]> 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
Illumos #3498 panic in arc_read()
[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 */
24
25 /* Portions Copyright 2010 Robert Milkowski */
26
27 #include <sys/cred.h>
28 #include <sys/zfs_context.h>
29 #include <sys/dmu_objset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_dataset.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_pool.h>
34 #include <sys/dsl_synctask.h>
35 #include <sys/dsl_deleg.h>
36 #include <sys/dnode.h>
37 #include <sys/dbuf.h>
38 #include <sys/zvol.h>
39 #include <sys/dmu_tx.h>
40 #include <sys/zap.h>
41 #include <sys/zil.h>
42 #include <sys/dmu_impl.h>
43 #include <sys/zfs_ioctl.h>
44 #include <sys/sa.h>
45 #include <sys/zfs_onexit.h>
46
47 /*
48 * Needed to close a window in dnode_move() that allows the objset to be freed
49 * before it can be safely accessed.
50 */
51 krwlock_t os_lock;
52
53 void
54 dmu_objset_init(void)
55 {
56 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
57 }
58
59 void
60 dmu_objset_fini(void)
61 {
62 rw_destroy(&os_lock);
63 }
64
65 spa_t *
66 dmu_objset_spa(objset_t *os)
67 {
68 return (os->os_spa);
69 }
70
71 zilog_t *
72 dmu_objset_zil(objset_t *os)
73 {
74 return (os->os_zil);
75 }
76
77 dsl_pool_t *
78 dmu_objset_pool(objset_t *os)
79 {
80 dsl_dataset_t *ds;
81
82 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
83 return (ds->ds_dir->dd_pool);
84 else
85 return (spa_get_dsl(os->os_spa));
86 }
87
88 dsl_dataset_t *
89 dmu_objset_ds(objset_t *os)
90 {
91 return (os->os_dsl_dataset);
92 }
93
94 dmu_objset_type_t
95 dmu_objset_type(objset_t *os)
96 {
97 return (os->os_phys->os_type);
98 }
99
100 void
101 dmu_objset_name(objset_t *os, char *buf)
102 {
103 dsl_dataset_name(os->os_dsl_dataset, buf);
104 }
105
106 uint64_t
107 dmu_objset_id(objset_t *os)
108 {
109 dsl_dataset_t *ds = os->os_dsl_dataset;
110
111 return (ds ? ds->ds_object : 0);
112 }
113
114 uint64_t
115 dmu_objset_syncprop(objset_t *os)
116 {
117 return (os->os_sync);
118 }
119
120 uint64_t
121 dmu_objset_logbias(objset_t *os)
122 {
123 return (os->os_logbias);
124 }
125
126 static void
127 checksum_changed_cb(void *arg, uint64_t newval)
128 {
129 objset_t *os = arg;
130
131 /*
132 * Inheritance should have been done by now.
133 */
134 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
135
136 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
137 }
138
139 static void
140 compression_changed_cb(void *arg, uint64_t newval)
141 {
142 objset_t *os = arg;
143
144 /*
145 * Inheritance and range checking should have been done by now.
146 */
147 ASSERT(newval != ZIO_COMPRESS_INHERIT);
148
149 os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
150 }
151
152 static void
153 copies_changed_cb(void *arg, uint64_t newval)
154 {
155 objset_t *os = arg;
156
157 /*
158 * Inheritance and range checking should have been done by now.
159 */
160 ASSERT(newval > 0);
161 ASSERT(newval <= spa_max_replication(os->os_spa));
162
163 os->os_copies = newval;
164 }
165
166 static void
167 dedup_changed_cb(void *arg, uint64_t newval)
168 {
169 objset_t *os = arg;
170 spa_t *spa = os->os_spa;
171 enum zio_checksum checksum;
172
173 /*
174 * Inheritance should have been done by now.
175 */
176 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
177
178 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
179
180 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
181 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
182 }
183
184 static void
185 primary_cache_changed_cb(void *arg, uint64_t newval)
186 {
187 objset_t *os = arg;
188
189 /*
190 * Inheritance and range checking should have been done by now.
191 */
192 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
193 newval == ZFS_CACHE_METADATA);
194
195 os->os_primary_cache = newval;
196 }
197
198 static void
199 secondary_cache_changed_cb(void *arg, uint64_t newval)
200 {
201 objset_t *os = arg;
202
203 /*
204 * Inheritance and range checking should have been done by now.
205 */
206 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
207 newval == ZFS_CACHE_METADATA);
208
209 os->os_secondary_cache = newval;
210 }
211
212 static void
213 sync_changed_cb(void *arg, uint64_t newval)
214 {
215 objset_t *os = arg;
216
217 /*
218 * Inheritance and range checking should have been done by now.
219 */
220 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
221 newval == ZFS_SYNC_DISABLED);
222
223 os->os_sync = newval;
224 if (os->os_zil)
225 zil_set_sync(os->os_zil, newval);
226 }
227
228 static void
229 logbias_changed_cb(void *arg, uint64_t newval)
230 {
231 objset_t *os = arg;
232
233 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
234 newval == ZFS_LOGBIAS_THROUGHPUT);
235 os->os_logbias = newval;
236 if (os->os_zil)
237 zil_set_logbias(os->os_zil, newval);
238 }
239
240 void
241 dmu_objset_byteswap(void *buf, size_t size)
242 {
243 objset_phys_t *osp = buf;
244
245 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
246 dnode_byteswap(&osp->os_meta_dnode);
247 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
248 osp->os_type = BSWAP_64(osp->os_type);
249 osp->os_flags = BSWAP_64(osp->os_flags);
250 if (size == sizeof (objset_phys_t)) {
251 dnode_byteswap(&osp->os_userused_dnode);
252 dnode_byteswap(&osp->os_groupused_dnode);
253 }
254 }
255
256 int
257 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
258 objset_t **osp)
259 {
260 objset_t *os;
261 int i, err;
262
263 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
264
265 os = kmem_zalloc(sizeof (objset_t), KM_PUSHPAGE);
266 os->os_dsl_dataset = ds;
267 os->os_spa = spa;
268 os->os_rootbp = bp;
269 if (!BP_IS_HOLE(os->os_rootbp)) {
270 uint32_t aflags = ARC_WAIT;
271 zbookmark_t zb;
272 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
273 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
274
275 if (DMU_OS_IS_L2CACHEABLE(os))
276 aflags |= ARC_L2CACHE;
277
278 dprintf_bp(os->os_rootbp, "reading %s", "");
279 err = arc_read(NULL, spa, os->os_rootbp,
280 arc_getbuf_func, &os->os_phys_buf,
281 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
282 if (err) {
283 kmem_free(os, sizeof (objset_t));
284 /* convert checksum errors into IO errors */
285 if (err == ECKSUM)
286 err = EIO;
287 return (err);
288 }
289
290 /* Increase the blocksize if we are permitted. */
291 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
292 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
293 arc_buf_t *buf = arc_buf_alloc(spa,
294 sizeof (objset_phys_t), &os->os_phys_buf,
295 ARC_BUFC_METADATA);
296 bzero(buf->b_data, sizeof (objset_phys_t));
297 bcopy(os->os_phys_buf->b_data, buf->b_data,
298 arc_buf_size(os->os_phys_buf));
299 (void) arc_buf_remove_ref(os->os_phys_buf,
300 &os->os_phys_buf);
301 os->os_phys_buf = buf;
302 }
303
304 os->os_phys = os->os_phys_buf->b_data;
305 os->os_flags = os->os_phys->os_flags;
306 } else {
307 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
308 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
309 os->os_phys_buf = arc_buf_alloc(spa, size,
310 &os->os_phys_buf, ARC_BUFC_METADATA);
311 os->os_phys = os->os_phys_buf->b_data;
312 bzero(os->os_phys, size);
313 }
314
315 /*
316 * Note: the changed_cb will be called once before the register
317 * func returns, thus changing the checksum/compression from the
318 * default (fletcher2/off). Snapshots don't need to know about
319 * checksum/compression/copies.
