upgrade_id != 0)
1499 return;
1500
1501 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1502 dsl_dataset_long_hold(dmu_objset_ds(os), upgrade_tag);
1503
1504 mutex_enter(&os->os_upgrade_lock);
1505 if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) {
1506 os->os_upgrade_exit = B_FALSE;
1507 os->os_upgrade_cb = cb;
1508 os->os_upgrade_id = taskq_dispatch(
1509 os->os_spa->spa_upgrade_taskq,
1510 dmu_objset_upgrade_task_cb, os, TQ_SLEEP);
1511 if (os->os_upgrade_id == TASKQID_INVALID) {
1512 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
1513 os->os_upgrade_status = ENOMEM;
1514 }
1515 } else {
1516 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
1517 }
1518 mutex_exit(&os->os_upgrade_lock);
1519 }
1520
1521 static void
1522 dmu_objset_upgrade_stop(objset_t *os)
1523 {
1524 mutex_enter(&os->os_upgrade_lock);
1525 os->os_upgrade_exit = B_TRUE;
1526 if (os->os_upgrade_id != 0) {
1527 taskqid_t id = os->os_upgrade_id;
1528
1529 os->os_upgrade_id = 0;
1530 mutex_exit(&os->os_upgrade_lock);
1531
1532 if ((taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id)) == 0) {
1533 dsl_dataset_long_rele(dmu_objset_ds(os), upgrade_tag);
1534 }
1535 txg_wait_synced(os->os_spa->spa_dsl_pool, 0);
1536 } else {
1537 mutex_exit(&os->os_upgrade_lock);
1538 }
1539 }
1540
1541 static void
1542 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx)
1543 {
1544 dnode_t *dn;
1545
1546 while ((dn = multilist_sublist_head(list)) != NULL) {
1547 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1548 ASSERT(dn->dn_dbuf->db_data_pending);
1549 /*
1550 * Initialize dn_zio outside dnode_sync() because the
1551 * meta-dnode needs to set it outside dnode_sync().
1552 */
1553 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1554 ASSERT(dn->dn_zio);
1555
1556 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1557 multilist_sublist_remove(list, dn);
1558
1559 /*
1560 * See the comment above dnode_rele_task() for an explanation
1561 * of why this dnode hold is always needed (even when not
1562 * doing user accounting).
1563 */
1564 multilist_t *newlist = &dn->dn_objset->os_synced_dnodes;
1565 (void) dnode_add_ref(dn, newlist);
1566 multilist_insert(newlist, dn);
1567
1568 dnode_sync(dn, tx);
1569 }
1570 }
1571
1572 static void
1573 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1574 {
1575 (void) abuf;
1576 blkptr_t *bp = zio->io_bp;
1577 objset_t *os = arg;
1578 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1579 uint64_t fill = 0;
1580
1581 ASSERT(!BP_IS_EMBEDDED(bp));
1582 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1583 ASSERT0(BP_GET_LEVEL(bp));
1584
1585 /*
1586 * Update rootbp fill count: it should be the number of objects
1587 * allocated in the object set (not counting the "special"
1588 * objects that are stored in the objset_phys_t -- the meta
1589 * dnode and user/group/project accounting objects).
1590 */
1591 for (int i = 0; i < dnp->dn_nblkptr; i++)
1592 fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1593
1594 BP_SET_FILL(bp, fill);
1595
1596 if (os->os_dsl_dataset != NULL)
1597 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG);
1598 *os->os_rootbp = *bp;
1599 if (os->os_dsl_dataset != NULL)
1600 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG);
1601 }
1602
1603 static void
1604 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1605 {
1606 (void) abuf;
1607 blkptr_t *bp = zio->io_bp;
1608 blkptr_t *bp_orig = &zio->io_bp_orig;
1609 objset_t *os = arg;
1610
1611 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1612 ASSERT(BP_EQUAL(bp, bp_orig));
1613 } else {
1614 dsl_dataset_t *ds = os->os_dsl_dataset;
1615 dmu_tx_t *tx = os->os_synctx;
1616
1617 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1618 dsl_dataset_block_born(ds, bp, tx);
1619 }
1620 kmem_free(bp, sizeof (*bp));
1621 }
1622
1623 typedef struct sync_dnodes_arg {
1624 multilist_t *sda_list;
1625 int sda_sublist_idx;
1626 multilist_t *sda_newlist;
1627 dmu_tx_t *sda_tx;
1628 } sync_dnodes_arg_t;
1629
1630 static void
1631 sync_dnodes_task(void *arg)
1632 {
1633 sync_dnodes_arg_t *sda = arg;
1634
1635 multilist_sublist_t *ms =
1636 multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx);
1637
1638 dmu_objset_sync_dnodes(ms, sda->sda_tx);
1639
1640 multilist_sublist_unlock(ms);
1641
1642 kmem_free(sda, sizeof (*sda));
1643 }
1644
1645
1646 /* called from dsl */
1647 void
1648 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1649 {
1650 int txgoff;
1651 zbookmark_phys_t zb;
1652 zio_prop_t zp;
1653 zio_t *zio;
1654 list_t *list;
1655 dbuf_dirty_record_t *dr;
1656 int num_sublists;
1657 multilist_t *ml;
1658 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP);
1659 *blkptr_copy = *os->os_rootbp;
1660
1661 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", (u_longlong_t)tx->tx_txg);
1662
1663 ASSERT(dmu_tx_is_syncing(tx));
1664 /* XXX the write_done callback should really give us the tx... */
1665 os->os_synctx = tx;
1666
1667 if (os->os_dsl_dataset == NULL) {
1668 /*
1669 * This is the MOS. If we have upgraded,
1670 * spa_max_replication() could change, so reset
1671 * os_copies here.
1672 */
1673 os->os_copies = spa_max_replication(os->os_spa);
1674 }
1675
1676 /*
1677 * Create the root block IO
1678 */
1679 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1680 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1681 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1682 arc_release(os->os_phys_buf, &os->os_phys_buf);
1683
1684 dmu_write_policy(os, NULL, 0, 0, &zp);
1685
1686 /*
1687 * If we are either claiming the ZIL or doing a raw receive, write
1688 * out the os_phys_buf raw. Neither of these actions will effect the
1689 * MAC at this point.
1690 */
1691 if (os->os_raw_receive ||
1692 os->os_next_write_raw[tx->tx_txg & TXG_MASK]) {
1693 ASSERT(os->os_encrypted);
1694 arc_convert_to_raw(os->os_phys_buf,
1695 os->os_dsl_dataset->ds_object, ZFS_HOST_BYTEORDER,
1696 DMU_OT_OBJSET, NULL, NULL, NULL);
1697 }
1698
1699 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1700 blkptr_copy, os->os_phys_buf, B_FALSE, dmu_os_is_l2cacheable(os),
1701 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done,
1702 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1703
1704 /*
1705 * Sync special dnodes - the parent IO for the sync is the root block
1706 */
1707 DMU_META_DNODE(os)->dn_zio = zio;
1708 dnode_sync(DMU_META_DNODE(os), tx);
1709
1710 os->os_phys->os_flags = os->os_flags;
1711
1712 if (DMU_USERUSED_DNODE(os) &&
1713 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1714 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1715 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1716 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1717 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1718 }
1719
1720 if (DMU_PROJECTUSED_DNODE(os) &&
1721 DMU_PROJECTUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1722 DMU_PROJECTUSED_DNODE(os)->dn_zio = zio;
1723 dnode_sync(DMU_PROJECTUSED_DNODE(os), tx);
1724 }
1725
1726 txgoff = tx->tx_txg & TXG_MASK;
1727
1728 /*
1729 * We must create the list here because it uses the
1730 * dn_dirty_link[] of this txg. But it may already
1731 * exist because we call dsl_dataset_sync() twice per txg.
