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 */
27
28 /* Portions Copyright 2010 Robert Milkowski */
29
30 #include <sys/cred.h>
31 #include <sys/zfs_context.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_prop.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dsl_deleg.h>
39 #include <sys/dnode.h>
40 #include <sys/dbuf.h>
41 #include <sys/zvol.h>
42 #include <sys/dmu_tx.h>
43 #include <sys/zap.h>
44 #include <sys/zil.h>
45 #include <sys/dmu_impl.h>
46 #include <sys/zfs_ioctl.h>
47 #include <sys/sa.h>
48 #include <sys/zfs_onexit.h>
49 #include <sys/dsl_destroy.h>
50
51 /*
52 * Needed to close a window in dnode_move() that allows the objset to be freed
53 * before it can be safely accessed.
54 */
55 krwlock_t os_lock;
56
57 void
58 dmu_objset_init(void)
59 {
60 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
61 }
62
63 void
64 dmu_objset_fini(void)
65 {
66 rw_destroy(&os_lock);
67 }
68
69 spa_t *
70 dmu_objset_spa(objset_t *os)
71 {
72 return (os->os_spa);
73 }
74
75 zilog_t *
76 dmu_objset_zil(objset_t *os)
77 {
78 return (os->os_zil);
79 }
80
81 dsl_pool_t *
82 dmu_objset_pool(objset_t *os)
83 {
84 dsl_dataset_t *ds;
85
86 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
87 return (ds->ds_dir->dd_pool);
88 else
89 return (spa_get_dsl(os->os_spa));
90 }
91
92 dsl_dataset_t *
93 dmu_objset_ds(objset_t *os)
94 {
95 return (os->os_dsl_dataset);
96 }
97
98 dmu_objset_type_t
99 dmu_objset_type(objset_t *os)
100 {
101 return (os->os_phys->os_type);
102 }
103
104 void
105 dmu_objset_name(objset_t *os, char *buf)
106 {
107 dsl_dataset_name(os->os_dsl_dataset, buf);
108 }
109
110 uint64_t
111 dmu_objset_id(objset_t *os)
112 {
113 dsl_dataset_t *ds = os->os_dsl_dataset;
114
115 return (ds ? ds->ds_object : 0);
116 }
117
118 zfs_sync_type_t
119 dmu_objset_syncprop(objset_t *os)
120 {
121 return (os->os_sync);
122 }
123
124 zfs_logbias_op_t
125 dmu_objset_logbias(objset_t *os)
126 {
127 return (os->os_logbias);
128 }
129
130 static void
131 checksum_changed_cb(void *arg, uint64_t newval)
132 {
133 objset_t *os = arg;
134
135 /*
136 * Inheritance should have been done by now.
137 */
138 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
139
140 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
141 }
142
143 static void
144 compression_changed_cb(void *arg, uint64_t newval)
145 {
146 objset_t *os = arg;
147
148 /*
149 * Inheritance and range checking should have been done by now.
150 */
151 ASSERT(newval != ZIO_COMPRESS_INHERIT);
152
153 os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
154 }
155
156 static void
157 copies_changed_cb(void *arg, uint64_t newval)
158 {
159 objset_t *os = arg;
160
161 /*
162 * Inheritance and range checking should have been done by now.
163 */
164 ASSERT(newval > 0);
165 ASSERT(newval <= spa_max_replication(os->os_spa));
166
167 os->os_copies = newval;
168 }
169
170 static void
171 dedup_changed_cb(void *arg, uint64_t newval)
172 {
173 objset_t *os = arg;
174 spa_t *spa = os->os_spa;
175 enum zio_checksum checksum;
176
177 /*
178 * Inheritance should have been done by now.
179 */
180 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
181
182 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
183
184 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
185 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
186 }
187
188 static void
189 primary_cache_changed_cb(void *arg, uint64_t newval)
190 {
191 objset_t *os = arg;
192
193 /*
194 * Inheritance and range checking should have been done by now.
195 */
196 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
197 newval == ZFS_CACHE_METADATA);
198
199 os->os_primary_cache = newval;
200 }
201
202 static void
203 secondary_cache_changed_cb(void *arg, uint64_t newval)
204 {
205 objset_t *os = arg;
206
207 /*
208 * Inheritance and range checking should have been done by now.
209 */
210 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
211 newval == ZFS_CACHE_METADATA);
212
213 os->os_secondary_cache = newval;
214 }
215
216 static void
217 sync_changed_cb(void *arg, uint64_t newval)
218 {
219 objset_t *os = arg;
220
221 /*
222 * Inheritance and range checking should have been done by now.
223 */
224 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
225 newval == ZFS_SYNC_DISABLED);
226
227 os->os_sync = newval;
228 if (os->os_zil)
229 zil_set_sync(os->os_zil, newval);
230 }
231
232 static void
233 redundant_metadata_changed_cb(void *arg, uint64_t newval)
234 {
235 objset_t *os = arg;
236
237 /*
238 * Inheritance and range checking should have been done by now.
