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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 /* Portions Copyright 2010 Robert Milkowski */
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>
39 #include <sys/dmu_tx.h>
42 #include <sys/dmu_impl.h>
43 #include <sys/zfs_ioctl.h>
45 #include <sys/zfs_onexit.h>
48 * Needed to close a window in dnode_move() that allows the objset to be freed
49 * before it can be safely accessed.
56 rw_init(&os_lock
, NULL
, RW_DEFAULT
, NULL
);
66 dmu_objset_spa(objset_t
*os
)
72 dmu_objset_zil(objset_t
*os
)
78 dmu_objset_pool(objset_t
*os
)
82 if ((ds
= os
->os_dsl_dataset
) != NULL
&& ds
->ds_dir
)
83 return (ds
->ds_dir
->dd_pool
);
85 return (spa_get_dsl(os
->os_spa
));
89 dmu_objset_ds(objset_t
*os
)
91 return (os
->os_dsl_dataset
);
95 dmu_objset_type(objset_t
*os
)
97 return (os
->os_phys
->os_type
);
101 dmu_objset_name(objset_t
*os
, char *buf
)
103 dsl_dataset_name(os
->os_dsl_dataset
, buf
);
107 dmu_objset_id(objset_t
*os
)
109 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
111 return (ds
? ds
->ds_object
: 0);
115 dmu_objset_syncprop(objset_t
*os
)
117 return (os
->os_sync
);
121 dmu_objset_logbias(objset_t
*os
)
123 return (os
->os_logbias
);
127 checksum_changed_cb(void *arg
, uint64_t newval
)
132 * Inheritance should have been done by now.
134 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
136 os
->os_checksum
= zio_checksum_select(newval
, ZIO_CHECKSUM_ON_VALUE
);
140 compression_changed_cb(void *arg
, uint64_t newval
)
145 * Inheritance and range checking should have been done by now.
147 ASSERT(newval
!= ZIO_COMPRESS_INHERIT
);
149 os
->os_compress
= zio_compress_select(newval
, ZIO_COMPRESS_ON_VALUE
);
153 copies_changed_cb(void *arg
, uint64_t newval
)
158 * Inheritance and range checking should have been done by now.
161 ASSERT(newval
<= spa_max_replication(os
->os_spa
));
163 os
->os_copies
= newval
;
167 dedup_changed_cb(void *arg
, uint64_t newval
)
170 spa_t
*spa
= os
->os_spa
;
171 enum zio_checksum checksum
;
174 * Inheritance should have been done by now.
176 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
178 checksum
= zio_checksum_dedup_select(spa
, newval
, ZIO_CHECKSUM_OFF
);
180 os
->os_dedup_checksum
= checksum
& ZIO_CHECKSUM_MASK
;
181 os
->os_dedup_verify
= !!(checksum
& ZIO_CHECKSUM_VERIFY
);
185 primary_cache_changed_cb(void *arg
, uint64_t newval
)
190 * Inheritance and range checking should have been done by now.
192 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
193 newval
== ZFS_CACHE_METADATA
);
195 os
->os_primary_cache
= newval
;
199 secondary_cache_changed_cb(void *arg
, uint64_t newval
)
204 * Inheritance and range checking should have been done by now.
206 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
207 newval
== ZFS_CACHE_METADATA
);
209 os
->os_secondary_cache
= newval
;
213 sync_changed_cb(void *arg
, uint64_t newval
)
218 * Inheritance and range checking should have been done by now.