1732 */
1733 if (os->os_synced_dnodes.ml_sublists == NULL) {
1734 multilist_create(&os->os_synced_dnodes, sizeof (dnode_t),
1735 offsetof(dnode_t, dn_dirty_link[txgoff]),
1736 dnode_multilist_index_func);
1737 } else {
1738 ASSERT3U(os->os_synced_dnodes.ml_offset, ==,
1739 offsetof(dnode_t, dn_dirty_link[txgoff]));
1740 }
1741
1742 ml = &os->os_dirty_dnodes[txgoff];
1743 num_sublists = multilist_get_num_sublists(ml);
1744 for (int i = 0; i < num_sublists; i++) {
1745 if (multilist_sublist_is_empty_idx(ml, i))
1746 continue;
1747 sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP);
1748 sda->sda_list = ml;
1749 sda->sda_sublist_idx = i;
1750 sda->sda_tx = tx;
1751 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1752 sync_dnodes_task, sda, 0);
1753 /* callback frees sda */
1754 }
1755 taskq_wait(dmu_objset_pool(os)->dp_sync_taskq);
1756
1757 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1758 while ((dr = list_head(list)) != NULL) {
1759 ASSERT0(dr->dr_dbuf->db_level);
1760 list_remove(list, dr);
1761 zio_nowait(dr->dr_zio);
1762 }
1763
1764 /* Enable dnode backfill if enough objects have been freed. */
1765 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) {
1766 os->os_rescan_dnodes = B_TRUE;
1767 os->os_freed_dnodes = 0;
1768 }
1769
1770 /*
1771 * Free intent log blocks up to this tx.
1772 */
1773 zil_sync(os->os_zil, tx);
1774 os->os_phys->os_zil_header = os->os_zil_header;
1775 zio_nowait(zio);
1776 }
1777
1778 boolean_t
1779 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1780 {
1781 return (!multilist_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]));
1782 }
1783
1784 static file_info_cb_t *file_cbs[DMU_OST_NUMTYPES];
1785
1786 void
1787 dmu_objset_register_type(dmu_objset_type_t ost, file_info_cb_t *cb)
1788 {
1789 file_cbs[ost] = cb;
1790 }
1791
1792 int
1793 dmu_get_file_info(objset_t *os, dmu_object_type_t bonustype, const void *data,
1794 zfs_file_info_t *zfi)
1795 {
1796 file_info_cb_t *cb = file_cbs[os->os_phys->os_type];
1797 if (cb == NULL)
1798 return (EINVAL);
1799 return (cb(bonustype, data, zfi));
1800 }
1801
1802 boolean_t
1803 dmu_objset_userused_enabled(objset_t *os)
1804 {
1805 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1806 file_cbs[os->os_phys->os_type] != NULL &&
1807 DMU_USERUSED_DNODE(os) != NULL);
1808 }
1809
1810 boolean_t
1811 dmu_objset_userobjused_enabled(objset_t *os)
1812 {
1813 return (dmu_objset_userused_enabled(os) &&
1814 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING));
1815 }
1816
1817 boolean_t
1818 dmu_objset_projectquota_enabled(objset_t *os)
1819 {
1820 return (file_cbs[os->os_phys->os_type] != NULL &&
1821 DMU_PROJECTUSED_DNODE(os) != NULL &&
1822 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_PROJECT_QUOTA));
1823 }
1824
1825 typedef struct userquota_node {
1826 /* must be in the first filed, see userquota_update_cache() */
1827 char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN];
1828 int64_t uqn_delta;
1829 avl_node_t uqn_node;
1830 } userquota_node_t;
1831
1832 typedef struct userquota_cache {
1833 avl_tree_t uqc_user_deltas;
1834 avl_tree_t uqc_group_deltas;
1835 avl_tree_t uqc_project_deltas;
1836 } userquota_cache_t;
1837
1838 static int
1839 userquota_compare(const void *l, const void *r)
1840 {
1841 const userquota_node_t *luqn = l;
1842 const userquota_node_t *ruqn = r;
1843 int rv;
1844
1845 /*
1846 * NB: can only access uqn_id because userquota_update_cache() doesn't
1847 * pass in an entire userquota_node_t.
1848 */
1849 rv = strcmp(luqn->uqn_id, ruqn->uqn_id);
1850
1851 return (TREE_ISIGN(rv));
1852 }
1853
1854 static void
1855 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx)
1856 {
1857 void *cookie;
1858 userquota_node_t *uqn;
1859
1860 ASSERT(dmu_tx_is_syncing(tx));
1861
1862 cookie = NULL;
1863 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas,
1864 &cookie)) != NULL) {
1865 /*
1866 * os_userused_lock protects against concurrent calls to
1867 * zap_increment_int(). It's needed because zap_increment_int()
1868 * is not thread-safe (i.e. not atomic).
1869 */
1870 mutex_enter(&os->os_userused_lock);
1871 VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT,
1872 uqn->uqn_id, uqn->uqn_delta, tx));
1873 mutex_exit(&os->os_userused_lock);
1874 kmem_free(uqn, sizeof (*uqn));
1875 }
1876 avl_destroy(&cache->uqc_user_deltas);
1877
1878 cookie = NULL;
1879 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas,
1880 &cookie)) != NULL) {
1881 mutex_enter(&os->os_userused_lock);
1882 VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT,
1883 uqn->uqn_id, uqn->uqn_delta, tx));
1884 mutex_exit(&os->os_userused_lock);
1885 kmem_free(uqn, sizeof (*uqn));
1886 }
1887 avl_destroy(&cache->uqc_group_deltas);
1888
1889 if (dmu_objset_projectquota_enabled(os)) {
1890 cookie = NULL;
1891 while ((uqn = avl_destroy_nodes(&cache->uqc_project_deltas,
1892 &cookie)) != NULL) {
1893 mutex_enter(&os->os_userused_lock);
1894 VERIFY0(zap_increment(os, DMU_PROJECTUSED_OBJECT,
1895 uqn->uqn_id, uqn->uqn_delta, tx));
1896 mutex_exit(&os->os_userused_lock);
1897 kmem_free(uqn, sizeof (*uqn));
1898 }
1899 avl_destroy(&cache->uqc_project_deltas);
1900 }
1901 }
1902
1903 static void
1904 userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta)
1905 {
1906 userquota_node_t *uqn;
1907 avl_index_t idx;
1908
1909 ASSERT(strlen(id) < sizeof (uqn->uqn_id));
1910 /*
1911 * Use id directly for searching because uqn_id is the first field of
1912 * userquota_node_t and fields after uqn_id won't be accessed in
1913 * avl_find().
