1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 */
28
29 /* Portions Copyright 2010 Robert Milkowski */
30
31 #include <sys/cred.h>
32 #include <sys/zfs_context.h>
33 #include <sys/dmu_objset.h>
34 #include <sys/dsl_dir.h>
35 #include <sys/dsl_dataset.h>
36 #include <sys/dsl_prop.h>
37 #include <sys/dsl_pool.h>
38 #include <sys/dsl_synctask.h>
39 #include <sys/dsl_deleg.h>
40 #include <sys/dnode.h>
41 #include <sys/dbuf.h>
42 #include <sys/zvol.h>
43 #include <sys/dmu_tx.h>
44 #include <sys/zap.h>
45 #include <sys/zil.h>
46 #include <sys/dmu_impl.h>
47 #include <sys/zfs_ioctl.h>
48 #include <sys/sa.h>
49 #include <sys/zfs_onexit.h>
50 #include <sys/dsl_destroy.h>
51
52 /*
53 * Needed to close a window in dnode_move() that allows the objset to be freed
54 * before it can be safely accessed.
55 */
56 krwlock_t os_lock;
57
58 void
59 dmu_objset_init(void)
60 {
61 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
62 }
63
64 void
65 dmu_objset_fini(void)
66 {
67 rw_destroy(&os_lock);
68 }
69
70 spa_t *
71 dmu_objset_spa(objset_t *os)
72 {
73 return (os->os_spa);
74 }
75
76 zilog_t *
77 dmu_objset_zil(objset_t *os)
78 {
79 return (os->os_zil);
80 }
81
82 dsl_pool_t *
83 dmu_objset_pool(objset_t *os)
84 {
85 dsl_dataset_t *ds;
86
87 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
88 return (ds->ds_dir->dd_pool);
89 else
90 return (spa_get_dsl(os->os_spa));
91 }
92
93 dsl_dataset_t *
94 dmu_objset_ds(objset_t *os)
95 {
96 return (os->os_dsl_dataset);
97 }
98
99 dmu_objset_type_t
100 dmu_objset_type(objset_t *os)
101 {
102 return (os->os_phys->os_type);
103 }
104
105 void
106 dmu_objset_name(objset_t *os, char *buf)
107 {
108 dsl_dataset_name(os->os_dsl_dataset, buf);
109 }
110
111 uint64_t
112 dmu_objset_id(objset_t *os)
113 {
114 dsl_dataset_t *ds = os->os_dsl_dataset;
115
116 return (ds ? ds->ds_object : 0);
117 }
118
119 zfs_sync_type_t
120 dmu_objset_syncprop(objset_t *os)
121 {
122 return (os->os_sync);
123 }
124
125 zfs_logbias_op_t
126 dmu_objset_logbias(objset_t *os)
127 {
128 return (os->os_logbias);
129 }
130
131 static void
132 checksum_changed_cb(void *arg, uint64_t newval)
133 {
134 objset_t *os = arg;
135
136 /*
137 * Inheritance should have been done by now.
138 */
139 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
140
141 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
142 }
143
144 static void
145 compression_changed_cb(void *arg, uint64_t newval)
146 {
147 objset_t *os = arg;
148
149 /*
150 * Inheritance and range checking should have been done by now.
151 */
152 ASSERT(newval != ZIO_COMPRESS_INHERIT);
153
154 os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
155 }
156
157 static void
158 copies_changed_cb(void *arg, uint64_t newval)
159 {
160 objset_t *os = arg;
161
162 /*
163 * Inheritance and range checking should have been done by now.
164 */
165 ASSERT(newval > 0);
166 ASSERT(newval <= spa_max_replication(os->os_spa));
167
168 os->os_copies = newval;
169 }
170
171 static void
172 dedup_changed_cb(void *arg, uint64_t newval)
173 {
174 objset_t *os = arg;
175 spa_t *spa = os->os_spa;
176 enum zio_checksum checksum;
177
178 /*
179 * Inheritance should have been done by now.
180 */
181 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
182
183 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
184
185 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
186 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
187 }
188
189 static void
190 primary_cache_changed_cb(void *arg, uint64_t newval)
191 {
192 objset_t *os = arg;
193
194 /*
195 * Inheritance and range checking should have been done by now.
196 */
197 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
198 newval == ZFS_CACHE_METADATA);
199
200 os->os_primary_cache = newval;
201 }
202
203 static void
204 secondary_cache_changed_cb(void *arg, uint64_t newval)
205 {
206 objset_t *os = arg;
207
208 /*
209 * Inheritance and range checking should have been done by now.
210 */
211 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
212 newval == ZFS_CACHE_METADATA);
213
214 os->os_secondary_cache = newval;
215 }
216
217 static void
218 sync_changed_cb(void *arg, uint64_t newval)
219 {
220 objset_t *os = arg;
221
222 /*
223 * Inheritance and range checking should have been done by now.
224 */
225 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
226 newval == ZFS_SYNC_DISABLED);
227
228 os->os_sync = newval;
229 if (os->os_zil)
230 zil_set_sync(os->os_zil, newval);
231 }
232
233 static void
234 redundant_metadata_changed_cb(void *arg, uint64_t newval)
235 {
236 objset_t *os = arg;
237
238 /*
239 * Inheritance and range checking should have been done by now.
