]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/dmu_objset.c
projects / mirror_zfs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
cstyle: Resolve C style issues
[mirror_zfs.git] / module / zfs / dmu_objset.c
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) 2013 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 */
26
27 /* Portions Copyright 2010 Robert Milkowski */
28
29 #include <sys/cred.h>
30 #include <sys/zfs_context.h>
31 #include <sys/dmu_objset.h>
32 #include <sys/dsl_dir.h>
33 #include <sys/dsl_dataset.h>
34 #include <sys/dsl_prop.h>
35 #include <sys/dsl_pool.h>
36 #include <sys/dsl_synctask.h>
37 #include <sys/dsl_deleg.h>
38 #include <sys/dnode.h>
39 #include <sys/dbuf.h>
40 #include <sys/zvol.h>
41 #include <sys/dmu_tx.h>
42 #include <sys/zap.h>
43 #include <sys/zil.h>
44 #include <sys/dmu_impl.h>
45 #include <sys/zfs_ioctl.h>
46 #include <sys/sa.h>
47 #include <sys/zfs_onexit.h>
48 #include <sys/dsl_destroy.h>
49
50 /*
51 * Needed to close a window in dnode_move() that allows the objset to be freed
52 * before it can be safely accessed.
53 */
54 krwlock_t os_lock;
55
56 void
57 dmu_objset_init(void)
58 {
59 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
60 }
61
62 void
63 dmu_objset_fini(void)
64 {
65 rw_destroy(&os_lock);
66 }
67
68 spa_t *
69 dmu_objset_spa(objset_t *os)
70 {
71 return (os->os_spa);
72 }
73
74 zilog_t *
75 dmu_objset_zil(objset_t *os)
76 {
77 return (os->os_zil);
78 }
79
80 dsl_pool_t *
81 dmu_objset_pool(objset_t *os)
82 {
83 dsl_dataset_t *ds;
84
85 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
86 return (ds->ds_dir->dd_pool);
87 else
88 return (spa_get_dsl(os->os_spa));
89 }
90
91 dsl_dataset_t *
92 dmu_objset_ds(objset_t *os)
93 {
94 return (os->os_dsl_dataset);
95 }
96
97 dmu_objset_type_t
98 dmu_objset_type(objset_t *os)
99 {
100 return (os->os_phys->os_type);
101 }
102
103 void
104 dmu_objset_name(objset_t *os, char *buf)
105 {
106 dsl_dataset_name(os->os_dsl_dataset, buf);
107 }
108
109 uint64_t
110 dmu_objset_id(objset_t *os)
111 {
112 dsl_dataset_t *ds = os->os_dsl_dataset;
113
114 return (ds ? ds->ds_object : 0);
115 }
116
117 uint64_t
118 dmu_objset_syncprop(objset_t *os)
119 {
120 return (os->os_sync);
121 }
122
123 uint64_t
124 dmu_objset_logbias(objset_t *os)
125 {
126 return (os->os_logbias);
127 }
128
129 static void
130 checksum_changed_cb(void *arg, uint64_t newval)
131 {
132 objset_t *os = arg;
133
134 /*
135 * Inheritance should have been done by now.
136 */
137 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
138
139 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
140 }
141
142 static void
143 compression_changed_cb(void *arg, uint64_t newval)
144 {
145 objset_t *os = arg;
146
147 /*
148 * Inheritance and range checking should have been done by now.
149 */
150 ASSERT(newval != ZIO_COMPRESS_INHERIT);
151
152 os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
153 }
154
155 static void
156 copies_changed_cb(void *arg, uint64_t newval)
157 {
158 objset_t *os = arg;
159
160 /*
161 * Inheritance and range checking should have been done by now.
162 */
163 ASSERT(newval > 0);
164 ASSERT(newval <= spa_max_replication(os->os_spa));
165
166 os->os_copies = newval;
167 }
168
169 static void
170 dedup_changed_cb(void *arg, uint64_t newval)
171 {
172 objset_t *os = arg;
173 spa_t *spa = os->os_spa;
174 enum zio_checksum checksum;
175
176 /*
177 * Inheritance should have been done by now.
178 */
179 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
180
181 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
182
183 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
184 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
185 }
186
187 static void
188 primary_cache_changed_cb(void *arg, uint64_t newval)
189 {
190 objset_t *os = arg;
191
192 /*
193 * Inheritance and range checking should have been done by now.
194 */
195 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
196 newval == ZFS_CACHE_METADATA);
197
198 os->os_primary_cache = newval;
199 }
200
201 static void
202 secondary_cache_changed_cb(void *arg, uint64_t newval)
203 {
204 objset_t *os = arg;
205
206 /*
207 * Inheritance and range checking should have been done by now.
208 */
209 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
210 newval == ZFS_CACHE_METADATA);
211
212 os->os_secondary_cache = newval;
213 }
214
215 static void
216 sync_changed_cb(void *arg, uint64_t newval)
217 {
218 objset_t *os = arg;
219
220 /*
221 * Inheritance and range checking should have been done by now.
222 */
223 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
224 newval == ZFS_SYNC_DISABLED);
225
226 os->os_sync = newval;
227 if (os->os_zil)
228 zil_set_sync(os->os_zil, newval);
229 }
230
231 static void
232 logbias_changed_cb(void *arg, uint64_t newval)
233 {
234 objset_t *os = arg;
235
236 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
237 newval == ZFS_LOGBIAS_THROUGHPUT);
238 os->os_logbias = newval;
239 if (os->os_zil)
240 zil_set_logbias(os->os_zil, newval);
241 }
242
243 void
244 dmu_objset_byteswap(void *buf, size_t size)
245 {
246 objset_phys_t *osp = buf;
247
248 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
249 dnode_byteswap(&osp->os_meta_dnode);
250 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
251 osp->os_type = BSWAP_64(osp->os_type);
252 osp->os_flags = BSWAP_64(osp->os_flags);
253 if (size == sizeof (objset_phys_t)) {
254 dnode_byteswap(&osp->os_userused_dnode);
255 dnode_byteswap(&osp->os_groupused_dnode);
256 }
257 }
258
259 int
260 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
261 objset_t **osp)
262 {
263 objset_t *os;
264 int i, err;
265
266 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
267
268 os = kmem_zalloc(sizeof (objset_t), KM_PUSHPAGE);
269 os->os_dsl_dataset = ds;
270 os->os_spa = spa;
271 os->os_rootbp = bp;
272 if (!BP_IS_HOLE(os->os_rootbp)) {
273 uint32_t aflags = ARC_WAIT;
274 zbookmark_t zb;
275 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
276 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
277
278 if (DMU_OS_IS_L2CACHEABLE(os))
279 aflags |= ARC_L2CACHE;
280 if (DMU_OS_IS_L2COMPRESSIBLE(os))
281 aflags |= ARC_L2COMPRESS;
282
283 dprintf_bp(os->os_rootbp, "reading %s", "");
284 err = arc_read(NULL, spa, os->os_rootbp,
285 arc_getbuf_func, &os->os_phys_buf,
286 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
287 if (err != 0) {
288 kmem_free(os, sizeof (objset_t));
289 /* convert checksum errors into IO errors */
290 if (err == ECKSUM)
291 err = SET_ERROR(EIO);
292 return (err);
293 }
294
295 /* Increase the blocksize if we are permitted. */
296 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
297 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
298 arc_buf_t *buf = arc_buf_alloc(spa,
299 sizeof (objset_phys_t), &os->os_phys_buf,
300 ARC_BUFC_METADATA);
301 bzero(buf->b_data, sizeof (objset_phys_t));
302 bcopy(os->os_phys_buf->b_data, buf->b_data,
303 arc_buf_size(os->os_phys_buf));
304 (void) arc_buf_remove_ref(os->os_phys_buf,
305 &os->os_phys_buf);
306 os->os_phys_buf = buf;
307 }
308
309 os->os_phys = os->os_phys_buf->b_data;
310 os->os_flags = os->os_phys->os_flags;
311 } else {
312 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
313 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
314 os->os_phys_buf = arc_buf_alloc(spa, size,
315 &os->os_phys_buf, ARC_BUFC_METADATA);
316 os->os_phys = os->os_phys_buf->b_data;
317 bzero(os->os_phys, size);
318 }
319
320 /*
321 * Note: the changed_cb will be called once before the register
322 * func returns, thus changing the checksum/compression from the
323 * default (fletcher2/off). Snapshots don't need to know about
324 * checksum/compression/copies.