1914 */
1915 uqn = avl_find(avl, (const void *)id, &idx);
1916 if (uqn == NULL) {
1917 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP);
1918 strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id));
1919 avl_insert(avl, uqn, idx);
1920 }
1921 uqn->uqn_delta += delta;
1922 }
1923
1924 static void
1925 do_userquota_update(objset_t *os, userquota_cache_t *cache, uint64_t used,
1926 uint64_t flags, uint64_t user, uint64_t group, uint64_t project,
1927 boolean_t subtract)
1928 {
1929 if (flags & DNODE_FLAG_USERUSED_ACCOUNTED) {
1930 int64_t delta = DNODE_MIN_SIZE + used;
1931 char name[20];
1932
1933 if (subtract)
1934 delta = -delta;
1935
1936 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)user);
1937 userquota_update_cache(&cache->uqc_user_deltas, name, delta);
1938
1939 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)group);
1940 userquota_update_cache(&cache->uqc_group_deltas, name, delta);
1941
1942 if (dmu_objset_projectquota_enabled(os)) {
1943 (void) snprintf(name, sizeof (name), "%llx",
1944 (longlong_t)project);
1945 userquota_update_cache(&cache->uqc_project_deltas,
1946 name, delta);
1947 }
1948 }
1949 }
1950
1951 static void
1952 do_userobjquota_update(objset_t *os, userquota_cache_t *cache, uint64_t flags,
1953 uint64_t user, uint64_t group, uint64_t project, boolean_t subtract)
1954 {
1955 if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) {
1956 char name[20 + DMU_OBJACCT_PREFIX_LEN];
1957 int delta = subtract ? -1 : 1;
1958
1959 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx",
1960 (longlong_t)user);
1961 userquota_update_cache(&cache->uqc_user_deltas, name, delta);
1962
1963 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx",
1964 (longlong_t)group);
1965 userquota_update_cache(&cache->uqc_group_deltas, name, delta);
1966
1967 if (dmu_objset_projectquota_enabled(os)) {
1968 (void) snprintf(name, sizeof (name),
1969 DMU_OBJACCT_PREFIX "%llx", (longlong_t)project);
1970 userquota_update_cache(&cache->uqc_project_deltas,
1971 name, delta);
1972 }
1973 }
1974 }
1975
1976 typedef struct userquota_updates_arg {
1977 objset_t *uua_os;
1978 int uua_sublist_idx;
1979 dmu_tx_t *uua_tx;
1980 } userquota_updates_arg_t;
1981
1982 static void
1983 userquota_updates_task(void *arg)
1984 {
1985 userquota_updates_arg_t *uua = arg;
1986 objset_t *os = uua->uua_os;
1987 dmu_tx_t *tx = uua->uua_tx;
1988 dnode_t *dn;
1989 userquota_cache_t cache = { { 0 } };
1990
1991 multilist_sublist_t *list =
1992 multilist_sublist_lock(&os->os_synced_dnodes, uua->uua_sublist_idx);
1993
1994 ASSERT(multilist_sublist_head(list) == NULL ||
1995 dmu_objset_userused_enabled(os));
1996 avl_create(&cache.uqc_user_deltas, userquota_compare,
1997 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1998 avl_create(&cache.uqc_group_deltas, userquota_compare,
1999 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
2000 if (dmu_objset_projectquota_enabled(os))
2001 avl_create(&cache.uqc_project_deltas, userquota_compare,
2002 sizeof (userquota_node_t), offsetof(userquota_node_t,
2003 uqn_node));
2004
2005 while ((dn = multilist_sublist_head(list)) != NULL) {
2006 int flags;
2007 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
2008 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
2009 dn->dn_phys->dn_flags &
2010 DNODE_FLAG_USERUSED_ACCOUNTED);
2011
2012 flags = dn->dn_id_flags;
2013 ASSERT(flags);
2014 if (flags & DN_ID_OLD_EXIST) {
2015 do_userquota_update(os, &cache, dn->dn_oldused,
2016 dn->dn_oldflags, dn->dn_olduid, dn->dn_oldgid,
2017 dn->dn_oldprojid, B_TRUE);
2018 do_userobjquota_update(os, &cache, dn->dn_oldflags,
2019 dn->dn_olduid, dn->dn_oldgid,
2020 dn->dn_oldprojid, B_TRUE);
2021 }
2022 if (flags & DN_ID_NEW_EXIST) {
2023 do_userquota_update(os, &cache,
2024 DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags,
2025 dn->dn_newuid, dn->dn_newgid,
2026 dn->dn_newprojid, B_FALSE);
2027 do_userobjquota_update(os, &cache,
2028 dn->dn_phys->dn_flags, dn->dn_newuid, dn->dn_newgid,
2029 dn->dn_newprojid, B_FALSE);
2030 }
2031
2032 mutex_enter(&dn->dn_mtx);
2033 dn->dn_oldused = 0;
2034 dn->dn_oldflags = 0;
2035 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
2036 dn->dn_olduid = dn->dn_newuid;
2037 dn->dn_oldgid = dn->dn_newgid;
2038 dn->dn_oldprojid = dn->dn_newprojid;
2039 dn->dn_id_flags |= DN_ID_OLD_EXIST;
2040 if (dn->dn_bonuslen == 0)
2041 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
2042 else
2043 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
2044 }
2045 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
2046 mutex_exit(&dn->dn_mtx);
2047
2048 multilist_sublist_remove(list, dn);
2049 dnode_rele(dn, &os->os_synced_dnodes);
2050 }
2051 do_userquota_cacheflush(os, &cache, tx);
2052 multilist_sublist_unlock(list);
2053 kmem_free(uua, sizeof (*uua));
2054 }
2055
2056 /*
2057 * Release dnode holds from dmu_objset_sync_dnodes(). When the dnode is being
2058 * synced (i.e. we have issued the zio's for blocks in the dnode), it can't be
2059 * evicted because the block containing the dnode can't be evicted until it is
2060 * written out. However, this hold is necessary to prevent the dnode_t from
2061 * being moved (via dnode_move()) while it's still referenced by
2062 * dbuf_dirty_record_t:dr_dnode. And dr_dnode is needed for
2063 * dirty_lightweight_leaf-type dirty records.
2064 *
2065 * If we are doing user-object accounting, the dnode_rele() happens from
2066 * userquota_updates_task() instead.
2067 */
2068 static void
2069 dnode_rele_task(void *arg)
2070 {
2071 userquota_updates_arg_t *uua = arg;
2072 objset_t *os = uua->uua_os;
2073
2074 multilist_sublist_t *list =
2075 multilist_sublist_lock(&os->os_synced_dnodes, uua->uua_sublist_idx);
2076
2077 dnode_t *dn;
2078 while ((dn = multilist_sublist_head(list)) != NULL) {
2079 multilist_sublist_remove(list, dn);
2080 dnode_rele(dn, &os->os_synced_dnodes);
2081 }
2082 multilist_sublist_unlock(list);
2083 kmem_free(uua, sizeof (*uua));
2084 }
2085
2086 /*
2087 * Return TRUE if userquota updates are needed.
