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 os->os_evicting = B_TRUE;
686 dmu_objset_evict_dbufs(os);
687
688 mutex_enter(&os->os_lock);
689 spa_evicting_os_register(os->os_spa, os);
690 if (list_is_empty(&os->os_dnodes)) {
691 mutex_exit(&os->os_lock);
692 dmu_objset_evict_done(os);
693 } else {
694 mutex_exit(&os->os_lock);
695 }
696 }
697
698 void
699 dmu_objset_evict_done(objset_t *os)
700 {
701 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
702
703 dnode_special_close(&os->os_meta_dnode);
704 if (DMU_USERUSED_DNODE(os)) {
705 dnode_special_close(&os->os_userused_dnode);
706 dnode_special_close(&os->os_groupused_dnode);
707 }
708 zil_free(os->os_zil);
709
710 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
711
712 /*
713 * This is a barrier to prevent the objset from going away in
714 * dnode_move() until we can safely ensure that the objset is still in
715 * use. We consider the objset valid before the barrier and invalid
716 * after the barrier.
717 */
718 rw_enter(&os_lock, RW_READER);
719 rw_exit(&os_lock);
720
721 mutex_destroy(&os->os_lock);
722 mutex_destroy(&os->os_obj_lock);
723 mutex_destroy(&os->os_user_ptr_lock);
724 spa_evicting_os_deregister(os->os_spa, os);
725 kmem_free(os, sizeof (objset_t));
726 }
727
728 timestruc_t
729 dmu_objset_snap_cmtime(objset_t *os)
730 {
731 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
732 }
733
734 /* called from dsl for meta-objset */
735 objset_t *
736 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
737 dmu_objset_type_t type, dmu_tx_t *tx)
738 {
739 objset_t *os;
740 dnode_t *mdn;
741
742 ASSERT(dmu_tx_is_syncing(tx));
743
744 if (ds != NULL)
745 VERIFY0(dmu_objset_from_ds(ds, &os));
746 else
747 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
748
749 mdn = DMU_META_DNODE(os);
750
751 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
752 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
753
754 /*
755 * We don't want to have to increase the meta-dnode's nlevels
756 * later, because then we could do it in quescing context while
757 * we are also accessing it in open context.
758 *
759 * This precaution is not necessary for the MOS (ds == NULL),
760 * because the MOS is only updated in syncing context.
761 * This is most fortunate: the MOS is the only objset that
762 * needs to be synced multiple times as spa_sync() iterates
763 * to convergence, so minimizing its dn_nlevels matters.
764 */
765 if (ds != NULL) {
766 int levels = 1;
767
768 /*
769 * Determine the number of levels necessary for the meta-dnode
770 * to contain DN_MAX_OBJECT dnodes.
771 */
772 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
773 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
774 DN_MAX_OBJECT * sizeof (dnode_phys_t))
775 levels++;
776
777 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
778 mdn->dn_nlevels = levels;
779 }
780
781 ASSERT(type != DMU_OST_NONE);
782 ASSERT(type != DMU_OST_ANY);
783 ASSERT(type < DMU_OST_NUMTYPES);
784 os->os_phys->os_type = type;
785 if (dmu_objset_userused_enabled(os)) {
786 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
787 os->os_flags = os->os_phys->os_flags;
788 }
789
790 dsl_dataset_dirty(ds, tx);
791
792 return (os);
793 }
794
795 typedef struct dmu_objset_create_arg {
796 const char *doca_name;
797 cred_t *doca_cred;
798 void (*doca_userfunc)(objset_t *os, void *arg,
799 cred_t *cr, dmu_tx_t *tx);
800 void *doca_userarg;
801 dmu_objset_type_t doca_type;
802 uint64_t doca_flags;
803 } dmu_objset_create_arg_t;
804
805 /*ARGSUSED*/
806 static int
807 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
808 {
809 dmu_objset_create_arg_t *doca = arg;
810 dsl_pool_t *dp = dmu_tx_pool(tx);
811 dsl_dir_t *pdd;
812 const char *tail;
813 int error;
814
815 if (strchr(doca->doca_name, '@') != NULL)
816 return (SET_ERROR(EINVAL));
817
818 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
819 if (error != 0)
820 return (error);
821 if (tail == NULL) {
822 dsl_dir_rele(pdd, FTAG);
823 return (SET_ERROR(EEXIST));
824 }
825 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
826 doca->doca_cred);
827 dsl_dir_rele(pdd, FTAG);
828
829 return (error);
830 }
831
832 static void
833 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
834 {
835 dmu_objset_create_arg_t *doca = arg;