239 */
240 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
241 newval == ZFS_REDUNDANT_METADATA_MOST);
242
243 os->os_redundant_metadata = newval;
244 }
245
246 static void
247 logbias_changed_cb(void *arg, uint64_t newval)
248 {
249 objset_t *os = arg;
250
251 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
252 newval == ZFS_LOGBIAS_THROUGHPUT);
253 os->os_logbias = newval;
254 if (os->os_zil)
255 zil_set_logbias(os->os_zil, newval);
256 }
257
258 void
259 dmu_objset_byteswap(void *buf, size_t size)
260 {
261 objset_phys_t *osp = buf;
262
263 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
264 dnode_byteswap(&osp->os_meta_dnode);
265 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
266 osp->os_type = BSWAP_64(osp->os_type);
267 osp->os_flags = BSWAP_64(osp->os_flags);
268 if (size == sizeof (objset_phys_t)) {
269 dnode_byteswap(&osp->os_userused_dnode);
270 dnode_byteswap(&osp->os_groupused_dnode);
271 }
272 }
273
274 int
275 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
276 objset_t **osp)
277 {
278 objset_t *os;
279 int i, err;
280
281 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
282
283 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
284 os->os_dsl_dataset = ds;
285 os->os_spa = spa;
286 os->os_rootbp = bp;
287 if (!BP_IS_HOLE(os->os_rootbp)) {
288 uint32_t aflags = ARC_WAIT;
289 zbookmark_t zb;
290 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
291 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
292
293 if (DMU_OS_IS_L2CACHEABLE(os))
294 aflags |= ARC_L2CACHE;
295 if (DMU_OS_IS_L2COMPRESSIBLE(os))
296 aflags |= ARC_L2COMPRESS;
297
298 dprintf_bp(os->os_rootbp, "reading %s", "");
299 err = arc_read(NULL, spa, os->os_rootbp,
300 arc_getbuf_func, &os->os_phys_buf,
301 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
302 if (err != 0) {
303 kmem_free(os, sizeof (objset_t));
304 /* convert checksum errors into IO errors */
305 if (err == ECKSUM)
306 err = SET_ERROR(EIO);
307 return (err);
308 }
309
310 /* Increase the blocksize if we are permitted. */
311 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
312 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
313 arc_buf_t *buf = arc_buf_alloc(spa,
314 sizeof (objset_phys_t), &os->os_phys_buf,
315 ARC_BUFC_METADATA);
316 bzero(buf->b_data, sizeof (objset_phys_t));
317 bcopy(os->os_phys_buf->b_data, buf->b_data,
318 arc_buf_size(os->os_phys_buf));
319 (void) arc_buf_remove_ref(os->os_phys_buf,
320 &os->os_phys_buf);
321 os->os_phys_buf = buf;
322 }
323
324 os->os_phys = os->os_phys_buf->b_data;
325 os->os_flags = os->os_phys->os_flags;
326 } else {
327 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
328 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
329 os->os_phys_buf = arc_buf_alloc(spa, size,
330 &os->os_phys_buf, ARC_BUFC_METADATA);
331 os->os_phys = os->os_phys_buf->b_data;
332 bzero(os->os_phys, size);
333 }
334
335 /*
336 * Note: the changed_cb will be called once before the register
337 * func returns, thus changing the checksum/compression from the
338 * default (fletcher2/off). Snapshots don't need to know about
339 * checksum/compression/copies.
340 */
341 if (ds != NULL) {
342 err = dsl_prop_register(ds,
343 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
344 primary_cache_changed_cb, os);
345 if (err == 0) {
346 err = dsl_prop_register(ds,
347 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
348 secondary_cache_changed_cb, os);
349 }
350 if (!dsl_dataset_is_snapshot(ds)) {
351 if (err == 0) {
352 err = dsl_prop_register(ds,
353 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
354 checksum_changed_cb, os);
355 }
356 if (err == 0) {
357 err = dsl_prop_register(ds,
358 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
359 compression_changed_cb, os);
360 }
361 if (err == 0) {
362 err = dsl_prop_register(ds,
363 zfs_prop_to_name(ZFS_PROP_COPIES),
364 copies_changed_cb, os);
365 }
366 if (err == 0) {
367 err = dsl_prop_register(ds,
368 zfs_prop_to_name(ZFS_PROP_DEDUP),
369 dedup_changed_cb, os);
370 }
371 if (err == 0) {
372 err = dsl_prop_register(ds,
373 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
374 logbias_changed_cb, os);
375 }
376 if (err == 0) {
377 err = dsl_prop_register(ds,
378 zfs_prop_to_name(ZFS_PROP_SYNC),
379 sync_changed_cb, os);
380 }
381 if (err == 0) {
382 err = dsl_prop_register(ds,
383 zfs_prop_to_name(
384 ZFS_PROP_REDUNDANT_METADATA),
385 redundant_metadata_changed_cb, os);
386 }
387 }
388 if (err != 0) {
389 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
390 &os->os_phys_buf));
391 kmem_free(os, sizeof (objset_t));
392 return (err);
393 }
394 } else {
395 /* It's the meta-objset. */
396 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
397 os->os_compress = ZIO_COMPRESS_LZJB;
398 os->os_copies = spa_max_replication(spa);
399 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
400 os->os_dedup_verify = B_FALSE;
401 os->os_logbias = ZFS_LOGBIAS_LATENCY;
402 os->os_sync = ZFS_SYNC_STANDARD;
403 os->os_primary_cache = ZFS_CACHE_ALL;
404 os->os_secondary_cache = ZFS_CACHE_ALL;
405 }
406
407 if (ds == NULL || !dsl_dataset_is_snapshot(ds))
408 os->os_zil_header = os->os_phys->os_zil_header;
409 os->os_zil = zil_alloc(os, &os->os_zil_header);
410
411 for (i = 0; i < TXG_SIZE; i++) {
412 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
413 offsetof(dnode_t, dn_dirty_link[i]));
414 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
415 offsetof(dnode_t, dn_dirty_link[i]));
416 }
417 list_create(&os->os_dnodes, sizeof (dnode_t),
418 offsetof(dnode_t, dn_link));
419 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
420 offsetof(dmu_buf_impl_t, db_link));
421
422 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
423 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
424 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
425
426 DMU_META_DNODE(os) = dnode_special_open(os,
427 &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
428 &os->os_meta_dnode);
429 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
430 DMU_USERUSED_DNODE(os) = dnode_special_open(os,
431 &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
432 &os->os_userused_dnode);
433 DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
434 &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
435 &os->os_groupused_dnode);
436 }
437
438 *osp = os;
439 return (0);
440 }
441
442 int
443 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
444 {
445 int err = 0;
446
447 mutex_enter(&ds->ds_opening_lock);
448 if (ds->ds_objset == NULL) {
449 objset_t *os;
450 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
451 ds, dsl_dataset_get_blkptr(ds), &os);
452
453 if (err == 0) {
454 mutex_enter(&ds->ds_lock);
455 ASSERT(ds->ds_objset == NULL);
456 ds->ds_objset = os;
457 mutex_exit(&ds->ds_lock);
458 }
459 }
460 *osp = ds->ds_objset;
461 mutex_exit(&ds->ds_opening_lock);
462 return (err);
463 }
464
465 /*
466 * Holds the pool while the objset is held. Therefore only one objset
467 * can be held at a time.
468 */
469 int
470 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
471 {
472 dsl_pool_t *dp;
473 dsl_dataset_t *ds;
474 int err;
475
476 err = dsl_pool_hold(name, tag, &dp);
477 if (err != 0)
478 return (err);
479 err = dsl_dataset_hold(dp, name, tag, &ds);
480 if (err != 0) {
481 dsl_pool_rele(dp, tag);
482 return (err);
483 }
484
485 err = dmu_objset_from_ds(ds, osp);
486 if (err != 0) {
487 dsl_dataset_rele(ds, tag);
488 dsl_pool_rele(dp, tag);
489 }
490
491 return (err);
492 }
493
494 /*
495 * dsl_pool must not be held when this is called.
496 * Upon successful return, there will be a longhold on the dataset,
497 * and the dsl_pool will not be held.