320 */
321 if (ds) {
322 err = dsl_prop_register(ds, "primarycache",
323 primary_cache_changed_cb, os);
324 if (err == 0)
325 err = dsl_prop_register(ds, "secondarycache",
326 secondary_cache_changed_cb, os);
327 if (!dsl_dataset_is_snapshot(ds)) {
328 if (err == 0)
329 err = dsl_prop_register(ds, "checksum",
330 checksum_changed_cb, os);
331 if (err == 0)
332 err = dsl_prop_register(ds, "compression",
333 compression_changed_cb, os);
334 if (err == 0)
335 err = dsl_prop_register(ds, "copies",
336 copies_changed_cb, os);
337 if (err == 0)
338 err = dsl_prop_register(ds, "dedup",
339 dedup_changed_cb, os);
340 if (err == 0)
341 err = dsl_prop_register(ds, "logbias",
342 logbias_changed_cb, os);
343 if (err == 0)
344 err = dsl_prop_register(ds, "sync",
345 sync_changed_cb, os);
346 }
347 if (err) {
348 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
349 &os->os_phys_buf) == 1);
350 kmem_free(os, sizeof (objset_t));
351 return (err);
352 }
353 } else if (ds == NULL) {
354 /* It's the meta-objset. */
355 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
356 os->os_compress = ZIO_COMPRESS_LZJB;
357 os->os_copies = spa_max_replication(spa);
358 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
359 os->os_dedup_verify = 0;
360 os->os_logbias = 0;
361 os->os_sync = 0;
362 os->os_primary_cache = ZFS_CACHE_ALL;
363 os->os_secondary_cache = ZFS_CACHE_ALL;
364 }
365
366 if (ds == NULL || !dsl_dataset_is_snapshot(ds))
367 os->os_zil_header = os->os_phys->os_zil_header;
368 os->os_zil = zil_alloc(os, &os->os_zil_header);
369
370 for (i = 0; i < TXG_SIZE; i++) {
371 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
372 offsetof(dnode_t, dn_dirty_link[i]));
373 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
374 offsetof(dnode_t, dn_dirty_link[i]));
375 }
376 list_create(&os->os_dnodes, sizeof (dnode_t),
377 offsetof(dnode_t, dn_link));
378 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
379 offsetof(dmu_buf_impl_t, db_link));
380
381 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
382 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
383 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
384
385 DMU_META_DNODE(os) = dnode_special_open(os,
386 &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
387 &os->os_meta_dnode);
388 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
389 DMU_USERUSED_DNODE(os) = dnode_special_open(os,
390 &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
391 &os->os_userused_dnode);
392 DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
393 &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
394 &os->os_groupused_dnode);
395 }
396
397 /*
398 * We should be the only thread trying to do this because we
399 * have ds_opening_lock
400 */
401 if (ds) {
402 mutex_enter(&ds->ds_lock);
403 ASSERT(ds->ds_objset == NULL);
404 ds->ds_objset = os;
405 mutex_exit(&ds->ds_lock);
406 }
407
408 *osp = os;
409 return (0);
410 }
411
412 int
413 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
414 {
415 int err = 0;
416
417 mutex_enter(&ds->ds_opening_lock);
418 *osp = ds->ds_objset;
419 if (*osp == NULL) {
420 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
421 ds, dsl_dataset_get_blkptr(ds), osp);
422 }
423 mutex_exit(&ds->ds_opening_lock);
424 return (err);
425 }
426
427 /* called from zpl */
428 int
429 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
430 {
431 dsl_dataset_t *ds;
432 int err;
433
434 err = dsl_dataset_hold(name, tag, &ds);
435 if (err)
436 return (err);
437
438 err = dmu_objset_from_ds(ds, osp);
439 if (err)
440 dsl_dataset_rele(ds, tag);
441
442 return (err);
443 }
444
445 /* called from zpl */
446 int
447 dmu_objset_own(const char *name, dmu_objset_type_t type,
448 boolean_t readonly, void *tag, objset_t **osp)
449 {
450 dsl_dataset_t *ds;
451 int err;
452
453 err = dsl_dataset_own(name, B_FALSE, tag, &ds);
454 if (err)
455 return (err);
456
457 err = dmu_objset_from_ds(ds, osp);
458 if (err) {
459 dsl_dataset_disown(ds, tag);
460 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
461 dmu_objset_disown(*osp, tag);
462 return (EINVAL);
463 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
464 dmu_objset_disown(*osp, tag);
465 return (EROFS);
466 }
467 return (err);
468 }
469
470 void
471 dmu_objset_rele(objset_t *os, void *tag)
472 {
473 dsl_dataset_rele(os->os_dsl_dataset, tag);
474 }
475
476 void
477 dmu_objset_disown(objset_t *os, void *tag)
478 {
479 dsl_dataset_disown(os->os_dsl_dataset, tag);
480 }
481
482 int
483 dmu_objset_evict_dbufs(objset_t *os)
484 {
485 dnode_t *dn;
486
487 mutex_enter(&os->os_lock);
488
489 /* process the mdn last, since the other dnodes have holds on it */
490 list_remove(&os->os_dnodes, DMU_META_DNODE(os));
491 list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
492
493 /*
494 * Find the first dnode with holds. We have to do this dance
495 * because dnode_add_ref() only works if you already have a
496 * hold. If there are no holds then it has no dbufs so OK to
497 * skip.
