1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 */
28
29 /* Portions Copyright 2010 Robert Milkowski */
30
31 #include <sys/cred.h>
32 #include <sys/zfs_context.h>
33 #include <sys/dmu_objset.h>
34 #include <sys/dsl_dir.h>
35 #include <sys/dsl_dataset.h>
36 #include <sys/dsl_prop.h>
37 #include <sys/dsl_pool.h>
38 #include <sys/dsl_synctask.h>
39 #include <sys/dsl_deleg.h>
40 #include <sys/dnode.h>
41 #include <sys/dbuf.h>
42 #include <sys/zvol.h>
43 #include <sys/dmu_tx.h>
44 #include <sys/zap.h>
45 #include <sys/zil.h>
46 #include <sys/dmu_impl.h>
47 #include <sys/zfs_ioctl.h>
48 #include <sys/sa.h>
49 #include <sys/zfs_onexit.h>
50 #include <sys/dsl_destroy.h>
51
52 /*
53 * Needed to close a window in dnode_move() that allows the objset to be freed
54 * before it can be safely accessed.
55 */
56 krwlock_t os_lock;
57
58 void
59 dmu_objset_init(void)
60 {
61 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
62 }
63
64 void
65 dmu_objset_fini(void)
66 {
67 rw_destroy(&os_lock);
68 }
69
70 spa_t *
71 dmu_objset_spa(objset_t *os)
72 {
73 return (os->os_spa);
74 }
75
76 zilog_t *
77 dmu_objset_zil(objset_t *os)
78 {
79 return (os->os_zil);
80 }
81
82 dsl_pool_t *
83 dmu_objset_pool(objset_t *os)
84 {
85 dsl_dataset_t *ds;
86
87 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
88 return (ds->ds_dir->dd_pool);
89 else
90 return (spa_get_dsl(os->os_spa));
91 }
92
93 dsl_dataset_t *
94 dmu_objset_ds(objset_t *os)
95 {
96 return (os->os_dsl_dataset);
97 }
98
99 dmu_objset_type_t
100 dmu_objset_type(objset_t *os)
101 {
102 return (os->os_phys->os_type);
103 }
104
105 void
106 dmu_objset_name(objset_t *os, char *buf)
107 {
108 dsl_dataset_name(os->os_dsl_dataset, buf);
109 }
110
111 uint64_t
112 dmu_objset_id(objset_t *os)
113 {
114 dsl_dataset_t *ds = os->os_dsl_dataset;
115
116 return (ds ? ds->ds_object : 0);
117 }
118
119 zfs_sync_type_t
120 dmu_objset_syncprop(objset_t *os)
121 {
122 return (os->os_sync);
123 }
124
125 zfs_logbias_op_t
126 dmu_objset_logbias(objset_t *os)
127 {
128 return (os->os_logbias);
129 }
130
131 static void
132 checksum_changed_cb(void *arg, uint64_t newval)
133 {
134 objset_t *os = arg;
135
136 /*
137 * Inheritance should have been done by now.
138 */
139 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
140
141 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
142 }
143
144 static void
145 compression_changed_cb(void *arg, uint64_t newval)
146 {
147 objset_t *os = arg;
148
149 /*
150 * Inheritance and range checking should have been done by now.
151 */
152 ASSERT(newval != ZIO_COMPRESS_INHERIT);
153
154 os->os_compress = zio_compress_select(os->os_spa, newval,
155 ZIO_COMPRESS_ON);
156 }
157
158 static void
159 copies_changed_cb(void *arg, uint64_t newval)
160 {
161 objset_t *os = arg;
162
163 /*
164 * Inheritance and range checking should have been done by now.
165 */
166 ASSERT(newval > 0);
167 ASSERT(newval <= spa_max_replication(os->os_spa));
168
169 os->os_copies = newval;
170 }
171
172 static void
173 dedup_changed_cb(void *arg, uint64_t newval)
174 {
175 objset_t *os = arg;
176 spa_t *spa = os->os_spa;
177 enum zio_checksum checksum;
178
179 /*
180 * Inheritance should have been done by now.
181 */
182 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
183
184 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
185
186 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
187 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
188 }
189
190 static void
191 primary_cache_changed_cb(void *arg, uint64_t newval)
192 {
193 objset_t *os = arg;
194
195 /*
196 * Inheritance and range checking should have been done by now.
197 */
198 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
199 newval == ZFS_CACHE_METADATA);
200
201 os->os_primary_cache = newval;
202 }
203
204 static void
205 secondary_cache_changed_cb(void *arg, uint64_t newval)
206 {
207 objset_t *os = arg;
208
209 /*
210 * Inheritance and range checking should have been done by now.
211 */
212 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
213 newval == ZFS_CACHE_METADATA);
214
215 os->os_secondary_cache = newval;
216 }
217
218 static void
219 sync_changed_cb(void *arg, uint64_t newval)
220 {
221 objset_t *os = arg;
222
223 /*
224 * Inheritance and range checking should have been done by now.
225 */
226 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
227 newval == ZFS_SYNC_DISABLED);
228
229 os->os_sync = newval;
230 if (os->os_zil)
231 zil_set_sync(os->os_zil, newval);
232 }
233
234 static void
235 redundant_metadata_changed_cb(void *arg, uint64_t newval)
236 {
237 objset_t *os = arg;
238
239 /*
240 * Inheritance and range checking should have been done by now.
