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 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
28 */
29
30 /* Portions Copyright 2010 Robert Milkowski */
31
32 #include <sys/cred.h>
33 #include <sys/zfs_context.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_dataset.h>
37 #include <sys/dsl_prop.h>
38 #include <sys/dsl_pool.h>
39 #include <sys/dsl_synctask.h>
40 #include <sys/dsl_deleg.h>
41 #include <sys/dnode.h>
42 #include <sys/dbuf.h>
43 #include <sys/zvol.h>
44 #include <sys/dmu_tx.h>
45 #include <sys/zap.h>
46 #include <sys/zil.h>
47 #include <sys/dmu_impl.h>
48 #include <sys/zfs_ioctl.h>
49 #include <sys/sa.h>
50 #include <sys/zfs_onexit.h>
51 #include <sys/dsl_destroy.h>
52
53 /*
54 * Needed to close a window in dnode_move() that allows the objset to be freed
55 * before it can be safely accessed.
56 */
57 krwlock_t os_lock;
58
59 void
60 dmu_objset_init(void)
61 {
62 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
63 }
64
65 void
66 dmu_objset_fini(void)
67 {
68 rw_destroy(&os_lock);
69 }
70
71 spa_t *
72 dmu_objset_spa(objset_t *os)
73 {
74 return (os->os_spa);
75 }
76
77 zilog_t *
78 dmu_objset_zil(objset_t *os)
79 {
80 return (os->os_zil);
81 }
82
83 dsl_pool_t *
84 dmu_objset_pool(objset_t *os)
85 {
86 dsl_dataset_t *ds;
87
88 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
89 return (ds->ds_dir->dd_pool);
90 else
91 return (spa_get_dsl(os->os_spa));
92 }
93
94 dsl_dataset_t *
95 dmu_objset_ds(objset_t *os)
96 {
97 return (os->os_dsl_dataset);
98 }
99
100 dmu_objset_type_t
101 dmu_objset_type(objset_t *os)
102 {
103 return (os->os_phys->os_type);
104 }
105
106 void
107 dmu_objset_name(objset_t *os, char *buf)
108 {
109 dsl_dataset_name(os->os_dsl_dataset, buf);
110 }
111
112 uint64_t
113 dmu_objset_id(objset_t *os)
114 {
115 dsl_dataset_t *ds = os->os_dsl_dataset;
116
117 return (ds ? ds->ds_object : 0);
118 }
119
120 zfs_sync_type_t
121 dmu_objset_syncprop(objset_t *os)
122 {
123 return (os->os_sync);
124 }
125
126 zfs_logbias_op_t
127 dmu_objset_logbias(objset_t *os)
128 {
129 return (os->os_logbias);
130 }
131
132 static void
133 checksum_changed_cb(void *arg, uint64_t newval)
134 {
135 objset_t *os = arg;
136
137 /*
138 * Inheritance should have been done by now.
139 */
140 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
141
142 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
143 }
144
145 static void
146 compression_changed_cb(void *arg, uint64_t newval)
147 {
148 objset_t *os = arg;
149
150 /*
151 * Inheritance and range checking should have been done by now.
152 */
153 ASSERT(newval != ZIO_COMPRESS_INHERIT);
154
155 os->os_compress = zio_compress_select(os->os_spa, newval,
156 ZIO_COMPRESS_ON);
157 }
158
159 static void
160 copies_changed_cb(void *arg, uint64_t newval)
161 {
162 objset_t *os = arg;
163
164 /*
165 * Inheritance and range checking should have been done by now.
166 */
167 ASSERT(newval > 0);
168 ASSERT(newval <= spa_max_replication(os->os_spa));
169
170 os->os_copies = newval;
171 }
172
173 static void
174 dedup_changed_cb(void *arg, uint64_t newval)
175 {
176 objset_t *os = arg;
177 spa_t *spa = os->os_spa;
178 enum zio_checksum checksum;
179
180 /*
181 * Inheritance should have been done by now.
182 */
183 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
184
185 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
186
187 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
188 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
189 }
190
191 static void
192 primary_cache_changed_cb(void *arg, uint64_t newval)
193 {
194 objset_t *os = arg;
195
196 /*
197 * Inheritance and range checking should have been done by now.
198 */
199 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
200 newval == ZFS_CACHE_METADATA);
201
202 os->os_primary_cache = newval;
203 }
204
205 static void
206 secondary_cache_changed_cb(void *arg, uint64_t newval)
207 {
208 objset_t *os = arg;
209
210 /*
211 * Inheritance and range checking should have been done by now.
212 */
213 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
214 newval == ZFS_CACHE_METADATA);
215
216 os->os_secondary_cache = newval;
217 }
218
219 static void
220 sync_changed_cb(void *arg, uint64_t newval)
221 {
222 objset_t *os = arg;
223
224 /*
225 * Inheritance and range checking should have been done by now.
226 */
227 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
228 newval == ZFS_SYNC_DISABLED);
229
230 os->os_sync = newval;
231 if (os->os_zil)
232 zil_set_sync(os->os_zil, newval);
233 }
234
235 static void
236 redundant_metadata_changed_cb(void *arg, uint64_t newval)
237 {
238 objset_t *os = arg;
239
240 /*
241 * Inheritance and range checking should have been done by now.