2088 */
2089 static boolean_t
2090 dmu_objset_do_userquota_updates_prep(objset_t *os, dmu_tx_t *tx)
2091 {
2092 if (!dmu_objset_userused_enabled(os))
2093 return (B_FALSE);
2094
2095 /*
2096 * If this is a raw receive just return and handle accounting
2097 * later when we have the keys loaded. We also don't do user
2098 * accounting during claiming since the datasets are not owned
2099 * for the duration of claiming and this txg should only be
2100 * used for recovery.
2101 */
2102 if (os->os_encrypted && dmu_objset_is_receiving(os))
2103 return (B_FALSE);
2104
2105 if (tx->tx_txg <= os->os_spa->spa_claim_max_txg)
2106 return (B_FALSE);
2107
2108 /* Allocate the user/group/project used objects if necessary. */
2109 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
2110 VERIFY0(zap_create_claim(os,
2111 DMU_USERUSED_OBJECT,
2112 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
2113 VERIFY0(zap_create_claim(os,
2114 DMU_GROUPUSED_OBJECT,
2115 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
2116 }
2117
2118 if (dmu_objset_projectquota_enabled(os) &&
2119 DMU_PROJECTUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
2120 VERIFY0(zap_create_claim(os, DMU_PROJECTUSED_OBJECT,
2121 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
2122 }
2123 return (B_TRUE);
2124 }
2125
2126 /*
2127 * Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and
2128 * also release the holds on the dnodes from dmu_objset_sync_dnodes().
2129 * The caller must taskq_wait(dp_sync_taskq).
2130 */
2131 void
2132 dmu_objset_sync_done(objset_t *os, dmu_tx_t *tx)
2133 {
2134 boolean_t need_userquota = dmu_objset_do_userquota_updates_prep(os, tx);
2135
2136 int num_sublists = multilist_get_num_sublists(&os->os_synced_dnodes);
2137 for (int i = 0; i < num_sublists; i++) {
2138 userquota_updates_arg_t *uua =
2139 kmem_alloc(sizeof (*uua), KM_SLEEP);
2140 uua->uua_os = os;
2141 uua->uua_sublist_idx = i;
2142 uua->uua_tx = tx;
2143
2144 /*
2145 * If we don't need to update userquotas, use
2146 * dnode_rele_task() to call dnode_rele()
2147 */
2148 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
2149 need_userquota ? userquota_updates_task : dnode_rele_task,
2150 uua, 0);
2151 /* callback frees uua */
2152 }
2153 }
2154
2155
2156 /*
2157 * Returns a pointer to data to find uid/gid from
2158 *
2159 * If a dirty record for transaction group that is syncing can't
2160 * be found then NULL is returned. In the NULL case it is assumed
2161 * the uid/gid aren't changing.
2162 */
2163 static void *
2164 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
2165 {
2166 dbuf_dirty_record_t *dr;
2167 void *data;
2168
2169 if (db->db_dirtycnt == 0)
2170 return (db->db.db_data); /* Nothing is changing */
2171
2172 dr = dbuf_find_dirty_eq(db, tx->tx_txg);
2173
2174 if (dr == NULL) {
2175 data = NULL;
2176 } else {
2177 if (dr->dr_dnode->dn_bonuslen == 0 &&
2178 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
2179 data = dr->dt.dl.dr_data->b_data;
2180 else
2181 data = dr->dt.dl.dr_data;
2182 }
2183
2184 return (data);
2185 }
2186
2187 void
2188 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
2189 {
2190 objset_t *os = dn->dn_objset;
2191 void *data = NULL;
2192 dmu_buf_impl_t *db = NULL;
2193 int flags = dn->dn_id_flags;
2194 int error;
2195 boolean_t have_spill = B_FALSE;
2196
2197 if (!dmu_objset_userused_enabled(dn->dn_objset))
2198 return;
2199
2200 /*
2201 * Raw receives introduce a problem with user accounting. Raw
2202 * receives cannot update the user accounting info because the
2203 * user ids and the sizes are encrypted. To guarantee that we
2204 * never end up with bad user accounting, we simply disable it
2205 * during raw receives. We also disable this for normal receives
2206 * so that an incremental raw receive may be done on top of an
2207 * existing non-raw receive.
2208 */
2209 if (os->os_encrypted && dmu_objset_is_receiving(os))
2210 return;
2211
2212 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
2213 DN_ID_CHKED_SPILL)))
2214 return;
2215
2216 if (before && dn->dn_bonuslen != 0)
2217 data = DN_BONUS(dn->dn_phys);
2218 else if (!before && dn->dn_bonuslen != 0) {
2219 if (dn->dn_bonus) {
2220 db = dn->dn_bonus;
2221 mutex_enter(&db->db_mtx);
2222 data = dmu_objset_userquota_find_data(db, tx);
2223 } else {
2224 data = DN_BONUS(dn->dn_phys);
2225 }
2226 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
2227 int rf = 0;
2228
2229 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
2230 rf |= DB_RF_HAVESTRUCT;
2231 error = dmu_spill_hold_by_dnode(dn,
2232 rf | DB_RF_MUST_SUCCEED,
2233 FTAG, (dmu_buf_t **)&db);
2234 ASSERT(error == 0);
2235 mutex_enter(&db->db_mtx);
2236 data = (before) ? db->db.db_data :
2237 dmu_objset_userquota_find_data(db, tx);
2238 have_spill = B_TRUE;
2239 } else {
2240 mutex_enter(&dn->dn_mtx);
2241 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
2242 mutex_exit(&dn->dn_mtx);
2243 return;
2244 }
2245
2246 /*
2247 * Must always call the callback in case the object
2248 * type has changed and that type isn't an object type to track
2249 */
2250 zfs_file_info_t zfi;
2251 error = file_cbs[os->os_phys->os_type](dn->dn_bonustype, data, &zfi);
2252
2253 if (before) {
2254 ASSERT(data);
2255 dn->dn_olduid = zfi.zfi_user;
2256 dn->dn_oldgid = zfi.zfi_group;
2257 dn->dn_oldprojid = zfi.zfi_project;
2258 } else if (data) {
2259 dn->dn_newuid = zfi.zfi_user;
2260 dn->dn_newgid = zfi.zfi_group;
2261 dn->dn_newprojid = zfi.zfi_project;
2262 }
2263
2264 /*
2265 * Preserve existing uid/gid when the callback can't determine
2266 * what the new uid/gid are and the callback returned EEXIST.
2267 * The EEXIST error tells us to just use the existing uid/gid.
2268 * If we don't know what the old values are then just assign
2269 * them to 0, since that is a new file being created.