241 */
242 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
243 newval == ZFS_REDUNDANT_METADATA_MOST);
244
245 os->os_redundant_metadata = newval;
246 }
247
248 static void
249 logbias_changed_cb(void *arg, uint64_t newval)
250 {
251 objset_t *os = arg;
252
253 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
254 newval == ZFS_LOGBIAS_THROUGHPUT);
255 os->os_logbias = newval;
256 if (os->os_zil)
257 zil_set_logbias(os->os_zil, newval);
258 }
259
260 static void
261 recordsize_changed_cb(void *arg, uint64_t newval)
262 {
263 objset_t *os = arg;
264
265 os->os_recordsize = newval;
266 }
267
268 void
269 dmu_objset_byteswap(void *buf, size_t size)
270 {
271 objset_phys_t *osp = buf;
272
273 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
274 dnode_byteswap(&osp->os_meta_dnode);
275 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
276 osp->os_type = BSWAP_64(osp->os_type);
277 osp->os_flags = BSWAP_64(osp->os_flags);
278 if (size == sizeof (objset_phys_t)) {
279 dnode_byteswap(&osp->os_userused_dnode);
280 dnode_byteswap(&osp->os_groupused_dnode);
281 }
282 }
283
284 int
285 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
286 objset_t **osp)
287 {
288 objset_t *os;
289 int i, err;
290
291 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
292
293 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
294 os->os_dsl_dataset = ds;
295 os->os_spa = spa;
296 os->os_rootbp = bp;
297 if (!BP_IS_HOLE(os->os_rootbp)) {
298 arc_flags_t aflags = ARC_FLAG_WAIT;
299 zbookmark_phys_t zb;
300 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
301 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
302
303 if (DMU_OS_IS_L2CACHEABLE(os))
304 aflags |= ARC_FLAG_L2CACHE;
305 if (DMU_OS_IS_L2COMPRESSIBLE(os))
306 aflags |= ARC_FLAG_L2COMPRESS;
307
308 dprintf_bp(os->os_rootbp, "reading %s", "");
309 err = arc_read(NULL, spa, os->os_rootbp,
310 arc_getbuf_func, &os->os_phys_buf,
311 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
312 if (err != 0) {
313 kmem_free(os, sizeof (objset_t));
314 /* convert checksum errors into IO errors */
315 if (err == ECKSUM)
316 err = SET_ERROR(EIO);
317 return (err);
318 }
319
320 /* Increase the blocksize if we are permitted. */
321 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
322 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
323 arc_buf_t *buf = arc_buf_alloc(spa,
324 sizeof (objset_phys_t), &os->os_phys_buf,
325 ARC_BUFC_METADATA);
326 bzero(buf->b_data, sizeof (objset_phys_t));
327 bcopy(os->os_phys_buf->b_data, buf->b_data,
328 arc_buf_size(os->os_phys_buf));
329 (void) arc_buf_remove_ref(os->os_phys_buf,
330 &os->os_phys_buf);
331 os->os_phys_buf = buf;
332 }
333
334 os->os_phys = os->os_phys_buf->b_data;
335 os->os_flags = os->os_phys->os_flags;
336 } else {
337 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
338 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
339 os->os_phys_buf = arc_buf_alloc(spa, size,
340 &os->os_phys_buf, ARC_BUFC_METADATA);
341 os->os_phys = os->os_phys_buf->b_data;
342 bzero(os->os_phys, size);
343 }
344
345 /*
346 * Note: the changed_cb will be called once before the register
347 * func returns, thus changing the checksum/compression from the
348 * default (fletcher2/off). Snapshots don't need to know about
349 * checksum/compression/copies.
350 */
351 if (ds != NULL) {
352 err = dsl_prop_register(ds,
353 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
354 primary_cache_changed_cb, os);
355 if (err == 0) {
356 err = dsl_prop_register(ds,
357 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
358 secondary_cache_changed_cb, os);
359 }
360 if (!ds->ds_is_snapshot) {
361 if (err == 0) {
362 err = dsl_prop_register(ds,
363 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
364 checksum_changed_cb, os);
365 }
366 if (err == 0) {
367 err = dsl_prop_register(ds,
368 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
369 compression_changed_cb, os);
370 }
371 if (err == 0) {
372 err = dsl_prop_register(ds,
373 zfs_prop_to_name(ZFS_PROP_COPIES),
374 copies_changed_cb, os);
375 }
376 if (err == 0) {
377 err = dsl_prop_register(ds,
378 zfs_prop_to_name(ZFS_PROP_DEDUP),
379 dedup_changed_cb, os);
380 }
381 if (err == 0) {
382 err = dsl_prop_register(ds,
383 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
384 logbias_changed_cb, os);
385 }
386 if (err == 0) {
387 err = dsl_prop_register(ds,
388 zfs_prop_to_name(ZFS_PROP_SYNC),
389 sync_changed_cb, os);
390 }
391 if (err == 0) {
392 err = dsl_prop_register(ds,
393 zfs_prop_to_name(
394 ZFS_PROP_REDUNDANT_METADATA),
395 redundant_metadata_changed_cb, os);
396 }
397 if (err == 0) {
398 err = dsl_prop_register(ds,
399 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
400 recordsize_changed_cb, os);
401 }
402 }
403 if (err != 0) {
404 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
405 &os->os_phys_buf));
406 kmem_free(os, sizeof (objset_t));
407 return (err);
408 }
409 } else {
410 /* It's the meta-objset. */
411 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
412 os->os_compress = ZIO_COMPRESS_ON;
413 os->os_copies = spa_max_replication(spa);
414 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
415 os->os_dedup_verify = B_FALSE;
416 os->os_logbias = ZFS_LOGBIAS_LATENCY;
417 os->os_sync = ZFS_SYNC_STANDARD;
418 os->os_primary_cache = ZFS_CACHE_ALL;
419 os->os_secondary_cache = ZFS_CACHE_ALL;
420 }
421
422 if (ds == NULL || !ds->ds_is_snapshot)
423 os->os_zil_header = os->os_phys->os_zil_header;
424 os->os_zil = zil_alloc(os, &os->os_zil_header);
425
426 for (i = 0; i < TXG_SIZE; i++) {
427 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
428 offsetof(dnode_t, dn_dirty_link[i]));
429 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
430 offsetof(dnode_t, dn_dirty_link[i]));
431 }
432 list_create(&os->os_dnodes, sizeof (dnode_t),
433 offsetof(dnode_t, dn_link));
434 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
435 offsetof(dmu_buf_impl_t, db_link));
436
437 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
438 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
439 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
440
441 dnode_special_open(os, &os->os_phys->os_meta_dnode,
442 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
443 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
444 dnode_special_open(os, &os->os_phys->os_userused_dnode,
445 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
446 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
447 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
448 }
449
450 *osp = os;
451 return (0);
452 }
453
454 int
455 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
456 {
457 int err = 0;
458
459 mutex_enter(&ds->ds_opening_lock);
460 if (ds->ds_objset == NULL) {
461 objset_t *os;
462 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
463 ds, dsl_dataset_get_blkptr(ds), &os);
464
465 if (err == 0) {
466 mutex_enter(&ds->ds_lock);
467 ASSERT(ds->ds_objset == NULL);
468 ds->ds_objset = os;
469 mutex_exit(&ds->ds_lock);
470 }
471 }
472 *osp = ds->ds_objset;
473 mutex_exit(&ds->ds_opening_lock);
474 return (err);
475 }
476
477 /*
478 * Holds the pool while the objset is held. Therefore only one objset
479 * can be held at a time.