498 */
499 for (dn = list_head(&os->os_dnodes);
500 dn && !dnode_add_ref(dn, FTAG);
501 dn = list_next(&os->os_dnodes, dn))
502 continue;
503
504 while (dn) {
505 dnode_t *next_dn = dn;
506
507 do {
508 next_dn = list_next(&os->os_dnodes, next_dn);
509 } while (next_dn && !dnode_add_ref(next_dn, FTAG));
510
511 mutex_exit(&os->os_lock);
512 dnode_evict_dbufs(dn);
513 dnode_rele(dn, FTAG);
514 mutex_enter(&os->os_lock);
515 dn = next_dn;
516 }
517 dn = list_head(&os->os_dnodes);
518 mutex_exit(&os->os_lock);
519 return (dn != DMU_META_DNODE(os));
520 }
521
522 void
523 dmu_objset_evict(objset_t *os)
524 {
525 dsl_dataset_t *ds = os->os_dsl_dataset;
526 int t;
527
528 for (t = 0; t < TXG_SIZE; t++)
529 ASSERT(!dmu_objset_is_dirty(os, t));
530
531 if (ds) {
532 if (!dsl_dataset_is_snapshot(ds)) {
533 VERIFY(0 == dsl_prop_unregister(ds, "checksum",
534 checksum_changed_cb, os));
535 VERIFY(0 == dsl_prop_unregister(ds, "compression",
536 compression_changed_cb, os));
537 VERIFY(0 == dsl_prop_unregister(ds, "copies",
538 copies_changed_cb, os));
539 VERIFY(0 == dsl_prop_unregister(ds, "dedup",
540 dedup_changed_cb, os));
541 VERIFY(0 == dsl_prop_unregister(ds, "logbias",
542 logbias_changed_cb, os));
543 VERIFY(0 == dsl_prop_unregister(ds, "sync",
544 sync_changed_cb, os));
545 }
546 VERIFY(0 == dsl_prop_unregister(ds, "primarycache",
547 primary_cache_changed_cb, os));
548 VERIFY(0 == dsl_prop_unregister(ds, "secondarycache",
549 secondary_cache_changed_cb, os));
550 }
551
552 if (os->os_sa)
553 sa_tear_down(os);
554
555 /*
556 * We should need only a single pass over the dnode list, since
557 * nothing can be added to the list at this point.
558 */
559 (void) dmu_objset_evict_dbufs(os);
560
561 dnode_special_close(&os->os_meta_dnode);
562 if (DMU_USERUSED_DNODE(os)) {
563 dnode_special_close(&os->os_userused_dnode);
564 dnode_special_close(&os->os_groupused_dnode);
565 }
566 zil_free(os->os_zil);
567
568 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
569
570 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf) == 1);
571
572 /*
573 * This is a barrier to prevent the objset from going away in
574 * dnode_move() until we can safely ensure that the objset is still in
575 * use. We consider the objset valid before the barrier and invalid
576 * after the barrier.
577 */
578 rw_enter(&os_lock, RW_READER);
579 rw_exit(&os_lock);
580
581 mutex_destroy(&os->os_lock);
582 mutex_destroy(&os->os_obj_lock);
583 mutex_destroy(&os->os_user_ptr_lock);
584 kmem_free(os, sizeof (objset_t));
585 }
586
587 timestruc_t
588 dmu_objset_snap_cmtime(objset_t *os)
589 {
590 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
591 }
592
593 /* called from dsl for meta-objset */
594 objset_t *
595 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
596 dmu_objset_type_t type, dmu_tx_t *tx)
597 {
598 objset_t *os;
599 dnode_t *mdn;
600
601 ASSERT(dmu_tx_is_syncing(tx));
602 if (ds != NULL)
603 VERIFY(0 == dmu_objset_from_ds(ds, &os));
604 else
605 VERIFY(0 == dmu_objset_open_impl(spa, NULL, bp, &os));
606
607 mdn = DMU_META_DNODE(os);
608
609 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
610 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
611
612 /*
613 * We don't want to have to increase the meta-dnode's nlevels
614 * later, because then we could do it in quescing context while
615 * we are also accessing it in open context.
616 *
617 * This precaution is not necessary for the MOS (ds == NULL),
618 * because the MOS is only updated in syncing context.
619 * This is most fortunate: the MOS is the only objset that
620 * needs to be synced multiple times as spa_sync() iterates
621 * to convergence, so minimizing its dn_nlevels matters.
622 */
623 if (ds != NULL) {
624 int levels = 1;
625
626 /*
627 * Determine the number of levels necessary for the meta-dnode
628 * to contain DN_MAX_OBJECT dnodes.
629 */
630 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
631 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
632 DN_MAX_OBJECT * sizeof (dnode_phys_t))
633 levels++;
634
635 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
636 mdn->dn_nlevels = levels;
637 }
638
639 ASSERT(type != DMU_OST_NONE);
640 ASSERT(type != DMU_OST_ANY);
641 ASSERT(type < DMU_OST_NUMTYPES);
642 os->os_phys->os_type = type;
643 if (dmu_objset_userused_enabled(os)) {
644 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
645 os->os_flags = os->os_phys->os_flags;
646 }
647
648 dsl_dataset_dirty(ds, tx);
649
650 return (os);
651 }
652
653 struct oscarg {
654 void (*userfunc)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx);
655 void *userarg;
656 dsl_dataset_t *clone_origin;
657 const char *lastname;
658 dmu_objset_type_t type;
659 uint64_t flags;
660 cred_t *cr;
661 };
662
663 /*ARGSUSED*/
664 static int
665 dmu_objset_create_check(void *arg1, void *arg2, dmu_tx_t *tx)
666 {
667 dsl_dir_t *dd = arg1;
668 struct oscarg *oa = arg2;
669 objset_t *mos = dd->dd_pool->dp_meta_objset;
670 int err;
671 uint64_t ddobj;
672
673 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
674 oa->lastname, sizeof (uint64_t), 1, &ddobj);
675 if (err != ENOENT)
676 return (err ? err : EEXIST);
677
678 if (oa->clone_origin != NULL) {
679 /* You can't clone across pools. */
680 if (oa->clone_origin->ds_dir->dd_pool != dd->dd_pool)
681 return (EXDEV);
682
683 /* You can only clone snapshots, not the head datasets. */
684 if (!dsl_dataset_is_snapshot(oa->clone_origin))
685 return (EINVAL);
686 }
687
688 return (0);
689 }
690
691 static void
692 dmu_objset_create_sync(void *arg1, void *arg2, dmu_tx_t *tx)
693 {
694 dsl_dir_t *dd = arg1;
695 spa_t *spa = dd->dd_pool->dp_spa;
696 struct oscarg *oa = arg2;
697 uint64_t obj;
698
699 ASSERT(dmu_tx_is_syncing(tx));
700
701 obj = dsl_dataset_create_sync(dd, oa->lastname,