325 */
326 if (ds) {
327 err = dsl_prop_register(ds,
328 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
329 primary_cache_changed_cb, os);
330 if (err == 0) {
331 err = dsl_prop_register(ds,
332 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
333 secondary_cache_changed_cb, os);
334 }
335 if (!dsl_dataset_is_snapshot(ds)) {
336 if (err == 0) {
337 err = dsl_prop_register(ds,
338 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
339 checksum_changed_cb, os);
340 }
341 if (err == 0) {
342 err = dsl_prop_register(ds,
343 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
344 compression_changed_cb, os);
345 }
346 if (err == 0) {
347 err = dsl_prop_register(ds,
348 zfs_prop_to_name(ZFS_PROP_COPIES),
349 copies_changed_cb, os);
350 }
351 if (err == 0) {
352 err = dsl_prop_register(ds,
353 zfs_prop_to_name(ZFS_PROP_DEDUP),
354 dedup_changed_cb, os);
355 }
356 if (err == 0) {
357 err = dsl_prop_register(ds,
358 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
359 logbias_changed_cb, os);
360 }
361 if (err == 0) {
362 err = dsl_prop_register(ds,
363 zfs_prop_to_name(ZFS_PROP_SYNC),
364 sync_changed_cb, os);
365 }
366 }
367 if (err != 0) {
368 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
369 &os->os_phys_buf));
370 kmem_free(os, sizeof (objset_t));
371 return (err);
372 }
373 } else if (ds == NULL) {
374 /* It's the meta-objset. */
375 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
376 os->os_compress = ZIO_COMPRESS_LZJB;
377 os->os_copies = spa_max_replication(spa);
378 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
379 os->os_dedup_verify = 0;
380 os->os_logbias = 0;
381 os->os_sync = 0;
382 os->os_primary_cache = ZFS_CACHE_ALL;
383 os->os_secondary_cache = ZFS_CACHE_ALL;
384 }
385
386 if (ds == NULL || !dsl_dataset_is_snapshot(ds))
387 os->os_zil_header = os->os_phys->os_zil_header;
388 os->os_zil = zil_alloc(os, &os->os_zil_header);
389
390 for (i = 0; i < TXG_SIZE; i++) {
391 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
392 offsetof(dnode_t, dn_dirty_link[i]));
393 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
394 offsetof(dnode_t, dn_dirty_link[i]));
395 }
396 list_create(&os->os_dnodes, sizeof (dnode_t),
397 offsetof(dnode_t, dn_link));
398 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
399 offsetof(dmu_buf_impl_t, db_link));
400
401 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
402 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
403 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
404
405 DMU_META_DNODE(os) = dnode_special_open(os,
406 &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
407 &os->os_meta_dnode);
408 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
409 DMU_USERUSED_DNODE(os) = dnode_special_open(os,
410 &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
411 &os->os_userused_dnode);
412 DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
413 &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
414 &os->os_groupused_dnode);
415 }
416
417 /*
418 * We should be the only thread trying to do this because we
419 * have ds_opening_lock
420 */
421 if (ds) {
422 mutex_enter(&ds->ds_lock);
423 ASSERT(ds->ds_objset == NULL);
424 ds->ds_objset = os;
425 mutex_exit(&ds->ds_lock);
426 }
427
428 *osp = os;
429 return (0);
430 }
431
432 int
433 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
434 {
435 int err = 0;
436
437 mutex_enter(&ds->ds_opening_lock);
438 *osp = ds->ds_objset;
439 if (*osp == NULL) {
440 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
441 ds, dsl_dataset_get_blkptr(ds), osp);
442 }
443 mutex_exit(&ds->ds_opening_lock);
444 return (err);
445 }
446
447 /*
448 * Holds the pool while the objset is held. Therefore only one objset
449 * can be held at a time.
450 */
451 int
452 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
453 {
454 dsl_pool_t *dp;
455 dsl_dataset_t *ds;
456 int err;
457
458 err = dsl_pool_hold(name, tag, &dp);
459 if (err != 0)
460 return (err);
461 err = dsl_dataset_hold(dp, name, tag, &ds);
462 if (err != 0) {
463 dsl_pool_rele(dp, tag);
464 return (err);
465 }
466
467 err = dmu_objset_from_ds(ds, osp);
468 if (err != 0) {
469 dsl_dataset_rele(ds, tag);
470 dsl_pool_rele(dp, tag);
471 }
472
473 return (err);
474 }
475
476 /*
477 * dsl_pool must not be held when this is called.
478 * Upon successful return, there will be a longhold on the dataset,
479 * and the dsl_pool will not be held.