240 */
241 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
242 newval == ZFS_REDUNDANT_METADATA_MOST);
243
244 os->os_redundant_metadata = newval;
245 }
246
247 static void
248 logbias_changed_cb(void *arg, uint64_t newval)
249 {
250 objset_t *os = arg;
251
252 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
253 newval == ZFS_LOGBIAS_THROUGHPUT);
254 os->os_logbias = newval;
255 if (os->os_zil)
256 zil_set_logbias(os->os_zil, newval);
257 }
258
259 static void
260 recordsize_changed_cb(void *arg, uint64_t newval)
261 {
262 objset_t *os = arg;
263
264 os->os_recordsize = newval;
265 }
266
267 void
268 dmu_objset_byteswap(void *buf, size_t size)
269 {
270 objset_phys_t *osp = buf;
271
272 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
273 dnode_byteswap(&osp->os_meta_dnode);
274 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
275 osp->os_type = BSWAP_64(osp->os_type);
276 osp->os_flags = BSWAP_64(osp->os_flags);
277 if (size == sizeof (objset_phys_t)) {
278 dnode_byteswap(&osp->os_userused_dnode);
279 dnode_byteswap(&osp->os_groupused_dnode);
280 }
281 }
282
283 int
284 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
285 objset_t **osp)
286 {
287 objset_t *os;
288 int i, err;
289
290 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
291
292 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
293 os->os_dsl_dataset = ds;
294 os->os_spa = spa;
295 os->os_rootbp = bp;
296 if (!BP_IS_HOLE(os->os_rootbp)) {
297 arc_flags_t aflags = ARC_FLAG_WAIT;
298 zbookmark_phys_t zb;
299 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
300 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
301
302 if (DMU_OS_IS_L2CACHEABLE(os))
303 aflags |= ARC_FLAG_L2CACHE;
304 if (DMU_OS_IS_L2COMPRESSIBLE(os))
305 aflags |= ARC_FLAG_L2COMPRESS;
306
307 dprintf_bp(os->os_rootbp, "reading %s", "");
308 err = arc_read(NULL, spa, os->os_rootbp,
309 arc_getbuf_func, &os->os_phys_buf,
310 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
311 if (err != 0) {
312 kmem_free(os, sizeof (objset_t));
313 /* convert checksum errors into IO errors */
314 if (err == ECKSUM)
315 err = SET_ERROR(EIO);
316 return (err);
317 }
318
319 /* Increase the blocksize if we are permitted. */
320 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
321 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
322 arc_buf_t *buf = arc_buf_alloc(spa,
323 sizeof (objset_phys_t), &os->os_phys_buf,
324 ARC_BUFC_METADATA);
325 bzero(buf->b_data, sizeof (objset_phys_t));
326 bcopy(os->os_phys_buf->b_data, buf->b_data,
327 arc_buf_size(os->os_phys_buf));
328 (void) arc_buf_remove_ref(os->os_phys_buf,
329 &os->os_phys_buf);
330 os->os_phys_buf = buf;
331 }
332
333 os->os_phys = os->os_phys_buf->b_data;
334 os->os_flags = os->os_phys->os_flags;
335 } else {
336 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
337 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
338 os->os_phys_buf = arc_buf_alloc(spa, size,
339 &os->os_phys_buf, ARC_BUFC_METADATA);
340 os->os_phys = os->os_phys_buf->b_data;
341 bzero(os->os_phys, size);
342 }
343
344 /*
345 * Note: the changed_cb will be called once before the register
346 * func returns, thus changing the checksum/compression from the
347 * default (fletcher2/off). Snapshots don't need to know about
348 * checksum/compression/copies.
349 */
350 if (ds != NULL) {
351 err = dsl_prop_register(ds,
352 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
353 primary_cache_changed_cb, os);
354 if (err == 0) {
355 err = dsl_prop_register(ds,
356 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
357 secondary_cache_changed_cb, os);
358 }
359 if (!ds->ds_is_snapshot) {
360 if (err == 0) {
361 err = dsl_prop_register(ds,
362 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
363 checksum_changed_cb, os);
364 }
365 if (err == 0) {
366 err = dsl_prop_register(ds,
367 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
368 compression_changed_cb, os);
369 }
370 if (err == 0) {
371 err = dsl_prop_register(ds,
372 zfs_prop_to_name(ZFS_PROP_COPIES),
373 copies_changed_cb, os);
374 }
375 if (err == 0) {
376 err = dsl_prop_register(ds,
377 zfs_prop_to_name(ZFS_PROP_DEDUP),
378 dedup_changed_cb, os);
379 }
380 if (err == 0) {
381 err = dsl_prop_register(ds,
382 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
383 logbias_changed_cb, os);
384 }
385 if (err == 0) {
386 err = dsl_prop_register(ds,
387 zfs_prop_to_name(ZFS_PROP_SYNC),
388 sync_changed_cb, os);
389 }
390 if (err == 0) {
391 err = dsl_prop_register(ds,
392 zfs_prop_to_name(
393 ZFS_PROP_REDUNDANT_METADATA),
394 redundant_metadata_changed_cb, os);
395 }
396 if (err == 0) {
397 err = dsl_prop_register(ds,
398 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
399 recordsize_changed_cb, os);
400 }
401 }
402 if (err != 0) {
403 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
404 &os->os_phys_buf));
405 kmem_free(os, sizeof (objset_t));
406 return (err);
407 }
408 } else {
409 /* It's the meta-objset. */
410 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
411 os->os_compress = ZIO_COMPRESS_LZJB;
412 os->os_copies = spa_max_replication(spa);
413 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
414 os->os_dedup_verify = B_FALSE;
415 os->os_logbias = ZFS_LOGBIAS_LATENCY;
416 os->os_sync = ZFS_SYNC_STANDARD;
417 os->os_primary_cache = ZFS_CACHE_ALL;
418 os->os_secondary_cache = ZFS_CACHE_ALL;
419 }
420
421 if (ds == NULL || !ds->ds_is_snapshot)
422 os->os_zil_header = os->os_phys->os_zil_header;
423 os->os_zil = zil_alloc(os, &os->os_zil_header);
424
425 for (i = 0; i < TXG_SIZE; i++) {
426 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
427 offsetof(dnode_t, dn_dirty_link[i]));
428 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
429 offsetof(dnode_t, dn_dirty_link[i]));
430 }
431 list_create(&os->os_dnodes, sizeof (dnode_t),
432 offsetof(dnode_t, dn_link));
433 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
434 offsetof(dmu_buf_impl_t, db_link));
435
436 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
437 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
438 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
439
440 dnode_special_open(os, &os->os_phys->os_meta_dnode,
441 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
442 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
443 dnode_special_open(os, &os->os_phys->os_userused_dnode,
444 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
445 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
446 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
447 }
448
449 *osp = os;
450 return (0);
451 }
452
453 int
454 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
455 {
456 int err = 0;
457
458 mutex_enter(&ds->ds_opening_lock);
459 if (ds->ds_objset == NULL) {
460 objset_t *os;
461 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
462 ds, dsl_dataset_get_blkptr(ds), &os);
463
464 if (err == 0) {
465 mutex_enter(&ds->ds_lock);
466 ASSERT(ds->ds_objset == NULL);
467 ds->ds_objset = os;
468 mutex_exit(&ds->ds_lock);
469 }
470 }
471 *osp = ds->ds_objset;
472 mutex_exit(&ds->ds_opening_lock);
473 return (err);
474 }
475
476 /*
477 * Holds the pool while the objset is held. Therefore only one objset
478 * can be held at a time.
479 */
480 int
481 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
482 {
483 dsl_pool_t *dp;
484 dsl_dataset_t *ds;
485 int err;
486
487 err = dsl_pool_hold(name, tag, &dp);
488 if (err != 0)
489 return (err);
490 err = dsl_dataset_hold(dp, name, tag, &ds);
491 if (err != 0) {
492 dsl_pool_rele(dp, tag);
493 return (err);
494 }
495
496 err = dmu_objset_from_ds(ds, osp);
497 if (err != 0) {
498 dsl_dataset_rele(ds, tag);
499 dsl_pool_rele(dp, tag);
500 }
501
502 return (err);
503 }
504
505 /*
506 * dsl_pool must not be held when this is called.
507 * Upon successful return, there will be a longhold on the dataset,
508 * and the dsl_pool will not be held.