2270 */
2271 if (!before && data == NULL && error == EEXIST) {
2272 if (flags & DN_ID_OLD_EXIST) {
2273 dn->dn_newuid = dn->dn_olduid;
2274 dn->dn_newgid = dn->dn_oldgid;
2275 dn->dn_newprojid = dn->dn_oldprojid;
2276 } else {
2277 dn->dn_newuid = 0;
2278 dn->dn_newgid = 0;
2279 dn->dn_newprojid = ZFS_DEFAULT_PROJID;
2280 }
2281 error = 0;
2282 }
2283
2284 if (db)
2285 mutex_exit(&db->db_mtx);
2286
2287 mutex_enter(&dn->dn_mtx);
2288 if (error == 0 && before)
2289 dn->dn_id_flags |= DN_ID_OLD_EXIST;
2290 if (error == 0 && !before)
2291 dn->dn_id_flags |= DN_ID_NEW_EXIST;
2292
2293 if (have_spill) {
2294 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
2295 } else {
2296 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
2297 }
2298 mutex_exit(&dn->dn_mtx);
2299 if (have_spill)
2300 dmu_buf_rele((dmu_buf_t *)db, FTAG);
2301 }
2302
2303 boolean_t
2304 dmu_objset_userspace_present(objset_t *os)
2305 {
2306 return (os->os_phys->os_flags &
2307 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
2308 }
2309
2310 boolean_t
2311 dmu_objset_userobjspace_present(objset_t *os)
2312 {
2313 return (os->os_phys->os_flags &
2314 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE);
2315 }
2316
2317 boolean_t
2318 dmu_objset_projectquota_present(objset_t *os)
2319 {
2320 return (os->os_phys->os_flags &
2321 OBJSET_FLAG_PROJECTQUOTA_COMPLETE);
2322 }
2323
2324 static int
2325 dmu_objset_space_upgrade(objset_t *os)
2326 {
2327 uint64_t obj;
2328 int err = 0;
2329
2330 /*
2331 * We simply need to mark every object dirty, so that it will be
2332 * synced out and now accounted. If this is called
2333 * concurrently, or if we already did some work before crashing,
2334 * that's fine, since we track each object's accounted state
2335 * independently.
2336 */
2337
2338 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
2339 dmu_tx_t *tx;
2340 dmu_buf_t *db;
2341 int objerr;
2342
2343 mutex_enter(&os->os_upgrade_lock);
2344 if (os->os_upgrade_exit)
2345 err = SET_ERROR(EINTR);
2346 mutex_exit(&os->os_upgrade_lock);
2347 if (err != 0)
2348 return (err);
2349
2350 if (issig(JUSTLOOKING) && issig(FORREAL))
2351 return (SET_ERROR(EINTR));
2352
2353 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
2354 if (objerr != 0)
2355 continue;
2356 tx = dmu_tx_create(os);
2357 dmu_tx_hold_bonus(tx, obj);
2358 objerr = dmu_tx_assign(tx, TXG_WAIT);
2359 if (objerr != 0) {
2360 dmu_buf_rele(db, FTAG);
2361 dmu_tx_abort(tx);
2362 continue;
2363 }
2364 dmu_buf_will_dirty(db, tx);
2365 dmu_buf_rele(db, FTAG);
2366 dmu_tx_commit(tx);
2367 }
2368 return (0);
2369 }
2370
2371 static int
2372 dmu_objset_userspace_upgrade_cb(objset_t *os)
2373 {
2374 int err = 0;
2375
2376 if (dmu_objset_userspace_present(os))
2377 return (0);
2378 if (dmu_objset_is_snapshot(os))
2379 return (SET_ERROR(EINVAL));
2380 if (!dmu_objset_userused_enabled(os))
2381 return (SET_ERROR(ENOTSUP));
2382
2383 err = dmu_objset_space_upgrade(os);
2384 if (err)
2385 return (err);
2386
2387 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
2388 txg_wait_synced(dmu_objset_pool(os), 0);
2389 return (0);
2390 }
2391
2392 void
2393 dmu_objset_userspace_upgrade(objset_t *os)
2394 {
2395 dmu_objset_upgrade(os, dmu_objset_userspace_upgrade_cb);
2396 }
2397
2398 static int
2399 dmu_objset_id_quota_upgrade_cb(objset_t *os)
2400 {
2401 int err = 0;
2402
2403 if (dmu_objset_userobjspace_present(os) &&
2404 dmu_objset_projectquota_present(os))
2405 return (0);
2406 if (dmu_objset_is_snapshot(os))
2407 return (SET_ERROR(EINVAL));
2408 if (!dmu_objset_userused_enabled(os))
2409 return (SET_ERROR(ENOTSUP));
2410 if (!dmu_objset_projectquota_enabled(os) &&
2411 dmu_objset_userobjspace_present(os))
2412 return (SET_ERROR(ENOTSUP));
2413
2414 if (dmu_objset_userobjused_enabled(os))
2415 dmu_objset_ds(os)->ds_feature_activation[
2416 SPA_FEATURE_USEROBJ_ACCOUNTING] = (void *)B_TRUE;
2417 if (dmu_objset_projectquota_enabled(os))
2418 dmu_objset_ds(os)->ds_feature_activation[
2419 SPA_FEATURE_PROJECT_QUOTA] = (void *)B_TRUE;
2420
2421 err = dmu_objset_space_upgrade(os);
2422 if (err)
2423 return (err);
2424
2425 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
2426 if (dmu_objset_userobjused_enabled(os))
2427 os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE;
2428 if (dmu_objset_projectquota_enabled(os))
2429 os->os_flags |= OBJSET_FLAG_PROJECTQUOTA_COMPLETE;
2430
2431 txg_wait_synced(dmu_objset_pool(os), 0);
2432 return (0);
2433 }
2434
2435 void
2436 dmu_objset_id_quota_upgrade(objset_t *os)
2437 {
2438 dmu_objset_upgrade(os, dmu_objset_id_quota_upgrade_cb);
2439 }
2440
2441 boolean_t
2442 dmu_objset_userobjspace_upgradable(objset_t *os)
2443 {
2444 return (dmu_objset_type(os) == DMU_OST_ZFS &&
2445 !dmu_objset_is_snapshot(os) &&
2446 dmu_objset_userobjused_enabled(os) &&
2447 !dmu_objset_userobjspace_present(os) &&
2448 spa_writeable(dmu_objset_spa(os)));
2449 }
2450
2451 boolean_t
2452 dmu_objset_projectquota_upgradable(objset_t *os)
2453 {
2454 return (dmu_objset_type(os) == DMU_OST_ZFS &&
2455 !dmu_objset_is_snapshot(os) &&
2456 dmu_objset_projectquota_enabled(os) &&
2457 !dmu_objset_projectquota_present(os) &&
2458 spa_writeable(dmu_objset_spa(os)));
2459 }
2460
2461 void
2462 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
2463 uint64_t *usedobjsp, uint64_t *availobjsp)
2464 {
2465 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
2466 usedobjsp, availobjsp);
2467 }
2468
2469 uint64_t
2470 dmu_objset_fsid_guid(objset_t *os)
2471 {
2472 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
2473 }
2474
2475 void
2476 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
2477 {
2478 stat->dds_type = os->os_phys->os_type;
2479 if (os->os_dsl_dataset)
2480 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
2481 }
2482
2483 void
2484 dmu_objset_stats(objset_t *os, nvlist_t *nv)
2485 {
2486 ASSERT(os->os_dsl_dataset ||
2487 os->os_phys->os_type == DMU_OST_META);
2488
2489 if (os->os_dsl_dataset != NULL)
2490 dsl_dataset_stats(os->os_dsl_dataset, nv);
2491
2492 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
2493 os->os_phys->os_type);
2494 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
2495 dmu_objset_userspace_present(os));