480 */
481 int
482 dmu_objset_own(const char *name, dmu_objset_type_t type,
483 boolean_t readonly, void *tag, objset_t **osp)
484 {
485 dsl_pool_t *dp;
486 dsl_dataset_t *ds;
487 int err;
488
489 err = dsl_pool_hold(name, FTAG, &dp);
490 if (err != 0)
491 return (err);
492 err = dsl_dataset_own(dp, name, tag, &ds);
493 if (err != 0) {
494 dsl_pool_rele(dp, FTAG);
495 return (err);
496 }
497
498 err = dmu_objset_from_ds(ds, osp);
499 dsl_pool_rele(dp, FTAG);
500 if (err != 0) {
501 dsl_dataset_disown(ds, tag);
502 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
503 dsl_dataset_disown(ds, tag);
504 return (SET_ERROR(EINVAL));
505 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
506 dsl_dataset_disown(ds, tag);
507 return (SET_ERROR(EROFS));
508 }
509 return (err);
510 }
511
512 void
513 dmu_objset_rele(objset_t *os, void *tag)
514 {
515 dsl_pool_t *dp = dmu_objset_pool(os);
516 dsl_dataset_rele(os->os_dsl_dataset, tag);
517 dsl_pool_rele(dp, tag);
518 }
519
520 /*
521 * When we are called, os MUST refer to an objset associated with a dataset
522 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
523 * == tag. We will then release and reacquire ownership of the dataset while
524 * holding the pool config_rwlock to avoid intervening namespace or ownership
525 * changes may occur.
526 *
527 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
528 * release the hold on its dataset and acquire a new one on the dataset of the
529 * same name so that it can be partially torn down and reconstructed.
530 */
531 void
532 dmu_objset_refresh_ownership(objset_t *os, void *tag)
533 {
534 dsl_pool_t *dp;
535 dsl_dataset_t *ds, *newds;
536 char name[MAXNAMELEN];
537
538 ds = os->os_dsl_dataset;
539 VERIFY3P(ds, !=, NULL);
540 VERIFY3P(ds->ds_owner, ==, tag);
541 VERIFY(dsl_dataset_long_held(ds));
542
543 dsl_dataset_name(ds, name);
544 dp = dmu_objset_pool(os);
545 dsl_pool_config_enter(dp, FTAG);
546 dmu_objset_disown(os, tag);
547 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
548 VERIFY3P(newds, ==, os->os_dsl_dataset);
549 dsl_pool_config_exit(dp, FTAG);
550 }
551
552 void
553 dmu_objset_disown(objset_t *os, void *tag)
554 {
555 dsl_dataset_disown(os->os_dsl_dataset, tag);
556 }
557
558 void
559 dmu_objset_evict_dbufs(objset_t *os)
560 {
561 dnode_t *dn;
562
563 mutex_enter(&os->os_lock);
564
565 /* process the mdn last, since the other dnodes have holds on it */
566 list_remove(&os->os_dnodes, DMU_META_DNODE(os));
567 list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
568
569 /*
570 * Find the first dnode with holds. We have to do this dance
571 * because dnode_add_ref() only works if you already have a
572 * hold. If there are no holds then it has no dbufs so OK to
573 * skip.
574 */
575 for (dn = list_head(&os->os_dnodes);
576 dn && !dnode_add_ref(dn, FTAG);
577 dn = list_next(&os->os_dnodes, dn))
578 continue;
579
580 while (dn) {
581 dnode_t *next_dn = dn;
582
583 do {
584 next_dn = list_next(&os->os_dnodes, next_dn);
585 } while (next_dn && !dnode_add_ref(next_dn, FTAG));
586
587 mutex_exit(&os->os_lock);
588 dnode_evict_dbufs(dn);
589 dnode_rele(dn, FTAG);
590 mutex_enter(&os->os_lock);
591 dn = next_dn;
592 }
593 mutex_exit(&os->os_lock);
594 }
595
596 void
597 dmu_objset_evict(objset_t *os)
598 {
599 int t;
600
601 dsl_dataset_t *ds = os->os_dsl_dataset;
602
603 for (t = 0; t < TXG_SIZE; t++)
604 ASSERT(!dmu_objset_is_dirty(os, t));
605
606 if (ds) {
607 if (!dsl_dataset_is_snapshot(ds)) {
608 VERIFY0(dsl_prop_unregister(ds,
609 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
610 checksum_changed_cb, os));
611 VERIFY0(dsl_prop_unregister(ds,
612 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
613 compression_changed_cb, os));
614 VERIFY0(dsl_prop_unregister(ds,
615 zfs_prop_to_name(ZFS_PROP_COPIES),
616 copies_changed_cb, os));
617 VERIFY0(dsl_prop_unregister(ds,
618 zfs_prop_to_name(ZFS_PROP_DEDUP),
619 dedup_changed_cb, os));
620 VERIFY0(dsl_prop_unregister(ds,
621 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
622 logbias_changed_cb, os));
623 VERIFY0(dsl_prop_unregister(ds,
624 zfs_prop_to_name(ZFS_PROP_SYNC),
625 sync_changed_cb, os));
626 }
627 VERIFY0(dsl_prop_unregister(ds,
628 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
629 primary_cache_changed_cb, os));
630 VERIFY0(dsl_prop_unregister(ds,
631 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
632 secondary_cache_changed_cb, os));
633 }
634
635 if (os->os_sa)
636 sa_tear_down(os);
637
638 dmu_objset_evict_dbufs(os);
639
640 dnode_special_close(&os->os_meta_dnode);
641 if (DMU_USERUSED_DNODE(os)) {
642 dnode_special_close(&os->os_userused_dnode);
643 dnode_special_close(&os->os_groupused_dnode);
644 }
645 zil_free(os->os_zil);
646
647 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
648
649 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
650
651 /*
652 * This is a barrier to prevent the objset from going away in
653 * dnode_move() until we can safely ensure that the objset is still in
654 * use. We consider the objset valid before the barrier and invalid
655 * after the barrier.
656 */
657 rw_enter(&os_lock, RW_READER);
658 rw_exit(&os_lock);
659
660 mutex_destroy(&os->os_lock);
661 mutex_destroy(&os->os_obj_lock);
662 mutex_destroy(&os->os_user_ptr_lock);
663 kmem_free(os, sizeof (objset_t));
664 }
665
666 timestruc_t
667 dmu_objset_snap_cmtime(objset_t *os)
668 {
669 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
670 }
671
672 /* called from dsl for meta-objset */
673 objset_t *
674 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
675 dmu_objset_type_t type, dmu_tx_t *tx)
676 {
677 objset_t *os;
678 dnode_t *mdn;
679
680 ASSERT(dmu_tx_is_syncing(tx));
681
682 if (ds != NULL)
683 VERIFY0(dmu_objset_from_ds(ds, &os));
684 else
685 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
686
687 mdn = DMU_META_DNODE(os);
688
689 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
690 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
691
692 /*
693 * We don't want to have to increase the meta-dnode's nlevels
694 * later, because then we could do it in quescing context while
695 * we are also accessing it in open context.
696 *
697 * This precaution is not necessary for the MOS (ds == NULL),
698 * because the MOS is only updated in syncing context.
699 * This is most fortunate: the MOS is the only objset that
700 * needs to be synced multiple times as spa_sync() iterates
701 * to convergence, so minimizing its dn_nlevels matters.
702 */
703 if (ds != NULL) {
704 int levels = 1;
705
706 /*
707 * Determine the number of levels necessary for the meta-dnode
708 * to contain DN_MAX_OBJECT dnodes.