480 */
481 int
482 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
483 {
484 dsl_pool_t *dp;
485 dsl_dataset_t *ds;
486 int err;
487
488 err = dsl_pool_hold(name, tag, &dp);
489 if (err != 0)
490 return (err);
491 err = dsl_dataset_hold(dp, name, tag, &ds);
492 if (err != 0) {
493 dsl_pool_rele(dp, tag);
494 return (err);
495 }
496
497 err = dmu_objset_from_ds(ds, osp);
498 if (err != 0) {
499 dsl_dataset_rele(ds, tag);
500 dsl_pool_rele(dp, tag);
501 }
502
503 return (err);
504 }
505
506 /*
507 * dsl_pool must not be held when this is called.
508 * Upon successful return, there will be a longhold on the dataset,
509 * and the dsl_pool will not be held.
510 */
511 int
512 dmu_objset_own(const char *name, dmu_objset_type_t type,
513 boolean_t readonly, void *tag, objset_t **osp)
514 {
515 dsl_pool_t *dp;
516 dsl_dataset_t *ds;
517 int err;
518
519 err = dsl_pool_hold(name, FTAG, &dp);
520 if (err != 0)
521 return (err);
522 err = dsl_dataset_own(dp, name, tag, &ds);
523 if (err != 0) {
524 dsl_pool_rele(dp, FTAG);
525 return (err);
526 }
527
528 err = dmu_objset_from_ds(ds, osp);
529 dsl_pool_rele(dp, FTAG);
530 if (err != 0) {
531 dsl_dataset_disown(ds, tag);
532 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
533 dsl_dataset_disown(ds, tag);
534 return (SET_ERROR(EINVAL));
535 } else if (!readonly && ds->ds_is_snapshot) {
536 dsl_dataset_disown(ds, tag);
537 return (SET_ERROR(EROFS));
538 }
539 return (err);
540 }
541
542 void
543 dmu_objset_rele(objset_t *os, void *tag)
544 {
545 dsl_pool_t *dp = dmu_objset_pool(os);
546 dsl_dataset_rele(os->os_dsl_dataset, tag);
547 dsl_pool_rele(dp, tag);
548 }
549
550 /*
551 * When we are called, os MUST refer to an objset associated with a dataset
552 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
553 * == tag. We will then release and reacquire ownership of the dataset while
554 * holding the pool config_rwlock to avoid intervening namespace or ownership
555 * changes may occur.
556 *
557 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
558 * release the hold on its dataset and acquire a new one on the dataset of the
559 * same name so that it can be partially torn down and reconstructed.
560 */
561 void
562 dmu_objset_refresh_ownership(objset_t *os, void *tag)
563 {
564 dsl_pool_t *dp;
565 dsl_dataset_t *ds, *newds;
566 char name[MAXNAMELEN];
567
568 ds = os->os_dsl_dataset;
569 VERIFY3P(ds, !=, NULL);
570 VERIFY3P(ds->ds_owner, ==, tag);
571 VERIFY(dsl_dataset_long_held(ds));
572
573 dsl_dataset_name(ds, name);
574 dp = dmu_objset_pool(os);
575 dsl_pool_config_enter(dp, FTAG);
576 dmu_objset_disown(os, tag);
577 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
578 VERIFY3P(newds, ==, os->os_dsl_dataset);
579 dsl_pool_config_exit(dp, FTAG);
580 }
581
582 void
583 dmu_objset_disown(objset_t *os, void *tag)
584 {
585 dsl_dataset_disown(os->os_dsl_dataset, tag);
586 }
587
588 void
589 dmu_objset_evict_dbufs(objset_t *os)
590 {
591 dnode_t dn_marker;
592 dnode_t *dn;
593
594 mutex_enter(&os->os_lock);
595 dn = list_head(&os->os_dnodes);
596 while (dn != NULL) {
597 /*
598 * Skip dnodes without holds. We have to do this dance
599 * because dnode_add_ref() only works if there is already a
600 * hold. If the dnode has no holds, then it has no dbufs.
601 */
602 if (dnode_add_ref(dn, FTAG)) {
603 list_insert_after(&os->os_dnodes, dn, &dn_marker);
604 mutex_exit(&os->os_lock);
605
606 dnode_evict_dbufs(dn);
607 dnode_rele(dn, FTAG);
608
609 mutex_enter(&os->os_lock);
610 dn = list_next(&os->os_dnodes, &dn_marker);
611 list_remove(&os->os_dnodes, &dn_marker);
612 } else {
613 dn = list_next(&os->os_dnodes, dn);
614 }
615 }
616 mutex_exit(&os->os_lock);
617
618 if (DMU_USERUSED_DNODE(os) != NULL) {
619 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
620 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
621 }
622 dnode_evict_dbufs(DMU_META_DNODE(os));
623 }
624
625 /*
626 * Objset eviction processing is split into into two pieces.
627 * The first marks the objset as evicting, evicts any dbufs that
628 * have a refcount of zero, and then queues up the objset for the
629 * second phase of eviction. Once os->os_dnodes has been cleared by
630 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
631 * The second phase closes the special dnodes, dequeues the objset from
632 * the list of those undergoing eviction, and finally frees the objset.