498 */
499 int
500 dmu_objset_own(const char *name, dmu_objset_type_t type,
501 boolean_t readonly, void *tag, objset_t **osp)
502 {
503 dsl_pool_t *dp;
504 dsl_dataset_t *ds;
505 int err;
506
507 err = dsl_pool_hold(name, FTAG, &dp);
508 if (err != 0)
509 return (err);
510 err = dsl_dataset_own(dp, name, tag, &ds);
511 if (err != 0) {
512 dsl_pool_rele(dp, FTAG);
513 return (err);
514 }
515
516 err = dmu_objset_from_ds(ds, osp);
517 dsl_pool_rele(dp, FTAG);
518 if (err != 0) {
519 dsl_dataset_disown(ds, tag);
520 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
521 dsl_dataset_disown(ds, tag);
522 return (SET_ERROR(EINVAL));
523 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
524 dsl_dataset_disown(ds, tag);
525 return (SET_ERROR(EROFS));
526 }
527 return (err);
528 }
529
530 void
531 dmu_objset_rele(objset_t *os, void *tag)
532 {
533 dsl_pool_t *dp = dmu_objset_pool(os);
534 dsl_dataset_rele(os->os_dsl_dataset, tag);
535 dsl_pool_rele(dp, tag);
536 }
537
538 /*
539 * When we are called, os MUST refer to an objset associated with a dataset
540 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
541 * == tag. We will then release and reacquire ownership of the dataset while
542 * holding the pool config_rwlock to avoid intervening namespace or ownership
543 * changes may occur.
544 *
545 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
546 * release the hold on its dataset and acquire a new one on the dataset of the
547 * same name so that it can be partially torn down and reconstructed.
548 */
549 void
550 dmu_objset_refresh_ownership(objset_t *os, void *tag)
551 {
552 dsl_pool_t *dp;
553 dsl_dataset_t *ds, *newds;
554 char name[MAXNAMELEN];
555
556 ds = os->os_dsl_dataset;
557 VERIFY3P(ds, !=, NULL);
558 VERIFY3P(ds->ds_owner, ==, tag);
559 VERIFY(dsl_dataset_long_held(ds));
560
561 dsl_dataset_name(ds, name);
562 dp = dmu_objset_pool(os);
563 dsl_pool_config_enter(dp, FTAG);
564 dmu_objset_disown(os, tag);
565 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
566 VERIFY3P(newds, ==, os->os_dsl_dataset);
567 dsl_pool_config_exit(dp, FTAG);
568 }
569
570 void
571 dmu_objset_disown(objset_t *os, void *tag)
572 {
573 dsl_dataset_disown(os->os_dsl_dataset, tag);
574 }
575
576 void
577 dmu_objset_evict_dbufs(objset_t *os)
578 {
579 dnode_t *dn;
580
581 mutex_enter(&os->os_lock);
582
583 /* process the mdn last, since the other dnodes have holds on it */
584 list_remove(&os->os_dnodes, DMU_META_DNODE(os));
585 list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
586
587 /*
588 * Find the first dnode with holds. We have to do this dance
589 * because dnode_add_ref() only works if you already have a
590 * hold. If there are no holds then it has no dbufs so OK to
591 * skip.
592 */
593 for (dn = list_head(&os->os_dnodes);
594 dn && !dnode_add_ref(dn, FTAG);
595 dn = list_next(&os->os_dnodes, dn))
596 continue;
597
598 while (dn) {
599 dnode_t *next_dn = dn;
600
601 do {
602 next_dn = list_next(&os->os_dnodes, next_dn);
603 } while (next_dn && !dnode_add_ref(next_dn, FTAG));
604
605 mutex_exit(&os->os_lock);
606 dnode_evict_dbufs(dn);
607 dnode_rele(dn, FTAG);
608 mutex_enter(&os->os_lock);
609 dn = next_dn;
610 }
611 mutex_exit(&os->os_lock);
612 }
613
614 void
615 dmu_objset_evict(objset_t *os)
616 {
617 dsl_dataset_t *ds = os->os_dsl_dataset;
618
619 for (int t = 0; t < TXG_SIZE; t++)
620 ASSERT(!dmu_objset_is_dirty(os, t));
621
622 if (ds) {
623 if (!dsl_dataset_is_snapshot(ds)) {
624 VERIFY0(dsl_prop_unregister(ds,
625 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
626 checksum_changed_cb, os));
627 VERIFY0(dsl_prop_unregister(ds,
628 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
629 compression_changed_cb, os));
630 VERIFY0(dsl_prop_unregister(ds,
631 zfs_prop_to_name(ZFS_PROP_COPIES),
632 copies_changed_cb, os));
633 VERIFY0(dsl_prop_unregister(ds,
634 zfs_prop_to_name(ZFS_PROP_DEDUP),
635 dedup_changed_cb, os));
636 VERIFY0(dsl_prop_unregister(ds,
637 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
638 logbias_changed_cb, os));
639 VERIFY0(dsl_prop_unregister(ds,
640 zfs_prop_to_name(ZFS_PROP_SYNC),
641 sync_changed_cb, os));
642 VERIFY0(dsl_prop_unregister(ds,
643 zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA),
644 redundant_metadata_changed_cb, os));
645 }
646 VERIFY0(dsl_prop_unregister(ds,
647 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
648 primary_cache_changed_cb, os));
649 VERIFY0(dsl_prop_unregister(ds,
650 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
651 secondary_cache_changed_cb, os));
652 }
653
654 if (os->os_sa)
655 sa_tear_down(os);
656
657 dmu_objset_evict_dbufs(os);
658
659 dnode_special_close(&os->os_meta_dnode);
660 if (DMU_USERUSED_DNODE(os)) {
661 dnode_special_close(&os->os_userused_dnode);
662 dnode_special_close(&os->os_groupused_dnode);
663 }
664 zil_free(os->os_zil);
665
666 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
667
668 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
669
670 /*
671 * This is a barrier to prevent the objset from going away in
672 * dnode_move() until we can safely ensure that the objset is still in
673 * use. We consider the objset valid before the barrier and invalid
674 * after the barrier.
675 */
676 rw_enter(&os_lock, RW_READER);
677 rw_exit(&os_lock);
678
679 mutex_destroy(&os->os_lock);
680 mutex_destroy(&os->os_obj_lock);
681 mutex_destroy(&os->os_user_ptr_lock);
682 kmem_free(os, sizeof (objset_t));
683 }
684
685 timestruc_t
686 dmu_objset_snap_cmtime(objset_t *os)
687 {
688 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
689 }
690
691 /* called from dsl for meta-objset */
692 objset_t *
693 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
694 dmu_objset_type_t type, dmu_tx_t *tx)
695 {
696 objset_t *os;
697 dnode_t *mdn;
698
699 ASSERT(dmu_tx_is_syncing(tx));
700
701 if (ds != NULL)
702 VERIFY0(dmu_objset_from_ds(ds, &os));
703 else
704 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
705
706 mdn = DMU_META_DNODE(os);
707
708 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
709 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
710
711 /*
712 * We don't want to have to increase the meta-dnode's nlevels
713 * later, because then we could do it in quescing context while
714 * we are also accessing it in open context.