709 */
710 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
711 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
712 DN_MAX_OBJECT * sizeof (dnode_phys_t))
713 levels++;
714
715 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
716 mdn->dn_nlevels = levels;
717 }
718
719 ASSERT(type != DMU_OST_NONE);
720 ASSERT(type != DMU_OST_ANY);
721 ASSERT(type < DMU_OST_NUMTYPES);
722 os->os_phys->os_type = type;
723 if (dmu_objset_userused_enabled(os)) {
724 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
725 os->os_flags = os->os_phys->os_flags;
726 }
727
728 dsl_dataset_dirty(ds, tx);
729
730 return (os);
731 }
732
733 typedef struct dmu_objset_create_arg {
734 const char *doca_name;
735 cred_t *doca_cred;
736 void (*doca_userfunc)(objset_t *os, void *arg,
737 cred_t *cr, dmu_tx_t *tx);
738 void *doca_userarg;
739 dmu_objset_type_t doca_type;
740 uint64_t doca_flags;
741 } dmu_objset_create_arg_t;
742
743 /*ARGSUSED*/
744 static int
745 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
746 {
747 dmu_objset_create_arg_t *doca = arg;
748 dsl_pool_t *dp = dmu_tx_pool(tx);
749 dsl_dir_t *pdd;
750 const char *tail;
751 int error;
752
753 if (strchr(doca->doca_name, '@') != NULL)
754 return (SET_ERROR(EINVAL));
755
756 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
757 if (error != 0)
758 return (error);
759 if (tail == NULL) {
760 dsl_dir_rele(pdd, FTAG);
761 return (SET_ERROR(EEXIST));
762 }
763 dsl_dir_rele(pdd, FTAG);
764
765 return (0);
766 }
767
768 static void
769 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
770 {
771 dmu_objset_create_arg_t *doca = arg;
772 dsl_pool_t *dp = dmu_tx_pool(tx);
773 dsl_dir_t *pdd;
774 const char *tail;
775 dsl_dataset_t *ds;
776 uint64_t obj;
777 blkptr_t *bp;
778 objset_t *os;
779
780 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
781
782 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
783 doca->doca_cred, tx);
784
785 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
786 bp = dsl_dataset_get_blkptr(ds);
787 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
788 ds, bp, doca->doca_type, tx);
789
790 if (doca->doca_userfunc != NULL) {
791 doca->doca_userfunc(os, doca->doca_userarg,
792 doca->doca_cred, tx);
793 }
794
795 spa_history_log_internal_ds(ds, "create", tx, "");
796 dsl_dataset_rele(ds, FTAG);
797 dsl_dir_rele(pdd, FTAG);
798 }
799
800 int
801 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
802 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
803 {
804 dmu_objset_create_arg_t doca;
805
806 doca.doca_name = name;
807 doca.doca_cred = CRED();
808 doca.doca_flags = flags;
809 doca.doca_userfunc = func;
810 doca.doca_userarg = arg;
811 doca.doca_type = type;
812
813 return (dsl_sync_task(name,
814 dmu_objset_create_check, dmu_objset_create_sync, &doca, 5));
815 }
816
817 typedef struct dmu_objset_clone_arg {
818 const char *doca_clone;
819 const char *doca_origin;
820 cred_t *doca_cred;
821 } dmu_objset_clone_arg_t;
822
823 /*ARGSUSED*/
824 static int
825 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
826 {
827 dmu_objset_clone_arg_t *doca = arg;
828 dsl_dir_t *pdd;
829 const char *tail;
830 int error;
831 dsl_dataset_t *origin;
832 dsl_pool_t *dp = dmu_tx_pool(tx);
833
834 if (strchr(doca->doca_clone, '@') != NULL)
835 return (SET_ERROR(EINVAL));
836
837 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
838 if (error != 0)
839 return (error);
840 if (tail == NULL) {
841 dsl_dir_rele(pdd, FTAG);
842 return (SET_ERROR(EEXIST));
843 }
844 /* You can't clone across pools. */
845 if (pdd->dd_pool != dp) {
846 dsl_dir_rele(pdd, FTAG);
847 return (SET_ERROR(EXDEV));
848 }
849 dsl_dir_rele(pdd, FTAG);
850
851 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
852 if (error != 0)
853 return (error);
854
855 /* You can't clone across pools. */
856 if (origin->ds_dir->dd_pool != dp) {
857 dsl_dataset_rele(origin, FTAG);
858 return (SET_ERROR(EXDEV));
859 }
860
861 /* You can only clone snapshots, not the head datasets. */
862 if (!dsl_dataset_is_snapshot(origin)) {
863 dsl_dataset_rele(origin, FTAG);
864 return (SET_ERROR(EINVAL));
865 }
866 dsl_dataset_rele(origin, FTAG);
867
868 return (0);
869 }
870
871 static void
872 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
873 {
874 dmu_objset_clone_arg_t *doca = arg;
875 dsl_pool_t *dp = dmu_tx_pool(tx);
876 dsl_dir_t *pdd;
877 const char *tail;
878 dsl_dataset_t *origin, *ds;
879 uint64_t obj;
880 char namebuf[MAXNAMELEN];
881
882 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
883 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
884
885 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
886 doca->doca_cred, tx);
887
888 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
889 dsl_dataset_name(origin, namebuf);
890 spa_history_log_internal_ds(ds, "clone", tx,
891 "origin=%s (%llu)", namebuf, origin->ds_object);
892 dsl_dataset_rele(ds, FTAG);
893 dsl_dataset_rele(origin, FTAG);
894 dsl_dir_rele(pdd, FTAG);
895 }
896
897 int
898 dmu_objset_clone(const char *clone, const char *origin)
899 {
900 dmu_objset_clone_arg_t doca;
901
902 doca.doca_clone = clone;
903 doca.doca_origin = origin;
904 doca.doca_cred = CRED();
905
906 return (dsl_sync_task(clone,
907 dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 5));
908 }
909
910 int
911 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
912 {
913 int err;
914 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
915 nvlist_t *snaps = fnvlist_alloc();
916
917 fnvlist_add_boolean(snaps, longsnap);
918 strfree(longsnap);
919 err = dsl_dataset_snapshot(snaps, NULL, NULL);
920 fnvlist_free(snaps);
921 return (err);
922 }
923
924 static void
925 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
926 {
927 dnode_t *dn;
928
929 while ((dn = list_head(list))) {
930 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
931 ASSERT(dn->dn_dbuf->db_data_pending);
932 /*
933 * Initialize dn_zio outside dnode_sync() because the
934 * meta-dnode needs to set it ouside dnode_sync().