242 */
243 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
244 newval == ZFS_REDUNDANT_METADATA_MOST);
245
246 os->os_redundant_metadata = newval;
247 }
248
249 static void
250 logbias_changed_cb(void *arg, uint64_t newval)
251 {
252 objset_t *os = arg;
253
254 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
255 newval == ZFS_LOGBIAS_THROUGHPUT);
256 os->os_logbias = newval;
257 if (os->os_zil)
258 zil_set_logbias(os->os_zil, newval);
259 }
260
261 static void
262 recordsize_changed_cb(void *arg, uint64_t newval)
263 {
264 objset_t *os = arg;
265
266 os->os_recordsize = newval;
267 }
268
269 void
270 dmu_objset_byteswap(void *buf, size_t size)
271 {
272 objset_phys_t *osp = buf;
273
274 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
275 dnode_byteswap(&osp->os_meta_dnode);
276 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
277 osp->os_type = BSWAP_64(osp->os_type);
278 osp->os_flags = BSWAP_64(osp->os_flags);
279 if (size == sizeof (objset_phys_t)) {
280 dnode_byteswap(&osp->os_userused_dnode);
281 dnode_byteswap(&osp->os_groupused_dnode);
282 }
283 }
284
285 int
286 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
287 objset_t **osp)
288 {
289 objset_t *os;
290 int i, err;
291
292 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
293
294 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
295 os->os_dsl_dataset = ds;
296 os->os_spa = spa;
297 os->os_rootbp = bp;
298 if (!BP_IS_HOLE(os->os_rootbp)) {
299 arc_flags_t aflags = ARC_FLAG_WAIT;
300 zbookmark_phys_t zb;
301 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
302 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
303
304 if (DMU_OS_IS_L2CACHEABLE(os))
305 aflags |= ARC_FLAG_L2CACHE;
306 if (DMU_OS_IS_L2COMPRESSIBLE(os))
307 aflags |= ARC_FLAG_L2COMPRESS;
308
309 dprintf_bp(os->os_rootbp, "reading %s", "");
310 err = arc_read(NULL, spa, os->os_rootbp,
311 arc_getbuf_func, &os->os_phys_buf,
312 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
313 if (err != 0) {
314 kmem_free(os, sizeof (objset_t));
315 /* convert checksum errors into IO errors */
316 if (err == ECKSUM)
317 err = SET_ERROR(EIO);
318 return (err);
319 }
320
321 /* Increase the blocksize if we are permitted. */
322 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
323 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
324 arc_buf_t *buf = arc_buf_alloc(spa,
325 sizeof (objset_phys_t), &os->os_phys_buf,
326 ARC_BUFC_METADATA);
327 bzero(buf->b_data, sizeof (objset_phys_t));
328 bcopy(os->os_phys_buf->b_data, buf->b_data,
329 arc_buf_size(os->os_phys_buf));
330 (void) arc_buf_remove_ref(os->os_phys_buf,
331 &os->os_phys_buf);
332 os->os_phys_buf = buf;
333 }
334
335 os->os_phys = os->os_phys_buf->b_data;
336 os->os_flags = os->os_phys->os_flags;
337 } else {
338 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
339 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
340 os->os_phys_buf = arc_buf_alloc(spa, size,
341 &os->os_phys_buf, ARC_BUFC_METADATA);
342 os->os_phys = os->os_phys_buf->b_data;
343 bzero(os->os_phys, size);
344 }
345
346 /*
347 * Note: the changed_cb will be called once before the register
348 * func returns, thus changing the checksum/compression from the
349 * default (fletcher2/off). Snapshots don't need to know about
350 * checksum/compression/copies.
351 */
352 if (ds != NULL) {
353 err = dsl_prop_register(ds,
354 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
355 primary_cache_changed_cb, os);
356 if (err == 0) {
357 err = dsl_prop_register(ds,
358 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
359 secondary_cache_changed_cb, os);
360 }
361 if (!ds->ds_is_snapshot) {
362 if (err == 0) {
363 err = dsl_prop_register(ds,
364 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
365 checksum_changed_cb, os);
366 }
367 if (err == 0) {
368 err = dsl_prop_register(ds,
369 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
370 compression_changed_cb, os);
371 }
372 if (err == 0) {
373 err = dsl_prop_register(ds,
374 zfs_prop_to_name(ZFS_PROP_COPIES),
375 copies_changed_cb, os);
376 }
377 if (err == 0) {
378 err = dsl_prop_register(ds,
379 zfs_prop_to_name(ZFS_PROP_DEDUP),
380 dedup_changed_cb, os);
381 }
382 if (err == 0) {
383 err = dsl_prop_register(ds,
384 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
385 logbias_changed_cb, os);
386 }
387 if (err == 0) {
388 err = dsl_prop_register(ds,
389 zfs_prop_to_name(ZFS_PROP_SYNC),
390 sync_changed_cb, os);
391 }
392 if (err == 0) {
393 err = dsl_prop_register(ds,
394 zfs_prop_to_name(
395 ZFS_PROP_REDUNDANT_METADATA),
396 redundant_metadata_changed_cb, os);
397 }
398 if (err == 0) {
399 err = dsl_prop_register(ds,
400 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
401 recordsize_changed_cb, os);
402 }
403 }
404 if (err != 0) {
405 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
406 &os->os_phys_buf));
407 kmem_free(os, sizeof (objset_t));
408 return (err);
409 }
410 } else {
411 /* It's the meta-objset. */
412 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
413 os->os_compress = ZIO_COMPRESS_ON;
414 os->os_copies = spa_max_replication(spa);
415 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
416 os->os_dedup_verify = B_FALSE;
417 os->os_logbias = ZFS_LOGBIAS_LATENCY;
418 os->os_sync = ZFS_SYNC_STANDARD;
419 os->os_primary_cache = ZFS_CACHE_ALL;
420 os->os_secondary_cache = ZFS_CACHE_ALL;
421 }
422
423 if (ds == NULL || !ds->ds_is_snapshot)
424 os->os_zil_header = os->os_phys->os_zil_header;
425 os->os_zil = zil_alloc(os, &os->os_zil_header);
426
427 for (i = 0; i < TXG_SIZE; i++) {
428 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
429 offsetof(dnode_t, dn_dirty_link[i]));
430 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
431 offsetof(dnode_t, dn_dirty_link[i]));
432 }
433 list_create(&os->os_dnodes, sizeof (dnode_t),
434 offsetof(dnode_t, dn_link));
435 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
436 offsetof(dmu_buf_impl_t, db_link));
437
438 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
439 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
440 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
441
442 dnode_special_open(os, &os->os_phys->os_meta_dnode,
443 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
444 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
445 dnode_special_open(os, &os->os_phys->os_userused_dnode,
446 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
447 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
448 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
449 }
450
451 *osp = os;
452 return (0);
453 }
454
455 int
456 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
457 {
458 int err = 0;
459
460 mutex_enter(&ds->ds_opening_lock);
461 if (ds->ds_objset == NULL) {
462 objset_t *os;
463 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
464 ds, dsl_dataset_get_blkptr(ds), &os);
465
466 if (err == 0) {
467 mutex_enter(&ds->ds_lock);
468 ASSERT(ds->ds_objset == NULL);
469 ds->ds_objset = os;
470 mutex_exit(&ds->ds_lock);
471 }
472 }
473 *osp = ds->ds_objset;
474 mutex_exit(&ds->ds_opening_lock);
475 return (err);
476 }
477
478 /*
479 * Holds the pool while the objset is held. Therefore only one objset
480 * can be held at a time.
481 */
482 int
483 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
484 {
485 dsl_pool_t *dp;
486 dsl_dataset_t *ds;
487 int err;
488
489 err = dsl_pool_hold(name, tag, &dp);
490 if (err != 0)
491 return (err);
492 err = dsl_dataset_hold(dp, name, tag, &ds);
493 if (err != 0) {
494 dsl_pool_rele(dp, tag);
495 return (err);
496 }
497
498 err = dmu_objset_from_ds(ds, osp);
499 if (err != 0) {
500 dsl_dataset_rele(ds, tag);
501 dsl_pool_rele(dp, tag);
502 }
503
504 return (err);
505 }
506
507 /*
508 * dsl_pool must not be held when this is called.