633 *
634 * NOTE: Due to asynchronous eviction processing (invocation of
635 * dnode_buf_pageout()), it is possible for the meta dnode for the
636 * objset to have no holds even though os->os_dnodes is not empty.
637 */
638 void
639 dmu_objset_evict(objset_t *os)
640 {
641 dsl_dataset_t *ds = os->os_dsl_dataset;
642
643 for (int t = 0; t < TXG_SIZE; t++)
644 ASSERT(!dmu_objset_is_dirty(os, t));
645
646 if (ds) {
647 if (!ds->ds_is_snapshot) {
648 VERIFY0(dsl_prop_unregister(ds,
649 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
650 checksum_changed_cb, os));
651 VERIFY0(dsl_prop_unregister(ds,
652 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
653 compression_changed_cb, os));
654 VERIFY0(dsl_prop_unregister(ds,
655 zfs_prop_to_name(ZFS_PROP_COPIES),
656 copies_changed_cb, os));
657 VERIFY0(dsl_prop_unregister(ds,
658 zfs_prop_to_name(ZFS_PROP_DEDUP),
659 dedup_changed_cb, os));
660 VERIFY0(dsl_prop_unregister(ds,
661 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
662 logbias_changed_cb, os));
663 VERIFY0(dsl_prop_unregister(ds,
664 zfs_prop_to_name(ZFS_PROP_SYNC),
665 sync_changed_cb, os));
666 VERIFY0(dsl_prop_unregister(ds,
667 zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA),
668 redundant_metadata_changed_cb, os));
669 VERIFY0(dsl_prop_unregister(ds,
670 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
671 recordsize_changed_cb, os));
672 }
673 VERIFY0(dsl_prop_unregister(ds,
674 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
675 primary_cache_changed_cb, os));
676 VERIFY0(dsl_prop_unregister(ds,
677 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
678 secondary_cache_changed_cb, os));
679 }
680
681 if (os->os_sa)
682 sa_tear_down(os);
683
684 os->os_evicting = B_TRUE;
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 /* You can't clone across pools. */
909 if (pdd->dd_pool != dp) {
910 dsl_dir_rele(pdd, FTAG);
911 return (SET_ERROR(EXDEV));
912 }
913 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
914 doca->doca_cred);
915 if (error != 0) {
916 dsl_dir_rele(pdd, FTAG);
917 return (SET_ERROR(EDQUOT));
918 }
919 dsl_dir_rele(pdd, FTAG);
920
921 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
922 if (error != 0)
923 return (error);
924
925 /* You can't clone across pools. */
926 if (origin->ds_dir->dd_pool != dp) {
927 dsl_dataset_rele(origin, FTAG);
928 return (SET_ERROR(EXDEV));
929 }
930
931 /* You can only clone snapshots, not the head datasets. */
932 if (!origin->ds_is_snapshot) {
933 dsl_dataset_rele(origin, FTAG);
934 return (SET_ERROR(EINVAL));
935 }
936 dsl_dataset_rele(origin, FTAG);
937
938 return (0);
939 }
940
941 static void
942 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
943 {
944 dmu_objset_clone_arg_t *doca = arg;
945 dsl_pool_t *dp = dmu_tx_pool(tx);
946 dsl_dir_t *pdd;
947 const char *tail;
948 dsl_dataset_t *origin, *ds;
949 uint64_t obj;
950 char namebuf[MAXNAMELEN];
951
952 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
953 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
954
955 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
956 doca->doca_cred, tx);
957
958 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
959 dsl_dataset_name(origin, namebuf);
960 spa_history_log_internal_ds(ds, "clone", tx,
961 "origin=%s (%llu)", namebuf, origin->ds_object);
962 dsl_dataset_rele(ds, FTAG);
963 dsl_dataset_rele(origin, FTAG);
964 dsl_dir_rele(pdd, FTAG);
965 }
966
967 int
968 dmu_objset_clone(const char *clone, const char *origin)
969 {
970 dmu_objset_clone_arg_t doca;
971
972 doca.doca_clone = clone;
973 doca.doca_origin = origin;
974 doca.doca_cred = CRED();
975
976 return (dsl_sync_task(clone,
977 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
978 5, ZFS_SPACE_CHECK_NORMAL));
979 }
980
981 int
982 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
983 {
984 int err;
985 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
986 nvlist_t *snaps = fnvlist_alloc();
987
988 fnvlist_add_boolean(snaps, longsnap);
989 strfree(longsnap);
990 err = dsl_dataset_snapshot(snaps, NULL, NULL);
991 fnvlist_free(snaps);
992 return (err);
993 }
994
995 static void
996 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
997 {
998 dnode_t *dn;
999
1000 while (dn = list_head(list)) {
1001 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1002 ASSERT(dn->dn_dbuf->db_data_pending);
1003 /*
1004 * Initialize dn_zio outside dnode_sync() because the
1005 * meta-dnode needs to set it ouside dnode_sync().
1006 */
1007 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1008 ASSERT(dn->dn_zio);
1009
1010 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1011 list_remove(list, dn);
1012
1013 if (newlist) {
1014 (void) dnode_add_ref(dn, newlist);
1015 list_insert_tail(newlist, dn);
1016 }
1017
1018 dnode_sync(dn, tx);
1019 }
1020 }
1021
1022 /* ARGSUSED */
1023 static void
1024 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1025 {
1026 blkptr_t *bp = zio->io_bp;
1027 objset_t *os = arg;
1028 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1029
1030 ASSERT(!BP_IS_EMBEDDED(bp));
1031 ASSERT3P(bp, ==, os->os_rootbp);
1032 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1033 ASSERT0(BP_GET_LEVEL(bp));
1034
1035 /*
1036 * Update rootbp fill count: it should be the number of objects
1037 * allocated in the object set (not counting the "special"
1038 * objects that are stored in the objset_phys_t -- the meta
1039 * dnode and user/group accounting objects).