715 *
716 * This precaution is not necessary for the MOS (ds == NULL),
717 * because the MOS is only updated in syncing context.
718 * This is most fortunate: the MOS is the only objset that
719 * needs to be synced multiple times as spa_sync() iterates
720 * to convergence, so minimizing its dn_nlevels matters.
721 */
722 if (ds != NULL) {
723 int levels = 1;
724
725 /*
726 * Determine the number of levels necessary for the meta-dnode
727 * to contain DN_MAX_OBJECT dnodes.
728 */
729 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
730 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
731 DN_MAX_OBJECT * sizeof (dnode_phys_t))
732 levels++;
733
734 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
735 mdn->dn_nlevels = levels;
736 }
737
738 ASSERT(type != DMU_OST_NONE);
739 ASSERT(type != DMU_OST_ANY);
740 ASSERT(type < DMU_OST_NUMTYPES);
741 os->os_phys->os_type = type;
742 if (dmu_objset_userused_enabled(os)) {
743 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
744 os->os_flags = os->os_phys->os_flags;
745 }
746
747 dsl_dataset_dirty(ds, tx);
748
749 return (os);
750 }
751
752 typedef struct dmu_objset_create_arg {
753 const char *doca_name;
754 cred_t *doca_cred;
755 void (*doca_userfunc)(objset_t *os, void *arg,
756 cred_t *cr, dmu_tx_t *tx);
757 void *doca_userarg;
758 dmu_objset_type_t doca_type;
759 uint64_t doca_flags;
760 } dmu_objset_create_arg_t;
761
762 /*ARGSUSED*/
763 static int
764 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
765 {
766 dmu_objset_create_arg_t *doca = arg;
767 dsl_pool_t *dp = dmu_tx_pool(tx);
768 dsl_dir_t *pdd;
769 const char *tail;
770 int error;
771
772 if (strchr(doca->doca_name, '@') != NULL)
773 return (SET_ERROR(EINVAL));
774
775 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
776 if (error != 0)
777 return (error);
778 if (tail == NULL) {
779 dsl_dir_rele(pdd, FTAG);
780 return (SET_ERROR(EEXIST));
781 }
782 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
783 doca->doca_cred);
784 dsl_dir_rele(pdd, FTAG);
785
786 return (error);
787 }
788
789 static void
790 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
791 {
792 dmu_objset_create_arg_t *doca = arg;
793 dsl_pool_t *dp = dmu_tx_pool(tx);
794 dsl_dir_t *pdd;
795 const char *tail;
796 dsl_dataset_t *ds;
797 uint64_t obj;
798 blkptr_t *bp;
799 objset_t *os;
800
801 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
802
803 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
804 doca->doca_cred, tx);
805
806 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
807 bp = dsl_dataset_get_blkptr(ds);
808 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
809 ds, bp, doca->doca_type, tx);
810
811 if (doca->doca_userfunc != NULL) {
812 doca->doca_userfunc(os, doca->doca_userarg,
813 doca->doca_cred, tx);
814 }
815
816 spa_history_log_internal_ds(ds, "create", tx, "");
817 dsl_dataset_rele(ds, FTAG);
818 dsl_dir_rele(pdd, FTAG);
819 }
820
821 int
822 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
823 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
824 {
825 dmu_objset_create_arg_t doca;
826
827 doca.doca_name = name;
828 doca.doca_cred = CRED();
829 doca.doca_flags = flags;
830 doca.doca_userfunc = func;
831 doca.doca_userarg = arg;
832 doca.doca_type = type;
833
834 return (dsl_sync_task(name,
835 dmu_objset_create_check, dmu_objset_create_sync, &doca, 5));
836 }
837
838 typedef struct dmu_objset_clone_arg {
839 const char *doca_clone;
840 const char *doca_origin;
841 cred_t *doca_cred;
842 } dmu_objset_clone_arg_t;
843
844 /*ARGSUSED*/
845 static int
846 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
847 {
848 dmu_objset_clone_arg_t *doca = arg;
849 dsl_dir_t *pdd;
850 const char *tail;
851 int error;
852 dsl_dataset_t *origin;
853 dsl_pool_t *dp = dmu_tx_pool(tx);
854
855 if (strchr(doca->doca_clone, '@') != NULL)
856 return (SET_ERROR(EINVAL));
857
858 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
859 if (error != 0)
860 return (error);
861 if (tail == NULL) {
862 dsl_dir_rele(pdd, FTAG);
863 return (SET_ERROR(EEXIST));
864 }
865 /* You can't clone across pools. */
866 if (pdd->dd_pool != dp) {
867 dsl_dir_rele(pdd, FTAG);
868 return (SET_ERROR(EXDEV));
869 }
870 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
871 doca->doca_cred);
872 if (error != 0) {
873 dsl_dir_rele(pdd, FTAG);
874 return (SET_ERROR(EDQUOT));
875 }
876 dsl_dir_rele(pdd, FTAG);
877
878 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
879 if (error != 0)
880 return (error);
881
882 /* You can't clone across pools. */
883 if (origin->ds_dir->dd_pool != dp) {
884 dsl_dataset_rele(origin, FTAG);
885 return (SET_ERROR(EXDEV));
886 }
887
888 /* You can only clone snapshots, not the head datasets. */
889 if (!dsl_dataset_is_snapshot(origin)) {
890 dsl_dataset_rele(origin, FTAG);
891 return (SET_ERROR(EINVAL));
892 }
893 dsl_dataset_rele(origin, FTAG);
894
895 return (0);
896 }
897
898 static void
899 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
900 {
901 dmu_objset_clone_arg_t *doca = arg;
902 dsl_pool_t *dp = dmu_tx_pool(tx);
903 dsl_dir_t *pdd;
904 const char *tail;
905 dsl_dataset_t *origin, *ds;
906 uint64_t obj;
907 char namebuf[MAXNAMELEN];
908
909 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
910 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
911
912 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
913 doca->doca_cred, tx);
914
915 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
916 dsl_dataset_name(origin, namebuf);
917 spa_history_log_internal_ds(ds, "clone", tx,
918 "origin=%s (%llu)", namebuf, origin->ds_object);
919 dsl_dataset_rele(ds, FTAG);
920 dsl_dataset_rele(origin, FTAG);
921 dsl_dir_rele(pdd, FTAG);
922 }
923
924 int
925 dmu_objset_clone(const char *clone, const char *origin)
926 {
927 dmu_objset_clone_arg_t doca;
928
929 doca.doca_clone = clone;
930 doca.doca_origin = origin;
931 doca.doca_cred = CRED();
932
933 return (dsl_sync_task(clone,
934 dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 5));
935 }
936
937 int
938 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
939 {
940 int err;
941 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
942 nvlist_t *snaps = fnvlist_alloc();
943
944 fnvlist_add_boolean(snaps, longsnap);
945 strfree(longsnap);
946 err = dsl_dataset_snapshot(snaps, NULL, NULL);
947 fnvlist_free(snaps);
948 return (err);
949 }
950
951 static void
952 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
953 {
954 dnode_t *dn;
955
956 while (dn = list_head(list)) {
957 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
958 ASSERT(dn->dn_dbuf->db_data_pending);
959 /*
960 * Initialize dn_zio outside dnode_sync() because the
961 * meta-dnode needs to set it ouside dnode_sync().