509 * Upon successful return, there will be a longhold on the dataset,
510 * and the dsl_pool will not be held.
511 */
512 int
513 dmu_objset_own(const char *name, dmu_objset_type_t type,
514 boolean_t readonly, void *tag, objset_t **osp)
515 {
516 dsl_pool_t *dp;
517 dsl_dataset_t *ds;
518 int err;
519
520 err = dsl_pool_hold(name, FTAG, &dp);
521 if (err != 0)
522 return (err);
523 err = dsl_dataset_own(dp, name, tag, &ds);
524 if (err != 0) {
525 dsl_pool_rele(dp, FTAG);
526 return (err);
527 }
528
529 err = dmu_objset_from_ds(ds, osp);
530 dsl_pool_rele(dp, FTAG);
531 if (err != 0) {
532 dsl_dataset_disown(ds, tag);
533 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
534 dsl_dataset_disown(ds, tag);
535 return (SET_ERROR(EINVAL));
536 } else if (!readonly && ds->ds_is_snapshot) {
537 dsl_dataset_disown(ds, tag);
538 return (SET_ERROR(EROFS));
539 }
540 return (err);
541 }
542
543 void
544 dmu_objset_rele(objset_t *os, void *tag)
545 {
546 dsl_pool_t *dp = dmu_objset_pool(os);
547 dsl_dataset_rele(os->os_dsl_dataset, tag);
548 dsl_pool_rele(dp, tag);
549 }
550
551 /*
552 * When we are called, os MUST refer to an objset associated with a dataset
553 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
554 * == tag. We will then release and reacquire ownership of the dataset while
555 * holding the pool config_rwlock to avoid intervening namespace or ownership
556 * changes may occur.
557 *
558 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
559 * release the hold on its dataset and acquire a new one on the dataset of the
560 * same name so that it can be partially torn down and reconstructed.
561 */
562 void
563 dmu_objset_refresh_ownership(objset_t *os, void *tag)
564 {
565 dsl_pool_t *dp;
566 dsl_dataset_t *ds, *newds;
567 char name[MAXNAMELEN];
568
569 ds = os->os_dsl_dataset;
570 VERIFY3P(ds, !=, NULL);
571 VERIFY3P(ds->ds_owner, ==, tag);
572 VERIFY(dsl_dataset_long_held(ds));
573
574 dsl_dataset_name(ds, name);
575 dp = dmu_objset_pool(os);
576 dsl_pool_config_enter(dp, FTAG);
577 dmu_objset_disown(os, tag);
578 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
579 VERIFY3P(newds, ==, os->os_dsl_dataset);
580 dsl_pool_config_exit(dp, FTAG);
581 }
582
583 void
584 dmu_objset_disown(objset_t *os, void *tag)
585 {
586 dsl_dataset_disown(os->os_dsl_dataset, tag);
587 }
588
589 void
590 dmu_objset_evict_dbufs(objset_t *os)
591 {
592 dnode_t dn_marker;
593 dnode_t *dn;
594
595 mutex_enter(&os->os_lock);
596 dn = list_head(&os->os_dnodes);
597 while (dn != NULL) {
598 /*
599 * Skip dnodes without holds. We have to do this dance
600 * because dnode_add_ref() only works if there is already a
601 * hold. If the dnode has no holds, then it has no dbufs.
602 */
603 if (dnode_add_ref(dn, FTAG)) {
604 list_insert_after(&os->os_dnodes, dn, &dn_marker);
605 mutex_exit(&os->os_lock);
606
607 dnode_evict_dbufs(dn);
608 dnode_rele(dn, FTAG);
609
610 mutex_enter(&os->os_lock);
611 dn = list_next(&os->os_dnodes, &dn_marker);
612 list_remove(&os->os_dnodes, &dn_marker);
613 } else {
614 dn = list_next(&os->os_dnodes, dn);
615 }
616 }
617 mutex_exit(&os->os_lock);
618
619 if (DMU_USERUSED_DNODE(os) != NULL) {
620 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
621 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
622 }
623 dnode_evict_dbufs(DMU_META_DNODE(os));
624 }
625
626 /*
627 * Objset eviction processing is split into into two pieces.
628 * The first marks the objset as evicting, evicts any dbufs that
629 * have a refcount of zero, and then queues up the objset for the
630 * second phase of eviction. Once os->os_dnodes has been cleared by
631 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
632 * The second phase closes the special dnodes, dequeues the objset from
633 * the list of those undergoing eviction, and finally frees the objset.
634 *
635 * NOTE: Due to asynchronous eviction processing (invocation of
636 * dnode_buf_pageout()), it is possible for the meta dnode for the
637 * objset to have no holds even though os->os_dnodes is not empty.
638 */
639 void
640 dmu_objset_evict(objset_t *os)
641 {
642 dsl_dataset_t *ds = os->os_dsl_dataset;
643
644 for (int t = 0; t < TXG_SIZE; t++)
645 ASSERT(!dmu_objset_is_dirty(os, t));
646
647 if (ds) {
648 if (!ds->ds_is_snapshot) {
649 VERIFY0(dsl_prop_unregister(ds,
650 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
651 checksum_changed_cb, os));
652 VERIFY0(dsl_prop_unregister(ds,
653 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
654 compression_changed_cb, os));
655 VERIFY0(dsl_prop_unregister(ds,
656 zfs_prop_to_name(ZFS_PROP_COPIES),
657 copies_changed_cb, os));
658 VERIFY0(dsl_prop_unregister(ds,
659 zfs_prop_to_name(ZFS_PROP_DEDUP),
660 dedup_changed_cb, os));
661 VERIFY0(dsl_prop_unregister(ds,
662 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
663 logbias_changed_cb, os));
664 VERIFY0(dsl_prop_unregister(ds,
665 zfs_prop_to_name(ZFS_PROP_SYNC),
666 sync_changed_cb, os));
667 VERIFY0(dsl_prop_unregister(ds,
668 zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA),
669 redundant_metadata_changed_cb, os));
670 VERIFY0(dsl_prop_unregister(ds,
671 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
672 recordsize_changed_cb, os));
673 }
674 VERIFY0(dsl_prop_unregister(ds,
675 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
676 primary_cache_changed_cb, os));
677 VERIFY0(dsl_prop_unregister(ds,
678 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
679 secondary_cache_changed_cb, os));
680 }
681
682 if (os->os_sa)
683 sa_tear_down(os);
684
685 dmu_objset_evict_dbufs(os);
686
687 mutex_enter(&os->os_lock);
688 spa_evicting_os_register(os->os_spa, os);
689 if (list_is_empty(&os->os_dnodes)) {
690 mutex_exit(&os->os_lock);
691 dmu_objset_evict_done(os);
692 } else {
693 mutex_exit(&os->os_lock);
694 }
695 }
696
697 void
698 dmu_objset_evict_done(objset_t *os)
699 {
700 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
701
702 dnode_special_close(&os->os_meta_dnode);
703 if (DMU_USERUSED_DNODE(os)) {
704 dnode_special_close(&os->os_userused_dnode);
705 dnode_special_close(&os->os_groupused_dnode);
706 }
707 zil_free(os->os_zil);
708
709 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
710
711 /*
712 * This is a barrier to prevent the objset from going away in
713 * dnode_move() until we can safely ensure that the objset is still in
714 * use. We consider the objset valid before the barrier and invalid
715 * after the barrier.