509 */
510 int
511 dmu_objset_own(const char *name, dmu_objset_type_t type,
512 boolean_t readonly, void *tag, objset_t **osp)
513 {
514 dsl_pool_t *dp;
515 dsl_dataset_t *ds;
516 int err;
517
518 err = dsl_pool_hold(name, FTAG, &dp);
519 if (err != 0)
520 return (err);
521 err = dsl_dataset_own(dp, name, tag, &ds);
522 if (err != 0) {
523 dsl_pool_rele(dp, FTAG);
524 return (err);
525 }
526
527 err = dmu_objset_from_ds(ds, osp);
528 dsl_pool_rele(dp, FTAG);
529 if (err != 0) {
530 dsl_dataset_disown(ds, tag);
531 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
532 dsl_dataset_disown(ds, tag);
533 return (SET_ERROR(EINVAL));
534 } else if (!readonly && ds->ds_is_snapshot) {
535 dsl_dataset_disown(ds, tag);
536 return (SET_ERROR(EROFS));
537 }
538 return (err);
539 }
540
541 void
542 dmu_objset_rele(objset_t *os, void *tag)
543 {
544 dsl_pool_t *dp = dmu_objset_pool(os);
545 dsl_dataset_rele(os->os_dsl_dataset, tag);
546 dsl_pool_rele(dp, tag);
547 }
548
549 /*
550 * When we are called, os MUST refer to an objset associated with a dataset
551 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
552 * == tag. We will then release and reacquire ownership of the dataset while
553 * holding the pool config_rwlock to avoid intervening namespace or ownership
554 * changes may occur.
555 *
556 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
557 * release the hold on its dataset and acquire a new one on the dataset of the
558 * same name so that it can be partially torn down and reconstructed.
559 */
560 void
561 dmu_objset_refresh_ownership(objset_t *os, void *tag)
562 {
563 dsl_pool_t *dp;
564 dsl_dataset_t *ds, *newds;
565 char name[MAXNAMELEN];
566
567 ds = os->os_dsl_dataset;
568 VERIFY3P(ds, !=, NULL);
569 VERIFY3P(ds->ds_owner, ==, tag);
570 VERIFY(dsl_dataset_long_held(ds));
571
572 dsl_dataset_name(ds, name);
573 dp = dmu_objset_pool(os);
574 dsl_pool_config_enter(dp, FTAG);
575 dmu_objset_disown(os, tag);
576 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
577 VERIFY3P(newds, ==, os->os_dsl_dataset);
578 dsl_pool_config_exit(dp, FTAG);
579 }
580
581 void
582 dmu_objset_disown(objset_t *os, void *tag)
583 {
584 dsl_dataset_disown(os->os_dsl_dataset, tag);
585 }
586
587 void
588 dmu_objset_evict_dbufs(objset_t *os)
589 {
590 dnode_t dn_marker;
591 dnode_t *dn;
592
593 mutex_enter(&os->os_lock);
594 dn = list_head(&os->os_dnodes);
595 while (dn != NULL) {
596 /*
597 * Skip dnodes without holds. We have to do this dance
598 * because dnode_add_ref() only works if there is already a
599 * hold. If the dnode has no holds, then it has no dbufs.
600 */
601 if (dnode_add_ref(dn, FTAG)) {
602 list_insert_after(&os->os_dnodes, dn, &dn_marker);
603 mutex_exit(&os->os_lock);
604
605 dnode_evict_dbufs(dn);
606 dnode_rele(dn, FTAG);
607
608 mutex_enter(&os->os_lock);
609 dn = list_next(&os->os_dnodes, &dn_marker);
610 list_remove(&os->os_dnodes, &dn_marker);
611 } else {
612 dn = list_next(&os->os_dnodes, dn);
613 }
614 }
615 mutex_exit(&os->os_lock);
616
617 if (DMU_USERUSED_DNODE(os) != NULL) {
618 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
619 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
620 }
621 dnode_evict_dbufs(DMU_META_DNODE(os));
622 }
623
624 /*
625 * Objset eviction processing is split into into two pieces.
626 * The first marks the objset as evicting, evicts any dbufs that
627 * have a refcount of zero, and then queues up the objset for the
628 * second phase of eviction. Once os->os_dnodes has been cleared by
629 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
630 * The second phase closes the special dnodes, dequeues the objset from
631 * the list of those undergoing eviction, and finally frees the objset.
632 *
633 * NOTE: Due to asynchronous eviction processing (invocation of
634 * dnode_buf_pageout()), it is possible for the meta dnode for the
635 * objset to have no holds even though os->os_dnodes is not empty.
636 */
637 void
638 dmu_objset_evict(objset_t *os)
639 {
640 dsl_dataset_t *ds = os->os_dsl_dataset;
641
642 for (int t = 0; t < TXG_SIZE; t++)
643 ASSERT(!dmu_objset_is_dirty(os, t));
644
645 if (ds) {
646 if (!ds->ds_is_snapshot) {
647 VERIFY0(dsl_prop_unregister(ds,
648 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
649 checksum_changed_cb, os));
650 VERIFY0(dsl_prop_unregister(ds,
651 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
652 compression_changed_cb, os));
653 VERIFY0(dsl_prop_unregister(ds,
654 zfs_prop_to_name(ZFS_PROP_COPIES),
655 copies_changed_cb, os));
656 VERIFY0(dsl_prop_unregister(ds,
657 zfs_prop_to_name(ZFS_PROP_DEDUP),
658 dedup_changed_cb, os));
659 VERIFY0(dsl_prop_unregister(ds,
660 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
661 logbias_changed_cb, os));
662 VERIFY0(dsl_prop_unregister(ds,
663 zfs_prop_to_name(ZFS_PROP_SYNC),
664 sync_changed_cb, os));
665 VERIFY0(dsl_prop_unregister(ds,
666 zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA),
667 redundant_metadata_changed_cb, os));
668 VERIFY0(dsl_prop_unregister(ds,
669 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
670 recordsize_changed_cb, os));
671 }
672 VERIFY0(dsl_prop_unregister(ds,
673 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
674 primary_cache_changed_cb, os));
675 VERIFY0(dsl_prop_unregister(ds,
676 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
677 secondary_cache_changed_cb, os));
678 }
679
680 if (os->os_sa)
681 sa_tear_down(os);
682
683 os->os_evicting = B_TRUE;
684 dmu_objset_evict_dbufs(os);
685
686 mutex_enter(&os->os_lock);
687 spa_evicting_os_register(os->os_spa, os);
688 if (list_is_empty(&os->os_dnodes)) {
689 mutex_exit(&os->os_lock);
690 dmu_objset_evict_done(os);
691 } else {
692 mutex_exit(&os->os_lock);
693 }
694 }
695
696 void
697 dmu_objset_evict_done(objset_t *os)
698 {
699 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
700
701 dnode_special_close(&os->os_meta_dnode);
702 if (DMU_USERUSED_DNODE(os)) {
703 dnode_special_close(&os->os_userused_dnode);
704 dnode_special_close(&os->os_groupused_dnode);
705 }
706 zil_free(os->os_zil);
707
708 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
709
710 /*
711 * This is a barrier to prevent the objset from going away in
712 * dnode_move() until we can safely ensure that the objset is still in
713 * use. We consider the objset valid before the barrier and invalid
714 * after the barrier.