1040 */
1041 bp->blk_fill = 0;
1042 for (int i = 0; i < dnp->dn_nblkptr; i++)
1043 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1044 }
1045
1046 /* ARGSUSED */
1047 static void
1048 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1049 {
1050 blkptr_t *bp = zio->io_bp;
1051 blkptr_t *bp_orig = &zio->io_bp_orig;
1052 objset_t *os = arg;
1053
1054 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1055 ASSERT(BP_EQUAL(bp, bp_orig));
1056 } else {
1057 dsl_dataset_t *ds = os->os_dsl_dataset;
1058 dmu_tx_t *tx = os->os_synctx;
1059
1060 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1061 dsl_dataset_block_born(ds, bp, tx);
1062 }
1063 }
1064
1065 /* called from dsl */
1066 void
1067 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1068 {
1069 int txgoff;
1070 zbookmark_phys_t zb;
1071 zio_prop_t zp;
1072 zio_t *zio;
1073 list_t *list;
1074 list_t *newlist = NULL;
1075 dbuf_dirty_record_t *dr;
1076
1077 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1078
1079 ASSERT(dmu_tx_is_syncing(tx));
1080 /* XXX the write_done callback should really give us the tx... */
1081 os->os_synctx = tx;
1082
1083 if (os->os_dsl_dataset == NULL) {
1084 /*
1085 * This is the MOS. If we have upgraded,
1086 * spa_max_replication() could change, so reset
1087 * os_copies here.
1088 */
1089 os->os_copies = spa_max_replication(os->os_spa);
1090 }
1091
1092 /*
1093 * Create the root block IO
1094 */
1095 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1096 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1097 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1098 arc_release(os->os_phys_buf, &os->os_phys_buf);
1099
1100 dmu_write_policy(os, NULL, 0, 0, &zp);
1101
1102 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1103 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1104 DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1105 NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1106 ZIO_FLAG_MUSTSUCCEED, &zb);
1107
1108 /*
1109 * Sync special dnodes - the parent IO for the sync is the root block
1110 */
1111 DMU_META_DNODE(os)->dn_zio = zio;
1112 dnode_sync(DMU_META_DNODE(os), tx);
1113
1114 os->os_phys->os_flags = os->os_flags;
1115
1116 if (DMU_USERUSED_DNODE(os) &&
1117 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1118 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1119 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1120 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1121 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1122 }
1123
1124 txgoff = tx->tx_txg & TXG_MASK;
1125
1126 if (dmu_objset_userused_enabled(os)) {
1127 newlist = &os->os_synced_dnodes;
1128 /*
1129 * We must create the list here because it uses the
1130 * dn_dirty_link[] of this txg.
1131 */
1132 list_create(newlist, sizeof (dnode_t),
1133 offsetof(dnode_t, dn_dirty_link[txgoff]));
1134 }
1135
1136 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1137 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1138
1139 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1140 while (dr = list_head(list)) {
1141 ASSERT0(dr->dr_dbuf->db_level);
1142 list_remove(list, dr);
1143 if (dr->dr_zio)
1144 zio_nowait(dr->dr_zio);
1145 }
1146 /*
1147 * Free intent log blocks up to this tx.
1148 */
1149 zil_sync(os->os_zil, tx);
1150 os->os_phys->os_zil_header = os->os_zil_header;
1151 zio_nowait(zio);
1152 }
1153
1154 boolean_t
1155 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1156 {
1157 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1158 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1159 }
1160
1161 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1162
1163 void
1164 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1165 {
1166 used_cbs[ost] = cb;
1167 }
1168
1169 boolean_t
1170 dmu_objset_userused_enabled(objset_t *os)
1171 {
1172 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1173 used_cbs[os->os_phys->os_type] != NULL &&
1174 DMU_USERUSED_DNODE(os) != NULL);
1175 }
1176
1177 static void
1178 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1179 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1180 {
1181 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1182 int64_t delta = DNODE_SIZE + used;
1183 if (subtract)
1184 delta = -delta;
1185 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1186 user, delta, tx));
1187 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1188 group, delta, tx));
1189 }
1190 }
1191
1192 void
1193 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1194 {
1195 dnode_t *dn;
1196 list_t *list = &os->os_synced_dnodes;
1197
1198 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1199
1200 while (dn = list_head(list)) {
1201 int flags;
1202 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1203 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1204 dn->dn_phys->dn_flags &
1205 DNODE_FLAG_USERUSED_ACCOUNTED);
1206
1207 /* Allocate the user/groupused objects if necessary. */
1208 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1209 VERIFY(0 == zap_create_claim(os,
1210 DMU_USERUSED_OBJECT,
1211 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1212 VERIFY(0 == zap_create_claim(os,
1213 DMU_GROUPUSED_OBJECT,
1214 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1215 }
1216
1217 /*
1218 * We intentionally modify the zap object even if the
1219 * net delta is zero. Otherwise
1220 * the block of the zap obj could be shared between
1221 * datasets but need to be different between them after
1222 * a bprewrite.
1223 */
1224
1225 flags = dn->dn_id_flags;
1226 ASSERT(flags);
1227 if (flags & DN_ID_OLD_EXIST) {
1228 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1229 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1230 }
1231 if (flags & DN_ID_NEW_EXIST) {
1232 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1233 dn->dn_phys->dn_flags, dn->dn_newuid,
1234 dn->dn_newgid, B_FALSE, tx);
1235 }
1236
1237 mutex_enter(&dn->dn_mtx);
1238 dn->dn_oldused = 0;
1239 dn->dn_oldflags = 0;
1240 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1241 dn->dn_olduid = dn->dn_newuid;
1242 dn->dn_oldgid = dn->dn_newgid;
1243 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1244 if (dn->dn_bonuslen == 0)
1245 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1246 else
1247 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1248 }
1249 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1250 mutex_exit(&dn->dn_mtx);
1251
1252 list_remove(list, dn);
1253 dnode_rele(dn, list);
1254 }
1255 }
1256
1257 /*
1258 * Returns a pointer to data to find uid/gid from
1259 *
1260 * If a dirty record for transaction group that is syncing can't
1261 * be found then NULL is returned. In the NULL case it is assumed
1262 * the uid/gid aren't changing.