716 */
717 rw_enter(&os_lock, RW_READER);
718 rw_exit(&os_lock);
719
720 mutex_destroy(&os->os_lock);
721 mutex_destroy(&os->os_obj_lock);
722 mutex_destroy(&os->os_user_ptr_lock);
723 spa_evicting_os_deregister(os->os_spa, os);
724 kmem_free(os, sizeof (objset_t));
725 }
726
727 timestruc_t
728 dmu_objset_snap_cmtime(objset_t *os)
729 {
730 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
731 }
732
733 /* called from dsl for meta-objset */
734 objset_t *
735 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
736 dmu_objset_type_t type, dmu_tx_t *tx)
737 {
738 objset_t *os;
739 dnode_t *mdn;
740
741 ASSERT(dmu_tx_is_syncing(tx));
742
743 if (ds != NULL)
744 VERIFY0(dmu_objset_from_ds(ds, &os));
745 else
746 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
747
748 mdn = DMU_META_DNODE(os);
749
750 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
751 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
752
753 /*
754 * We don't want to have to increase the meta-dnode's nlevels
755 * later, because then we could do it in quescing context while
756 * we are also accessing it in open context.
757 *
758 * This precaution is not necessary for the MOS (ds == NULL),
759 * because the MOS is only updated in syncing context.
760 * This is most fortunate: the MOS is the only objset that
761 * needs to be synced multiple times as spa_sync() iterates
762 * to convergence, so minimizing its dn_nlevels matters.
763 */
764 if (ds != NULL) {
765 int levels = 1;
766
767 /*
768 * Determine the number of levels necessary for the meta-dnode
769 * to contain DN_MAX_OBJECT dnodes.
770 */
771 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
772 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
773 DN_MAX_OBJECT * sizeof (dnode_phys_t))
774 levels++;
775
776 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
777 mdn->dn_nlevels = levels;
778 }
779
780 ASSERT(type != DMU_OST_NONE);
781 ASSERT(type != DMU_OST_ANY);
782 ASSERT(type < DMU_OST_NUMTYPES);
783 os->os_phys->os_type = type;
784 if (dmu_objset_userused_enabled(os)) {
785 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
786 os->os_flags = os->os_phys->os_flags;
787 }
788
789 dsl_dataset_dirty(ds, tx);
790
791 return (os);
792 }
793
794 typedef struct dmu_objset_create_arg {
795 const char *doca_name;
796 cred_t *doca_cred;
797 void (*doca_userfunc)(objset_t *os, void *arg,
798 cred_t *cr, dmu_tx_t *tx);
799 void *doca_userarg;
800 dmu_objset_type_t doca_type;
801 uint64_t doca_flags;
802 } dmu_objset_create_arg_t;
803
804 /*ARGSUSED*/
805 static int
806 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
807 {
808 dmu_objset_create_arg_t *doca = arg;
809 dsl_pool_t *dp = dmu_tx_pool(tx);
810 dsl_dir_t *pdd;
811 const char *tail;
812 int error;
813
814 if (strchr(doca->doca_name, '@') != NULL)
815 return (SET_ERROR(EINVAL));
816
817 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
818 if (error != 0)
819 return (error);
820 if (tail == NULL) {
821 dsl_dir_rele(pdd, FTAG);
822 return (SET_ERROR(EEXIST));
823 }
824 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
825 doca->doca_cred);
826 dsl_dir_rele(pdd, FTAG);
827
828 return (error);
829 }
830
831 static void
832 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
833 {
834 dmu_objset_create_arg_t *doca = arg;
835 dsl_pool_t *dp = dmu_tx_pool(tx);
836 dsl_dir_t *pdd;
837 const char *tail;
838 dsl_dataset_t *ds;
839 uint64_t obj;
840 blkptr_t *bp;
841 objset_t *os;
842
843 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
844
845 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
846 doca->doca_cred, tx);
847
848 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
849 bp = dsl_dataset_get_blkptr(ds);
850 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
851 ds, bp, doca->doca_type, tx);
852
853 if (doca->doca_userfunc != NULL) {
854 doca->doca_userfunc(os, doca->doca_userarg,
855 doca->doca_cred, tx);
856 }
857
858 spa_history_log_internal_ds(ds, "create", tx, "");
859 dsl_dataset_rele(ds, FTAG);
860 dsl_dir_rele(pdd, FTAG);
861 }
862
863 int
864 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
865 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
866 {
867 dmu_objset_create_arg_t doca;
868
869 doca.doca_name = name;
870 doca.doca_cred = CRED();
871 doca.doca_flags = flags;
872 doca.doca_userfunc = func;
873 doca.doca_userarg = arg;
874 doca.doca_type = type;
875
876 return (dsl_sync_task(name,
877 dmu_objset_create_check, dmu_objset_create_sync, &doca,
878 5, ZFS_SPACE_CHECK_NORMAL));
879 }
880
881 typedef struct dmu_objset_clone_arg {
882 const char *doca_clone;
883 const char *doca_origin;
884 cred_t *doca_cred;
885 } dmu_objset_clone_arg_t;
886
887 /*ARGSUSED*/
888 static int
889 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
890 {
891 dmu_objset_clone_arg_t *doca = arg;
892 dsl_dir_t *pdd;
893 const char *tail;
894 int error;
895 dsl_dataset_t *origin;
896 dsl_pool_t *dp = dmu_tx_pool(tx);
897
898 if (strchr(doca->doca_clone, '@') != NULL)
899 return (SET_ERROR(EINVAL));
900
901 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
902 if (error != 0)
903 return (error);
904 if (tail == NULL) {
905 dsl_dir_rele(pdd, FTAG);
906 return (SET_ERROR(EEXIST));
907 }
908
909 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
910 doca->doca_cred);
911 if (error != 0) {
912 dsl_dir_rele(pdd, FTAG);
913 return (SET_ERROR(EDQUOT));
914 }
915 dsl_dir_rele(pdd, FTAG);
916
917 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
918 if (error != 0)
919 return (error);
920
921 /* You can only clone snapshots, not the head datasets. */
922 if (!origin->ds_is_snapshot) {
923 dsl_dataset_rele(origin, FTAG);
924 return (SET_ERROR(EINVAL));
925 }
926 dsl_dataset_rele(origin, FTAG);
927
928 return (0);
929 }
930
931 static void
932 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
933 {
934 dmu_objset_clone_arg_t *doca = arg;
935 dsl_pool_t *dp = dmu_tx_pool(tx);
936 dsl_dir_t *pdd;
937 const char *tail;
938 dsl_dataset_t *origin, *ds;
939 uint64_t obj;
940 char namebuf[MAXNAMELEN];
941
942 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
943 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
944
945 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
946 doca->doca_cred, tx);
947
948 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
949 dsl_dataset_name(origin, namebuf);
950 spa_history_log_internal_ds(ds, "clone", tx,
951 "origin=%s (%llu)", namebuf, origin->ds_object);
952 dsl_dataset_rele(ds, FTAG);
953 dsl_dataset_rele(origin, FTAG);
954 dsl_dir_rele(pdd, FTAG);
955 }
956
957 int
958 dmu_objset_clone(const char *clone, const char *origin)
959 {
960 dmu_objset_clone_arg_t doca;
961
962 doca.doca_clone = clone;
963 doca.doca_origin = origin;
964 doca.doca_cred = CRED();
965
966 return (dsl_sync_task(clone,
967 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
968 5, ZFS_SPACE_CHECK_NORMAL));
969 }
970
971 int
972 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
973 {
974 int err;
975 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
976 nvlist_t *snaps = fnvlist_alloc();
977
978 fnvlist_add_boolean(snaps, longsnap);
979 strfree(longsnap);
980 err = dsl_dataset_snapshot(snaps, NULL, NULL);
981 fnvlist_free(snaps);
982 return (err);
983 }
984
985 static void
986 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
987 {
988 dnode_t *dn;
989
990 while (dn = list_head(list)) {
991 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
992 ASSERT(dn->dn_dbuf->db_data_pending);
993 /*
994 * Initialize dn_zio outside dnode_sync() because the
995 * meta-dnode needs to set it ouside dnode_sync().