836 dsl_pool_t *dp = dmu_tx_pool(tx);
837 dsl_dir_t *pdd;
838 const char *tail;
839 dsl_dataset_t *ds;
840 uint64_t obj;
841 blkptr_t *bp;
842 objset_t *os;
843
844 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
845
846 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
847 doca->doca_cred, tx);
848
849 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
850 bp = dsl_dataset_get_blkptr(ds);
851 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
852 ds, bp, doca->doca_type, tx);
853
854 if (doca->doca_userfunc != NULL) {
855 doca->doca_userfunc(os, doca->doca_userarg,
856 doca->doca_cred, tx);
857 }
858
859 spa_history_log_internal_ds(ds, "create", tx, "");
860 dsl_dataset_rele(ds, FTAG);
861 dsl_dir_rele(pdd, FTAG);
862 }
863
864 int
865 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
866 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
867 {
868 dmu_objset_create_arg_t doca;
869
870 doca.doca_name = name;
871 doca.doca_cred = CRED();
872 doca.doca_flags = flags;
873 doca.doca_userfunc = func;
874 doca.doca_userarg = arg;
875 doca.doca_type = type;
876
877 return (dsl_sync_task(name,
878 dmu_objset_create_check, dmu_objset_create_sync, &doca,
879 5, ZFS_SPACE_CHECK_NORMAL));
880 }
881
882 typedef struct dmu_objset_clone_arg {
883 const char *doca_clone;
884 const char *doca_origin;
885 cred_t *doca_cred;
886 } dmu_objset_clone_arg_t;
887
888 /*ARGSUSED*/
889 static int
890 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
891 {
892 dmu_objset_clone_arg_t *doca = arg;
893 dsl_dir_t *pdd;
894 const char *tail;
895 int error;
896 dsl_dataset_t *origin;
897 dsl_pool_t *dp = dmu_tx_pool(tx);
898
899 if (strchr(doca->doca_clone, '@') != NULL)
900 return (SET_ERROR(EINVAL));
901
902 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
903 if (error != 0)
904 return (error);
905 if (tail == NULL) {
906 dsl_dir_rele(pdd, FTAG);
907 return (SET_ERROR(EEXIST));
908 }
909
910 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
911 doca->doca_cred);
912 if (error != 0) {
913 dsl_dir_rele(pdd, FTAG);
914 return (SET_ERROR(EDQUOT));
915 }
916 dsl_dir_rele(pdd, FTAG);
917
918 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
919 if (error != 0)
920 return (error);
921
922 /* You can only clone snapshots, not the head datasets. */
923 if (!origin->ds_is_snapshot) {
924 dsl_dataset_rele(origin, FTAG);
925 return (SET_ERROR(EINVAL));
926 }
927 dsl_dataset_rele(origin, FTAG);
928
929 return (0);
930 }
931
932 static void
933 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
934 {
935 dmu_objset_clone_arg_t *doca = arg;
936 dsl_pool_t *dp = dmu_tx_pool(tx);
937 dsl_dir_t *pdd;
938 const char *tail;
939 dsl_dataset_t *origin, *ds;
940 uint64_t obj;
941 char namebuf[MAXNAMELEN];
942
943 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
944 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
945
946 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
947 doca->doca_cred, tx);
948
949 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
950 dsl_dataset_name(origin, namebuf);
951 spa_history_log_internal_ds(ds, "clone", tx,
952 "origin=%s (%llu)", namebuf, origin->ds_object);
953 dsl_dataset_rele(ds, FTAG);
954 dsl_dataset_rele(origin, FTAG);
955 dsl_dir_rele(pdd, FTAG);
956 }
957
958 int
959 dmu_objset_clone(const char *clone, const char *origin)
960 {
961 dmu_objset_clone_arg_t doca;
962
963 doca.doca_clone = clone;
964 doca.doca_origin = origin;
965 doca.doca_cred = CRED();
966
967 return (dsl_sync_task(clone,
968 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
969 5, ZFS_SPACE_CHECK_NORMAL));
970 }
971
972 int
973 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
974 {
975 int err;
976 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
977 nvlist_t *snaps = fnvlist_alloc();
978
979 fnvlist_add_boolean(snaps, longsnap);
980 strfree(longsnap);
981 err = dsl_dataset_snapshot(snaps, NULL, NULL);
982 fnvlist_free(snaps);
983 return (err);
984 }
985
986 static void
987 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
988 {
989 dnode_t *dn;
990
991 while (dn = list_head(list)) {
992 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
993 ASSERT(dn->dn_dbuf->db_data_pending);
994 /*
995 * Initialize dn_zio outside dnode_sync() because the
996 * meta-dnode needs to set it ouside dnode_sync().