2496 }
2497
2498 int
2499 dmu_objset_is_snapshot(objset_t *os)
2500 {
2501 if (os->os_dsl_dataset != NULL)
2502 return (os->os_dsl_dataset->ds_is_snapshot);
2503 else
2504 return (B_FALSE);
2505 }
2506
2507 int
2508 dmu_snapshot_realname(objset_t *os, const char *name, char *real, int maxlen,
2509 boolean_t *conflict)
2510 {
2511 dsl_dataset_t *ds = os->os_dsl_dataset;
2512 uint64_t ignored;
2513
2514 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
2515 return (SET_ERROR(ENOENT));
2516
2517 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
2518 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
2519 MT_NORMALIZE, real, maxlen, conflict));
2520 }
2521
2522 int
2523 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
2524 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
2525 {
2526 dsl_dataset_t *ds = os->os_dsl_dataset;
2527 zap_cursor_t cursor;
2528 zap_attribute_t attr;
2529
2530 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
2531
2532 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
2533 return (SET_ERROR(ENOENT));
2534
2535 zap_cursor_init_serialized(&cursor,
2536 ds->ds_dir->dd_pool->dp_meta_objset,
2537 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
2538
2539 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
2540 zap_cursor_fini(&cursor);
2541 return (SET_ERROR(ENOENT));
2542 }
2543
2544 if (strlen(attr.za_name) + 1 > namelen) {
2545 zap_cursor_fini(&cursor);
2546 return (SET_ERROR(ENAMETOOLONG));
2547 }
2548
2549 (void) strlcpy(name, attr.za_name, namelen);
2550 if (idp)
2551 *idp = attr.za_first_integer;
2552 if (case_conflict)
2553 *case_conflict = attr.za_normalization_conflict;
2554 zap_cursor_advance(&cursor);
2555 *offp = zap_cursor_serialize(&cursor);
2556 zap_cursor_fini(&cursor);
2557
2558 return (0);
2559 }
2560
2561 int
2562 dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value)
2563 {
2564 return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value));
2565 }
2566
2567 int
2568 dmu_dir_list_next(objset_t *os, int namelen, char *name,
2569 uint64_t *idp, uint64_t *offp)
2570 {
2571 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
2572 zap_cursor_t cursor;
2573 zap_attribute_t attr;
2574
2575 /* there is no next dir on a snapshot! */
2576 if (os->os_dsl_dataset->ds_object !=
2577 dsl_dir_phys(dd)->dd_head_dataset_obj)
2578 return (SET_ERROR(ENOENT));
2579
2580 zap_cursor_init_serialized(&cursor,
2581 dd->dd_pool->dp_meta_objset,
2582 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
2583
2584 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
2585 zap_cursor_fini(&cursor);
2586 return (SET_ERROR(ENOENT));
2587 }
2588
2589 if (strlen(attr.za_name) + 1 > namelen) {
2590 zap_cursor_fini(&cursor);
2591 return (SET_ERROR(ENAMETOOLONG));
2592 }
2593
2594 (void) strlcpy(name, attr.za_name, namelen);
2595 if (idp)
2596 *idp = attr.za_first_integer;
2597 zap_cursor_advance(&cursor);
2598 *offp = zap_cursor_serialize(&cursor);
2599 zap_cursor_fini(&cursor);
2600
2601 return (0);
2602 }
2603
2604 typedef struct dmu_objset_find_ctx {
2605 taskq_t *dc_tq;
2606 dsl_pool_t *dc_dp;
2607 uint64_t dc_ddobj;
2608 char *dc_ddname; /* last component of ddobj's name */
2609 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
2610 void *dc_arg;
2611 int dc_flags;
2612 kmutex_t *dc_error_lock;
2613 int *dc_error;
2614 } dmu_objset_find_ctx_t;
2615
2616 static void
2617 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
2618 {
2619 dsl_pool_t *dp = dcp->dc_dp;
2620 dsl_dir_t *dd;
2621 dsl_dataset_t *ds;
2622 zap_cursor_t zc;
2623 zap_attribute_t *attr;
2624 uint64_t thisobj;
2625 int err = 0;
2626
2627 /* don't process if there already was an error */
2628 if (*dcp->dc_error != 0)
2629 goto out;
2630
2631 /*
2632 * Note: passing the name (dc_ddname) here is optional, but it
2633 * improves performance because we don't need to call
2634 * zap_value_search() to determine the name.
2635 */
2636 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd);
2637 if (err != 0)
2638 goto out;
2639
2640 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2641 if (dd->dd_myname[0] == '$') {
2642 dsl_dir_rele(dd, FTAG);
2643 goto out;
2644 }
2645
2646 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2647 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2648
2649 /*
2650 * Iterate over all children.
2651 */
2652 if (dcp->dc_flags & DS_FIND_CHILDREN) {
2653 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2654 dsl_dir_phys(dd)->dd_child_dir_zapobj);
2655 zap_cursor_retrieve(&zc, attr) == 0;
2656 (void) zap_cursor_advance(&zc)) {
2657 ASSERT3U(attr->za_integer_length, ==,
2658 sizeof (uint64_t));
2659 ASSERT3U(attr->za_num_integers, ==, 1);
2660
2661 dmu_objset_find_ctx_t *child_dcp =
2662 kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
2663 *child_dcp = *dcp;
2664 child_dcp->dc_ddobj = attr->za_first_integer;
2665 child_dcp->dc_ddname = spa_strdup(attr->za_name);
2666 if (dcp->dc_tq != NULL)
2667 (void) taskq_dispatch(dcp->dc_tq,
2668 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
2669 else
2670 dmu_objset_find_dp_impl(child_dcp);
2671 }
2672 zap_cursor_fini(&zc);
2673 }
2674
2675 /*
2676 * Iterate over all snapshots.
2677 */
2678 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
2679 dsl_dataset_t *ds;
2680 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2681
2682 if (err == 0) {
2683 uint64_t snapobj;
2684
2685 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2686 dsl_dataset_rele(ds, FTAG);
2687
2688 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2689 zap_cursor_retrieve(&zc, attr) == 0;
2690 (void) zap_cursor_advance(&zc)) {
2691 ASSERT3U(attr->za_integer_length, ==,
2692 sizeof (uint64_t));
2693 ASSERT3U(attr->za_num_integers, ==, 1);
2694
2695 err = dsl_dataset_hold_obj(dp,
2696 attr->za_first_integer, FTAG, &ds);
2697 if (err != 0)
2698 break;
2699 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2700 dsl_dataset_rele(ds, FTAG);
2701 if (err != 0)
2702 break;
2703 }
2704 zap_cursor_fini(&zc);
2705 }
2706 }
2707
2708 kmem_free(attr, sizeof (zap_attribute_t));
2709
2710 if (err != 0) {
2711 dsl_dir_rele(dd, FTAG);
2712 goto out;
2713 }
2714
2715 /*
2716 * Apply to self.