996 */
997 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
998 ASSERT(dn->dn_zio);
999
1000 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1001 list_remove(list, dn);
1002
1003 if (newlist) {
1004 (void) dnode_add_ref(dn, newlist);
1005 list_insert_tail(newlist, dn);
1006 }
1007
1008 dnode_sync(dn, tx);
1009 }
1010 }
1011
1012 /* ARGSUSED */
1013 static void
1014 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1015 {
1016 blkptr_t *bp = zio->io_bp;
1017 objset_t *os = arg;
1018 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1019
1020 ASSERT(!BP_IS_EMBEDDED(bp));
1021 ASSERT3P(bp, ==, os->os_rootbp);
1022 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1023 ASSERT0(BP_GET_LEVEL(bp));
1024
1025 /*
1026 * Update rootbp fill count: it should be the number of objects
1027 * allocated in the object set (not counting the "special"
1028 * objects that are stored in the objset_phys_t -- the meta
1029 * dnode and user/group accounting objects).
1030 */
1031 bp->blk_fill = 0;
1032 for (int i = 0; i < dnp->dn_nblkptr; i++)
1033 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1034 }
1035
1036 /* ARGSUSED */
1037 static void
1038 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1039 {
1040 blkptr_t *bp = zio->io_bp;
1041 blkptr_t *bp_orig = &zio->io_bp_orig;
1042 objset_t *os = arg;
1043
1044 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1045 ASSERT(BP_EQUAL(bp, bp_orig));
1046 } else {
1047 dsl_dataset_t *ds = os->os_dsl_dataset;
1048 dmu_tx_t *tx = os->os_synctx;
1049
1050 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1051 dsl_dataset_block_born(ds, bp, tx);
1052 }
1053 }
1054
1055 /* called from dsl */
1056 void
1057 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1058 {
1059 int txgoff;
1060 zbookmark_phys_t zb;
1061 zio_prop_t zp;
1062 zio_t *zio;
1063 list_t *list;
1064 list_t *newlist = NULL;
1065 dbuf_dirty_record_t *dr;
1066
1067 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1068
1069 ASSERT(dmu_tx_is_syncing(tx));
1070 /* XXX the write_done callback should really give us the tx... */
1071 os->os_synctx = tx;
1072
1073 if (os->os_dsl_dataset == NULL) {
1074 /*
1075 * This is the MOS. If we have upgraded,
1076 * spa_max_replication() could change, so reset
1077 * os_copies here.
1078 */
1079 os->os_copies = spa_max_replication(os->os_spa);
1080 }
1081
1082 /*
1083 * Create the root block IO
1084 */
1085 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1086 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1087 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1088 arc_release(os->os_phys_buf, &os->os_phys_buf);
1089
1090 dmu_write_policy(os, NULL, 0, 0, &zp);
1091
1092 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1093 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1094 DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1095 NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1096 ZIO_FLAG_MUSTSUCCEED, &zb);
1097
1098 /*
1099 * Sync special dnodes - the parent IO for the sync is the root block
1100 */
1101 DMU_META_DNODE(os)->dn_zio = zio;
1102 dnode_sync(DMU_META_DNODE(os), tx);
1103
1104 os->os_phys->os_flags = os->os_flags;
1105
1106 if (DMU_USERUSED_DNODE(os) &&
1107 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1108 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1109 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1110 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1111 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1112 }
1113
1114 txgoff = tx->tx_txg & TXG_MASK;
1115
1116 if (dmu_objset_userused_enabled(os)) {
1117 newlist = &os->os_synced_dnodes;
1118 /*
1119 * We must create the list here because it uses the
1120 * dn_dirty_link[] of this txg.
1121 */
1122 list_create(newlist, sizeof (dnode_t),
1123 offsetof(dnode_t, dn_dirty_link[txgoff]));
1124 }
1125
1126 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1127 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1128
1129 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1130 while (dr = list_head(list)) {
1131 ASSERT0(dr->dr_dbuf->db_level);
1132 list_remove(list, dr);
1133 if (dr->dr_zio)
1134 zio_nowait(dr->dr_zio);
1135 }
1136 /*
1137 * Free intent log blocks up to this tx.