997 */
998 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
999 ASSERT(dn->dn_zio);
1000
1001 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1002 list_remove(list, dn);
1003
1004 if (newlist) {
1005 (void) dnode_add_ref(dn, newlist);
1006 list_insert_tail(newlist, dn);
1007 }
1008
1009 dnode_sync(dn, tx);
1010 }
1011 }
1012
1013 /* ARGSUSED */
1014 static void
1015 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1016 {
1017 blkptr_t *bp = zio->io_bp;
1018 objset_t *os = arg;
1019 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1020
1021 ASSERT(!BP_IS_EMBEDDED(bp));
1022 ASSERT3P(bp, ==, os->os_rootbp);
1023 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1024 ASSERT0(BP_GET_LEVEL(bp));
1025
1026 /*
1027 * Update rootbp fill count: it should be the number of objects
1028 * allocated in the object set (not counting the "special"
1029 * objects that are stored in the objset_phys_t -- the meta
1030 * dnode and user/group accounting objects).
1031 */
1032 bp->blk_fill = 0;
1033 for (int i = 0; i < dnp->dn_nblkptr; i++)
1034 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1035 }
1036
1037 /* ARGSUSED */
1038 static void
1039 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1040 {
1041 blkptr_t *bp = zio->io_bp;
1042 blkptr_t *bp_orig = &zio->io_bp_orig;
1043 objset_t *os = arg;
1044
1045 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1046 ASSERT(BP_EQUAL(bp, bp_orig));
1047 } else {
1048 dsl_dataset_t *ds = os->os_dsl_dataset;
1049 dmu_tx_t *tx = os->os_synctx;
1050
1051 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1052 dsl_dataset_block_born(ds, bp, tx);
1053 }
1054 }
1055
1056 /* called from dsl */
1057 void
1058 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1059 {
1060 int txgoff;
1061 zbookmark_phys_t zb;
1062 zio_prop_t zp;
1063 zio_t *zio;
1064 list_t *list;
1065 list_t *newlist = NULL;
1066 dbuf_dirty_record_t *dr;
1067
1068 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1069
1070 ASSERT(dmu_tx_is_syncing(tx));
1071 /* XXX the write_done callback should really give us the tx... */
1072 os->os_synctx = tx;
1073
1074 if (os->os_dsl_dataset == NULL) {
1075 /*
1076 * This is the MOS. If we have upgraded,
1077 * spa_max_replication() could change, so reset
1078 * os_copies here.
1079 */
1080 os->os_copies = spa_max_replication(os->os_spa);
1081 }
1082
1083 /*
1084 * Create the root block IO
1085 */
1086 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1087 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1088 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1089 arc_release(os->os_phys_buf, &os->os_phys_buf);
1090
1091 dmu_write_policy(os, NULL, 0, 0, &zp);
1092
1093 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1094 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1095 DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1096 NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1097 ZIO_FLAG_MUSTSUCCEED, &zb);
1098
1099 /*
1100 * Sync special dnodes - the parent IO for the sync is the root block
1101 */
1102 DMU_META_DNODE(os)->dn_zio = zio;
1103 dnode_sync(DMU_META_DNODE(os), tx);
1104
1105 os->os_phys->os_flags = os->os_flags;
1106
1107 if (DMU_USERUSED_DNODE(os) &&
1108 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1109 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1110 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1111 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1112 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1113 }
1114
1115 txgoff = tx->tx_txg & TXG_MASK;
1116
1117 if (dmu_objset_userused_enabled(os)) {
1118 newlist = &os->os_synced_dnodes;
1119 /*
1120 * We must create the list here because it uses the
1121 * dn_dirty_link[] of this txg.