220 ASSERT(newval
== ZFS_SYNC_STANDARD
|| newval
== ZFS_SYNC_ALWAYS
||
221 newval
== ZFS_SYNC_DISABLED
);
223 os
->os_sync
= newval
;
225 zil_set_sync(os
->os_zil
, newval
);
229 logbias_changed_cb(void *arg
, uint64_t newval
)
233 ASSERT(newval
== ZFS_LOGBIAS_LATENCY
||
234 newval
== ZFS_LOGBIAS_THROUGHPUT
);
235 os
->os_logbias
= newval
;
237 zil_set_logbias(os
->os_zil
, newval
);
241 dmu_objset_byteswap(void *buf
, size_t size
)
243 objset_phys_t
*osp
= buf
;
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
);
257 dmu_objset_open_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
263 ASSERT(ds
== NULL
|| MUTEX_HELD(&ds
->ds_opening_lock
));
265 os
= kmem_zalloc(sizeof (objset_t
), KM_SLEEP
);
266 os
->os_dsl_dataset
= ds
;
269 if (!BP_IS_HOLE(os
->os_rootbp
)) {
270 uint32_t aflags
= ARC_WAIT
;
272 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
273 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
275 if (DMU_OS_IS_L2CACHEABLE(os
))
276 aflags
|= ARC_L2CACHE
;
278 dprintf_bp(os
->os_rootbp
, "reading %s", "");
280 * XXX when bprewrite scrub can change the bp,
281 * and this is called from dmu_objset_open_ds_os, the bp
282 * could change, and we'll need a lock.
284 err
= dsl_read_nolock(NULL
, spa
, os
->os_rootbp
,
285 arc_getbuf_func
, &os
->os_phys_buf
,
286 ZIO_PRIORITY_SYNC_READ
, ZIO_FLAG_CANFAIL
, &aflags
, &zb
);
288 kmem_free(os
, sizeof (objset_t
));
289 /* convert checksum errors into IO errors */
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
,
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
,
306 os
->os_phys_buf
= buf
;
309 os
->os_phys
= os
->os_phys_buf
->b_data
;
310 os
->os_flags
= os
->os_phys
->os_flags
;
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
);
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.
327 err
= dsl_prop_register(ds
, "primarycache",
328 primary_cache_changed_cb
, os
);
330 err
= dsl_prop_register(ds
, "secondarycache",
331 secondary_cache_changed_cb
, os
);
332 if (!dsl_dataset_is_snapshot(ds
)) {
334 err
= dsl_prop_register(ds
, "checksum",
335 checksum_changed_cb
, os
);
337 err
= dsl_prop_register(ds
, "compression",
338 compression_changed_cb
, os
);
340 err
= dsl_prop_register(ds
, "copies",
341 copies_changed_cb
, os
);
343 err
= dsl_prop_register(ds
, "dedup",
344 dedup_changed_cb
, os
);
346 err
= dsl_prop_register(ds
, "logbias",
347 logbias_changed_cb
, os
);
349 err
= dsl_prop_register(ds
, "sync",
350 sync_changed_cb
, os
);
353 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
,
354 &os
->os_phys_buf
) == 1);
355 kmem_free(os
, sizeof (objset_t
));
358 } else if (ds
== NULL
) {
359 /* It's the meta-objset. */
360 os
->os_checksum
= ZIO_CHECKSUM_FLETCHER_4
;
361 os
->os_compress
= ZIO_COMPRESS_LZJB
;
362 os
->os_copies
= spa_max_replication(spa
);
363 os
->os_dedup_checksum
= ZIO_CHECKSUM_OFF
;
364 os
->os_dedup_verify
= 0;
367 os
->os_primary_cache
= ZFS_CACHE_ALL
;
368 os
->os_secondary_cache
= ZFS_CACHE_ALL
;
371 if (ds
== NULL
|| !dsl_dataset_is_snapshot(ds
))
372 os
->os_zil_header
= os
->os_phys
->os_zil_header
;
373 os
->os_zil
= zil_alloc(os
, &os
->os_zil_header
);
375 for (i
= 0; i
< TXG_SIZE
; i
++) {
376 list_create(&os
->os_dirty_dnodes
[i
], sizeof (dnode_t
),
377 offsetof(dnode_t
, dn_dirty_link
[i
]));