1263 */
1264 static void *
1265 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1266 {
1267 dbuf_dirty_record_t *dr, **drp;
1268 void *data;
1269
1270 if (db->db_dirtycnt == 0)
1271 return (db->db.db_data); /* Nothing is changing */
1272
1273 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1274 if (dr->dr_txg == tx->tx_txg)
1275 break;
1276
1277 if (dr == NULL) {
1278 data = NULL;
1279 } else {
1280 dnode_t *dn;
1281
1282 DB_DNODE_ENTER(dr->dr_dbuf);
1283 dn = DB_DNODE(dr->dr_dbuf);
1284
1285 if (dn->dn_bonuslen == 0 &&
1286 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1287 data = dr->dt.dl.dr_data->b_data;
1288 else
1289 data = dr->dt.dl.dr_data;
1290
1291 DB_DNODE_EXIT(dr->dr_dbuf);
1292 }
1293
1294 return (data);
1295 }
1296
1297 void
1298 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1299 {
1300 objset_t *os = dn->dn_objset;
1301 void *data = NULL;
1302 dmu_buf_impl_t *db = NULL;
1303 uint64_t *user = NULL;
1304 uint64_t *group = NULL;
1305 int flags = dn->dn_id_flags;
1306 int error;
1307 boolean_t have_spill = B_FALSE;
1308
1309 if (!dmu_objset_userused_enabled(dn->dn_objset))
1310 return;
1311
1312 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1313 DN_ID_CHKED_SPILL)))
1314 return;
1315
1316 if (before && dn->dn_bonuslen != 0)
1317 data = DN_BONUS(dn->dn_phys);
1318 else if (!before && dn->dn_bonuslen != 0) {
1319 if (dn->dn_bonus) {
1320 db = dn->dn_bonus;
1321 mutex_enter(&db->db_mtx);
1322 data = dmu_objset_userquota_find_data(db, tx);
1323 } else {
1324 data = DN_BONUS(dn->dn_phys);
1325 }
1326 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1327 int rf = 0;
1328
1329 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1330 rf |= DB_RF_HAVESTRUCT;
1331 error = dmu_spill_hold_by_dnode(dn,
1332 rf | DB_RF_MUST_SUCCEED,
1333 FTAG, (dmu_buf_t **)&db);
1334 ASSERT(error == 0);
1335 mutex_enter(&db->db_mtx);
1336 data = (before) ? db->db.db_data :
1337 dmu_objset_userquota_find_data(db, tx);
1338 have_spill = B_TRUE;
1339 } else {
1340 mutex_enter(&dn->dn_mtx);
1341 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1342 mutex_exit(&dn->dn_mtx);
1343 return;
1344 }
1345
1346 if (before) {
1347 ASSERT(data);
1348 user = &dn->dn_olduid;
1349 group = &dn->dn_oldgid;
1350 } else if (data) {
1351 user = &dn->dn_newuid;
1352 group = &dn->dn_newgid;
1353 }
1354
1355 /*
1356 * Must always call the callback in case the object
1357 * type has changed and that type isn't an object type to track
1358 */
1359 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1360 user, group);
1361
1362 /*
1363 * Preserve existing uid/gid when the callback can't determine
1364 * what the new uid/gid are and the callback returned EEXIST.
1365 * The EEXIST error tells us to just use the existing uid/gid.
1366 * If we don't know what the old values are then just assign
1367 * them to 0, since that is a new file being created.
1368 */
1369 if (!before && data == NULL && error == EEXIST) {
1370 if (flags & DN_ID_OLD_EXIST) {
1371 dn->dn_newuid = dn->dn_olduid;
1372 dn->dn_newgid = dn->dn_oldgid;
1373 } else {
1374 dn->dn_newuid = 0;
1375 dn->dn_newgid = 0;
1376 }
1377 error = 0;
1378 }
1379
1380 if (db)
1381 mutex_exit(&db->db_mtx);
1382
1383 mutex_enter(&dn->dn_mtx);
1384 if (error == 0 && before)
1385 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1386 if (error == 0 && !before)
1387 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1388
1389 if (have_spill) {
1390 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1391 } else {
1392 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1393 }
1394 mutex_exit(&dn->dn_mtx);
1395 if (have_spill)
1396 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1397 }
1398
1399 boolean_t
1400 dmu_objset_userspace_present(objset_t *os)
1401 {
1402 return (os->os_phys->os_flags &
1403 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1404 }
1405
1406 int
1407 dmu_objset_userspace_upgrade(objset_t *os)
1408 {
1409 uint64_t obj;
1410 int err = 0;
1411
1412 if (dmu_objset_userspace_present(os))
1413 return (0);
1414 if (!dmu_objset_userused_enabled(os))
1415 return (SET_ERROR(ENOTSUP));
1416 if (dmu_objset_is_snapshot(os))
1417 return (SET_ERROR(EINVAL));
1418
1419 /*
1420 * We simply need to mark every object dirty, so that it will be
1421 * synced out and now accounted. If this is called
1422 * concurrently, or if we already did some work before crashing,
1423 * that's fine, since we track each object's accounted state
1424 * independently.