1122 */
1123 list_create(newlist, sizeof (dnode_t),
1124 offsetof(dnode_t, dn_dirty_link[txgoff]));
1125 }
1126
1127 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1128 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1129
1130 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1131 while (dr = list_head(list)) {
1132 ASSERT0(dr->dr_dbuf->db_level);
1133 list_remove(list, dr);
1134 if (dr->dr_zio)
1135 zio_nowait(dr->dr_zio);
1136 }
1137 /*
1138 * Free intent log blocks up to this tx.
1139 */
1140 zil_sync(os->os_zil, tx);
1141 os->os_phys->os_zil_header = os->os_zil_header;
1142 zio_nowait(zio);
1143 }
1144
1145 boolean_t
1146 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1147 {
1148 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1149 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1150 }
1151
1152 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1153
1154 void
1155 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1156 {
1157 used_cbs[ost] = cb;
1158 }
1159
1160 boolean_t
1161 dmu_objset_userused_enabled(objset_t *os)
1162 {
1163 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1164 used_cbs[os->os_phys->os_type] != NULL &&
1165 DMU_USERUSED_DNODE(os) != NULL);
1166 }
1167
1168 static void
1169 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1170 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1171 {
1172 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1173 int64_t delta = DNODE_SIZE + used;
1174 if (subtract)
1175 delta = -delta;
1176 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1177 user, delta, tx));
1178 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1179 group, delta, tx));
1180 }
1181 }
1182
1183 void
1184 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1185 {
1186 dnode_t *dn;
1187 list_t *list = &os->os_synced_dnodes;
1188
1189 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1190
1191 while (dn = list_head(list)) {
1192 int flags;
1193 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1194 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1195 dn->dn_phys->dn_flags &
1196 DNODE_FLAG_USERUSED_ACCOUNTED);
1197
1198 /* Allocate the user/groupused objects if necessary. */
1199 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1200 VERIFY(0 == zap_create_claim(os,
1201 DMU_USERUSED_OBJECT,
1202 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1203 VERIFY(0 == zap_create_claim(os,
1204 DMU_GROUPUSED_OBJECT,
1205 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1206 }
1207
1208 /*
1209 * We intentionally modify the zap object even if the
1210 * net delta is zero. Otherwise
1211 * the block of the zap obj could be shared between
1212 * datasets but need to be different between them after
1213 * a bprewrite.
1214 */
1215
1216 flags = dn->dn_id_flags;
1217 ASSERT(flags);
1218 if (flags & DN_ID_OLD_EXIST) {
1219 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1220 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1221 }
1222 if (flags & DN_ID_NEW_EXIST) {
1223 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1224 dn->dn_phys->dn_flags, dn->dn_newuid,
1225 dn->dn_newgid, B_FALSE, tx);
1226 }
1227
1228 mutex_enter(&dn->dn_mtx);
1229 dn->dn_oldused = 0;
1230 dn->dn_oldflags = 0;
1231 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1232 dn->dn_olduid = dn->dn_newuid;
1233 dn->dn_oldgid = dn->dn_newgid;
1234 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1235 if (dn->dn_bonuslen == 0)
1236 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1237 else
1238 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1239 }
1240 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1241 mutex_exit(&dn->dn_mtx);
1242
1243 list_remove(list, dn);
1244 dnode_rele(dn, list);
1245 }
1246 }
1247
1248 /*
1249 * Returns a pointer to data to find uid/gid from
1250 *
1251 * If a dirty record for transaction group that is syncing can't
1252 * be found then NULL is returned. In the NULL case it is assumed
1253 * the uid/gid aren't changing.