935 */
936 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
937 ASSERT(dn->dn_zio);
938
939 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
940 list_remove(list, dn);
941
942 if (newlist) {
943 (void) dnode_add_ref(dn, newlist);
944 list_insert_tail(newlist, dn);
945 }
946
947 dnode_sync(dn, tx);
948 }
949 }
950
951 /* ARGSUSED */
952 static void
953 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
954 {
955 int i;
956
957 blkptr_t *bp = zio->io_bp;
958 objset_t *os = arg;
959 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
960
961 ASSERT3P(bp, ==, os->os_rootbp);
962 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
963 ASSERT0(BP_GET_LEVEL(bp));
964
965 /*
966 * Update rootbp fill count: it should be the number of objects
967 * allocated in the object set (not counting the "special"
968 * objects that are stored in the objset_phys_t -- the meta
969 * dnode and user/group accounting objects).
970 */
971 bp->blk_fill = 0;
972 for (i = 0; i < dnp->dn_nblkptr; i++)
973 bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
974 }
975
976 /* ARGSUSED */
977 static void
978 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
979 {
980 blkptr_t *bp = zio->io_bp;
981 blkptr_t *bp_orig = &zio->io_bp_orig;
982 objset_t *os = arg;
983
984 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
985 ASSERT(BP_EQUAL(bp, bp_orig));
986 } else {
987 dsl_dataset_t *ds = os->os_dsl_dataset;
988 dmu_tx_t *tx = os->os_synctx;
989
990 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
991 dsl_dataset_block_born(ds, bp, tx);
992 }
993 }
994
995 /* called from dsl */
996 void
997 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
998 {
999 int txgoff;
1000 zbookmark_t zb;
1001 zio_prop_t zp;
1002 zio_t *zio;
1003 list_t *list;
1004 list_t *newlist = NULL;
1005 dbuf_dirty_record_t *dr;
1006
1007 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1008
1009 ASSERT(dmu_tx_is_syncing(tx));
1010 /* XXX the write_done callback should really give us the tx... */
1011 os->os_synctx = tx;
1012
1013 if (os->os_dsl_dataset == NULL) {
1014 /*
1015 * This is the MOS. If we have upgraded,
1016 * spa_max_replication() could change, so reset
1017 * os_copies here.
1018 */
1019 os->os_copies = spa_max_replication(os->os_spa);
1020 }
1021
1022 /*
1023 * Create the root block IO
1024 */
1025 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1026 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1027 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1028 arc_release(os->os_phys_buf, &os->os_phys_buf);
1029
1030 dmu_write_policy(os, NULL, 0, 0, &zp);
1031
1032 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1033 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1034 DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1035 NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1036 ZIO_FLAG_MUSTSUCCEED, &zb);
1037
1038 /*
1039 * Sync special dnodes - the parent IO for the sync is the root block
1040 */
1041 DMU_META_DNODE(os)->dn_zio = zio;
1042 dnode_sync(DMU_META_DNODE(os), tx);
1043
1044 os->os_phys->os_flags = os->os_flags;
1045
1046 if (DMU_USERUSED_DNODE(os) &&
1047 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1048 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1049 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1050 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1051 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1052 }
1053
1054 txgoff = tx->tx_txg & TXG_MASK;
1055
1056 if (dmu_objset_userused_enabled(os)) {
1057 newlist = &os->os_synced_dnodes;
1058 /*
1059 * We must create the list here because it uses the
1060 * dn_dirty_link[] of this txg.
1061 */
1062 list_create(newlist, sizeof (dnode_t),
1063 offsetof(dnode_t, dn_dirty_link[txgoff]));
1064 }
1065
1066 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1067 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1068
1069 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1070 while ((dr = list_head(list))) {
1071 ASSERT0(dr->dr_dbuf->db_level);
1072 list_remove(list, dr);
1073 if (dr->dr_zio)
1074 zio_nowait(dr->dr_zio);
1075 }
1076 /*
1077 * Free intent log blocks up to this tx.
1078 */
1079 zil_sync(os->os_zil, tx);
1080 os->os_phys->os_zil_header = os->os_zil_header;
1081 zio_nowait(zio);
1082 }
1083
1084 boolean_t
1085 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1086 {
1087 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1088 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1089 }
1090
1091 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1092
1093 void
1094 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1095 {
1096 used_cbs[ost] = cb;
1097 }
1098
1099 boolean_t
1100 dmu_objset_userused_enabled(objset_t *os)
1101 {
1102 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1103 used_cbs[os->os_phys->os_type] != NULL &&
1104 DMU_USERUSED_DNODE(os) != NULL);
1105 }
1106
1107 static void
1108 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1109 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1110 {
1111 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1112 int64_t delta = DNODE_SIZE + used;
1113 if (subtract)
1114 delta = -delta;
1115 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1116 user, delta, tx));
1117 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1118 group, delta, tx));
1119 }
1120 }
1121
1122 void
1123 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1124 {
1125 dnode_t *dn;
1126 list_t *list = &os->os_synced_dnodes;
1127
1128 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1129
1130 while ((dn = list_head(list))) {
1131 int flags;
1132 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1133 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1134 dn->dn_phys->dn_flags &
1135 DNODE_FLAG_USERUSED_ACCOUNTED);
1136
1137 /* Allocate the user/groupused objects if necessary. */
1138 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1139 VERIFY(0 == zap_create_claim(os,
1140 DMU_USERUSED_OBJECT,
1141 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1142 VERIFY(0 == zap_create_claim(os,
1143 DMU_GROUPUSED_OBJECT,
1144 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1145 }
1146
1147 /*
1148 * We intentionally modify the zap object even if the
1149 * net delta is zero. Otherwise
1150 * the block of the zap obj could be shared between
1151 * datasets but need to be different between them after
1152 * a bprewrite.
1153 */
1154
1155 flags = dn->dn_id_flags;
1156 ASSERT(flags);
1157 if (flags & DN_ID_OLD_EXIST) {
1158 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1159 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1160 }
1161 if (flags & DN_ID_NEW_EXIST) {
1162 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1163 dn->dn_phys->dn_flags, dn->dn_newuid,
1164 dn->dn_newgid, B_FALSE, tx);
1165 }
1166
1167 mutex_enter(&dn->dn_mtx);
1168 dn->dn_oldused = 0;
1169 dn->dn_oldflags = 0;
1170 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1171 dn->dn_olduid = dn->dn_newuid;
1172 dn->dn_oldgid = dn->dn_newgid;
1173 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1174 if (dn->dn_bonuslen == 0)
1175 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1176 else
1177 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1178 }
1179 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1180 mutex_exit(&dn->dn_mtx);
1181
1182 list_remove(list, dn);
1183 dnode_rele(dn, list);
1184 }
1185 }
1186
1187 /*
1188 * Returns a pointer to data to find uid/gid from
1189 *
1190 * If a dirty record for transaction group that is syncing can't
1191 * be found then NULL is returned. In the NULL case it is assumed
1192 * the uid/gid aren't changing.