715 */
716 rw_enter(&os_lock, RW_READER);
717 rw_exit(&os_lock);
718
719 mutex_destroy(&os->os_lock);
720 mutex_destroy(&os->os_obj_lock);
721 mutex_destroy(&os->os_user_ptr_lock);
722 spa_evicting_os_deregister(os->os_spa, os);
723 kmem_free(os, sizeof (objset_t));
724 }
725
726 timestruc_t
727 dmu_objset_snap_cmtime(objset_t *os)
728 {
729 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
730 }
731
732 /* called from dsl for meta-objset */
733 objset_t *
734 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
735 dmu_objset_type_t type, dmu_tx_t *tx)
736 {
737 objset_t *os;
738 dnode_t *mdn;
739
740 ASSERT(dmu_tx_is_syncing(tx));
741
742 if (ds != NULL)
743 VERIFY0(dmu_objset_from_ds(ds, &os));
744 else
745 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
746
747 mdn = DMU_META_DNODE(os);
748
749 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
750 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
751
752 /*
753 * We don't want to have to increase the meta-dnode's nlevels
754 * later, because then we could do it in quescing context while
755 * we are also accessing it in open context.
756 *
757 * This precaution is not necessary for the MOS (ds == NULL),
758 * because the MOS is only updated in syncing context.
759 * This is most fortunate: the MOS is the only objset that
760 * needs to be synced multiple times as spa_sync() iterates
761 * to convergence, so minimizing its dn_nlevels matters.
762 */
763 if (ds != NULL) {
764 int levels = 1;
765
766 /*
767 * Determine the number of levels necessary for the meta-dnode
768 * to contain DN_MAX_OBJECT dnodes.
769 */
770 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
771 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
772 DN_MAX_OBJECT * sizeof (dnode_phys_t))
773 levels++;
774
775 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
776 mdn->dn_nlevels = levels;
777 }
778
779 ASSERT(type != DMU_OST_NONE);
780 ASSERT(type != DMU_OST_ANY);
781 ASSERT(type < DMU_OST_NUMTYPES);
782 os->os_phys->os_type = type;
783 if (dmu_objset_userused_enabled(os)) {
784 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
785 os->os_flags = os->os_phys->os_flags;
786 }
787
788 dsl_dataset_dirty(ds, tx);
789
790 return (os);
791 }
792
793 typedef struct dmu_objset_create_arg {
794 const char *doca_name;
795 cred_t *doca_cred;
796 void (*doca_userfunc)(objset_t *os, void *arg,
797 cred_t *cr, dmu_tx_t *tx);
798 void *doca_userarg;
799 dmu_objset_type_t doca_type;
800 uint64_t doca_flags;
801 } dmu_objset_create_arg_t;
802
803 /*ARGSUSED*/
804 static int
805 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
806 {
807 dmu_objset_create_arg_t *doca = arg;
808 dsl_pool_t *dp = dmu_tx_pool(tx);
809 dsl_dir_t *pdd;
810 const char *tail;
811 int error;
812
813 if (strchr(doca->doca_name, '@') != NULL)
814 return (SET_ERROR(EINVAL));
815
816 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
817 if (error != 0)
818 return (error);
819 if (tail == NULL) {
820 dsl_dir_rele(pdd, FTAG);
821 return (SET_ERROR(EEXIST));
822 }
823 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
824 doca->doca_cred);
825 dsl_dir_rele(pdd, FTAG);
826
827 return (error);
828 }
829
830 static void
831 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
832 {
833 dmu_objset_create_arg_t *doca = arg;
834 dsl_pool_t *dp = dmu_tx_pool(tx);
835 dsl_dir_t *pdd;
836 const char *tail;
837 dsl_dataset_t *ds;
838 uint64_t obj;
839 blkptr_t *bp;
840 objset_t *os;
841
842 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
843
844 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
845 doca->doca_cred, tx);
846
847 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
848 bp = dsl_dataset_get_blkptr(ds);
849 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
850 ds, bp, doca->doca_type, tx);
851
852 if (doca->doca_userfunc != NULL) {
853 doca->doca_userfunc(os, doca->doca_userarg,
854 doca->doca_cred, tx);
855 }
856
857 spa_history_log_internal_ds(ds, "create", tx, "");
858 dsl_dataset_rele(ds, FTAG);
859 dsl_dir_rele(pdd, FTAG);
860 }
861
862 int
863 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
864 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
865 {
866 dmu_objset_create_arg_t doca;
867
868 doca.doca_name = name;
869 doca.doca_cred = CRED();
870 doca.doca_flags = flags;
871 doca.doca_userfunc = func;
872 doca.doca_userarg = arg;
873 doca.doca_type = type;
874
875 return (dsl_sync_task(name,
876 dmu_objset_create_check, dmu_objset_create_sync, &doca,
877 5, ZFS_SPACE_CHECK_NORMAL));
878 }
879
880 typedef struct dmu_objset_clone_arg {
881 const char *doca_clone;
882 const char *doca_origin;
883 cred_t *doca_cred;
884 } dmu_objset_clone_arg_t;
885
886 /*ARGSUSED*/
887 static int
888 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
889 {
890 dmu_objset_clone_arg_t *doca = arg;
891 dsl_dir_t *pdd;
892 const char *tail;
893 int error;
894 dsl_dataset_t *origin;
895 dsl_pool_t *dp = dmu_tx_pool(tx);
896
897 if (strchr(doca->doca_clone, '@') != NULL)
898 return (SET_ERROR(EINVAL));
899
900 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
901 if (error != 0)
902 return (error);
903 if (tail == NULL) {
904 dsl_dir_rele(pdd, FTAG);
905 return (SET_ERROR(EEXIST));
906 }
907 /* You can't clone across pools. */
908 if (pdd->dd_pool != dp) {
909 dsl_dir_rele(pdd, FTAG);
910 return (SET_ERROR(EXDEV));
911 }
912 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
913 doca->doca_cred);
914 if (error != 0) {
915 dsl_dir_rele(pdd, FTAG);
916 return (SET_ERROR(EDQUOT));
917 }
918 dsl_dir_rele(pdd, FTAG);
919
920 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
921 if (error != 0)
922 return (error);
923
924 /* You can't clone across pools. */
925 if (origin->ds_dir->dd_pool != dp) {
926 dsl_dataset_rele(origin, FTAG);
927 return (SET_ERROR(EXDEV));
928 }
929
930 /* You can only clone snapshots, not the head datasets. */
931 if (!origin->ds_is_snapshot) {
932 dsl_dataset_rele(origin, FTAG);
933 return (SET_ERROR(EINVAL));
934 }
935 dsl_dataset_rele(origin, FTAG);
936
937 return (0);
938 }
939
940 static void