1254 */
1255 static void *
1256 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1257 {
1258 dbuf_dirty_record_t *dr, **drp;
1259 void *data;
1260
1261 if (db->db_dirtycnt == 0)
1262 return (db->db.db_data); /* Nothing is changing */
1263
1264 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1265 if (dr->dr_txg == tx->tx_txg)
1266 break;
1267
1268 if (dr == NULL) {
1269 data = NULL;
1270 } else {
1271 dnode_t *dn;
1272
1273 DB_DNODE_ENTER(dr->dr_dbuf);
1274 dn = DB_DNODE(dr->dr_dbuf);
1275
1276 if (dn->dn_bonuslen == 0 &&
1277 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1278 data = dr->dt.dl.dr_data->b_data;
1279 else
1280 data = dr->dt.dl.dr_data;
1281
1282 DB_DNODE_EXIT(dr->dr_dbuf);
1283 }
1284
1285 return (data);
1286 }
1287
1288 void
1289 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1290 {
1291 objset_t *os = dn->dn_objset;
1292 void *data = NULL;
1293 dmu_buf_impl_t *db = NULL;
1294 uint64_t *user = NULL;
1295 uint64_t *group = NULL;
1296 int flags = dn->dn_id_flags;
1297 int error;
1298 boolean_t have_spill = B_FALSE;
1299
1300 if (!dmu_objset_userused_enabled(dn->dn_objset))
1301 return;
1302
1303 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1304 DN_ID_CHKED_SPILL)))
1305 return;
1306
1307 if (before && dn->dn_bonuslen != 0)
1308 data = DN_BONUS(dn->dn_phys);
1309 else if (!before && dn->dn_bonuslen != 0) {
1310 if (dn->dn_bonus) {
1311 db = dn->dn_bonus;
1312 mutex_enter(&db->db_mtx);
1313 data = dmu_objset_userquota_find_data(db, tx);
1314 } else {
1315 data = DN_BONUS(dn->dn_phys);
1316 }
1317 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1318 int rf = 0;
1319
1320 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1321 rf |= DB_RF_HAVESTRUCT;
1322 error = dmu_spill_hold_by_dnode(dn,
1323 rf | DB_RF_MUST_SUCCEED,
1324 FTAG, (dmu_buf_t **)&db);
1325 ASSERT(error == 0);
1326 mutex_enter(&db->db_mtx);
1327 data = (before) ? db->db.db_data :
1328 dmu_objset_userquota_find_data(db, tx);
1329 have_spill = B_TRUE;
1330 } else {
1331 mutex_enter(&dn->dn_mtx);
1332 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1333 mutex_exit(&dn->dn_mtx);
1334 return;
1335 }
1336
1337 if (before) {
1338 ASSERT(data);
1339 user = &dn->dn_olduid;
1340 group = &dn->dn_oldgid;
1341 } else if (data) {
1342 user = &dn->dn_newuid;
1343 group = &dn->dn_newgid;
1344 }
1345
1346 /*
1347 * Must always call the callback in case the object
1348 * type has changed and that type isn't an object type to track
1349 */
1350 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1351 user, group);
1352
1353 /*
1354 * Preserve existing uid/gid when the callback can't determine
1355 * what the new uid/gid are and the callback returned EEXIST.
1356 * The EEXIST error tells us to just use the existing uid/gid.
1357 * If we don't know what the old values are then just assign
1358 * them to 0, since that is a new file being created.
1359 */
1360 if (!before && data == NULL && error == EEXIST) {
1361 if (flags & DN_ID_OLD_EXIST) {
1362 dn->dn_newuid = dn->dn_olduid;
1363 dn->dn_newgid = dn->dn_oldgid;
1364 } else {
1365 dn->dn_newuid = 0;
1366 dn->dn_newgid = 0;
1367 }
1368 error = 0;
1369 }
1370
1371 if (db)
1372 mutex_exit(&db->db_mtx);
1373
1374 mutex_enter(&dn->dn_mtx);
1375 if (error == 0 && before)
1376 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1377 if (error == 0 && !before)
1378 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1379
1380 if (have_spill) {
1381 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1382 } else {
1383 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1384 }
1385 mutex_exit(&dn->dn_mtx);
1386 if (have_spill)
1387 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1388 }
1389
1390 boolean_t
1391 dmu_objset_userspace_present(objset_t *os)
1392 {
1393 return (os->os_phys->os_flags &
1394 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1395 }
1396
1397 int
1398 dmu_objset_userspace_upgrade(objset_t *os)
1399 {
1400 uint64_t obj;
1401 int err = 0;
1402
1403 if (dmu_objset_userspace_present(os))
1404 return (0);
1405 if (!dmu_objset_userused_enabled(os))
1406 return (SET_ERROR(ENOTSUP));
1407 if (dmu_objset_is_snapshot(os))
1408 return (SET_ERROR(EINVAL));
1409
1410 /*
1411 * We simply need to mark every object dirty, so that it will be
1412 * synced out and now accounted. If this is called
1413 * concurrently, or if we already did some work before crashing,
1414 * that's fine, since we track each object's accounted state
1415 * independently.