1138 */
1139 zil_sync(os->os_zil, tx);
1140 os->os_phys->os_zil_header = os->os_zil_header;
1141 zio_nowait(zio);
1142 }
1143
1144 boolean_t
1145 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1146 {
1147 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1148 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1149 }
1150
1151 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1152
1153 void
1154 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1155 {
1156 used_cbs[ost] = cb;
1157 }
1158
1159 boolean_t
1160 dmu_objset_userused_enabled(objset_t *os)
1161 {
1162 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1163 used_cbs[os->os_phys->os_type] != NULL &&
1164 DMU_USERUSED_DNODE(os) != NULL);
1165 }
1166
1167 static void
1168 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1169 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1170 {
1171 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1172 int64_t delta = DNODE_SIZE + used;
1173 if (subtract)
1174 delta = -delta;
1175 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1176 user, delta, tx));
1177 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1178 group, delta, tx));
1179 }
1180 }
1181
1182 void
1183 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1184 {
1185 dnode_t *dn;
1186 list_t *list = &os->os_synced_dnodes;
1187
1188 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1189
1190 while (dn = list_head(list)) {
1191 int flags;
1192 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1193 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1194 dn->dn_phys->dn_flags &
1195 DNODE_FLAG_USERUSED_ACCOUNTED);
1196
1197 /* Allocate the user/groupused objects if necessary. */
1198 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1199 VERIFY(0 == zap_create_claim(os,
1200 DMU_USERUSED_OBJECT,
1201 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1202 VERIFY(0 == zap_create_claim(os,
1203 DMU_GROUPUSED_OBJECT,
1204 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1205 }
1206
1207 /*
1208 * We intentionally modify the zap object even if the
1209 * net delta is zero. Otherwise
1210 * the block of the zap obj could be shared between
1211 * datasets but need to be different between them after
1212 * a bprewrite.
1213 */
1214
1215 flags = dn->dn_id_flags;
1216 ASSERT(flags);
1217 if (flags & DN_ID_OLD_EXIST) {
1218 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1219 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1220 }
1221 if (flags & DN_ID_NEW_EXIST) {
1222 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1223 dn->dn_phys->dn_flags, dn->dn_newuid,
1224 dn->dn_newgid, B_FALSE, tx);
1225 }
1226
1227 mutex_enter(&dn->dn_mtx);
1228 dn->dn_oldused = 0;
1229 dn->dn_oldflags = 0;
1230 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1231 dn->dn_olduid = dn->dn_newuid;
1232 dn->dn_oldgid = dn->dn_newgid;
1233 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1234 if (dn->dn_bonuslen == 0)
1235 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1236 else
1237 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1238 }
1239 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1240 mutex_exit(&dn->dn_mtx);
1241
1242 list_remove(list, dn);
1243 dnode_rele(dn, list);
1244 }
1245 }
1246
1247 /*
1248 * Returns a pointer to data to find uid/gid from
1249 *
1250 * If a dirty record for transaction group that is syncing can't
1251 * be found then NULL is returned. In the NULL case it is assumed
1252 * the uid/gid aren't changing.
1253 */
1254 static void *
1255 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1256 {
1257 dbuf_dirty_record_t *dr, **drp;
1258 void *data;
1259
1260 if (db->db_dirtycnt == 0)
1261 return (db->db.db_data); /* Nothing is changing */
1262
1263 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1264 if (dr->dr_txg == tx->tx_txg)
1265 break;
1266
1267 if (dr == NULL) {
1268 data = NULL;
1269 } else {
1270 dnode_t *dn;
1271
1272 DB_DNODE_ENTER(dr->dr_dbuf);
1273 dn = DB_DNODE(dr->dr_dbuf);
1274
1275 if (dn->dn_bonuslen == 0 &&
1276 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1277 data = dr->dt.dl.dr_data->b_data;
1278 else
1279 data = dr->dt.dl.dr_data;
1280
1281 DB_DNODE_EXIT(dr->dr_dbuf);
1282 }
1283
1284 return (data);
1285 }
1286
1287 void
1288 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1289 {
1290 objset_t *os = dn->dn_objset;
1291 void *data = NULL;
1292 dmu_buf_impl_t *db = NULL;
1293 uint64_t *user = NULL;
1294 uint64_t *group = NULL;
1295 int flags = dn->dn_id_flags;
1296 int error;
1297 boolean_t have_spill = B_FALSE;
1298
1299 if (!dmu_objset_userused_enabled(dn->dn_objset))
1300 return;
1301
1302 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1303 DN_ID_CHKED_SPILL)))
1304 return;
1305
1306 if (before && dn->dn_bonuslen != 0)
1307 data = DN_BONUS(dn->dn_phys);
1308 else if (!before && dn->dn_bonuslen != 0) {
1309 if (dn->dn_bonus) {
1310 db = dn->dn_bonus;
1311 mutex_enter(&db->db_mtx);
1312 data = dmu_objset_userquota_find_data(db, tx);
1313 } else {
1314 data = DN_BONUS(dn->dn_phys);
1315 }
1316 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1317 int rf = 0;
1318
1319 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1320 rf |= DB_RF_HAVESTRUCT;
1321 error = dmu_spill_hold_by_dnode(dn,
1322 rf | DB_RF_MUST_SUCCEED,
1323 FTAG, (dmu_buf_t **)&db);
1324 ASSERT(error == 0);
1325 mutex_enter(&db->db_mtx);
1326 data = (before) ? db->db.db_data :
1327 dmu_objset_userquota_find_data(db, tx);
1328 have_spill = B_TRUE;
1329 } else {
1330 mutex_enter(&dn->dn_mtx);
1331 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1332 mutex_exit(&dn->dn_mtx);
1333 return;
1334 }
1335
1336 if (before) {
1337 ASSERT(data);
1338 user = &dn->dn_olduid;
1339 group = &dn->dn_oldgid;
1340 } else if (data) {
1341 user = &dn->dn_newuid;
1342 group = &dn->dn_newgid;
1343 }
1344
1345 /*
1346 * Must always call the callback in case the object
1347 * type has changed and that type isn't an object type to track
1348 */
1349 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1350 user, group);
1351
1352 /*
1353 * Preserve existing uid/gid when the callback can't determine
1354 * what the new uid/gid are and the callback returned EEXIST.
1355 * The EEXIST error tells us to just use the existing uid/gid.
1356 * If we don't know what the old values are then just assign
1357 * them to 0, since that is a new file being created.
1358 */
1359 if (!before && data == NULL && error == EEXIST) {
1360 if (flags & DN_ID_OLD_EXIST) {
1361 dn->dn_newuid = dn->dn_olduid;
1362 dn->dn_newgid = dn->dn_oldgid;
1363 } else {
1364 dn->dn_newuid = 0;
1365 dn->dn_newgid = 0;
1366 }
1367 error = 0;
1368 }
1369
1370 if (db)
1371 mutex_exit(&db->db_mtx);
1372
1373 mutex_enter(&dn->dn_mtx);
1374 if (error == 0 && before)
1375 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1376 if (error == 0 && !before)
1377 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1378
1379 if (have_spill) {
1380 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1381 } else {
1382 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1383 }
1384 mutex_exit(&dn->dn_mtx);
1385 if (have_spill)
1386 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1387 }
1388
1389 boolean_t
1390 dmu_objset_userspace_present(objset_t *os)
1391 {
1392 return (os->os_phys->os_flags &
1393 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1394 }
1395
1396 int
1397 dmu_objset_userspace_upgrade(objset_t *os)
1398 {
1399 uint64_t obj;
1400 int err = 0;
1401
1402 if (dmu_objset_userspace_present(os))
1403 return (0);
1404 if (!dmu_objset_userused_enabled(os))
1405 return (SET_ERROR(ENOTSUP));
1406 if (dmu_objset_is_snapshot(os))
1407 return (SET_ERROR(EINVAL));
1408
1409 /*
1410 * We simply need to mark every object dirty, so that it will be
1411 * synced out and now accounted. If this is called
1412 * concurrently, or if we already did some work before crashing,
1413 * that's fine, since we track each object's accounted state
1414 * independently.