941 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
942 {
943 dmu_objset_clone_arg_t *doca = arg;
944 dsl_pool_t *dp = dmu_tx_pool(tx);
945 dsl_dir_t *pdd;
946 const char *tail;
947 dsl_dataset_t *origin, *ds;
948 uint64_t obj;
949 char namebuf[MAXNAMELEN];
950
951 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
952 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
953
954 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
955 doca->doca_cred, tx);
956
957 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
958 dsl_dataset_name(origin, namebuf);
959 spa_history_log_internal_ds(ds, "clone", tx,
960 "origin=%s (%llu)", namebuf, origin->ds_object);
961 dsl_dataset_rele(ds, FTAG);
962 dsl_dataset_rele(origin, FTAG);
963 dsl_dir_rele(pdd, FTAG);
964 }
965
966 int
967 dmu_objset_clone(const char *clone, const char *origin)
968 {
969 dmu_objset_clone_arg_t doca;
970
971 doca.doca_clone = clone;
972 doca.doca_origin = origin;
973 doca.doca_cred = CRED();
974
975 return (dsl_sync_task(clone,
976 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
977 5, ZFS_SPACE_CHECK_NORMAL));
978 }
979
980 int
981 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
982 {
983 int err;
984 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
985 nvlist_t *snaps = fnvlist_alloc();
986
987 fnvlist_add_boolean(snaps, longsnap);
988 strfree(longsnap);
989 err = dsl_dataset_snapshot(snaps, NULL, NULL);
990 fnvlist_free(snaps);
991 return (err);
992 }
993
994 static void
995 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
996 {
997 dnode_t *dn;
998
999 while (dn = list_head(list)) {
1000 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1001 ASSERT(dn->dn_dbuf->db_data_pending);
1002 /*
1003 * Initialize dn_zio outside dnode_sync() because the
1004 * meta-dnode needs to set it ouside dnode_sync().
1005 */
1006 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1007 ASSERT(dn->dn_zio);
1008
1009 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1010 list_remove(list, dn);
1011
1012 if (newlist) {
1013 (void) dnode_add_ref(dn, newlist);
1014 list_insert_tail(newlist, dn);
1015 }
1016
1017 dnode_sync(dn, tx);
1018 }
1019 }
1020
1021 /* ARGSUSED */
1022 static void
1023 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1024 {
1025 blkptr_t *bp = zio->io_bp;
1026 objset_t *os = arg;
1027 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1028
1029 ASSERT(!BP_IS_EMBEDDED(bp));
1030 ASSERT3P(bp, ==, os->os_rootbp);
1031 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1032 ASSERT0(BP_GET_LEVEL(bp));
1033
1034 /*
1035 * Update rootbp fill count: it should be the number of objects
1036 * allocated in the object set (not counting the "special"
1037 * objects that are stored in the objset_phys_t -- the meta
1038 * dnode and user/group accounting objects).
1039 */
1040 bp->blk_fill = 0;
1041 for (int i = 0; i < dnp->dn_nblkptr; i++)
1042 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1043 }
1044
1045 /* ARGSUSED */
1046 static void
1047 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1048 {
1049 blkptr_t *bp = zio->io_bp;
1050 blkptr_t *bp_orig = &zio->io_bp_orig;
1051 objset_t *os = arg;
1052
1053 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1054 ASSERT(BP_EQUAL(bp, bp_orig));
1055 } else {
1056 dsl_dataset_t *ds = os->os_dsl_dataset;
1057 dmu_tx_t *tx = os->os_synctx;
1058
1059 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1060 dsl_dataset_block_born(ds, bp, tx);
1061 }
1062 }
1063
1064 /* called from dsl */
1065 void
1066 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1067 {
1068 int txgoff;
1069 zbookmark_phys_t zb;
1070 zio_prop_t zp;
1071 zio_t *zio;
1072 list_t *list;
1073 list_t *newlist = NULL;
1074 dbuf_dirty_record_t *dr;
1075
1076 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1077
1078 ASSERT(dmu_tx_is_syncing(tx));
1079 /* XXX the write_done callback should really give us the tx... */
1080 os->os_synctx = tx;
1081
1082 if (os->os_dsl_dataset == NULL) {
1083 /*
1084 * This is the MOS. If we have upgraded,
1085 * spa_max_replication() could change, so reset
1086 * os_copies here.
1087 */
1088 os->os_copies = spa_max_replication(os->os_spa);
1089 }
1090
1091 /*
1092 * Create the root block IO
1093 */
1094 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1095 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1096 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1097 arc_release(os->os_phys_buf, &os->os_phys_buf);
1098
1099 dmu_write_policy(os, NULL, 0, 0, &zp);
1100
1101 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1102 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1103 DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1104 NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1105 ZIO_FLAG_MUSTSUCCEED, &zb);
1106
1107 /*
1108 * Sync special dnodes - the parent IO for the sync is the root block
1109 */
1110 DMU_META_DNODE(os)->dn_zio = zio;
1111 dnode_sync(DMU_META_DNODE(os), tx);
1112
1113 os->os_phys->os_flags = os->os_flags;
1114
1115 if (DMU_USERUSED_DNODE(os) &&
1116 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1117 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1118 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1119 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1120 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1121 }
1122
1123 txgoff = tx->tx_txg & TXG_MASK;
1124
1125 if (dmu_objset_userused_enabled(os)) {
1126 newlist = &os->os_synced_dnodes;
1127 /*
1128 * We must create the list here because it uses the
1129 * dn_dirty_link[] of this txg.
1130 */
1131 list_create(newlist, sizeof (dnode_t),
1132 offsetof(dnode_t, dn_dirty_link[txgoff]));
1133 }
1134
1135 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1136 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1137
1138 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1139 while (dr = list_head(list)) {
1140 ASSERT0(dr->dr_dbuf->db_level);
1141 list_remove(list, dr);
1142 if (dr->dr_zio)
1143 zio_nowait(dr->dr_zio);
1144 }
1145 /*
1146 * Free intent log blocks up to this tx.