1416 */
1417
1418 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1419 dmu_tx_t *tx;
1420 dmu_buf_t *db;
1421 int objerr;
1422
1423 if (issig(JUSTLOOKING) && issig(FORREAL))
1424 return (SET_ERROR(EINTR));
1425
1426 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1427 if (objerr != 0)
1428 continue;
1429 tx = dmu_tx_create(os);
1430 dmu_tx_hold_bonus(tx, obj);
1431 objerr = dmu_tx_assign(tx, TXG_WAIT);
1432 if (objerr != 0) {
1433 dmu_tx_abort(tx);
1434 continue;
1435 }
1436 dmu_buf_will_dirty(db, tx);
1437 dmu_buf_rele(db, FTAG);
1438 dmu_tx_commit(tx);
1439 }
1440
1441 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1442 txg_wait_synced(dmu_objset_pool(os), 0);
1443 return (0);
1444 }
1445
1446 void
1447 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1448 uint64_t *usedobjsp, uint64_t *availobjsp)
1449 {
1450 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1451 usedobjsp, availobjsp);
1452 }
1453
1454 uint64_t
1455 dmu_objset_fsid_guid(objset_t *os)
1456 {
1457 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1458 }
1459
1460 void
1461 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1462 {
1463 stat->dds_type = os->os_phys->os_type;
1464 if (os->os_dsl_dataset)
1465 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1466 }
1467
1468 void
1469 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1470 {
1471 ASSERT(os->os_dsl_dataset ||
1472 os->os_phys->os_type == DMU_OST_META);
1473
1474 if (os->os_dsl_dataset != NULL)
1475 dsl_dataset_stats(os->os_dsl_dataset, nv);
1476
1477 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1478 os->os_phys->os_type);
1479 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1480 dmu_objset_userspace_present(os));
1481 }
1482
1483 int
1484 dmu_objset_is_snapshot(objset_t *os)
1485 {
1486 if (os->os_dsl_dataset != NULL)
1487 return (os->os_dsl_dataset->ds_is_snapshot);
1488 else
1489 return (B_FALSE);
1490 }
1491
1492 int
1493 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1494 boolean_t *conflict)
1495 {
1496 dsl_dataset_t *ds = os->os_dsl_dataset;
1497 uint64_t ignored;
1498
1499 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1500 return (SET_ERROR(ENOENT));
1501
1502 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1503 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1504 MT_FIRST, real, maxlen, conflict));
1505 }
1506
1507 int
1508 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1509 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1510 {
1511 dsl_dataset_t *ds = os->os_dsl_dataset;
1512 zap_cursor_t cursor;
1513 zap_attribute_t attr;
1514
1515 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1516
1517 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1518 return (SET_ERROR(ENOENT));
1519
1520 zap_cursor_init_serialized(&cursor,
1521 ds->ds_dir->dd_pool->dp_meta_objset,
1522 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1523
1524 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1525 zap_cursor_fini(&cursor);
1526 return (SET_ERROR(ENOENT));
1527 }
1528
1529 if (strlen(attr.za_name) + 1 > namelen) {
1530 zap_cursor_fini(&cursor);
1531 return (SET_ERROR(ENAMETOOLONG));
1532 }
1533
1534 (void) strcpy(name, attr.za_name);
1535 if (idp)
1536 *idp = attr.za_first_integer;
1537 if (case_conflict)
1538 *case_conflict = attr.za_normalization_conflict;
1539 zap_cursor_advance(&cursor);
1540 *offp = zap_cursor_serialize(&cursor);
1541 zap_cursor_fini(&cursor);
1542
1543 return (0);
1544 }
1545
1546 int
1547 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1548 uint64_t *idp, uint64_t *offp)
1549 {
1550 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1551 zap_cursor_t cursor;
1552 zap_attribute_t attr;
1553
1554 /* there is no next dir on a snapshot! */
1555 if (os->os_dsl_dataset->ds_object !=
1556 dsl_dir_phys(dd)->dd_head_dataset_obj)
1557 return (SET_ERROR(ENOENT));
1558
1559 zap_cursor_init_serialized(&cursor,
1560 dd->dd_pool->dp_meta_objset,
1561 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1562
1563 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1564 zap_cursor_fini(&cursor);
1565 return (SET_ERROR(ENOENT));
1566 }
1567
1568 if (strlen(attr.za_name) + 1 > namelen) {
1569 zap_cursor_fini(&cursor);
1570 return (SET_ERROR(ENAMETOOLONG));
1571 }
1572
1573 (void) strcpy(name, attr.za_name);
1574 if (idp)
1575 *idp = attr.za_first_integer;
1576 zap_cursor_advance(&cursor);
1577 *offp = zap_cursor_serialize(&cursor);
1578 zap_cursor_fini(&cursor);
1579
1580 return (0);
1581 }
1582
1583 /*
1584 * Find objsets under and including ddobj, call func(ds) on each.