962 */
963 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
964 ASSERT(dn->dn_zio);
965
966 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
967 list_remove(list, dn);
968
969 if (newlist) {
970 (void) dnode_add_ref(dn, newlist);
971 list_insert_tail(newlist, dn);
972 }
973
974 dnode_sync(dn, tx);
975 }
976 }
977
978 /* ARGSUSED */
979 static void
980 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
981 {
982 blkptr_t *bp = zio->io_bp;
983 objset_t *os = arg;
984 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
985
986 ASSERT(!BP_IS_EMBEDDED(bp));
987 ASSERT3P(bp, ==, os->os_rootbp);
988 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
989 ASSERT0(BP_GET_LEVEL(bp));
990
991 /*
992 * Update rootbp fill count: it should be the number of objects
993 * allocated in the object set (not counting the "special"
994 * objects that are stored in the objset_phys_t -- the meta
995 * dnode and user/group accounting objects).
996 */
997 bp->blk_fill = 0;
998 for (int i = 0; i < dnp->dn_nblkptr; i++)
999 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1000 }
1001
1002 /* ARGSUSED */
1003 static void
1004 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1005 {
1006 blkptr_t *bp = zio->io_bp;
1007 blkptr_t *bp_orig = &zio->io_bp_orig;
1008 objset_t *os = arg;
1009
1010 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1011 ASSERT(BP_EQUAL(bp, bp_orig));
1012 } else {
1013 dsl_dataset_t *ds = os->os_dsl_dataset;
1014 dmu_tx_t *tx = os->os_synctx;
1015
1016 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1017 dsl_dataset_block_born(ds, bp, tx);
1018 }
1019 }
1020
1021 /* called from dsl */
1022 void
1023 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1024 {
1025 int txgoff;
1026 zbookmark_t zb;
1027 zio_prop_t zp;
1028 zio_t *zio;
1029 list_t *list;
1030 list_t *newlist = NULL;
1031 dbuf_dirty_record_t *dr;
1032
1033 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1034
1035 ASSERT(dmu_tx_is_syncing(tx));
1036 /* XXX the write_done callback should really give us the tx... */
1037 os->os_synctx = tx;
1038
1039 if (os->os_dsl_dataset == NULL) {
1040 /*
1041 * This is the MOS. If we have upgraded,
1042 * spa_max_replication() could change, so reset
1043 * os_copies here.
1044 */
1045 os->os_copies = spa_max_replication(os->os_spa);
1046 }
1047
1048 /*
1049 * Create the root block IO
1050 */
1051 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1052 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1053 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1054 arc_release(os->os_phys_buf, &os->os_phys_buf);
1055
1056 dmu_write_policy(os, NULL, 0, 0, &zp);
1057
1058 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1059 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1060 DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1061 NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1062 ZIO_FLAG_MUSTSUCCEED, &zb);
1063
1064 /*
1065 * Sync special dnodes - the parent IO for the sync is the root block
1066 */
1067 DMU_META_DNODE(os)->dn_zio = zio;
1068 dnode_sync(DMU_META_DNODE(os), tx);
1069
1070 os->os_phys->os_flags = os->os_flags;
1071
1072 if (DMU_USERUSED_DNODE(os) &&
1073 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1074 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1075 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1076 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1077 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1078 }
1079
1080 txgoff = tx->tx_txg & TXG_MASK;
1081
1082 if (dmu_objset_userused_enabled(os)) {
1083 newlist = &os->os_synced_dnodes;
1084 /*
1085 * We must create the list here because it uses the
1086 * dn_dirty_link[] of this txg.
1087 */
1088 list_create(newlist, sizeof (dnode_t),
1089 offsetof(dnode_t, dn_dirty_link[txgoff]));
1090 }
1091
1092 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1093 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1094
1095 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1096 while (dr = list_head(list)) {
1097 ASSERT0(dr->dr_dbuf->db_level);
1098 list_remove(list, dr);
1099 if (dr->dr_zio)
1100 zio_nowait(dr->dr_zio);
1101 }
1102 /*
1103 * Free intent log blocks up to this tx.
1104 */
1105 zil_sync(os->os_zil, tx);
1106 os->os_phys->os_zil_header = os->os_zil_header;
1107 zio_nowait(zio);
1108 }
1109
1110 boolean_t
1111 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1112 {
1113 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1114 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1115 }
1116
1117 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1118
1119 void
1120 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1121 {
1122 used_cbs[ost] = cb;
1123 }
1124
1125 boolean_t
1126 dmu_objset_userused_enabled(objset_t *os)
1127 {
1128 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1129 used_cbs[os->os_phys->os_type] != NULL &&
1130 DMU_USERUSED_DNODE(os) != NULL);
1131 }
1132
1133 static void
1134 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1135 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1136 {
1137 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1138 int64_t delta = DNODE_SIZE + used;
1139 if (subtract)
1140 delta = -delta;
1141 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1142 user, delta, tx));
1143 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1144 group, delta, tx));
1145 }
1146 }
1147
1148 void
1149 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1150 {
1151 dnode_t *dn;
1152 list_t *list = &os->os_synced_dnodes;
1153
1154 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1155
1156 while (dn = list_head(list)) {
1157 int flags;
1158 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1159 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1160 dn->dn_phys->dn_flags &
1161 DNODE_FLAG_USERUSED_ACCOUNTED);
1162
1163 /* Allocate the user/groupused objects if necessary. */
1164 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1165 VERIFY(0 == zap_create_claim(os,
1166 DMU_USERUSED_OBJECT,
1167 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1168 VERIFY(0 == zap_create_claim(os,
1169 DMU_GROUPUSED_OBJECT,
1170 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1171 }
1172
1173 /*
1174 * We intentionally modify the zap object even if the
1175 * net delta is zero. Otherwise
1176 * the block of the zap obj could be shared between
1177 * datasets but need to be different between them after
1178 * a bprewrite.