1415 */
1416
1417 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1418 dmu_tx_t *tx;
1419 dmu_buf_t *db;
1420 int objerr;
1421
1422 if (issig(JUSTLOOKING) && issig(FORREAL))
1423 return (SET_ERROR(EINTR));
1424
1425 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1426 if (objerr != 0)
1427 continue;
1428 tx = dmu_tx_create(os);
1429 dmu_tx_hold_bonus(tx, obj);
1430 objerr = dmu_tx_assign(tx, TXG_WAIT);
1431 if (objerr != 0) {
1432 dmu_tx_abort(tx);
1433 continue;
1434 }
1435 dmu_buf_will_dirty(db, tx);
1436 dmu_buf_rele(db, FTAG);
1437 dmu_tx_commit(tx);
1438 }
1439
1440 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1441 txg_wait_synced(dmu_objset_pool(os), 0);
1442 return (0);
1443 }
1444
1445 void
1446 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1447 uint64_t *usedobjsp, uint64_t *availobjsp)
1448 {
1449 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1450 usedobjsp, availobjsp);
1451 }
1452
1453 uint64_t
1454 dmu_objset_fsid_guid(objset_t *os)
1455 {
1456 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1457 }
1458
1459 void
1460 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1461 {
1462 stat->dds_type = os->os_phys->os_type;
1463 if (os->os_dsl_dataset)
1464 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1465 }
1466
1467 void
1468 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1469 {
1470 ASSERT(os->os_dsl_dataset ||
1471 os->os_phys->os_type == DMU_OST_META);
1472
1473 if (os->os_dsl_dataset != NULL)
1474 dsl_dataset_stats(os->os_dsl_dataset, nv);
1475
1476 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1477 os->os_phys->os_type);
1478 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1479 dmu_objset_userspace_present(os));
1480 }
1481
1482 int
1483 dmu_objset_is_snapshot(objset_t *os)
1484 {
1485 if (os->os_dsl_dataset != NULL)
1486 return (os->os_dsl_dataset->ds_is_snapshot);
1487 else
1488 return (B_FALSE);
1489 }
1490
1491 int
1492 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1493 boolean_t *conflict)
1494 {
1495 dsl_dataset_t *ds = os->os_dsl_dataset;
1496 uint64_t ignored;
1497
1498 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1499 return (SET_ERROR(ENOENT));
1500
1501 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1502 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1503 MT_FIRST, real, maxlen, conflict));
1504 }
1505
1506 int
1507 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1508 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1509 {
1510 dsl_dataset_t *ds = os->os_dsl_dataset;
1511 zap_cursor_t cursor;
1512 zap_attribute_t attr;
1513
1514 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1515
1516 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1517 return (SET_ERROR(ENOENT));
1518
1519 zap_cursor_init_serialized(&cursor,
1520 ds->ds_dir->dd_pool->dp_meta_objset,
1521 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1522
1523 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1524 zap_cursor_fini(&cursor);
1525 return (SET_ERROR(ENOENT));
1526 }
1527
1528 if (strlen(attr.za_name) + 1 > namelen) {
1529 zap_cursor_fini(&cursor);
1530 return (SET_ERROR(ENAMETOOLONG));
1531 }
1532
1533 (void) strcpy(name, attr.za_name);
1534 if (idp)
1535 *idp = attr.za_first_integer;
1536 if (case_conflict)
1537 *case_conflict = attr.za_normalization_conflict;
1538 zap_cursor_advance(&cursor);
1539 *offp = zap_cursor_serialize(&cursor);
1540 zap_cursor_fini(&cursor);
1541
1542 return (0);
1543 }
1544
1545 int
1546 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1547 uint64_t *idp, uint64_t *offp)
1548 {
1549 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1550 zap_cursor_t cursor;
1551 zap_attribute_t attr;
1552
1553 /* there is no next dir on a snapshot! */
1554 if (os->os_dsl_dataset->ds_object !=
1555 dsl_dir_phys(dd)->dd_head_dataset_obj)
1556 return (SET_ERROR(ENOENT));
1557
1558 zap_cursor_init_serialized(&cursor,
1559 dd->dd_pool->dp_meta_objset,
1560 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1561
1562 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1563 zap_cursor_fini(&cursor);
1564 return (SET_ERROR(ENOENT));
1565 }
1566
1567 if (strlen(attr.za_name) + 1 > namelen) {
1568 zap_cursor_fini(&cursor);
1569 return (SET_ERROR(ENAMETOOLONG));
1570 }
1571
1572 (void) strcpy(name, attr.za_name);
1573 if (idp)
1574 *idp = attr.za_first_integer;
1575 zap_cursor_advance(&cursor);
1576 *offp = zap_cursor_serialize(&cursor);
1577 zap_cursor_fini(&cursor);
1578
1579 return (0);
1580 }
1581
1582 /*
1583 * Find objsets under and including ddobj, call func(ds) on each.
1584 */
1585 int
1586 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1587 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1588 {
1589 dsl_dir_t *dd;
1590 dsl_dataset_t *ds;
1591 zap_cursor_t zc;
1592 zap_attribute_t *attr;
1593 uint64_t thisobj;
1594 int err;
1595
1596 ASSERT(dsl_pool_config_held(dp));
1597
1598 err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1599 if (err != 0)
1600 return (err);
1601
1602 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1603 if (dd->dd_myname[0] == '$') {
1604 dsl_dir_rele(dd, FTAG);
1605 return (0);
1606 }
1607
1608 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1609 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1610
1611 /*
1612 * Iterate over all children.
1613 */
1614 if (flags & DS_FIND_CHILDREN) {
1615 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1616 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1617 zap_cursor_retrieve(&zc, attr) == 0;
1618 (void) zap_cursor_advance(&zc)) {
1619 ASSERT3U(attr->za_integer_length, ==,
1620 sizeof (uint64_t));
1621 ASSERT3U(attr->za_num_integers, ==, 1);
1622
1623 err = dmu_objset_find_dp(dp, attr->za_first_integer,
1624 func, arg, flags);
1625 if (err != 0)
1626 break;
1627 }
1628 zap_cursor_fini(&zc);
1629
1630 if (err != 0) {
1631 dsl_dir_rele(dd, FTAG);
1632 kmem_free(attr, sizeof (zap_attribute_t));
1633 return (err);
1634 }
1635 }
1636
1637 /*
1638 * Iterate over all snapshots.