702 oa->clone_origin, oa->flags, oa->cr, tx);
703
704 if (oa->clone_origin == NULL) {
705 dsl_pool_t *dp = dd->dd_pool;
706 dsl_dataset_t *ds;
707 blkptr_t *bp;
708 objset_t *os;
709
710 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
711 bp = dsl_dataset_get_blkptr(ds);
712 ASSERT(BP_IS_HOLE(bp));
713
714 os = dmu_objset_create_impl(spa, ds, bp, oa->type, tx);
715
716 if (oa->userfunc)
717 oa->userfunc(os, oa->userarg, oa->cr, tx);
718 dsl_dataset_rele(ds, FTAG);
719 }
720
721 spa_history_log_internal(LOG_DS_CREATE, spa, tx, "dataset = %llu", obj);
722 }
723
724 int
725 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
726 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
727 {
728 dsl_dir_t *pdd;
729 const char *tail;
730 int err = 0;
731 struct oscarg oa = { 0 };
732
733 ASSERT(strchr(name, '@') == NULL);
734 err = dsl_dir_open(name, FTAG, &pdd, &tail);
735 if (err)
736 return (err);
737 if (tail == NULL) {
738 dsl_dir_close(pdd, FTAG);
739 return (EEXIST);
740 }
741
742 oa.userfunc = func;
743 oa.userarg = arg;
744 oa.lastname = tail;
745 oa.type = type;
746 oa.flags = flags;
747 oa.cr = CRED();
748
749 err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
750 dmu_objset_create_sync, pdd, &oa, 5);
751 dsl_dir_close(pdd, FTAG);
752 return (err);
753 }
754
755 int
756 dmu_objset_clone(const char *name, dsl_dataset_t *clone_origin, uint64_t flags)
757 {
758 dsl_dir_t *pdd;
759 const char *tail;
760 int err = 0;
761 struct oscarg oa = { 0 };
762
763 ASSERT(strchr(name, '@') == NULL);
764 err = dsl_dir_open(name, FTAG, &pdd, &tail);
765 if (err)
766 return (err);
767 if (tail == NULL) {
768 dsl_dir_close(pdd, FTAG);
769 return (EEXIST);
770 }
771
772 oa.lastname = tail;
773 oa.clone_origin = clone_origin;
774 oa.flags = flags;
775 oa.cr = CRED();
776
777 err = dsl_sync_task_do(pdd->dd_pool, dmu_objset_create_check,
778 dmu_objset_create_sync, pdd, &oa, 5);
779 dsl_dir_close(pdd, FTAG);
780 return (err);
781 }
782
783 int
784 dmu_objset_destroy(const char *name, boolean_t defer)
785 {
786 dsl_dataset_t *ds;
787 int error;
788
789 error = dsl_dataset_own(name, B_TRUE, FTAG, &ds);
790 if (error == 0) {
791 error = dsl_dataset_destroy(ds, FTAG, defer);
792 /* dsl_dataset_destroy() closes the ds. */
793 }
794
795 return (error);
796 }
797
798 struct snaparg {
799 dsl_sync_task_group_t *dstg;
800 char *snapname;
801 char *htag;
802 char failed[MAXPATHLEN];
803 boolean_t recursive;
804 boolean_t needsuspend;
805 boolean_t temporary;
806 nvlist_t *props;
807 struct dsl_ds_holdarg *ha; /* only needed in the temporary case */
808 dsl_dataset_t *newds;
809 };
810
811 static int
812 snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
813 {
814 objset_t *os = arg1;
815 struct snaparg *sn = arg2;
816 int error;
817
818 /* The props have already been checked by zfs_check_userprops(). */
819
820 error = dsl_dataset_snapshot_check(os->os_dsl_dataset,
821 sn->snapname, tx);
822 if (error)
823 return (error);
824
825 if (sn->temporary) {
826 /*
827 * Ideally we would just call
828 * dsl_dataset_user_hold_check() and
829 * dsl_dataset_destroy_check() here. However the
830 * dataset we want to hold and destroy is the snapshot
831 * that we just confirmed we can create, but it won't
832 * exist until after these checks are run. Do any
833 * checks we can here and if more checks are added to
834 * those routines in the future, similar checks may be
835 * necessary here.
836 */
837 if (spa_version(os->os_spa) < SPA_VERSION_USERREFS)
838 return (ENOTSUP);
839 /*
840 * Not checking number of tags because the tag will be
841 * unique, as it will be the only tag.
842 */
843 if (strlen(sn->htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
844 return (E2BIG);
845
846 sn->ha = kmem_alloc(sizeof(struct dsl_ds_holdarg), KM_PUSHPAGE);
847 sn->ha->temphold = B_TRUE;
848 sn->ha->htag = sn->htag;
849 }
850 return (error);
851 }
852
853 static void
854 snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
855 {
856 objset_t *os = arg1;
857 dsl_dataset_t *ds = os->os_dsl_dataset;
858 struct snaparg *sn = arg2;
859
860 dsl_dataset_snapshot_sync(ds, sn->snapname, tx);
861
862 if (sn->props) {
863 dsl_props_arg_t pa;
864 pa.pa_props = sn->props;
865 pa.pa_source = ZPROP_SRC_LOCAL;
866 dsl_props_set_sync(ds->ds_prev, &pa, tx);
867 }
868
869 if (sn->temporary) {
870 struct dsl_ds_destroyarg da;
871
872 dsl_dataset_user_hold_sync(ds->ds_prev, sn->ha, tx);
873 kmem_free(sn->ha, sizeof (struct dsl_ds_holdarg));
874 sn->ha = NULL;
875 sn->newds = ds->ds_prev;
876
877 da.ds = ds->ds_prev;
878 da.defer = B_TRUE;
879 dsl_dataset_destroy_sync(&da, FTAG, tx);
880 }
881 }
882
883 static int
884 dmu_objset_snapshot_one(const char *name, void *arg)
885 {
886 struct snaparg *sn = arg;
887 objset_t *os;
888 int err;
889 char *cp;
890
891 /*
892 * If the objset starts with a '%', then ignore it unless it was
893 * explicitly named (ie, not recursive). These hidden datasets
894 * are always inconsistent, and by not opening them here, we can
895 * avoid a race with dsl_dir_destroy_check().
896 */
897 cp = strrchr(name, '/');
898 if (cp && cp[1] == '%' && sn->recursive)
899 return (0);
900
901 (void) strcpy(sn->failed, name);
902
903 /*
904 * Check permissions if we are doing a recursive snapshot. The
905 * permission checks for the starting dataset have already been
906 * performed in zfs_secpolicy_snapshot()
907 */
908 if (sn->recursive && (err = zfs_secpolicy_snapshot_perms(name, CRED())))
909 return (err);
910
911 err = dmu_objset_hold(name, sn, &os);
912 if (err != 0)
913 return (err);
914
915 /*
916 * If the objset is in an inconsistent state (eg, in the process
917 * of being destroyed), don't snapshot it. As with %hidden
918 * datasets, we return EBUSY if this name was explicitly
919 * requested (ie, not recursive), and otherwise ignore it.