1425 */
1426
1427 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1428 dmu_tx_t *tx;
1429 dmu_buf_t *db;
1430 int objerr;
1431
1432 if (issig(JUSTLOOKING) && issig(FORREAL))
1433 return (SET_ERROR(EINTR));
1434
1435 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1436 if (objerr != 0)
1437 continue;
1438 tx = dmu_tx_create(os);
1439 dmu_tx_hold_bonus(tx, obj);
1440 objerr = dmu_tx_assign(tx, TXG_WAIT);
1441 if (objerr != 0) {
1442 dmu_tx_abort(tx);
1443 continue;
1444 }
1445 dmu_buf_will_dirty(db, tx);
1446 dmu_buf_rele(db, FTAG);
1447 dmu_tx_commit(tx);
1448 }
1449
1450 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1451 txg_wait_synced(dmu_objset_pool(os), 0);
1452 return (0);
1453 }
1454
1455 void
1456 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1457 uint64_t *usedobjsp, uint64_t *availobjsp)
1458 {
1459 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1460 usedobjsp, availobjsp);
1461 }
1462
1463 uint64_t
1464 dmu_objset_fsid_guid(objset_t *os)
1465 {
1466 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1467 }
1468
1469 void
1470 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1471 {
1472 stat->dds_type = os->os_phys->os_type;
1473 if (os->os_dsl_dataset)
1474 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1475 }
1476
1477 void
1478 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1479 {
1480 ASSERT(os->os_dsl_dataset ||
1481 os->os_phys->os_type == DMU_OST_META);
1482
1483 if (os->os_dsl_dataset != NULL)
1484 dsl_dataset_stats(os->os_dsl_dataset, nv);
1485
1486 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1487 os->os_phys->os_type);
1488 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1489 dmu_objset_userspace_present(os));
1490 }
1491
1492 int
1493 dmu_objset_is_snapshot(objset_t *os)
1494 {
1495 if (os->os_dsl_dataset != NULL)
1496 return (os->os_dsl_dataset->ds_is_snapshot);
1497 else
1498 return (B_FALSE);
1499 }
1500
1501 int
1502 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1503 boolean_t *conflict)
1504 {
1505 dsl_dataset_t *ds = os->os_dsl_dataset;
1506 uint64_t ignored;
1507
1508 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1509 return (SET_ERROR(ENOENT));
1510
1511 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1512 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1513 MT_FIRST, real, maxlen, conflict));
1514 }
1515
1516 int
1517 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1518 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1519 {
1520 dsl_dataset_t *ds = os->os_dsl_dataset;
1521 zap_cursor_t cursor;
1522 zap_attribute_t attr;
1523
1524 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1525
1526 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1527 return (SET_ERROR(ENOENT));
1528
1529 zap_cursor_init_serialized(&cursor,
1530 ds->ds_dir->dd_pool->dp_meta_objset,
1531 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1532
1533 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1534 zap_cursor_fini(&cursor);
1535 return (SET_ERROR(ENOENT));
1536 }
1537
1538 if (strlen(attr.za_name) + 1 > namelen) {
1539 zap_cursor_fini(&cursor);
1540 return (SET_ERROR(ENAMETOOLONG));
1541 }
1542
1543 (void) strcpy(name, attr.za_name);
1544 if (idp)
1545 *idp = attr.za_first_integer;
1546 if (case_conflict)
1547 *case_conflict = attr.za_normalization_conflict;
1548 zap_cursor_advance(&cursor);
1549 *offp = zap_cursor_serialize(&cursor);
1550 zap_cursor_fini(&cursor);
1551
1552 return (0);
1553 }
1554
1555 int
1556 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1557 uint64_t *idp, uint64_t *offp)
1558 {
1559 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1560 zap_cursor_t cursor;
1561 zap_attribute_t attr;
1562
1563 /* there is no next dir on a snapshot! */
1564 if (os->os_dsl_dataset->ds_object !=
1565 dsl_dir_phys(dd)->dd_head_dataset_obj)
1566 return (SET_ERROR(ENOENT));
1567
1568 zap_cursor_init_serialized(&cursor,
1569 dd->dd_pool->dp_meta_objset,
1570 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1571
1572 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1573 zap_cursor_fini(&cursor);
1574 return (SET_ERROR(ENOENT));
1575 }
1576
1577 if (strlen(attr.za_name) + 1 > namelen) {
1578 zap_cursor_fini(&cursor);
1579 return (SET_ERROR(ENAMETOOLONG));
1580 }
1581
1582 (void) strcpy(name, attr.za_name);
1583 if (idp)
1584 *idp = attr.za_first_integer;
1585 zap_cursor_advance(&cursor);
1586 *offp = zap_cursor_serialize(&cursor);
1587 zap_cursor_fini(&cursor);
1588
1589 return (0);
1590 }
1591
1592 /*
1593 * Find objsets under and including ddobj, call func(ds) on each.
1594 */
1595 int
1596 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1597 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1598 {
1599 dsl_dir_t *dd;
1600 dsl_dataset_t *ds;
1601 zap_cursor_t zc;
1602 zap_attribute_t *attr;
1603 uint64_t thisobj;
1604 int err;
1605
1606 ASSERT(dsl_pool_config_held(dp));
1607
1608 err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1609 if (err != 0)
1610 return (err);
1611
1612 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1613 if (dd->dd_myname[0] == '$') {
1614 dsl_dir_rele(dd, FTAG);
1615 return (0);
1616 }
1617
1618 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1619 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1620
1621 /*
1622 * Iterate over all children.
1623 */
1624 if (flags & DS_FIND_CHILDREN) {
1625 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1626 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1627 zap_cursor_retrieve(&zc, attr) == 0;
1628 (void) zap_cursor_advance(&zc)) {
1629 ASSERT3U(attr->za_integer_length, ==,
1630 sizeof (uint64_t));
1631 ASSERT3U(attr->za_num_integers, ==, 1);
1632
1633 err = dmu_objset_find_dp(dp, attr->za_first_integer,
1634 func, arg, flags);
1635 if (err != 0)
1636 break;
1637 }
1638 zap_cursor_fini(&zc);
1639
1640 if (err != 0) {
1641 dsl_dir_rele(dd, FTAG);
1642 kmem_free(attr, sizeof (zap_attribute_t));
1643 return (err);
1644 }
1645 }
1646
1647 /*
1648 * Iterate over all snapshots.