1585 */
1586 int
1587 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1588 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1589 {
1590 dsl_dir_t *dd;
1591 dsl_dataset_t *ds;
1592 zap_cursor_t zc;
1593 zap_attribute_t *attr;
1594 uint64_t thisobj;
1595 int err;
1596
1597 ASSERT(dsl_pool_config_held(dp));
1598
1599 err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1600 if (err != 0)
1601 return (err);
1602
1603 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1604 if (dd->dd_myname[0] == '$') {
1605 dsl_dir_rele(dd, FTAG);
1606 return (0);
1607 }
1608
1609 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1610 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1611
1612 /*
1613 * Iterate over all children.
1614 */
1615 if (flags & DS_FIND_CHILDREN) {
1616 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1617 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1618 zap_cursor_retrieve(&zc, attr) == 0;
1619 (void) zap_cursor_advance(&zc)) {
1620 ASSERT3U(attr->za_integer_length, ==,
1621 sizeof (uint64_t));
1622 ASSERT3U(attr->za_num_integers, ==, 1);
1623
1624 err = dmu_objset_find_dp(dp, attr->za_first_integer,
1625 func, arg, flags);
1626 if (err != 0)
1627 break;
1628 }
1629 zap_cursor_fini(&zc);
1630
1631 if (err != 0) {
1632 dsl_dir_rele(dd, FTAG);
1633 kmem_free(attr, sizeof (zap_attribute_t));
1634 return (err);
1635 }
1636 }
1637
1638 /*
1639 * Iterate over all snapshots.
1640 */
1641 if (flags & DS_FIND_SNAPSHOTS) {
1642 dsl_dataset_t *ds;
1643 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1644
1645 if (err == 0) {
1646 uint64_t snapobj;
1647
1648 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1649 dsl_dataset_rele(ds, FTAG);
1650
1651 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1652 zap_cursor_retrieve(&zc, attr) == 0;
1653 (void) zap_cursor_advance(&zc)) {
1654 ASSERT3U(attr->za_integer_length, ==,
1655 sizeof (uint64_t));
1656 ASSERT3U(attr->za_num_integers, ==, 1);
1657
1658 err = dsl_dataset_hold_obj(dp,
1659 attr->za_first_integer, FTAG, &ds);
1660 if (err != 0)
1661 break;
1662 err = func(dp, ds, arg);
1663 dsl_dataset_rele(ds, FTAG);
1664 if (err != 0)
1665 break;
1666 }
1667 zap_cursor_fini(&zc);
1668 }
1669 }
1670
1671 dsl_dir_rele(dd, FTAG);
1672 kmem_free(attr, sizeof (zap_attribute_t));
1673
1674 if (err != 0)
1675 return (err);
1676
1677 /*
1678 * Apply to self.
1679 */
1680 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1681 if (err != 0)
1682 return (err);
1683 err = func(dp, ds, arg);
1684 dsl_dataset_rele(ds, FTAG);
1685 return (err);
1686 }
1687
1688 /*
1689 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1690 * The dp_config_rwlock must not be held when this is called, and it
1691 * will not be held when the callback is called.
1692 * Therefore this function should only be used when the pool is not changing
1693 * (e.g. in syncing context), or the callback can deal with the possible races.