920 */
921 if (os->os_dsl_dataset->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) {
922 dmu_objset_rele(os, sn);
923 return (sn->recursive ? 0 : EBUSY);
924 }
925
926 if (sn->needsuspend) {
927 err = zil_suspend(dmu_objset_zil(os));
928 if (err) {
929 dmu_objset_rele(os, sn);
930 return (err);
931 }
932 }
933 dsl_sync_task_create(sn->dstg, snapshot_check, snapshot_sync,
934 os, sn, 3);
935
936 return (0);
937 }
938
939 int
940 dmu_objset_snapshot(char *fsname, char *snapname, char *tag,
941 nvlist_t *props, boolean_t recursive, boolean_t temporary, int cleanup_fd)
942 {
943 dsl_sync_task_t *dst;
944 struct snaparg *sn;
945 spa_t *spa;
946 minor_t minor;
947 int err;
948
949 sn = kmem_alloc(sizeof (struct snaparg), KM_SLEEP);
950 (void) strcpy(sn->failed, fsname);
951
952 err = spa_open(fsname, &spa, FTAG);
953 if (err) {
954 kmem_free(sn, sizeof (struct snaparg));
955 return (err);
956 }
957
958 if (temporary) {
959 if (cleanup_fd < 0) {
960 spa_close(spa, FTAG);
961 return (EINVAL);
962 }
963 if ((err = zfs_onexit_fd_hold(cleanup_fd, &minor)) != 0) {
964 spa_close(spa, FTAG);
965 return (err);
966 }
967 }
968
969 sn->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
970 sn->snapname = snapname;
971 sn->htag = tag;
972 sn->props = props;
973 sn->recursive = recursive;
974 sn->needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
975 sn->temporary = temporary;
976 sn->ha = NULL;
977 sn->newds = NULL;
978
979 if (recursive) {
980 err = dmu_objset_find(fsname,
981 dmu_objset_snapshot_one, sn, DS_FIND_CHILDREN);
982 } else {
983 err = dmu_objset_snapshot_one(fsname, sn);
984 }
985
986 if (err == 0)
987 err = dsl_sync_task_group_wait(sn->dstg);
988
989 for (dst = list_head(&sn->dstg->dstg_tasks); dst;
990 dst = list_next(&sn->dstg->dstg_tasks, dst)) {
991 objset_t *os = dst->dst_arg1;
992 dsl_dataset_t *ds = os->os_dsl_dataset;
993 if (dst->dst_err) {
994 dsl_dataset_name(ds, sn->failed);
995 } else if (temporary) {
996 dsl_register_onexit_hold_cleanup(sn->newds, tag, minor);
997 }
998 if (sn->needsuspend)
999 zil_resume(dmu_objset_zil(os));
1000 dmu_objset_rele(os, sn);
1001 }
1002
1003 if (err)
1004 (void) strcpy(fsname, sn->failed);
1005 if (temporary)
1006 zfs_onexit_fd_rele(cleanup_fd);
1007 dsl_sync_task_group_destroy(sn->dstg);
1008 spa_close(spa, FTAG);
1009 kmem_free(sn, sizeof (struct snaparg));
1010 return (err);
1011 }
1012
1013 static void
1014 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
1015 {
1016 dnode_t *dn;
1017
1018 while ((dn = list_head(list))) {
1019 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1020 ASSERT(dn->dn_dbuf->db_data_pending);
1021 /*
1022 * Initialize dn_zio outside dnode_sync() because the
1023 * meta-dnode needs to set it ouside dnode_sync().
1024 */
1025 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1026 ASSERT(dn->dn_zio);
1027
1028 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1029 list_remove(list, dn);
1030
1031 if (newlist) {
1032 (void) dnode_add_ref(dn, newlist);
1033 list_insert_tail(newlist, dn);
1034 }
1035
1036 dnode_sync(dn, tx);
1037 }
1038 }
1039
1040 /* ARGSUSED */
1041 static void
1042 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1043 {
1044 int i;
1045
1046 blkptr_t *bp = zio->io_bp;
1047 objset_t *os = arg;
1048 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1049
1050 ASSERT(bp == os->os_rootbp);
1051 ASSERT(BP_GET_TYPE(bp) == DMU_OT_OBJSET);
1052 ASSERT(BP_GET_LEVEL(bp) == 0);
1053
1054 /*
1055 * Update rootbp fill count: it should be the number of objects
1056 * allocated in the object set (not counting the "special"
1057 * objects that are stored in the objset_phys_t -- the meta
1058 * dnode and user/group accounting objects).
1059 */
1060 bp->blk_fill = 0;
1061 for (i = 0; i < dnp->dn_nblkptr; i++)
1062 bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
1063 }
1064
1065 /* ARGSUSED */
1066 static void
1067 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1068 {
1069 blkptr_t *bp = zio->io_bp;
1070 blkptr_t *bp_orig = &zio->io_bp_orig;
1071 objset_t *os = arg;
1072
1073 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1074 ASSERT(BP_EQUAL(bp, bp_orig));
1075 } else {
1076 dsl_dataset_t *ds = os->os_dsl_dataset;
1077 dmu_tx_t *tx = os->os_synctx;
1078
1079 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1080 dsl_dataset_block_born(ds, bp, tx);
1081 }
1082 }
1083
1084 /* called from dsl */
1085 void
1086 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1087 {
1088 int txgoff;
1089 zbookmark_t zb;
1090 zio_prop_t zp;
1091 zio_t *zio;
1092 list_t *list;
1093 list_t *newlist = NULL;
1094 dbuf_dirty_record_t *dr;
1095
1096 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1097
1098 ASSERT(dmu_tx_is_syncing(tx));
1099 /* XXX the write_done callback should really give us the tx... */
1100 os->os_synctx = tx;
1101
1102 if (os->os_dsl_dataset == NULL) {
1103 /*
1104 * This is the MOS. If we have upgraded,
1105 * spa_max_replication() could change, so reset
1106 * os_copies here.
1107 */
1108 os->os_copies = spa_max_replication(os->os_spa);
1109 }
1110
1111 /*
1112 * Create the root block IO
1113 */
1114 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1115 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1116 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1117 arc_release(os->os_phys_buf, &os->os_phys_buf);
1118
1119 dmu_write_policy(os, NULL, 0, 0, &zp);
1120
1121 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1122 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), &zp,
1123 dmu_objset_write_ready, dmu_objset_write_done, os,
1124 ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1125
1126 /*
1127 * Sync special dnodes - the parent IO for the sync is the root block
1128 */
1129 DMU_META_DNODE(os)->dn_zio = zio;
1130 dnode_sync(DMU_META_DNODE(os), tx);
1131
1132 os->os_phys->os_flags = os->os_flags;
1133
1134 if (DMU_USERUSED_DNODE(os) &&
1135 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1136 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1137 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1138 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1139 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1140 }
1141
1142 txgoff = tx->tx_txg & TXG_MASK;
1143
1144 if (dmu_objset_userused_enabled(os)) {
1145 newlist = &os->os_synced_dnodes;
1146 /*
1147 * We must create the list here because it uses the
1148 * dn_dirty_link[] of this txg.