1147 */
1148 zil_sync(os->os_zil, tx);
1149 os->os_phys->os_zil_header = os->os_zil_header;
1150 zio_nowait(zio);
1151 }
1152
1153 boolean_t
1154 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1155 {
1156 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1157 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1158 }
1159
1160 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1161
1162 void
1163 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1164 {
1165 used_cbs[ost] = cb;
1166 }
1167
1168 boolean_t
1169 dmu_objset_userused_enabled(objset_t *os)
1170 {
1171 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1172 used_cbs[os->os_phys->os_type] != NULL &&
1173 DMU_USERUSED_DNODE(os) != NULL);
1174 }
1175
1176 static void
1177 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1178 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1179 {
1180 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1181 int64_t delta = DNODE_SIZE + used;
1182 if (subtract)
1183 delta = -delta;
1184 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1185 user, delta, tx));
1186 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1187 group, delta, tx));
1188 }
1189 }
1190
1191 void
1192 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1193 {
1194 dnode_t *dn;
1195 list_t *list = &os->os_synced_dnodes;
1196
1197 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1198
1199 while (dn = list_head(list)) {
1200 int flags;
1201 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1202 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1203 dn->dn_phys->dn_flags &
1204 DNODE_FLAG_USERUSED_ACCOUNTED);
1205
1206 /* Allocate the user/groupused objects if necessary. */
1207 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1208 VERIFY(0 == zap_create_claim(os,
1209 DMU_USERUSED_OBJECT,
1210 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1211 VERIFY(0 == zap_create_claim(os,
1212 DMU_GROUPUSED_OBJECT,
1213 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1214 }
1215
1216 /*
1217 * We intentionally modify the zap object even if the
1218 * net delta is zero. Otherwise
1219 * the block of the zap obj could be shared between
1220 * datasets but need to be different between them after
1221 * a bprewrite.
1222 */
1223
1224 flags = dn->dn_id_flags;
1225 ASSERT(flags);
1226 if (flags & DN_ID_OLD_EXIST) {
1227 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1228 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1229 }
1230 if (flags & DN_ID_NEW_EXIST) {
1231 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1232 dn->dn_phys->dn_flags, dn->dn_newuid,
1233 dn->dn_newgid, B_FALSE, tx);
1234 }
1235
1236 mutex_enter(&dn->dn_mtx);
1237 dn->dn_oldused = 0;
1238 dn->dn_oldflags = 0;
1239 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1240 dn->dn_olduid = dn->dn_newuid;
1241 dn->dn_oldgid = dn->dn_newgid;
1242 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1243 if (dn->dn_bonuslen == 0)
1244 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1245 else
1246 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1247 }
1248 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1249 mutex_exit(&dn->dn_mtx);
1250
1251 list_remove(list, dn);
1252 dnode_rele(dn, list);
1253 }
1254 }
1255
1256 /*
1257 * Returns a pointer to data to find uid/gid from
1258 *
1259 * If a dirty record for transaction group that is syncing can't
1260 * be found then NULL is returned. In the NULL case it is assumed
1261 * the uid/gid aren't changing.
1262 */
1263 static void *
1264 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1265 {
1266 dbuf_dirty_record_t *dr, **drp;
1267 void *data;
1268
1269 if (db->db_dirtycnt == 0)
1270 return (db->db.db_data); /* Nothing is changing */
1271
1272 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1273 if (dr->dr_txg == tx->tx_txg)
1274 break;
1275
1276 if (dr == NULL) {
1277 data = NULL;
1278 } else {
1279 dnode_t *dn;
1280
1281 DB_DNODE_ENTER(dr->dr_dbuf);
1282 dn = DB_DNODE(dr->dr_dbuf);
1283
1284 if (dn->dn_bonuslen == 0 &&
1285 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1286 data = dr->dt.dl.dr_data->b_data;
1287 else
1288 data = dr->dt.dl.dr_data;
1289
1290 DB_DNODE_EXIT(dr->dr_dbuf);
1291 }
1292
1293 return (data);
1294 }
1295
1296 void
1297 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1298 {
1299 objset_t *os = dn->dn_objset;
1300 void *data = NULL;
1301 dmu_buf_impl_t *db = NULL;
1302 uint64_t *user = NULL;
1303 uint64_t *group = NULL;
1304 int flags = dn->dn_id_flags;
1305 int error;
1306 boolean_t have_spill = B_FALSE;
1307
1308 if (!dmu_objset_userused_enabled(dn->dn_objset))
1309 return;
1310
1311 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1312 DN_ID_CHKED_SPILL)))
1313 return;
1314
1315 if (before && dn->dn_bonuslen != 0)
1316 data = DN_BONUS(dn->dn_phys);
1317 else if (!before && dn->dn_bonuslen != 0) {
1318 if (dn->dn_bonus) {
1319 db = dn->dn_bonus;
1320 mutex_enter(&db->db_mtx);
1321 data = dmu_objset_userquota_find_data(db, tx);
1322 } else {
1323 data = DN_BONUS(dn->dn_phys);
1324 }
1325 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1326 int rf = 0;
1327
1328 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1329 rf |= DB_RF_HAVESTRUCT;
1330 error = dmu_spill_hold_by_dnode(dn,
1331 rf | DB_RF_MUST_SUCCEED,
1332 FTAG, (dmu_buf_t **)&db);
1333 ASSERT(error == 0);
1334 mutex_enter(&db->db_mtx);
1335 data = (before) ? db->db.db_data :
1336 dmu_objset_userquota_find_data(db, tx);
1337 have_spill = B_TRUE;
1338 } else {
1339 mutex_enter(&dn->dn_mtx);
1340 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1341 mutex_exit(&dn->dn_mtx);
1342 return;
1343 }
1344
1345 if (before) {
1346 ASSERT(data);
1347 user = &dn->dn_olduid;
1348 group = &dn->dn_oldgid;
1349 } else if (data) {
1350 user = &dn->dn_newuid;
1351 group = &dn->dn_newgid;
1352 }
1353
1354 /*
1355 * Must always call the callback in case the object
1356 * type has changed and that type isn't an object type to track
1357 */
1358 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1359 user, group);
1360
1361 /*
1362 * Preserve existing uid/gid when the callback can't determine
1363 * what the new uid/gid are and the callback returned EEXIST.
1364 * The EEXIST error tells us to just use the existing uid/gid.
1365 * If we don't know what the old values are then just assign
1366 * them to 0, since that is a new file being created.
1367 */
1368 if (!before && data == NULL && error == EEXIST) {
1369 if (flags & DN_ID_OLD_EXIST) {
1370 dn->dn_newuid = dn->dn_olduid;
1371 dn->dn_newgid = dn->dn_oldgid;
1372 } else {
1373 dn->dn_newuid = 0;
1374 dn->dn_newgid = 0;
1375 }
1376 error = 0;
1377 }
1378
1379 if (db)
1380 mutex_exit(&db->db_mtx);
1381
1382 mutex_enter(&dn->dn_mtx);
1383 if (error == 0 && before)
1384 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1385 if (error == 0 && !before)
1386 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1387
1388 if (have_spill) {
1389 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1390 } else {
1391 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1392 }
1393 mutex_exit(&dn->dn_mtx);
1394 if (have_spill)
1395 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1396 }
1397
1398 boolean_t
1399 dmu_objset_userspace_present(objset_t *os)
1400 {
1401 return (os->os_phys->os_flags &
1402 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1403 }
1404
1405 int
1406 dmu_objset_userspace_upgrade(objset_t *os)
1407 {
1408 uint64_t obj;
1409 int err = 0;
1410
1411 if (dmu_objset_userspace_present(os))
1412 return (0);
1413 if (!dmu_objset_userused_enabled(os))
1414 return (SET_ERROR(ENOTSUP));
1415 if (dmu_objset_is_snapshot(os))
1416 return (SET_ERROR(EINVAL));
1417
1418 /*
1419 * We simply need to mark every object dirty, so that it will be
1420 * synced out and now accounted. If this is called
1421 * concurrently, or if we already did some work before crashing,
1422 * that's fine, since we track each object's accounted state
1423 * independently.