1694 */
1695 static int
1696 dmu_objset_find_impl(spa_t *spa, const char *name,
1697 int func(const char *, void *), void *arg, int flags)
1698 {
1699 dsl_dir_t *dd;
1700 dsl_pool_t *dp = spa_get_dsl(spa);
1701 dsl_dataset_t *ds;
1702 zap_cursor_t zc;
1703 zap_attribute_t *attr;
1704 char *child;
1705 uint64_t thisobj;
1706 int err;
1707
1708 dsl_pool_config_enter(dp, FTAG);
1709
1710 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1711 if (err != 0) {
1712 dsl_pool_config_exit(dp, FTAG);
1713 return (err);
1714 }
1715
1716 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1717 if (dd->dd_myname[0] == '$') {
1718 dsl_dir_rele(dd, FTAG);
1719 dsl_pool_config_exit(dp, FTAG);
1720 return (0);
1721 }
1722
1723 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1724 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1725
1726 /*
1727 * Iterate over all children.
1728 */
1729 if (flags & DS_FIND_CHILDREN) {
1730 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1731 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1732 zap_cursor_retrieve(&zc, attr) == 0;
1733 (void) zap_cursor_advance(&zc)) {
1734 ASSERT3U(attr->za_integer_length, ==,
1735 sizeof (uint64_t));
1736 ASSERT3U(attr->za_num_integers, ==, 1);
1737
1738 child = kmem_asprintf("%s/%s", name, attr->za_name);
1739 dsl_pool_config_exit(dp, FTAG);
1740 err = dmu_objset_find_impl(spa, child,
1741 func, arg, flags);
1742 dsl_pool_config_enter(dp, FTAG);
1743 strfree(child);
1744 if (err != 0)
1745 break;
1746 }
1747 zap_cursor_fini(&zc);
1748
1749 if (err != 0) {
1750 dsl_dir_rele(dd, FTAG);
1751 dsl_pool_config_exit(dp, FTAG);
1752 kmem_free(attr, sizeof (zap_attribute_t));
1753 return (err);
1754 }
1755 }
1756
1757 /*
1758 * Iterate over all snapshots.
1759 */
1760 if (flags & DS_FIND_SNAPSHOTS) {
1761 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1762
1763 if (err == 0) {
1764 uint64_t snapobj;
1765
1766 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1767 dsl_dataset_rele(ds, FTAG);
1768
1769 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1770 zap_cursor_retrieve(&zc, attr) == 0;
1771 (void) zap_cursor_advance(&zc)) {
1772 ASSERT3U(attr->za_integer_length, ==,
1773 sizeof (uint64_t));
1774 ASSERT3U(attr->za_num_integers, ==, 1);
1775
1776 child = kmem_asprintf("%s@%s",
1777 name, attr->za_name);
1778 dsl_pool_config_exit(dp, FTAG);
1779 err = func(child, arg);
1780 dsl_pool_config_enter(dp, FTAG);
1781 strfree(child);
1782 if (err != 0)
1783 break;
1784 }
1785 zap_cursor_fini(&zc);
1786 }
1787 }
1788
1789 dsl_dir_rele(dd, FTAG);
1790 kmem_free(attr, sizeof (zap_attribute_t));
1791 dsl_pool_config_exit(dp, FTAG);
1792
1793 if (err != 0)
1794 return (err);
1795
1796 /* Apply to self. */
1797 return (func(name, arg));
1798 }
1799
1800 /*
1801 * See comment above dmu_objset_find_impl().
1802 */
1803 int
1804 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1805 int flags)
1806 {
1807 spa_t *spa;
1808 int error;
1809
1810 error = spa_open(name, &spa, FTAG);
1811 if (error != 0)
1812 return (error);
1813 error = dmu_objset_find_impl(spa, name, func, arg, flags);
1814 spa_close(spa, FTAG);
1815 return (error);
1816 }
1817
1818 void
1819 dmu_objset_set_user(objset_t *os, void *user_ptr)
1820 {
1821 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1822 os->os_user_ptr = user_ptr;
1823 }
1824
1825 void *
1826 dmu_objset_get_user(objset_t *os)
1827 {
1828 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1829 return (os->os_user_ptr);
1830 }
1831
1832 /*
1833 * Determine name of filesystem, given name of snapshot.
1834 * buf must be at least MAXNAMELEN bytes
1835 */
1836 int
1837 dmu_fsname(const char *snapname, char *buf)
1838 {
1839 char *atp = strchr(snapname, '@');
1840 if (atp == NULL)
1841 return (SET_ERROR(EINVAL));
1842 if (atp - snapname >= MAXNAMELEN)
1843 return (SET_ERROR(ENAMETOOLONG));
1844 (void) strlcpy(buf, snapname, atp - snapname + 1);
1845 return (0);
1846 }