1649 */
1650 if (flags & DS_FIND_SNAPSHOTS) {
1651 dsl_dataset_t *ds;
1652 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1653
1654 if (err == 0) {
1655 uint64_t snapobj;
1656
1657 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1658 dsl_dataset_rele(ds, FTAG);
1659
1660 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1661 zap_cursor_retrieve(&zc, attr) == 0;
1662 (void) zap_cursor_advance(&zc)) {
1663 ASSERT3U(attr->za_integer_length, ==,
1664 sizeof (uint64_t));
1665 ASSERT3U(attr->za_num_integers, ==, 1);
1666
1667 err = dsl_dataset_hold_obj(dp,
1668 attr->za_first_integer, FTAG, &ds);
1669 if (err != 0)
1670 break;
1671 err = func(dp, ds, arg);
1672 dsl_dataset_rele(ds, FTAG);
1673 if (err != 0)
1674 break;
1675 }
1676 zap_cursor_fini(&zc);
1677 }
1678 }
1679
1680 dsl_dir_rele(dd, FTAG);
1681 kmem_free(attr, sizeof (zap_attribute_t));
1682
1683 if (err != 0)
1684 return (err);
1685
1686 /*
1687 * Apply to self.
1688 */
1689 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1690 if (err != 0)
1691 return (err);
1692 err = func(dp, ds, arg);
1693 dsl_dataset_rele(ds, FTAG);
1694 return (err);
1695 }
1696
1697 /*
1698 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1699 * The dp_config_rwlock must not be held when this is called, and it
1700 * will not be held when the callback is called.
1701 * Therefore this function should only be used when the pool is not changing
1702 * (e.g. in syncing context), or the callback can deal with the possible races.
1703 */
1704 static int
1705 dmu_objset_find_impl(spa_t *spa, const char *name,
1706 int func(const char *, void *), void *arg, int flags)
1707 {
1708 dsl_dir_t *dd;
1709 dsl_pool_t *dp = spa_get_dsl(spa);
1710 dsl_dataset_t *ds;
1711 zap_cursor_t zc;
1712 zap_attribute_t *attr;
1713 char *child;
1714 uint64_t thisobj;
1715 int err;
1716
1717 dsl_pool_config_enter(dp, FTAG);
1718
1719 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1720 if (err != 0) {
1721 dsl_pool_config_exit(dp, FTAG);
1722 return (err);
1723 }
1724
1725 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1726 if (dd->dd_myname[0] == '$') {
1727 dsl_dir_rele(dd, FTAG);
1728 dsl_pool_config_exit(dp, FTAG);
1729 return (0);
1730 }
1731
1732 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1733 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1734
1735 /*
1736 * Iterate over all children.
1737 */
1738 if (flags & DS_FIND_CHILDREN) {
1739 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1740 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1741 zap_cursor_retrieve(&zc, attr) == 0;
1742 (void) zap_cursor_advance(&zc)) {
1743 ASSERT3U(attr->za_integer_length, ==,
1744 sizeof (uint64_t));
1745 ASSERT3U(attr->za_num_integers, ==, 1);
1746
1747 child = kmem_asprintf("%s/%s", name, attr->za_name);
1748 dsl_pool_config_exit(dp, FTAG);
1749 err = dmu_objset_find_impl(spa, child,
1750 func, arg, flags);
1751 dsl_pool_config_enter(dp, FTAG);
1752 strfree(child);
1753 if (err != 0)
1754 break;
1755 }
1756 zap_cursor_fini(&zc);
1757
1758 if (err != 0) {
1759 dsl_dir_rele(dd, FTAG);
1760 dsl_pool_config_exit(dp, FTAG);
1761 kmem_free(attr, sizeof (zap_attribute_t));
1762 return (err);
1763 }
1764 }
1765
1766 /*
1767 * Iterate over all snapshots.
1768 */
1769 if (flags & DS_FIND_SNAPSHOTS) {
1770 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1771
1772 if (err == 0) {
1773 uint64_t snapobj;
1774
1775 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1776 dsl_dataset_rele(ds, FTAG);
1777
1778 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1779 zap_cursor_retrieve(&zc, attr) == 0;
1780 (void) zap_cursor_advance(&zc)) {
1781 ASSERT3U(attr->za_integer_length, ==,
1782 sizeof (uint64_t));
1783 ASSERT3U(attr->za_num_integers, ==, 1);
1784
1785 child = kmem_asprintf("%s@%s",
1786 name, attr->za_name);
1787 dsl_pool_config_exit(dp, FTAG);
1788 err = func(child, arg);
1789 dsl_pool_config_enter(dp, FTAG);
1790 strfree(child);
1791 if (err != 0)
1792 break;
1793 }
1794 zap_cursor_fini(&zc);
1795 }
1796 }
1797
1798 dsl_dir_rele(dd, FTAG);
1799 kmem_free(attr, sizeof (zap_attribute_t));
1800 dsl_pool_config_exit(dp, FTAG);
1801
1802 if (err != 0)
1803 return (err);
1804
1805 /* Apply to self. */
1806 return (func(name, arg));
1807 }
1808
1809 /*
1810 * See comment above dmu_objset_find_impl().
1811 */
1812 int
1813 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1814 int flags)
1815 {
1816 spa_t *spa;
1817 int error;
1818
1819 error = spa_open(name, &spa, FTAG);
1820 if (error != 0)
1821 return (error);
1822 error = dmu_objset_find_impl(spa, name, func, arg, flags);
1823 spa_close(spa, FTAG);
1824 return (error);
1825 }
1826
1827 void
1828 dmu_objset_set_user(objset_t *os, void *user_ptr)
1829 {
1830 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1831 os->os_user_ptr = user_ptr;
1832 }
1833
1834 void *
1835 dmu_objset_get_user(objset_t *os)
1836 {
1837 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1838 return (os->os_user_ptr);
1839 }
1840
1841 /*
1842 * Determine name of filesystem, given name of snapshot.
1843 * buf must be at least MAXNAMELEN bytes
1844 */
1845 int
1846 dmu_fsname(const char *snapname, char *buf)
1847 {
1848 char *atp = strchr(snapname, '@');
1849 if (atp == NULL)
1850 return (SET_ERROR(EINVAL));
1851 if (atp - snapname >= MAXNAMELEN)
1852 return (SET_ERROR(ENAMETOOLONG));
1853 (void) strlcpy(buf, snapname, atp - snapname + 1);
1854 return (0);
1855 }