1179 */
1180
1181 flags = dn->dn_id_flags;
1182 ASSERT(flags);
1183 if (flags & DN_ID_OLD_EXIST) {
1184 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1185 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1186 }
1187 if (flags & DN_ID_NEW_EXIST) {
1188 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1189 dn->dn_phys->dn_flags, dn->dn_newuid,
1190 dn->dn_newgid, B_FALSE, tx);
1191 }
1192
1193 mutex_enter(&dn->dn_mtx);
1194 dn->dn_oldused = 0;
1195 dn->dn_oldflags = 0;
1196 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1197 dn->dn_olduid = dn->dn_newuid;
1198 dn->dn_oldgid = dn->dn_newgid;
1199 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1200 if (dn->dn_bonuslen == 0)
1201 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1202 else
1203 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1204 }
1205 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1206 mutex_exit(&dn->dn_mtx);
1207
1208 list_remove(list, dn);
1209 dnode_rele(dn, list);
1210 }
1211 }
1212
1213 /*
1214 * Returns a pointer to data to find uid/gid from
1215 *
1216 * If a dirty record for transaction group that is syncing can't
1217 * be found then NULL is returned. In the NULL case it is assumed
1218 * the uid/gid aren't changing.
1219 */
1220 static void *
1221 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1222 {
1223 dbuf_dirty_record_t *dr, **drp;
1224 void *data;
1225
1226 if (db->db_dirtycnt == 0)
1227 return (db->db.db_data); /* Nothing is changing */
1228
1229 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1230 if (dr->dr_txg == tx->tx_txg)
1231 break;
1232
1233 if (dr == NULL) {
1234 data = NULL;
1235 } else {
1236 dnode_t *dn;
1237
1238 DB_DNODE_ENTER(dr->dr_dbuf);
1239 dn = DB_DNODE(dr->dr_dbuf);
1240
1241 if (dn->dn_bonuslen == 0 &&
1242 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1243 data = dr->dt.dl.dr_data->b_data;
1244 else
1245 data = dr->dt.dl.dr_data;
1246
1247 DB_DNODE_EXIT(dr->dr_dbuf);
1248 }
1249
1250 return (data);
1251 }
1252
1253 void
1254 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1255 {
1256 objset_t *os = dn->dn_objset;
1257 void *data = NULL;
1258 dmu_buf_impl_t *db = NULL;
1259 uint64_t *user = NULL;
1260 uint64_t *group = NULL;
1261 int flags = dn->dn_id_flags;
1262 int error;
1263 boolean_t have_spill = B_FALSE;
1264
1265 if (!dmu_objset_userused_enabled(dn->dn_objset))
1266 return;
1267
1268 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1269 DN_ID_CHKED_SPILL)))
1270 return;
1271
1272 if (before && dn->dn_bonuslen != 0)
1273 data = DN_BONUS(dn->dn_phys);
1274 else if (!before && dn->dn_bonuslen != 0) {
1275 if (dn->dn_bonus) {
1276 db = dn->dn_bonus;
1277 mutex_enter(&db->db_mtx);
1278 data = dmu_objset_userquota_find_data(db, tx);
1279 } else {
1280 data = DN_BONUS(dn->dn_phys);
1281 }
1282 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1283 int rf = 0;
1284
1285 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1286 rf |= DB_RF_HAVESTRUCT;
1287 error = dmu_spill_hold_by_dnode(dn,
1288 rf | DB_RF_MUST_SUCCEED,
1289 FTAG, (dmu_buf_t **)&db);
1290 ASSERT(error == 0);
1291 mutex_enter(&db->db_mtx);
1292 data = (before) ? db->db.db_data :
1293 dmu_objset_userquota_find_data(db, tx);
1294 have_spill = B_TRUE;
1295 } else {
1296 mutex_enter(&dn->dn_mtx);
1297 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1298 mutex_exit(&dn->dn_mtx);
1299 return;
1300 }
1301
1302 if (before) {
1303 ASSERT(data);
1304 user = &dn->dn_olduid;
1305 group = &dn->dn_oldgid;
1306 } else if (data) {
1307 user = &dn->dn_newuid;
1308 group = &dn->dn_newgid;
1309 }
1310
1311 /*
1312 * Must always call the callback in case the object
1313 * type has changed and that type isn't an object type to track
1314 */
1315 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1316 user, group);
1317
1318 /*
1319 * Preserve existing uid/gid when the callback can't determine
1320 * what the new uid/gid are and the callback returned EEXIST.
1321 * The EEXIST error tells us to just use the existing uid/gid.
1322 * If we don't know what the old values are then just assign
1323 * them to 0, since that is a new file being created.
1324 */
1325 if (!before && data == NULL && error == EEXIST) {
1326 if (flags & DN_ID_OLD_EXIST) {
1327 dn->dn_newuid = dn->dn_olduid;
1328 dn->dn_newgid = dn->dn_oldgid;
1329 } else {
1330 dn->dn_newuid = 0;
1331 dn->dn_newgid = 0;
1332 }
1333 error = 0;
1334 }
1335
1336 if (db)
1337 mutex_exit(&db->db_mtx);
1338
1339 mutex_enter(&dn->dn_mtx);
1340 if (error == 0 && before)
1341 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1342 if (error == 0 && !before)
1343 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1344
1345 if (have_spill) {
1346 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1347 } else {
1348 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1349 }
1350 mutex_exit(&dn->dn_mtx);
1351 if (have_spill)
1352 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1353 }
1354
1355 boolean_t
1356 dmu_objset_userspace_present(objset_t *os)
1357 {
1358 return (os->os_phys->os_flags &
1359 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1360 }
1361
1362 int
1363 dmu_objset_userspace_upgrade(objset_t *os)
1364 {
1365 uint64_t obj;
1366 int err = 0;
1367
1368 if (dmu_objset_userspace_present(os))
1369 return (0);
1370 if (!dmu_objset_userused_enabled(os))
1371 return (SET_ERROR(ENOTSUP));
1372 if (dmu_objset_is_snapshot(os))
1373 return (SET_ERROR(EINVAL));
1374
1375 /*
1376 * We simply need to mark every object dirty, so that it will be
1377 * synced out and now accounted. If this is called
1378 * concurrently, or if we already did some work before crashing,
1379 * that's fine, since we track each object's accounted state
1380 * independently.