1424 */
1425
1426 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1427 dmu_tx_t *tx;
1428 dmu_buf_t *db;
1429 int objerr;
1430
1431 if (issig(JUSTLOOKING) && issig(FORREAL))
1432 return (SET_ERROR(EINTR));
1433
1434 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1435 if (objerr != 0)
1436 continue;
1437 tx = dmu_tx_create(os);
1438 dmu_tx_hold_bonus(tx, obj);
1439 objerr = dmu_tx_assign(tx, TXG_WAIT);
1440 if (objerr != 0) {
1441 dmu_tx_abort(tx);
1442 continue;
1443 }
1444 dmu_buf_will_dirty(db, tx);
1445 dmu_buf_rele(db, FTAG);
1446 dmu_tx_commit(tx);
1447 }
1448
1449 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1450 txg_wait_synced(dmu_objset_pool(os), 0);
1451 return (0);
1452 }
1453
1454 void
1455 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1456 uint64_t *usedobjsp, uint64_t *availobjsp)
1457 {
1458 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1459 usedobjsp, availobjsp);
1460 }
1461
1462 uint64_t
1463 dmu_objset_fsid_guid(objset_t *os)
1464 {
1465 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1466 }
1467
1468 void
1469 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1470 {
1471 stat->dds_type = os->os_phys->os_type;
1472 if (os->os_dsl_dataset)
1473 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1474 }
1475
1476 void
1477 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1478 {
1479 ASSERT(os->os_dsl_dataset ||
1480 os->os_phys->os_type == DMU_OST_META);
1481
1482 if (os->os_dsl_dataset != NULL)
1483 dsl_dataset_stats(os->os_dsl_dataset, nv);
1484
1485 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1486 os->os_phys->os_type);
1487 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1488 dmu_objset_userspace_present(os));
1489 }
1490
1491 int
1492 dmu_objset_is_snapshot(objset_t *os)
1493 {
1494 if (os->os_dsl_dataset != NULL)
1495 return (os->os_dsl_dataset->ds_is_snapshot);
1496 else
1497 return (B_FALSE);
1498 }
1499
1500 int
1501 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1502 boolean_t *conflict)
1503 {
1504 dsl_dataset_t *ds = os->os_dsl_dataset;
1505 uint64_t ignored;
1506
1507 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1508 return (SET_ERROR(ENOENT));
1509
1510 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1511 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1512 MT_FIRST, real, maxlen, conflict));
1513 }
1514
1515 int
1516 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1517 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1518 {
1519 dsl_dataset_t *ds = os->os_dsl_dataset;
1520 zap_cursor_t cursor;
1521 zap_attribute_t attr;
1522
1523 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1524
1525 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1526 return (SET_ERROR(ENOENT));
1527
1528 zap_cursor_init_serialized(&cursor,
1529 ds->ds_dir->dd_pool->dp_meta_objset,
1530 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1531
1532 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1533 zap_cursor_fini(&cursor);
1534 return (SET_ERROR(ENOENT));
1535 }
1536
1537 if (strlen(attr.za_name) + 1 > namelen) {
1538 zap_cursor_fini(&cursor);
1539 return (SET_ERROR(ENAMETOOLONG));
1540 }
1541
1542 (void) strcpy(name, attr.za_name);
1543 if (idp)
1544 *idp = attr.za_first_integer;
1545 if (case_conflict)
1546 *case_conflict = attr.za_normalization_conflict;
1547 zap_cursor_advance(&cursor);
1548 *offp = zap_cursor_serialize(&cursor);
1549 zap_cursor_fini(&cursor);
1550
1551 return (0);
1552 }
1553
1554 int
1555 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1556 uint64_t *idp, uint64_t *offp)
1557 {
1558 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1559 zap_cursor_t cursor;
1560 zap_attribute_t attr;
1561
1562 /* there is no next dir on a snapshot! */
1563 if (os->os_dsl_dataset->ds_object !=
1564 dsl_dir_phys(dd)->dd_head_dataset_obj)
1565 return (SET_ERROR(ENOENT));
1566
1567 zap_cursor_init_serialized(&cursor,
1568 dd->dd_pool->dp_meta_objset,
1569 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1570
1571 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1572 zap_cursor_fini(&cursor);
1573 return (SET_ERROR(ENOENT));
1574 }
1575
1576 if (strlen(attr.za_name) + 1 > namelen) {
1577 zap_cursor_fini(&cursor);
1578 return (SET_ERROR(ENAMETOOLONG));
1579 }
1580
1581 (void) strcpy(name, attr.za_name);
1582 if (idp)
1583 *idp = attr.za_first_integer;
1584 zap_cursor_advance(&cursor);
1585 *offp = zap_cursor_serialize(&cursor);
1586 zap_cursor_fini(&cursor);
1587
1588 return (0);
1589 }
1590
1591 /*
1592 * Find objsets under and including ddobj, call func(ds) on each.
1593 */
1594 int
1595 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1596 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1597 {
1598 dsl_dir_t *dd;
1599 dsl_dataset_t *ds;
1600 zap_cursor_t zc;
1601 zap_attribute_t *attr;
1602 uint64_t thisobj;
1603 int err;
1604
1605 ASSERT(dsl_pool_config_held(dp));
1606
1607 err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1608 if (err != 0)
1609 return (err);
1610
1611 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1612 if (dd->dd_myname[0] == '$') {
1613 dsl_dir_rele(dd, FTAG);
1614 return (0);
1615 }
1616
1617 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1618 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1619
1620 /*
1621 * Iterate over all children.
1622 */
1623 if (flags & DS_FIND_CHILDREN) {
1624 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1625 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1626 zap_cursor_retrieve(&zc, attr) == 0;
1627 (void) zap_cursor_advance(&zc)) {
1628 ASSERT3U(attr->za_integer_length, ==,
1629 sizeof (uint64_t));
1630 ASSERT3U(attr->za_num_integers, ==, 1);
1631
1632 err = dmu_objset_find_dp(dp, attr->za_first_integer,
1633 func, arg, flags);
1634 if (err != 0)
1635 break;
1636 }
1637 zap_cursor_fini(&zc);
1638
1639 if (err != 0) {
1640 dsl_dir_rele(dd, FTAG);
1641 kmem_free(attr, sizeof (zap_attribute_t));
1642 return (err);
1643 }
1644 }
1645
1646 /*
1647 * Iterate over all snapshots.