1193 */
1194 static void *
1195 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1196 {
1197 dbuf_dirty_record_t *dr, **drp;
1198 void *data;
1199
1200 if (db->db_dirtycnt == 0)
1201 return (db->db.db_data); /* Nothing is changing */
1202
1203 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1204 if (dr->dr_txg == tx->tx_txg)
1205 break;
1206
1207 if (dr == NULL) {
1208 data = NULL;
1209 } else {
1210 dnode_t *dn;
1211
1212 DB_DNODE_ENTER(dr->dr_dbuf);
1213 dn = DB_DNODE(dr->dr_dbuf);
1214
1215 if (dn->dn_bonuslen == 0 &&
1216 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1217 data = dr->dt.dl.dr_data->b_data;
1218 else
1219 data = dr->dt.dl.dr_data;
1220
1221 DB_DNODE_EXIT(dr->dr_dbuf);
1222 }
1223
1224 return (data);
1225 }
1226
1227 void
1228 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1229 {
1230 objset_t *os = dn->dn_objset;
1231 void *data = NULL;
1232 dmu_buf_impl_t *db = NULL;
1233 uint64_t *user = NULL;
1234 uint64_t *group = NULL;
1235 int flags = dn->dn_id_flags;
1236 int error;
1237 boolean_t have_spill = B_FALSE;
1238
1239 if (!dmu_objset_userused_enabled(dn->dn_objset))
1240 return;
1241
1242 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1243 DN_ID_CHKED_SPILL)))
1244 return;
1245
1246 if (before && dn->dn_bonuslen != 0)
1247 data = DN_BONUS(dn->dn_phys);
1248 else if (!before && dn->dn_bonuslen != 0) {
1249 if (dn->dn_bonus) {
1250 db = dn->dn_bonus;
1251 mutex_enter(&db->db_mtx);
1252 data = dmu_objset_userquota_find_data(db, tx);
1253 } else {
1254 data = DN_BONUS(dn->dn_phys);
1255 }
1256 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1257 int rf = 0;
1258
1259 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1260 rf |= DB_RF_HAVESTRUCT;
1261 error = dmu_spill_hold_by_dnode(dn,
1262 rf | DB_RF_MUST_SUCCEED,
1263 FTAG, (dmu_buf_t **)&db);
1264 ASSERT(error == 0);
1265 mutex_enter(&db->db_mtx);
1266 data = (before) ? db->db.db_data :
1267 dmu_objset_userquota_find_data(db, tx);
1268 have_spill = B_TRUE;
1269 } else {
1270 mutex_enter(&dn->dn_mtx);
1271 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1272 mutex_exit(&dn->dn_mtx);
1273 return;
1274 }
1275
1276 if (before) {
1277 ASSERT(data);
1278 user = &dn->dn_olduid;
1279 group = &dn->dn_oldgid;
1280 } else if (data) {
1281 user = &dn->dn_newuid;
1282 group = &dn->dn_newgid;
1283 }
1284
1285 /*
1286 * Must always call the callback in case the object
1287 * type has changed and that type isn't an object type to track
1288 */
1289 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1290 user, group);
1291
1292 /*
1293 * Preserve existing uid/gid when the callback can't determine
1294 * what the new uid/gid are and the callback returned EEXIST.
1295 * The EEXIST error tells us to just use the existing uid/gid.
1296 * If we don't know what the old values are then just assign
1297 * them to 0, since that is a new file being created.
1298 */
1299 if (!before && data == NULL && error == EEXIST) {
1300 if (flags & DN_ID_OLD_EXIST) {
1301 dn->dn_newuid = dn->dn_olduid;
1302 dn->dn_newgid = dn->dn_oldgid;
1303 } else {
1304 dn->dn_newuid = 0;
1305 dn->dn_newgid = 0;
1306 }
1307 error = 0;
1308 }
1309
1310 if (db)
1311 mutex_exit(&db->db_mtx);
1312
1313 mutex_enter(&dn->dn_mtx);
1314 if (error == 0 && before)
1315 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1316 if (error == 0 && !before)
1317 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1318
1319 if (have_spill) {
1320 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1321 } else {
1322 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1323 }
1324 mutex_exit(&dn->dn_mtx);
1325 if (have_spill)
1326 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1327 }
1328
1329 boolean_t
1330 dmu_objset_userspace_present(objset_t *os)
1331 {
1332 return (os->os_phys->os_flags &
1333 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1334 }
1335
1336 int
1337 dmu_objset_userspace_upgrade(objset_t *os)
1338 {
1339 uint64_t obj;
1340 int err = 0;
1341
1342 if (dmu_objset_userspace_present(os))
1343 return (0);
1344 if (!dmu_objset_userused_enabled(os))
1345 return (SET_ERROR(ENOTSUP));
1346 if (dmu_objset_is_snapshot(os))
1347 return (SET_ERROR(EINVAL));
1348
1349 /*
1350 * We simply need to mark every object dirty, so that it will be
1351 * synced out and now accounted. If this is called
1352 * concurrently, or if we already did some work before crashing,
1353 * that's fine, since we track each object's accounted state
1354 * independently.