1381 */
1382
1383 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1384 dmu_tx_t *tx;
1385 dmu_buf_t *db;
1386 int objerr;
1387
1388 if (issig(JUSTLOOKING) && issig(FORREAL))
1389 return (SET_ERROR(EINTR));
1390
1391 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1392 if (objerr != 0)
1393 continue;
1394 tx = dmu_tx_create(os);
1395 dmu_tx_hold_bonus(tx, obj);
1396 objerr = dmu_tx_assign(tx, TXG_WAIT);
1397 if (objerr != 0) {
1398 dmu_tx_abort(tx);
1399 continue;
1400 }
1401 dmu_buf_will_dirty(db, tx);
1402 dmu_buf_rele(db, FTAG);
1403 dmu_tx_commit(tx);
1404 }
1405
1406 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1407 txg_wait_synced(dmu_objset_pool(os), 0);
1408 return (0);
1409 }
1410
1411 void
1412 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1413 uint64_t *usedobjsp, uint64_t *availobjsp)
1414 {
1415 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1416 usedobjsp, availobjsp);
1417 }
1418
1419 uint64_t
1420 dmu_objset_fsid_guid(objset_t *os)
1421 {
1422 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1423 }
1424
1425 void
1426 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1427 {
1428 stat->dds_type = os->os_phys->os_type;
1429 if (os->os_dsl_dataset)
1430 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1431 }
1432
1433 void
1434 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1435 {
1436 ASSERT(os->os_dsl_dataset ||
1437 os->os_phys->os_type == DMU_OST_META);
1438
1439 if (os->os_dsl_dataset != NULL)
1440 dsl_dataset_stats(os->os_dsl_dataset, nv);
1441
1442 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1443 os->os_phys->os_type);
1444 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1445 dmu_objset_userspace_present(os));
1446 }
1447
1448 int
1449 dmu_objset_is_snapshot(objset_t *os)
1450 {
1451 if (os->os_dsl_dataset != NULL)
1452 return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1453 else
1454 return (B_FALSE);
1455 }
1456
1457 int
1458 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1459 boolean_t *conflict)
1460 {
1461 dsl_dataset_t *ds = os->os_dsl_dataset;
1462 uint64_t ignored;
1463
1464 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1465 return (SET_ERROR(ENOENT));
1466
1467 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1468 ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1469 real, maxlen, conflict));
1470 }
1471
1472 int
1473 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1474 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1475 {
1476 dsl_dataset_t *ds = os->os_dsl_dataset;
1477 zap_cursor_t cursor;
1478 zap_attribute_t attr;
1479
1480 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1481
1482 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1483 return (SET_ERROR(ENOENT));
1484
1485 zap_cursor_init_serialized(&cursor,
1486 ds->ds_dir->dd_pool->dp_meta_objset,
1487 ds->ds_phys->ds_snapnames_zapobj, *offp);
1488
1489 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1490 zap_cursor_fini(&cursor);
1491 return (SET_ERROR(ENOENT));
1492 }
1493
1494 if (strlen(attr.za_name) + 1 > namelen) {
1495 zap_cursor_fini(&cursor);
1496 return (SET_ERROR(ENAMETOOLONG));
1497 }
1498
1499 (void) strcpy(name, attr.za_name);
1500 if (idp)
1501 *idp = attr.za_first_integer;
1502 if (case_conflict)
1503 *case_conflict = attr.za_normalization_conflict;
1504 zap_cursor_advance(&cursor);
1505 *offp = zap_cursor_serialize(&cursor);
1506 zap_cursor_fini(&cursor);
1507
1508 return (0);
1509 }
1510
1511 int
1512 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1513 uint64_t *idp, uint64_t *offp)
1514 {
1515 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1516 zap_cursor_t cursor;
1517 zap_attribute_t attr;
1518
1519 /* there is no next dir on a snapshot! */
1520 if (os->os_dsl_dataset->ds_object !=
1521 dd->dd_phys->dd_head_dataset_obj)
1522 return (SET_ERROR(ENOENT));
1523
1524 zap_cursor_init_serialized(&cursor,
1525 dd->dd_pool->dp_meta_objset,
1526 dd->dd_phys->dd_child_dir_zapobj, *offp);
1527
1528 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1529 zap_cursor_fini(&cursor);
1530 return (SET_ERROR(ENOENT));
1531 }
1532
1533 if (strlen(attr.za_name) + 1 > namelen) {
1534 zap_cursor_fini(&cursor);
1535 return (SET_ERROR(ENAMETOOLONG));
1536 }
1537
1538 (void) strcpy(name, attr.za_name);
1539 if (idp)
1540 *idp = attr.za_first_integer;
1541 zap_cursor_advance(&cursor);
1542 *offp = zap_cursor_serialize(&cursor);
1543 zap_cursor_fini(&cursor);
1544
1545 return (0);
1546 }
1547
1548 /*
1549 * Find objsets under and including ddobj, call func(ds) on each.
1550 */
1551 int
1552 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1553 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1554 {
1555 dsl_dir_t *dd;
1556 dsl_dataset_t *ds;
1557 zap_cursor_t zc;
1558 zap_attribute_t *attr;
1559 uint64_t thisobj;
1560 int err;
1561
1562 ASSERT(dsl_pool_config_held(dp));
1563
1564 err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1565 if (err != 0)
1566 return (err);
1567
1568 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1569 if (dd->dd_myname[0] == '$') {
1570 dsl_dir_rele(dd, FTAG);
1571 return (0);
1572 }
1573
1574 thisobj = dd->dd_phys->dd_head_dataset_obj;
1575 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1576
1577 /*
1578 * Iterate over all children.
1579 */
1580 if (flags & DS_FIND_CHILDREN) {
1581 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1582 dd->dd_phys->dd_child_dir_zapobj);
1583 zap_cursor_retrieve(&zc, attr) == 0;
1584 (void) zap_cursor_advance(&zc)) {
1585 ASSERT3U(attr->za_integer_length, ==,
1586 sizeof (uint64_t));
1587 ASSERT3U(attr->za_num_integers, ==, 1);
1588
1589 err = dmu_objset_find_dp(dp, attr->za_first_integer,
1590 func, arg, flags);
1591 if (err != 0)
1592 break;
1593 }
1594 zap_cursor_fini(&zc);
1595
1596 if (err != 0) {
1597 dsl_dir_rele(dd, FTAG);
1598 kmem_free(attr, sizeof (zap_attribute_t));
1599 return (err);
1600 }
1601 }
1602
1603 /*
1604 * Iterate over all snapshots.