1149 */
1150 list_create(newlist, sizeof (dnode_t),
1151 offsetof(dnode_t, dn_dirty_link[txgoff]));
1152 }
1153
1154 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1155 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1156
1157 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1158 while ((dr = list_head(list)) != NULL) {
1159 ASSERT(dr->dr_dbuf->db_level == 0);
1160 list_remove(list, dr);
1161 if (dr->dr_zio)
1162 zio_nowait(dr->dr_zio);
1163 }
1164 /*
1165 * Free intent log blocks up to this tx.
1166 */
1167 zil_sync(os->os_zil, tx);
1168 os->os_phys->os_zil_header = os->os_zil_header;
1169 zio_nowait(zio);
1170 }
1171
1172 boolean_t
1173 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1174 {
1175 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1176 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1177 }
1178
1179 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1180
1181 void
1182 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1183 {
1184 used_cbs[ost] = cb;
1185 }
1186
1187 boolean_t
1188 dmu_objset_userused_enabled(objset_t *os)
1189 {
1190 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1191 used_cbs[os->os_phys->os_type] != NULL &&
1192 DMU_USERUSED_DNODE(os) != NULL);
1193 }
1194
1195 static void
1196 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1197 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1198 {
1199 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1200 int64_t delta = DNODE_SIZE + used;
1201 if (subtract)
1202 delta = -delta;
1203 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1204 user, delta, tx));
1205 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1206 group, delta, tx));
1207 }
1208 }
1209
1210 void
1211 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1212 {
1213 dnode_t *dn;
1214 list_t *list = &os->os_synced_dnodes;
1215
1216 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1217
1218 while ((dn = list_head(list)) != NULL) {
1219 int flags;
1220 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1221 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1222 dn->dn_phys->dn_flags &
1223 DNODE_FLAG_USERUSED_ACCOUNTED);
1224
1225 /* Allocate the user/groupused objects if necessary. */
1226 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1227 VERIFY(0 == zap_create_claim(os,
1228 DMU_USERUSED_OBJECT,
1229 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1230 VERIFY(0 == zap_create_claim(os,
1231 DMU_GROUPUSED_OBJECT,
1232 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1233 }
1234
1235 /*
1236 * We intentionally modify the zap object even if the
1237 * net delta is zero. Otherwise
1238 * the block of the zap obj could be shared between
1239 * datasets but need to be different between them after
1240 * a bprewrite.
1241 */
1242
1243 flags = dn->dn_id_flags;
1244 ASSERT(flags);
1245 if (flags & DN_ID_OLD_EXIST) {
1246 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1247 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1248 }
1249 if (flags & DN_ID_NEW_EXIST) {
1250 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1251 dn->dn_phys->dn_flags, dn->dn_newuid,
1252 dn->dn_newgid, B_FALSE, tx);
1253 }
1254
1255 mutex_enter(&dn->dn_mtx);
1256 dn->dn_oldused = 0;
1257 dn->dn_oldflags = 0;
1258 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1259 dn->dn_olduid = dn->dn_newuid;
1260 dn->dn_oldgid = dn->dn_newgid;
1261 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1262 if (dn->dn_bonuslen == 0)
1263 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1264 else
1265 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1266 }
1267 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1268 mutex_exit(&dn->dn_mtx);
1269
1270 list_remove(list, dn);
1271 dnode_rele(dn, list);
1272 }
1273 }
1274
1275 /*
1276 * Returns a pointer to data to find uid/gid from
1277 *
1278 * If a dirty record for transaction group that is syncing can't
1279 * be found then NULL is returned. In the NULL case it is assumed
1280 * the uid/gid aren't changing.
1281 */
1282 static void *
1283 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1284 {
1285 dbuf_dirty_record_t *dr, **drp;
1286 void *data;
1287
1288 if (db->db_dirtycnt == 0)
1289 return (db->db.db_data); /* Nothing is changing */
1290
1291 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1292 if (dr->dr_txg == tx->tx_txg)
1293 break;
1294
1295 if (dr == NULL) {
1296 data = NULL;
1297 } else {
1298 dnode_t *dn;
1299
1300 DB_DNODE_ENTER(dr->dr_dbuf);
1301 dn = DB_DNODE(dr->dr_dbuf);
1302
1303 if (dn->dn_bonuslen == 0 &&
1304 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1305 data = dr->dt.dl.dr_data->b_data;
1306 else
1307 data = dr->dt.dl.dr_data;
1308
1309 DB_DNODE_EXIT(dr->dr_dbuf);
1310 }
1311
1312 return (data);
1313 }
1314
1315 void
1316 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1317 {
1318 objset_t *os = dn->dn_objset;
1319 void *data = NULL;
1320 dmu_buf_impl_t *db = NULL;
1321 uint64_t *user = NULL, *group = NULL;
1322 int flags = dn->dn_id_flags;
1323 int error;
1324 boolean_t have_spill = B_FALSE;
1325
1326 if (!dmu_objset_userused_enabled(dn->dn_objset))
1327 return;
1328
1329 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1330 DN_ID_CHKED_SPILL)))
1331 return;
1332
1333 if (before && dn->dn_bonuslen != 0)
1334 data = DN_BONUS(dn->dn_phys);
1335 else if (!before && dn->dn_bonuslen != 0) {
1336 if (dn->dn_bonus) {
1337 db = dn->dn_bonus;
1338 mutex_enter(&db->db_mtx);
1339 data = dmu_objset_userquota_find_data(db, tx);
1340 } else {
1341 data = DN_BONUS(dn->dn_phys);
1342 }
1343 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1344 int rf = 0;
1345
1346 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1347 rf |= DB_RF_HAVESTRUCT;
1348 error = dmu_spill_hold_by_dnode(dn,
1349 rf | DB_RF_MUST_SUCCEED,
1350 FTAG, (dmu_buf_t **)&db);
1351 ASSERT(error == 0);
1352 mutex_enter(&db->db_mtx);
1353 data = (before) ? db->db.db_data :
1354 dmu_objset_userquota_find_data(db, tx);
1355 have_spill = B_TRUE;
1356 } else {
1357 mutex_enter(&dn->dn_mtx);
1358 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1359 mutex_exit(&dn->dn_mtx);
1360 return;
1361 }
1362
1363 if (before) {
1364 ASSERT(data);
1365 user = &dn->dn_olduid;
1366 group = &dn->dn_oldgid;
1367 } else if (data) {
1368 user = &dn->dn_newuid;
1369 group = &dn->dn_newgid;
1370 }
1371
1372 /*
1373 * Must always call the callback in case the object
1374 * type has changed and that type isn't an object type to track
1375 */
1376 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1377 user, group);
1378
1379 /*
1380 * Preserve existing uid/gid when the callback can't determine
1381 * what the new uid/gid are and the callback returned EEXIST.