1355 */
1356
1357 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1358 dmu_tx_t *tx;
1359 dmu_buf_t *db;
1360 int objerr;
1361
1362 if (issig(JUSTLOOKING) && issig(FORREAL))
1363 return (SET_ERROR(EINTR));
1364
1365 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1366 if (objerr != 0)
1367 continue;
1368 tx = dmu_tx_create(os);
1369 dmu_tx_hold_bonus(tx, obj);
1370 objerr = dmu_tx_assign(tx, TXG_WAIT);
1371 if (objerr != 0) {
1372 dmu_tx_abort(tx);
1373 continue;
1374 }
1375 dmu_buf_will_dirty(db, tx);
1376 dmu_buf_rele(db, FTAG);
1377 dmu_tx_commit(tx);
1378 }
1379
1380 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1381 txg_wait_synced(dmu_objset_pool(os), 0);
1382 return (0);
1383 }
1384
1385 void
1386 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1387 uint64_t *usedobjsp, uint64_t *availobjsp)
1388 {
1389 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1390 usedobjsp, availobjsp);
1391 }
1392
1393 uint64_t
1394 dmu_objset_fsid_guid(objset_t *os)
1395 {
1396 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1397 }
1398
1399 void
1400 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1401 {
1402 stat->dds_type = os->os_phys->os_type;
1403 if (os->os_dsl_dataset)
1404 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1405 }
1406
1407 void
1408 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1409 {
1410 ASSERT(os->os_dsl_dataset ||
1411 os->os_phys->os_type == DMU_OST_META);
1412
1413 if (os->os_dsl_dataset != NULL)
1414 dsl_dataset_stats(os->os_dsl_dataset, nv);
1415
1416 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1417 os->os_phys->os_type);
1418 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1419 dmu_objset_userspace_present(os));
1420 }
1421
1422 int
1423 dmu_objset_is_snapshot(objset_t *os)
1424 {
1425 if (os->os_dsl_dataset != NULL)
1426 return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1427 else
1428 return (B_FALSE);
1429 }
1430
1431 int
1432 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1433 boolean_t *conflict)
1434 {
1435 dsl_dataset_t *ds = os->os_dsl_dataset;
1436 uint64_t ignored;
1437
1438 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1439 return (SET_ERROR(ENOENT));
1440
1441 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1442 ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1443 real, maxlen, conflict));
1444 }
1445
1446 int
1447 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1448 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1449 {
1450 dsl_dataset_t *ds = os->os_dsl_dataset;
1451 zap_cursor_t cursor;
1452 zap_attribute_t attr;
1453
1454 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1455
1456 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1457 return (SET_ERROR(ENOENT));
1458
1459 zap_cursor_init_serialized(&cursor,
1460 ds->ds_dir->dd_pool->dp_meta_objset,
1461 ds->ds_phys->ds_snapnames_zapobj, *offp);
1462
1463 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1464 zap_cursor_fini(&cursor);
1465 return (SET_ERROR(ENOENT));
1466 }
1467
1468 if (strlen(attr.za_name) + 1 > namelen) {
1469 zap_cursor_fini(&cursor);
1470 return (SET_ERROR(ENAMETOOLONG));
1471 }
1472
1473 (void) strcpy(name, attr.za_name);
1474 if (idp)
1475 *idp = attr.za_first_integer;
1476 if (case_conflict)
1477 *case_conflict = attr.za_normalization_conflict;
1478 zap_cursor_advance(&cursor);
1479 *offp = zap_cursor_serialize(&cursor);
1480 zap_cursor_fini(&cursor);
1481
1482 return (0);
1483 }
1484
1485 int
1486 dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value)
1487 {
1488 return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value));
1489 }
1490
1491 int
1492 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1493 uint64_t *idp, uint64_t *offp)
1494 {
1495 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1496 zap_cursor_t cursor;
1497 zap_attribute_t attr;
1498
1499 /* there is no next dir on a snapshot! */
1500 if (os->os_dsl_dataset->ds_object !=
1501 dd->dd_phys->dd_head_dataset_obj)
1502 return (SET_ERROR(ENOENT));
1503
1504 zap_cursor_init_serialized(&cursor,
1505 dd->dd_pool->dp_meta_objset,
1506 dd->dd_phys->dd_child_dir_zapobj, *offp);
1507
1508 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1509 zap_cursor_fini(&cursor);
1510 return (SET_ERROR(ENOENT));
1511 }
1512
1513 if (strlen(attr.za_name) + 1 > namelen) {
1514 zap_cursor_fini(&cursor);
1515 return (SET_ERROR(ENAMETOOLONG));
1516 }
1517
1518 (void) strcpy(name, attr.za_name);
1519 if (idp)
1520 *idp = attr.za_first_integer;
1521 zap_cursor_advance(&cursor);
1522 *offp = zap_cursor_serialize(&cursor);
1523 zap_cursor_fini(&cursor);
1524
1525 return (0);
1526 }
1527
1528 /*
1529 * Find objsets under and including ddobj, call func(ds) on each.
1530 */
1531 int
1532 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1533 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1534 {
1535 dsl_dir_t *dd;
1536 dsl_dataset_t *ds;
1537 zap_cursor_t zc;
1538 zap_attribute_t *attr;
1539 uint64_t thisobj;
1540 int err;
1541
1542 ASSERT(dsl_pool_config_held(dp));
1543
1544 err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1545 if (err != 0)
1546 return (err);
1547
1548 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1549 if (dd->dd_myname[0] == '$') {
1550 dsl_dir_rele(dd, FTAG);
1551 return (0);
1552 }
1553
1554 thisobj = dd->dd_phys->dd_head_dataset_obj;
1555 attr = kmem_alloc(sizeof (zap_attribute_t), KM_PUSHPAGE);
1556
1557 /*
1558 * Iterate over all children.
1559 */
1560 if (flags & DS_FIND_CHILDREN) {
1561 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1562 dd->dd_phys->dd_child_dir_zapobj);
1563 zap_cursor_retrieve(&zc, attr) == 0;
1564 (void) zap_cursor_advance(&zc)) {
1565 ASSERT3U(attr->za_integer_length, ==,
1566 sizeof (uint64_t));
1567 ASSERT3U(attr->za_num_integers, ==, 1);
1568
1569 err = dmu_objset_find_dp(dp, attr->za_first_integer,
1570 func, arg, flags);
1571 if (err != 0)
1572 break;
1573 }
1574 zap_cursor_fini(&zc);
1575
1576 if (err != 0) {
1577 dsl_dir_rele(dd, FTAG);
1578 kmem_free(attr, sizeof (zap_attribute_t));
1579 return (err);
1580 }
1581 }
1582
1583 /*
1584 * Iterate over all snapshots.
1585 */
1586 if (flags & DS_FIND_SNAPSHOTS) {
1587 dsl_dataset_t *ds;
1588 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1589
1590 if (err == 0) {
1591 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1592 dsl_dataset_rele(ds, FTAG);
1593
1594 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1595 zap_cursor_retrieve(&zc, attr) == 0;
1596 (void) zap_cursor_advance(&zc)) {
1597 ASSERT3U(attr->za_integer_length, ==,
1598 sizeof (uint64_t));
1599 ASSERT3U(attr->za_num_integers, ==, 1);
1600
1601 err = dsl_dataset_hold_obj(dp,
1602 attr->za_first_integer, FTAG, &ds);
1603 if (err != 0)
1604 break;
1605 err = func(dp, ds, arg);
1606 dsl_dataset_rele(ds, FTAG);
1607 if (err != 0)
1608 break;
1609 }
1610 zap_cursor_fini(&zc);
1611 }
1612 }
1613
1614 dsl_dir_rele(dd, FTAG);
1615 kmem_free(attr, sizeof (zap_attribute_t));
1616
1617 if (err != 0)
1618 return (err);
1619
1620 /*
1621 * Apply to self.
1622 */
1623 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1624 if (err != 0)
1625 return (err);
1626 err = func(dp, ds, arg);
1627 dsl_dataset_rele(ds, FTAG);
1628 return (err);
1629 }
1630
1631 /*
1632 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1633 * The dp_config_rwlock must not be held when this is called, and it
1634 * will not be held when the callback is called.
1635 * Therefore this function should only be used when the pool is not changing
1636 * (e.g. in syncing context), or the callback can deal with the possible races.