2717 */
2718 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2719
2720 /*
2721 * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2722 * that the dir will remain cached, and we won't have to re-instantiate
2723 * it (which could be expensive due to finding its name via
2724 * zap_value_search()).
2725 */
2726 dsl_dir_rele(dd, FTAG);
2727 if (err != 0)
2728 goto out;
2729 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2730 dsl_dataset_rele(ds, FTAG);
2731
2732 out:
2733 if (err != 0) {
2734 mutex_enter(dcp->dc_error_lock);
2735 /* only keep first error */
2736 if (*dcp->dc_error == 0)
2737 *dcp->dc_error = err;
2738 mutex_exit(dcp->dc_error_lock);
2739 }
2740
2741 if (dcp->dc_ddname != NULL)
2742 spa_strfree(dcp->dc_ddname);
2743 kmem_free(dcp, sizeof (*dcp));
2744 }
2745
2746 static void
2747 dmu_objset_find_dp_cb(void *arg)
2748 {
2749 dmu_objset_find_ctx_t *dcp = arg;
2750 dsl_pool_t *dp = dcp->dc_dp;
2751
2752 /*
2753 * We need to get a pool_config_lock here, as there are several
2754 * assert(pool_config_held) down the stack. Getting a lock via
2755 * dsl_pool_config_enter is risky, as it might be stalled by a
2756 * pending writer. This would deadlock, as the write lock can
2757 * only be granted when our parent thread gives up the lock.
2758 * The _prio interface gives us priority over a pending writer.
2759 */
2760 dsl_pool_config_enter_prio(dp, FTAG);
2761
2762 dmu_objset_find_dp_impl(dcp);
2763
2764 dsl_pool_config_exit(dp, FTAG);
2765 }
2766
2767 /*
2768 * Find objsets under and including ddobj, call func(ds) on each.
2769 * The order for the enumeration is completely undefined.
2770 * func is called with dsl_pool_config held.
2771 */
2772 int
2773 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
2774 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
2775 {
2776 int error = 0;
2777 taskq_t *tq = NULL;
2778 int ntasks;
2779 dmu_objset_find_ctx_t *dcp;
2780 kmutex_t err_lock;
2781
2782 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
2783 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
2784 dcp->dc_tq = NULL;
2785 dcp->dc_dp = dp;
2786 dcp->dc_ddobj = ddobj;
2787 dcp->dc_ddname = NULL;
2788 dcp->dc_func = func;
2789 dcp->dc_arg = arg;
2790 dcp->dc_flags = flags;
2791 dcp->dc_error_lock = &err_lock;
2792 dcp->dc_error = &error;
2793
2794 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
2795 /*
2796 * In case a write lock is held we can't make use of
2797 * parallelism, as down the stack of the worker threads
2798 * the lock is asserted via dsl_pool_config_held.
2799 * In case of a read lock this is solved by getting a read
2800 * lock in each worker thread, which isn't possible in case
2801 * of a writer lock. So we fall back to the synchronous path
2802 * here.
2803 * In the future it might be possible to get some magic into
2804 * dsl_pool_config_held in a way that it returns true for
2805 * the worker threads so that a single lock held from this
2806 * thread suffices. For now, stay single threaded.
2807 */
2808 dmu_objset_find_dp_impl(dcp);
2809 mutex_destroy(&err_lock);
2810
2811 return (error);
2812 }
2813
2814 ntasks = dmu_find_threads;
2815 if (ntasks == 0)
2816 ntasks = vdev_count_leaves(dp->dp_spa) * 4;
2817 tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks,
2818 INT_MAX, 0);
2819 if (tq == NULL) {
2820 kmem_free(dcp, sizeof (*dcp));
2821 mutex_destroy(&err_lock);
2822
2823 return (SET_ERROR(ENOMEM));
2824 }
2825 dcp->dc_tq = tq;
2826
2827 /* dcp will be freed by task */
2828 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
2829
2830 /*
2831 * PORTING: this code relies on the property of taskq_wait to wait
2832 * until no more tasks are queued and no more tasks are active. As
2833 * we always queue new tasks from within other tasks, task_wait
2834 * reliably waits for the full recursion to finish, even though we
2835 * enqueue new tasks after taskq_wait has been called.
2836 * On platforms other than illumos, taskq_wait may not have this
2837 * property.
2838 */
2839 taskq_wait(tq);
2840 taskq_destroy(tq);
2841 mutex_destroy(&err_lock);
2842
2843 return (error);
2844 }
2845
2846 /*
2847 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2848 * The dp_config_rwlock must not be held when this is called, and it
2849 * will not be held when the callback is called.
2850 * Therefore this function should only be used when the pool is not changing
2851 * (e.g. in syncing context), or the callback can deal with the possible races.
2852 */
2853 static int
2854 dmu_objset_find_impl(spa_t *spa, const char *name,
2855 int func(const char *, void *), void *arg, int flags)
2856 {
2857 dsl_dir_t *dd;
2858 dsl_pool_t *dp = spa_get_dsl(spa);
2859 dsl_dataset_t *ds;
2860 zap_cursor_t zc;
2861 zap_attribute_t *attr;
2862 char *child;
2863 uint64_t thisobj;
2864 int err;
2865
2866 dsl_pool_config_enter(dp, FTAG);
2867
2868 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
2869 if (err != 0) {
2870 dsl_pool_config_exit(dp, FTAG);
2871 return (err);
2872 }
2873
2874 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2875 if (dd->dd_myname[0] == '$') {
2876 dsl_dir_rele(dd, FTAG);
2877 dsl_pool_config_exit(dp, FTAG);
2878 return (0);
2879 }
2880
2881 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2882 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2883
2884 /*
2885 * Iterate over all children.
2886 */
2887 if (flags & DS_FIND_CHILDREN) {
2888 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2889 dsl_dir_phys(dd)->dd_child_dir_zapobj);
2890 zap_cursor_retrieve(&zc, attr) == 0;
2891 (void) zap_cursor_advance(&zc)) {
2892 ASSERT3U(attr->za_integer_length, ==,
2893 sizeof (uint64_t));
2894 ASSERT3U(attr->za_num_integers, ==, 1);
2895
2896 child = kmem_asprintf("%s/%s", name, attr->za_name);
2897 dsl_pool_config_exit(dp, FTAG);
2898 err = dmu_objset_find_impl(spa, child,
2899 func, arg, flags);
2900 dsl_pool_config_enter(dp, FTAG);
2901 kmem_strfree(child);
2902 if (err != 0)
2903 break;
2904 }
2905 zap_cursor_fini(&zc);
2906
2907 if (err != 0) {
2908 dsl_dir_rele(dd, FTAG);
2909 dsl_pool_config_exit(dp, FTAG);
2910 kmem_free(attr, sizeof (zap_attribute_t));
2911 return (err);
2912 }
2913 }
2914
2915 /*
2916 * Iterate over all snapshots.