1382 * The EEXIST error tells us to just use the existing uid/gid.
1383 * If we don't know what the old values are then just assign
1384 * them to 0, since that is a new file being created.
1385 */
1386 if (!before && data == NULL && error == EEXIST) {
1387 if (flags & DN_ID_OLD_EXIST) {
1388 dn->dn_newuid = dn->dn_olduid;
1389 dn->dn_newgid = dn->dn_oldgid;
1390 } else {
1391 dn->dn_newuid = 0;
1392 dn->dn_newgid = 0;
1393 }
1394 error = 0;
1395 }
1396
1397 if (db)
1398 mutex_exit(&db->db_mtx);
1399
1400 mutex_enter(&dn->dn_mtx);
1401 if (error == 0 && before)
1402 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1403 if (error == 0 && !before)
1404 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1405
1406 if (have_spill) {
1407 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1408 } else {
1409 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1410 }
1411 mutex_exit(&dn->dn_mtx);
1412 if (have_spill)
1413 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1414 }
1415
1416 boolean_t
1417 dmu_objset_userspace_present(objset_t *os)
1418 {
1419 return (os->os_phys->os_flags &
1420 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1421 }
1422
1423 int
1424 dmu_objset_userspace_upgrade(objset_t *os)
1425 {
1426 uint64_t obj;
1427 int err = 0;
1428
1429 if (dmu_objset_userspace_present(os))
1430 return (0);
1431 if (!dmu_objset_userused_enabled(os))
1432 return (ENOTSUP);
1433 if (dmu_objset_is_snapshot(os))
1434 return (EINVAL);
1435
1436 /*
1437 * We simply need to mark every object dirty, so that it will be
1438 * synced out and now accounted. If this is called
1439 * concurrently, or if we already did some work before crashing,
1440 * that's fine, since we track each object's accounted state
1441 * independently.
1442 */
1443
1444 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1445 dmu_tx_t *tx;
1446 dmu_buf_t *db;
1447 int objerr;
1448
1449 if (issig(JUSTLOOKING) && issig(FORREAL))
1450 return (EINTR);
1451
1452 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1453 if (objerr)
1454 continue;
1455 tx = dmu_tx_create(os);
1456 dmu_tx_hold_bonus(tx, obj);
1457 objerr = dmu_tx_assign(tx, TXG_WAIT);
1458 if (objerr) {
1459 dmu_tx_abort(tx);
1460 continue;
1461 }
1462 dmu_buf_will_dirty(db, tx);
1463 dmu_buf_rele(db, FTAG);
1464 dmu_tx_commit(tx);
1465 }
1466
1467 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1468 txg_wait_synced(dmu_objset_pool(os), 0);
1469 return (0);
1470 }
1471
1472 void
1473 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1474 uint64_t *usedobjsp, uint64_t *availobjsp)
1475 {
1476 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1477 usedobjsp, availobjsp);
1478 }
1479
1480 uint64_t
1481 dmu_objset_fsid_guid(objset_t *os)
1482 {
1483 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1484 }
1485
1486 void
1487 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1488 {
1489 stat->dds_type = os->os_phys->os_type;
1490 if (os->os_dsl_dataset)
1491 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1492 }
1493
1494 void
1495 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1496 {
1497 ASSERT(os->os_dsl_dataset ||
1498 os->os_phys->os_type == DMU_OST_META);
1499
1500 if (os->os_dsl_dataset != NULL)
1501 dsl_dataset_stats(os->os_dsl_dataset, nv);
1502
1503 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1504 os->os_phys->os_type);
1505 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1506 dmu_objset_userspace_present(os));
1507 }
1508
1509 int
1510 dmu_objset_is_snapshot(objset_t *os)
1511 {
1512 if (os->os_dsl_dataset != NULL)
1513 return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1514 else
1515 return (B_FALSE);
1516 }
1517
1518 int
1519 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1520 boolean_t *conflict)
1521 {
1522 dsl_dataset_t *ds = os->os_dsl_dataset;
1523 uint64_t ignored;
1524
1525 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1526 return (ENOENT);
1527
1528 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1529 ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1530 real, maxlen, conflict));
1531 }
1532
1533 int
1534 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1535 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1536 {
1537 dsl_dataset_t *ds = os->os_dsl_dataset;
1538 zap_cursor_t cursor;
1539 zap_attribute_t attr;
1540
1541 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1542 return (ENOENT);
1543
1544 zap_cursor_init_serialized(&cursor,
1545 ds->ds_dir->dd_pool->dp_meta_objset,
1546 ds->ds_phys->ds_snapnames_zapobj, *offp);
1547
1548 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1549 zap_cursor_fini(&cursor);
1550 return (ENOENT);
1551 }
1552
1553 if (strlen(attr.za_name) + 1 > namelen) {
1554 zap_cursor_fini(&cursor);
1555 return (ENAMETOOLONG);
1556 }
1557
1558 (void) strcpy(name, attr.za_name);
1559 if (idp)
1560 *idp = attr.za_first_integer;
1561 if (case_conflict)
1562 *case_conflict = attr.za_normalization_conflict;
1563 zap_cursor_advance(&cursor);
1564 *offp = zap_cursor_serialize(&cursor);
1565 zap_cursor_fini(&cursor);
1566
1567 return (0);
1568 }
1569
1570 int
1571 dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value)
1572 {
1573 return dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value);
1574 }
1575
1576 int
1577 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1578 uint64_t *idp, uint64_t *offp)
1579 {
1580 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1581 zap_cursor_t cursor;
1582 zap_attribute_t attr;
1583
1584 /* there is no next dir on a snapshot! */
1585 if (os->os_dsl_dataset->ds_object !=
1586 dd->dd_phys->dd_head_dataset_obj)
1587 return (ENOENT);
1588
1589 zap_cursor_init_serialized(&cursor,
1590 dd->dd_pool->dp_meta_objset,
1591 dd->dd_phys->dd_child_dir_zapobj, *offp);
1592
1593 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1594 zap_cursor_fini(&cursor);
1595 return (ENOENT);
1596 }
1597
1598 if (strlen(attr.za_name) + 1 > namelen) {
1599 zap_cursor_fini(&cursor);
1600 return (ENAMETOOLONG);
1601 }
1602
1603 (void) strcpy(name, attr.za_name);
1604 if (idp)
1605 *idp = attr.za_first_integer;
1606 zap_cursor_advance(&cursor);
1607 *offp = zap_cursor_serialize(&cursor);
1608 zap_cursor_fini(&cursor);
1609
1610 return (0);
1611 }
1612
1613 struct findarg {
1614 int (*func)(const char *, void *);
1615 void *arg;
1616 };
1617
1618 /* ARGSUSED */
1619 static int
1620 findfunc(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
1621 {
1622 struct findarg *fa = arg;
1623 return (fa->func(dsname, fa->arg));
1624 }
1625
1626 /*
1627 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1628 * Perhaps change all callers to use dmu_objset_find_spa()?