1648 */
1649 if (flags & DS_FIND_SNAPSHOTS) {
1650 dsl_dataset_t *ds;
1651 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1652
1653 if (err == 0) {
1654 uint64_t snapobj;
1655
1656 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1657 dsl_dataset_rele(ds, FTAG);
1658
1659 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1660 zap_cursor_retrieve(&zc, attr) == 0;
1661 (void) zap_cursor_advance(&zc)) {
1662 ASSERT3U(attr->za_integer_length, ==,
1663 sizeof (uint64_t));
1664 ASSERT3U(attr->za_num_integers, ==, 1);
1665
1666 err = dsl_dataset_hold_obj(dp,
1667 attr->za_first_integer, FTAG, &ds);
1668 if (err != 0)
1669 break;
1670 err = func(dp, ds, arg);
1671 dsl_dataset_rele(ds, FTAG);
1672 if (err != 0)
1673 break;
1674 }
1675 zap_cursor_fini(&zc);
1676 }
1677 }
1678
1679 dsl_dir_rele(dd, FTAG);
1680 kmem_free(attr, sizeof (zap_attribute_t));
1681
1682 if (err != 0)
1683 return (err);
1684
1685 /*
1686 * Apply to self.
1687 */
1688 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1689 if (err != 0)
1690 return (err);
1691 err = func(dp, ds, arg);
1692 dsl_dataset_rele(ds, FTAG);
1693 return (err);
1694 }
1695
1696 /*
1697 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1698 * The dp_config_rwlock must not be held when this is called, and it
1699 * will not be held when the callback is called.
1700 * Therefore this function should only be used when the pool is not changing
1701 * (e.g. in syncing context), or the callback can deal with the possible races.
1702 */
1703 static int
1704 dmu_objset_find_impl(spa_t *spa, const char *name,
1705 int func(const char *, void *), void *arg, int flags)
1706 {
1707 dsl_dir_t *dd;
1708 dsl_pool_t *dp = spa_get_dsl(spa);
1709 dsl_dataset_t *ds;
1710 zap_cursor_t zc;
1711 zap_attribute_t *attr;
1712 char *child;
1713 uint64_t thisobj;
1714 int err;
1715
1716 dsl_pool_config_enter(dp, FTAG);
1717
1718 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1719 if (err != 0) {
1720 dsl_pool_config_exit(dp, FTAG);
1721 return (err);
1722 }
1723
1724 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1725 if (dd->dd_myname[0] == '$') {
1726 dsl_dir_rele(dd, FTAG);
1727 dsl_pool_config_exit(dp, FTAG);
1728 return (0);
1729 }
1730
1731 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1732 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1733
1734 /*
1735 * Iterate over all children.
1736 */
1737 if (flags & DS_FIND_CHILDREN) {
1738 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1739 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1740 zap_cursor_retrieve(&zc, attr) == 0;
1741 (void) zap_cursor_advance(&zc)) {
1742 ASSERT3U(attr->za_integer_length, ==,
1743 sizeof (uint64_t));
1744 ASSERT3U(attr->za_num_integers, ==, 1);
1745
1746 child = kmem_asprintf("%s/%s", name, attr->za_name);
1747 dsl_pool_config_exit(dp, FTAG);
1748 err = dmu_objset_find_impl(spa, child,
1749 func, arg, flags);
1750 dsl_pool_config_enter(dp, FTAG);
1751 strfree(child);
1752 if (err != 0)
1753 break;
1754 }
1755 zap_cursor_fini(&zc);
1756
1757 if (err != 0) {
1758 dsl_dir_rele(dd, FTAG);
1759 dsl_pool_config_exit(dp, FTAG);
1760 kmem_free(attr, sizeof (zap_attribute_t));
1761 return (err);
1762 }
1763 }
1764
1765 /*
1766 * Iterate over all snapshots.
1767 */
1768 if (flags & DS_FIND_SNAPSHOTS) {
1769 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1770
1771 if (err == 0) {
1772 uint64_t snapobj;
1773
1774 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1775 dsl_dataset_rele(ds, FTAG);
1776
1777 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1778 zap_cursor_retrieve(&zc, attr) == 0;
1779 (void) zap_cursor_advance(&zc)) {
1780 ASSERT3U(attr->za_integer_length, ==,
1781 sizeof (uint64_t));
1782 ASSERT3U(attr->za_num_integers, ==, 1);
1783
1784 child = kmem_asprintf("%s@%s",
1785 name, attr->za_name);
1786 dsl_pool_config_exit(dp, FTAG);
1787 err = func(child, arg);
1788 dsl_pool_config_enter(dp, FTAG);
1789 strfree(child);
1790 if (err != 0)
1791 break;
1792 }
1793 zap_cursor_fini(&zc);
1794 }
1795 }
1796
1797 dsl_dir_rele(dd, FTAG);
1798 kmem_free(attr, sizeof (zap_attribute_t));
1799 dsl_pool_config_exit(dp, FTAG);
1800
1801 if (err != 0)
1802 return (err);
1803
1804 /* Apply to self. */
1805 return (func(name, arg));
1806 }
1807
1808 /*
1809 * See comment above dmu_objset_find_impl().
1810 */
1811 int
1812 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1813 int flags)
1814 {
1815 spa_t *spa;
1816 int error;
1817
1818 error = spa_open(name, &spa, FTAG);
1819 if (error != 0)
1820 return (error);
1821 error = dmu_objset_find_impl(spa, name, func, arg, flags);
1822 spa_close(spa, FTAG);
1823 return (error);
1824 }
1825
1826 void
1827 dmu_objset_set_user(objset_t *os, void *user_ptr)
1828 {
1829 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1830 os->os_user_ptr = user_ptr;
1831 }
1832
1833 void *
1834 dmu_objset_get_user(objset_t *os)
1835 {
1836 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1837 return (os->os_user_ptr);
1838 }
1839
1840 /*
1841 * Determine name of filesystem, given name of snapshot.
1842 * buf must be at least MAXNAMELEN bytes
1843 */
1844 int
1845 dmu_fsname(const char *snapname, char *buf)
1846 {
1847 char *atp = strchr(snapname, '@');
1848 if (atp == NULL)
1849 return (SET_ERROR(EINVAL));
1850 if (atp - snapname >= MAXNAMELEN)
1851 return (SET_ERROR(ENAMETOOLONG));
1852 (void) strlcpy(buf, snapname, atp - snapname + 1);
1853 return (0);
1854 }