2917 */
2918 if (flags & DS_FIND_SNAPSHOTS) {
2919 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2920
2921 if (err == 0) {
2922 uint64_t snapobj;
2923
2924 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2925 dsl_dataset_rele(ds, FTAG);
2926
2927 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2928 zap_cursor_retrieve(&zc, attr) == 0;
2929 (void) zap_cursor_advance(&zc)) {
2930 ASSERT3U(attr->za_integer_length, ==,
2931 sizeof (uint64_t));
2932 ASSERT3U(attr->za_num_integers, ==, 1);
2933
2934 child = kmem_asprintf("%s@%s",
2935 name, attr->za_name);
2936 dsl_pool_config_exit(dp, FTAG);
2937 err = func(child, arg);
2938 dsl_pool_config_enter(dp, FTAG);
2939 kmem_strfree(child);
2940 if (err != 0)
2941 break;
2942 }
2943 zap_cursor_fini(&zc);
2944 }
2945 }
2946
2947 dsl_dir_rele(dd, FTAG);
2948 kmem_free(attr, sizeof (zap_attribute_t));
2949 dsl_pool_config_exit(dp, FTAG);
2950
2951 if (err != 0)
2952 return (err);
2953
2954 /* Apply to self. */
2955 return (func(name, arg));
2956 }
2957
2958 /*
2959 * See comment above dmu_objset_find_impl().
2960 */
2961 int
2962 dmu_objset_find(const char *name, int func(const char *, void *), void *arg,
2963 int flags)
2964 {
2965 spa_t *spa;
2966 int error;
2967
2968 error = spa_open(name, &spa, FTAG);
2969 if (error != 0)
2970 return (error);
2971 error = dmu_objset_find_impl(spa, name, func, arg, flags);
2972 spa_close(spa, FTAG);
2973 return (error);
2974 }
2975
2976 boolean_t
2977 dmu_objset_incompatible_encryption_version(objset_t *os)
2978 {
2979 return (dsl_dir_incompatible_encryption_version(
2980 os->os_dsl_dataset->ds_dir));
2981 }
2982
2983 void
2984 dmu_objset_set_user(objset_t *os, void *user_ptr)
2985 {
2986 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2987 os->os_user_ptr = user_ptr;
2988 }
2989
2990 void *
2991 dmu_objset_get_user(objset_t *os)
2992 {
2993 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2994 return (os->os_user_ptr);
2995 }
2996
2997 /*
2998 * Determine name of filesystem, given name of snapshot.
2999 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
3000 */
3001 int
3002 dmu_fsname(const char *snapname, char *buf)
3003 {
3004 char *atp = strchr(snapname, '@');
3005 if (atp == NULL)
3006 return (SET_ERROR(EINVAL));
3007 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN)
3008 return (SET_ERROR(ENAMETOOLONG));
3009 (void) strlcpy(buf, snapname, atp - snapname + 1);
3010 return (0);
3011 }
3012
3013 /*
3014 * Call when we think we're going to write/free space in open context
3015 * to track the amount of dirty data in the open txg, which is also the
3016 * amount of memory that can not be evicted until this txg syncs.
3017 *
3018 * Note that there are two conditions where this can be called from
3019 * syncing context:
3020 *
3021 * [1] When we just created the dataset, in which case we go on with
3022 * updating any accounting of dirty data as usual.
3023 * [2] When we are dirtying MOS data, in which case we only update the
3024 * pool's accounting of dirty data.
3025 */
3026 void
3027 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx)
3028 {
3029 dsl_dataset_t *ds = os->os_dsl_dataset;
3030 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space);
3031
3032 if (ds != NULL) {
3033 dsl_dir_willuse_space(ds->ds_dir, aspace, tx);
3034 }
3035
3036 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx);
3037 }
3038
3039 #if defined(_KERNEL)
3040 EXPORT_SYMBOL(dmu_objset_zil);
3041 EXPORT_SYMBOL(dmu_objset_pool);
3042 EXPORT_SYMBOL(dmu_objset_ds);
3043 EXPORT_SYMBOL(dmu_objset_type);
3044 EXPORT_SYMBOL(dmu_objset_name);
3045 EXPORT_SYMBOL(dmu_objset_hold);
3046 EXPORT_SYMBOL(dmu_objset_hold_flags);
3047 EXPORT_SYMBOL(dmu_objset_own);
3048 EXPORT_SYMBOL(dmu_objset_rele);
3049 EXPORT_SYMBOL(dmu_objset_rele_flags);
3050 EXPORT_SYMBOL(dmu_objset_disown);
3051 EXPORT_SYMBOL(dmu_objset_from_ds);
3052 EXPORT_SYMBOL(dmu_objset_create);
3053 EXPORT_SYMBOL(dmu_objset_clone);
3054 EXPORT_SYMBOL(dmu_objset_stats);
3055 EXPORT_SYMBOL(dmu_objset_fast_stat);
3056 EXPORT_SYMBOL(dmu_objset_spa);
3057 EXPORT_SYMBOL(dmu_objset_space);
3058 EXPORT_SYMBOL(dmu_objset_fsid_guid);
3059 EXPORT_SYMBOL(dmu_objset_find);
3060 EXPORT_SYMBOL(dmu_objset_byteswap);
3061 EXPORT_SYMBOL(dmu_objset_evict_dbufs);
3062 EXPORT_SYMBOL(dmu_objset_snap_cmtime);
3063 EXPORT_SYMBOL(dmu_objset_dnodesize);
3064
3065 EXPORT_SYMBOL(dmu_objset_sync);
3066 EXPORT_SYMBOL(dmu_objset_is_dirty);
3067 EXPORT_SYMBOL(dmu_objset_create_impl_dnstats);
3068 EXPORT_SYMBOL(dmu_objset_create_impl);
3069 EXPORT_SYMBOL(dmu_objset_open_impl);
3070 EXPORT_SYMBOL(dmu_objset_evict);
3071 EXPORT_SYMBOL(dmu_objset_register_type);
3072 EXPORT_SYMBOL(dmu_objset_sync_done);
3073 EXPORT_SYMBOL(dmu_objset_userquota_get_ids);
3074 EXPORT_SYMBOL(dmu_objset_userused_enabled);
3075 EXPORT_SYMBOL(dmu_objset_userspace_upgrade);
3076 EXPORT_SYMBOL(dmu_objset_userspace_present);
3077 EXPORT_SYMBOL(dmu_objset_userobjused_enabled);
3078 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable);
3079 EXPORT_SYMBOL(dmu_objset_userobjspace_present);
3080 EXPORT_SYMBOL(dmu_objset_projectquota_enabled);
3081 EXPORT_SYMBOL(dmu_objset_projectquota_present);
3082 EXPORT_SYMBOL(dmu_objset_projectquota_upgradable);
3083 EXPORT_SYMBOL(dmu_objset_id_quota_upgrade);
3084 #endif