1629 */
1630 int
1631 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1632 int flags)
1633 {
1634 struct findarg fa;
1635 fa.func = func;
1636 fa.arg = arg;
1637 return (dmu_objset_find_spa(NULL, name, findfunc, &fa, flags));
1638 }
1639
1640 /*
1641 * Find all objsets under name, call func on each
1642 */
1643 int
1644 dmu_objset_find_spa(spa_t *spa, const char *name,
1645 int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags)
1646 {
1647 dsl_dir_t *dd;
1648 dsl_pool_t *dp;
1649 dsl_dataset_t *ds;
1650 zap_cursor_t zc;
1651 zap_attribute_t *attr;
1652 char *child;
1653 uint64_t thisobj;
1654 int err;
1655
1656 if (name == NULL)
1657 name = spa_name(spa);
1658 err = dsl_dir_open_spa(spa, name, FTAG, &dd, NULL);
1659 if (err)
1660 return (err);
1661
1662 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1663 if (dd->dd_myname[0] == '$') {
1664 dsl_dir_close(dd, FTAG);
1665 return (0);
1666 }
1667
1668 thisobj = dd->dd_phys->dd_head_dataset_obj;
1669 attr = kmem_alloc(sizeof (zap_attribute_t), KM_PUSHPAGE);
1670 dp = dd->dd_pool;
1671
1672 /*
1673 * Iterate over all children.
1674 */
1675 if (flags & DS_FIND_CHILDREN) {
1676 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1677 dd->dd_phys->dd_child_dir_zapobj);
1678 zap_cursor_retrieve(&zc, attr) == 0;
1679 (void) zap_cursor_advance(&zc)) {
1680 ASSERT(attr->za_integer_length == sizeof (uint64_t));
1681 ASSERT(attr->za_num_integers == 1);
1682
1683 child = kmem_asprintf("%s/%s", name, attr->za_name);
1684 err = dmu_objset_find_spa(spa, child, func, arg, flags);
1685 strfree(child);
1686 if (err)
1687 break;
1688 }
1689 zap_cursor_fini(&zc);
1690
1691 if (err) {
1692 dsl_dir_close(dd, FTAG);
1693 kmem_free(attr, sizeof (zap_attribute_t));
1694 return (err);
1695 }
1696 }
1697
1698 /*
1699 * Iterate over all snapshots.
1700 */
1701 if (flags & DS_FIND_SNAPSHOTS) {
1702 if (!dsl_pool_sync_context(dp))
1703 rw_enter(&dp->dp_config_rwlock, RW_READER);
1704 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1705 if (!dsl_pool_sync_context(dp))
1706 rw_exit(&dp->dp_config_rwlock);
1707
1708 if (err == 0) {
1709 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1710 dsl_dataset_rele(ds, FTAG);
1711
1712 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1713 zap_cursor_retrieve(&zc, attr) == 0;
1714 (void) zap_cursor_advance(&zc)) {
1715 ASSERT(attr->za_integer_length ==
1716 sizeof (uint64_t));
1717 ASSERT(attr->za_num_integers == 1);
1718
1719 child = kmem_asprintf("%s@%s",
1720 name, attr->za_name);
1721 err = func(spa, attr->za_first_integer,
1722 child, arg);
1723 strfree(child);
1724 if (err)
1725 break;
1726 }
1727 zap_cursor_fini(&zc);
1728 }
1729 }
1730
1731 dsl_dir_close(dd, FTAG);
1732 kmem_free(attr, sizeof (zap_attribute_t));
1733
1734 if (err)
1735 return (err);
1736
1737 /*
1738 * Apply to self if appropriate.
1739 */
1740 err = func(spa, thisobj, name, arg);
1741 return (err);
1742 }
1743
1744 /* ARGSUSED */
1745 int
1746 dmu_objset_prefetch(const char *name, void *arg)
1747 {
1748 dsl_dataset_t *ds;
1749
1750 if (dsl_dataset_hold(name, FTAG, &ds))
1751 return (0);
1752
1753 if (!BP_IS_HOLE(&ds->ds_phys->ds_bp)) {
1754 mutex_enter(&ds->ds_opening_lock);
1755 if (ds->ds_objset == NULL) {
1756 uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
1757 zbookmark_t zb;
1758
1759 SET_BOOKMARK(&zb, ds->ds_object, ZB_ROOT_OBJECT,
1760 ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1761
1762 (void) arc_read(NULL, dsl_dataset_get_spa(ds),
1763 &ds->ds_phys->ds_bp, NULL, NULL,
1764 ZIO_PRIORITY_ASYNC_READ,
1765 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
1766 &aflags, &zb);
1767 }
1768 mutex_exit(&ds->ds_opening_lock);
1769 }
1770
1771 dsl_dataset_rele(ds, FTAG);
1772 return (0);
1773 }
1774
1775 void
1776 dmu_objset_set_user(objset_t *os, void *user_ptr)
1777 {
1778 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1779 os->os_user_ptr = user_ptr;
1780 }
1781
1782 void *
1783 dmu_objset_get_user(objset_t *os)
1784 {
1785 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1786 return (os->os_user_ptr);
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_destroy);
1803 EXPORT_SYMBOL(dmu_objset_snapshot);
1804 EXPORT_SYMBOL(dmu_objset_stats);
1805 EXPORT_SYMBOL(dmu_objset_fast_stat);
1806 EXPORT_SYMBOL(dmu_objset_spa);
1807 EXPORT_SYMBOL(dmu_objset_space);
1808 EXPORT_SYMBOL(dmu_objset_fsid_guid);
1809 EXPORT_SYMBOL(dmu_objset_find);
1810 EXPORT_SYMBOL(dmu_objset_find_spa);
1811 EXPORT_SYMBOL(dmu_objset_prefetch);
1812 EXPORT_SYMBOL(dmu_objset_byteswap);
1813 EXPORT_SYMBOL(dmu_objset_evict_dbufs);
1814 EXPORT_SYMBOL(dmu_objset_snap_cmtime);
1815
1816 EXPORT_SYMBOL(dmu_objset_sync);
1817 EXPORT_SYMBOL(dmu_objset_is_dirty);
1818 EXPORT_SYMBOL(dmu_objset_create_impl);
1819 EXPORT_SYMBOL(dmu_objset_open_impl);
1820 EXPORT_SYMBOL(dmu_objset_evict);
1821 EXPORT_SYMBOL(dmu_objset_register_type);
1822 EXPORT_SYMBOL(dmu_objset_do_userquota_updates);
1823 EXPORT_SYMBOL(dmu_objset_userquota_get_ids);
1824 EXPORT_SYMBOL(dmu_objset_userused_enabled);
1825 EXPORT_SYMBOL(dmu_objset_userspace_upgrade);
1826 EXPORT_SYMBOL(dmu_objset_userspace_present);
1827 #endif
ZFS On Linux mirror for PVE