1605 */
1606 if (flags & DS_FIND_SNAPSHOTS) {
1607 dsl_dataset_t *ds;
1608 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1609
1610 if (err == 0) {
1611 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1612 dsl_dataset_rele(ds, FTAG);
1613
1614 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1615 zap_cursor_retrieve(&zc, attr) == 0;
1616 (void) zap_cursor_advance(&zc)) {
1617 ASSERT3U(attr->za_integer_length, ==,
1618 sizeof (uint64_t));
1619 ASSERT3U(attr->za_num_integers, ==, 1);
1620
1621 err = dsl_dataset_hold_obj(dp,
1622 attr->za_first_integer, FTAG, &ds);
1623 if (err != 0)
1624 break;
1625 err = func(dp, ds, arg);
1626 dsl_dataset_rele(ds, FTAG);
1627 if (err != 0)
1628 break;
1629 }
1630 zap_cursor_fini(&zc);
1631 }
1632 }
1633
1634 dsl_dir_rele(dd, FTAG);
1635 kmem_free(attr, sizeof (zap_attribute_t));
1636
1637 if (err != 0)
1638 return (err);
1639
1640 /*
1641 * Apply to self.
1642 */
1643 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1644 if (err != 0)
1645 return (err);
1646 err = func(dp, ds, arg);
1647 dsl_dataset_rele(ds, FTAG);
1648 return (err);
1649 }
1650
1651 /*
1652 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1653 * The dp_config_rwlock must not be held when this is called, and it
1654 * will not be held when the callback is called.
1655 * Therefore this function should only be used when the pool is not changing
1656 * (e.g. in syncing context), or the callback can deal with the possible races.
1657 */
1658 static int
1659 dmu_objset_find_impl(spa_t *spa, const char *name,
1660 int func(const char *, void *), void *arg, int flags)
1661 {
1662 dsl_dir_t *dd;
1663 dsl_pool_t *dp = spa_get_dsl(spa);
1664 dsl_dataset_t *ds;
1665 zap_cursor_t zc;
1666 zap_attribute_t *attr;
1667 char *child;
1668 uint64_t thisobj;
1669 int err;
1670
1671 dsl_pool_config_enter(dp, FTAG);
1672
1673 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1674 if (err != 0) {
1675 dsl_pool_config_exit(dp, FTAG);
1676 return (err);
1677 }
1678
1679 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1680 if (dd->dd_myname[0] == '$') {
1681 dsl_dir_rele(dd, FTAG);
1682 dsl_pool_config_exit(dp, FTAG);
1683 return (0);
1684 }
1685
1686 thisobj = dd->dd_phys->dd_head_dataset_obj;
1687 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1688
1689 /*
1690 * Iterate over all children.
1691 */
1692 if (flags & DS_FIND_CHILDREN) {
1693 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1694 dd->dd_phys->dd_child_dir_zapobj);
1695 zap_cursor_retrieve(&zc, attr) == 0;
1696 (void) zap_cursor_advance(&zc)) {
1697 ASSERT3U(attr->za_integer_length, ==,
1698 sizeof (uint64_t));
1699 ASSERT3U(attr->za_num_integers, ==, 1);
1700
1701 child = kmem_asprintf("%s/%s", name, attr->za_name);
1702 dsl_pool_config_exit(dp, FTAG);
1703 err = dmu_objset_find_impl(spa, child,
1704 func, arg, flags);
1705 dsl_pool_config_enter(dp, FTAG);
1706 strfree(child);
1707 if (err != 0)
1708 break;
1709 }
1710 zap_cursor_fini(&zc);
1711
1712 if (err != 0) {
1713 dsl_dir_rele(dd, FTAG);
1714 dsl_pool_config_exit(dp, FTAG);
1715 kmem_free(attr, sizeof (zap_attribute_t));
1716 return (err);
1717 }
1718 }
1719
1720 /*
1721 * Iterate over all snapshots.
1722 */
1723 if (flags & DS_FIND_SNAPSHOTS) {
1724 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1725
1726 if (err == 0) {
1727 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1728 dsl_dataset_rele(ds, FTAG);
1729
1730 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1731 zap_cursor_retrieve(&zc, attr) == 0;
1732 (void) zap_cursor_advance(&zc)) {
1733 ASSERT3U(attr->za_integer_length, ==,
1734 sizeof (uint64_t));
1735 ASSERT3U(attr->za_num_integers, ==, 1);
1736
1737 child = kmem_asprintf("%s@%s",
1738 name, attr->za_name);
1739 dsl_pool_config_exit(dp, FTAG);
1740 err = func(child, arg);
1741 dsl_pool_config_enter(dp, FTAG);
1742 strfree(child);
1743 if (err != 0)
1744 break;
1745 }
1746 zap_cursor_fini(&zc);
1747 }
1748 }
1749
1750 dsl_dir_rele(dd, FTAG);
1751 kmem_free(attr, sizeof (zap_attribute_t));
1752 dsl_pool_config_exit(dp, FTAG);
1753
1754 if (err != 0)
1755 return (err);
1756
1757 /* Apply to self. */
1758 return (func(name, arg));
1759 }
1760
1761 /*
1762 * See comment above dmu_objset_find_impl().
1763 */
1764 int
1765 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1766 int flags)
1767 {
1768 spa_t *spa;
1769 int error;
1770
1771 error = spa_open(name, &spa, FTAG);
1772 if (error != 0)
1773 return (error);
1774 error = dmu_objset_find_impl(spa, name, func, arg, flags);
1775 spa_close(spa, FTAG);
1776 return (error);
1777 }
1778
1779 void
1780 dmu_objset_set_user(objset_t *os, void *user_ptr)
1781 {
1782 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1783 os->os_user_ptr = user_ptr;
1784 }
1785
1786 void *
1787 dmu_objset_get_user(objset_t *os)
1788 {
1789 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1790 return (os->os_user_ptr);
1791 }
1792
1793 /*
1794 * Determine name of filesystem, given name of snapshot.
1795 * buf must be at least MAXNAMELEN bytes
1796 */
1797 int
1798 dmu_fsname(const char *snapname, char *buf)
1799 {
1800 char *atp = strchr(snapname, '@');
1801 if (atp == NULL)
1802 return (SET_ERROR(EINVAL));
1803 if (atp - snapname >= MAXNAMELEN)
1804 return (SET_ERROR(ENAMETOOLONG));
1805 (void) strlcpy(buf, snapname, atp - snapname + 1);
1806 return (0);
1807 }