1639 */
1640 if (flags & DS_FIND_SNAPSHOTS) {
1641 dsl_dataset_t *ds;
1642 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1643
1644 if (err == 0) {
1645 uint64_t snapobj;
1646
1647 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1648 dsl_dataset_rele(ds, FTAG);
1649
1650 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1651 zap_cursor_retrieve(&zc, attr) == 0;
1652 (void) zap_cursor_advance(&zc)) {
1653 ASSERT3U(attr->za_integer_length, ==,
1654 sizeof (uint64_t));
1655 ASSERT3U(attr->za_num_integers, ==, 1);
1656
1657 err = dsl_dataset_hold_obj(dp,
1658 attr->za_first_integer, FTAG, &ds);
1659 if (err != 0)
1660 break;
1661 err = func(dp, ds, arg);
1662 dsl_dataset_rele(ds, FTAG);
1663 if (err != 0)
1664 break;
1665 }
1666 zap_cursor_fini(&zc);
1667 }
1668 }
1669
1670 dsl_dir_rele(dd, FTAG);
1671 kmem_free(attr, sizeof (zap_attribute_t));
1672
1673 if (err != 0)
1674 return (err);
1675
1676 /*
1677 * Apply to self.
1678 */
1679 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1680 if (err != 0)
1681 return (err);
1682 err = func(dp, ds, arg);
1683 dsl_dataset_rele(ds, FTAG);
1684 return (err);
1685 }
1686
1687 /*
1688 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1689 * The dp_config_rwlock must not be held when this is called, and it
1690 * will not be held when the callback is called.
1691 * Therefore this function should only be used when the pool is not changing
1692 * (e.g. in syncing context), or the callback can deal with the possible races.
1693 */
1694 static int
1695 dmu_objset_find_impl(spa_t *spa, const char *name,
1696 int func(const char *, void *), void *arg, int flags)
1697 {
1698 dsl_dir_t *dd;
1699 dsl_pool_t *dp = spa_get_dsl(spa);
1700 dsl_dataset_t *ds;
1701 zap_cursor_t zc;
1702 zap_attribute_t *attr;
1703 char *child;
1704 uint64_t thisobj;
1705 int err;
1706
1707 dsl_pool_config_enter(dp, FTAG);
1708
1709 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1710 if (err != 0) {
1711 dsl_pool_config_exit(dp, FTAG);
1712 return (err);
1713 }
1714
1715 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1716 if (dd->dd_myname[0] == '$') {
1717 dsl_dir_rele(dd, FTAG);
1718 dsl_pool_config_exit(dp, FTAG);
1719 return (0);
1720 }
1721
1722 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1723 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1724
1725 /*
1726 * Iterate over all children.
1727 */
1728 if (flags & DS_FIND_CHILDREN) {
1729 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1730 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1731 zap_cursor_retrieve(&zc, attr) == 0;
1732 (void) zap_cursor_advance(&zc)) {
1733 ASSERT3U(attr->za_integer_length, ==,
1734 sizeof (uint64_t));
1735 ASSERT3U(attr->za_num_integers, ==, 1);
1736
1737 child = kmem_asprintf("%s/%s", name, attr->za_name);
1738 dsl_pool_config_exit(dp, FTAG);
1739 err = dmu_objset_find_impl(spa, child,
1740 func, arg, flags);
1741 dsl_pool_config_enter(dp, FTAG);
1742 strfree(child);
1743 if (err != 0)
1744 break;
1745 }
1746 zap_cursor_fini(&zc);
1747
1748 if (err != 0) {
1749 dsl_dir_rele(dd, FTAG);
1750 dsl_pool_config_exit(dp, FTAG);
1751 kmem_free(attr, sizeof (zap_attribute_t));
1752 return (err);
1753 }
1754 }
1755
1756 /*
1757 * Iterate over all snapshots.
1758 */
1759 if (flags & DS_FIND_SNAPSHOTS) {
1760 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1761
1762 if (err == 0) {
1763 uint64_t snapobj;
1764
1765 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1766 dsl_dataset_rele(ds, FTAG);
1767
1768 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1769 zap_cursor_retrieve(&zc, attr) == 0;
1770 (void) zap_cursor_advance(&zc)) {
1771 ASSERT3U(attr->za_integer_length, ==,
1772 sizeof (uint64_t));
1773 ASSERT3U(attr->za_num_integers, ==, 1);
1774
1775 child = kmem_asprintf("%s@%s",
1776 name, attr->za_name);
1777 dsl_pool_config_exit(dp, FTAG);
1778 err = func(child, arg);
1779 dsl_pool_config_enter(dp, FTAG);
1780 strfree(child);
1781 if (err != 0)
1782 break;
1783 }
1784 zap_cursor_fini(&zc);
1785 }
1786 }
1787
1788 dsl_dir_rele(dd, FTAG);
1789 kmem_free(attr, sizeof (zap_attribute_t));
1790 dsl_pool_config_exit(dp, FTAG);
1791
1792 if (err != 0)
1793 return (err);
1794
1795 /* Apply to self. */
1796 return (func(name, arg));
1797 }
1798
1799 /*
1800 * See comment above dmu_objset_find_impl().
1801 */
1802 int
1803 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1804 int flags)
1805 {
1806 spa_t *spa;
1807 int error;
1808
1809 error = spa_open(name, &spa, FTAG);
1810 if (error != 0)
1811 return (error);
1812 error = dmu_objset_find_impl(spa, name, func, arg, flags);
1813 spa_close(spa, FTAG);
1814 return (error);
1815 }
1816
1817 void
1818 dmu_objset_set_user(objset_t *os, void *user_ptr)
1819 {
1820 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1821 os->os_user_ptr = user_ptr;
1822 }
1823
1824 void *
1825 dmu_objset_get_user(objset_t *os)
1826 {
1827 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1828 return (os->os_user_ptr);
1829 }
1830
1831 /*
1832 * Determine name of filesystem, given name of snapshot.
1833 * buf must be at least MAXNAMELEN bytes
1834 */
1835 int
1836 dmu_fsname(const char *snapname, char *buf)
1837 {
1838 char *atp = strchr(snapname, '@');
1839 if (atp == NULL)
1840 return (SET_ERROR(EINVAL));
1841 if (atp - snapname >= MAXNAMELEN)
1842 return (SET_ERROR(ENAMETOOLONG));
1843 (void) strlcpy(buf, snapname, atp - snapname + 1);
1844 return (0);
1845 }