1637 */
1638 static int
1639 dmu_objset_find_impl(spa_t *spa, const char *name,
1640 int func(const char *, void *), void *arg, int flags)
1641 {
1642 dsl_dir_t *dd;
1643 dsl_pool_t *dp = spa_get_dsl(spa);
1644 dsl_dataset_t *ds;
1645 zap_cursor_t zc;
1646 zap_attribute_t *attr;
1647 char *child;
1648 uint64_t thisobj;
1649 int err;
1650
1651 dsl_pool_config_enter(dp, FTAG);
1652
1653 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1654 if (err != 0) {
1655 dsl_pool_config_exit(dp, FTAG);
1656 return (err);
1657 }
1658
1659 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1660 if (dd->dd_myname[0] == '$') {
1661 dsl_dir_rele(dd, FTAG);
1662 dsl_pool_config_exit(dp, FTAG);
1663 return (0);
1664 }
1665
1666 thisobj = dd->dd_phys->dd_head_dataset_obj;
1667 attr = kmem_alloc(sizeof (zap_attribute_t), KM_PUSHPAGE);
1668
1669 /*
1670 * Iterate over all children.
1671 */
1672 if (flags & DS_FIND_CHILDREN) {
1673 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1674 dd->dd_phys->dd_child_dir_zapobj);
1675 zap_cursor_retrieve(&zc, attr) == 0;
1676 (void) zap_cursor_advance(&zc)) {
1677 ASSERT3U(attr->za_integer_length, ==,
1678 sizeof (uint64_t));
1679 ASSERT3U(attr->za_num_integers, ==, 1);
1680
1681 child = kmem_asprintf("%s/%s", name, attr->za_name);
1682 dsl_pool_config_exit(dp, FTAG);
1683 err = dmu_objset_find_impl(spa, child,
1684 func, arg, flags);
1685 dsl_pool_config_enter(dp, FTAG);
1686 strfree(child);
1687 if (err != 0)
1688 break;
1689 }
1690 zap_cursor_fini(&zc);
1691
1692 if (err != 0) {
1693 dsl_dir_rele(dd, FTAG);
1694 dsl_pool_config_exit(dp, FTAG);
1695 kmem_free(attr, sizeof (zap_attribute_t));
1696 return (err);
1697 }
1698 }
1699
1700 /*
1701 * Iterate over all snapshots.
1702 */
1703 if (flags & DS_FIND_SNAPSHOTS) {
1704 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1705
1706 if (err == 0) {
1707 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1708 dsl_dataset_rele(ds, FTAG);
1709
1710 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1711 zap_cursor_retrieve(&zc, attr) == 0;
1712 (void) zap_cursor_advance(&zc)) {
1713 ASSERT3U(attr->za_integer_length, ==,
1714 sizeof (uint64_t));
1715 ASSERT3U(attr->za_num_integers, ==, 1);
1716
1717 child = kmem_asprintf("%s@%s",
1718 name, attr->za_name);
1719 dsl_pool_config_exit(dp, FTAG);
1720 err = func(child, arg);
1721 dsl_pool_config_enter(dp, FTAG);
1722 strfree(child);
1723 if (err != 0)
1724 break;
1725 }
1726 zap_cursor_fini(&zc);
1727 }
1728 }
1729
1730 dsl_dir_rele(dd, FTAG);
1731 kmem_free(attr, sizeof (zap_attribute_t));
1732 dsl_pool_config_exit(dp, FTAG);
1733
1734 if (err != 0)
1735 return (err);
1736
1737 /* Apply to self. */
1738 return (func(name, arg));
1739 }
1740
1741 /*
1742 * See comment above dmu_objset_find_impl().
1743 */
1744 int
1745 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1746 int flags)
1747 {
1748 spa_t *spa;
1749 int error;
1750
1751 error = spa_open(name, &spa, FTAG);
1752 if (error != 0)
1753 return (error);
1754 error = dmu_objset_find_impl(spa, name, func, arg, flags);
1755 spa_close(spa, FTAG);
1756 return (error);
1757 }
1758
1759 void
1760 dmu_objset_set_user(objset_t *os, void *user_ptr)
1761 {
1762 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1763 os->os_user_ptr = user_ptr;
1764 }
1765
1766 void *
1767 dmu_objset_get_user(objset_t *os)
1768 {
1769 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1770 return (os->os_user_ptr);
1771 }
1772
1773 /*
1774 * Determine name of filesystem, given name of snapshot.
1775 * buf must be at least MAXNAMELEN bytes
1776 */
1777 int
1778 dmu_fsname(const char *snapname, char *buf)
1779 {
1780 char *atp = strchr(snapname, '@');
1781 if (atp == NULL)
1782 return (SET_ERROR(EINVAL));
1783 if (atp - snapname >= MAXNAMELEN)
1784 return (SET_ERROR(ENAMETOOLONG));
1785 (void) strlcpy(buf, snapname, atp - snapname + 1);
1786 return (0);
1787 }
1788
1789 #if defined(_KERNEL) && defined(HAVE_SPL)
1790 EXPORT_SYMBOL(dmu_objset_zil);
1791 EXPORT_SYMBOL(dmu_objset_pool);
1792 EXPORT_SYMBOL(dmu_objset_ds);
1793 EXPORT_SYMBOL(dmu_objset_type);
1794 EXPORT_SYMBOL(dmu_objset_name);
1795 EXPORT_SYMBOL(dmu_objset_hold);
1796 EXPORT_SYMBOL(dmu_objset_own);
1797 EXPORT_SYMBOL(dmu_objset_rele);
1798 EXPORT_SYMBOL(dmu_objset_disown);
1799 EXPORT_SYMBOL(dmu_objset_from_ds);
1800 EXPORT_SYMBOL(dmu_objset_create);
1801 EXPORT_SYMBOL(dmu_objset_clone);
1802 EXPORT_SYMBOL(dmu_objset_stats);
1803 EXPORT_SYMBOL(dmu_objset_fast_stat);
1804 EXPORT_SYMBOL(dmu_objset_spa);
1805 EXPORT_SYMBOL(dmu_objset_space);
1806 EXPORT_SYMBOL(dmu_objset_fsid_guid);
1807 EXPORT_SYMBOL(dmu_objset_find);
1808 EXPORT_SYMBOL(dmu_objset_byteswap);
1809 EXPORT_SYMBOL(dmu_objset_evict_dbufs);
1810 EXPORT_SYMBOL(dmu_objset_snap_cmtime);
1811
1812 EXPORT_SYMBOL(dmu_objset_sync);
1813 EXPORT_SYMBOL(dmu_objset_is_dirty);
1814 EXPORT_SYMBOL(dmu_objset_create_impl);
1815 EXPORT_SYMBOL(dmu_objset_open_impl);
1816 EXPORT_SYMBOL(dmu_objset_evict);
1817 EXPORT_SYMBOL(dmu_objset_register_type);
1818 EXPORT_SYMBOL(dmu_objset_do_userquota_updates);
1819 EXPORT_SYMBOL(dmu_objset_userquota_get_ids);
1820 EXPORT_SYMBOL(dmu_objset_userused_enabled);
1821 EXPORT_SYMBOL(dmu_objset_userspace_upgrade);
1822 EXPORT_SYMBOL(dmu_objset_userspace_present);
1823 #endif
ZFS On Linux mirror for PVE