378 list_create(&os
->os_free_dnodes
[i
], sizeof (dnode_t
),
379 offsetof(dnode_t
, dn_dirty_link
[i
]));
381 list_create(&os
->os_dnodes
, sizeof (dnode_t
),
382 offsetof(dnode_t
, dn_link
));
383 list_create(&os
->os_downgraded_dbufs
, sizeof (dmu_buf_impl_t
),
384 offsetof(dmu_buf_impl_t
, db_link
));
386 mutex_init(&os
->os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
387 mutex_init(&os
->os_obj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
388 mutex_init(&os
->os_user_ptr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
390 DMU_META_DNODE(os
) = dnode_special_open(os
,
391 &os
->os_phys
->os_meta_dnode
, DMU_META_DNODE_OBJECT
,
393 if (arc_buf_size(os
->os_phys_buf
) >= sizeof (objset_phys_t
)) {
394 DMU_USERUSED_DNODE(os
) = dnode_special_open(os
,
395 &os
->os_phys
->os_userused_dnode
, DMU_USERUSED_OBJECT
,
396 &os
->os_userused_dnode
);
397 DMU_GROUPUSED_DNODE(os
) = dnode_special_open(os
,
398 &os
->os_phys
->os_groupused_dnode
, DMU_GROUPUSED_OBJECT
,
399 &os
->os_groupused_dnode
);
403 * We should be the only thread trying to do this because we
404 * have ds_opening_lock
407 mutex_enter(&ds
->ds_lock
);
408 ASSERT(ds
->ds_objset
== NULL
);
410 mutex_exit(&ds
->ds_lock
);
418 dmu_objset_from_ds(dsl_dataset_t
*ds
, objset_t
**osp
)
422 mutex_enter(&ds
->ds_opening_lock
);
423 *osp
= ds
->ds_objset
;
425 err
= dmu_objset_open_impl(dsl_dataset_get_spa(ds
),
426 ds
, dsl_dataset_get_blkptr(ds
), osp
);
428 mutex_exit(&ds
->ds_opening_lock
);
432 /* called from zpl */
434 dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
)
439 err
= dsl_dataset_hold(name
, tag
, &ds
);
443 err
= dmu_objset_from_ds(ds
, osp
);
445 dsl_dataset_rele(ds
, tag
);
450 /* called from zpl */
452 dmu_objset_own(const char *name
, dmu_objset_type_t type
,
453 boolean_t readonly
, void *tag
, objset_t
**osp
)
458 err
= dsl_dataset_own(name
, B_FALSE
, tag
, &ds
);
462 err
= dmu_objset_from_ds(ds
, osp
);
464 dsl_dataset_disown(ds
, tag
);
465 } else if (type
!= DMU_OST_ANY
&& type
!= (*osp
)->os_phys
->os_type
) {
466 dmu_objset_disown(*osp
, tag
);
468 } else if (!readonly
&& dsl_dataset_is_snapshot(ds
)) {
469 dmu_objset_disown(*osp
, tag
);
476 dmu_objset_rele(objset_t
*os
, void *tag
)
478 dsl_dataset_rele(os
->os_dsl_dataset
, tag
);
482 dmu_objset_disown(objset_t
*os
, void *tag
)
484 dsl_dataset_disown(os
->os_dsl_dataset
, tag
);
488 dmu_objset_evict_dbufs(objset_t
*os
)
492 mutex_enter(&os
->os_lock
);
494 /* process the mdn last, since the other dnodes have holds on it */
495 list_remove(&os
->os_dnodes
, DMU_META_DNODE(os
));
496 list_insert_tail(&os
->os_dnodes
, DMU_META_DNODE(os
));
499 * Find the first dnode with holds. We have to do this dance
500 * because dnode_add_ref() only works if you already have a
501 * hold. If there are no holds then it has no dbufs so OK to
504 for (dn
= list_head(&os
->os_dnodes
);
505 dn
&& !dnode_add_ref(dn
, FTAG
);
506 dn
= list_next(&os
->os_dnodes
, dn
))
510 dnode_t
*next_dn
= dn
;
513 next_dn
= list_next(&os
->os_dnodes
, next_dn
);
514 } while (next_dn
&& !dnode_add_ref(next_dn
, FTAG
));
516 mutex_exit(&os
->os_lock
);
517 dnode_evict_dbufs(dn
);
518 dnode_rele(dn
, FTAG
);
519 mutex_enter(&os
->os_lock
);
522 dn
= list_head(&os
->os_dnodes
);
523 mutex_exit(&os
->os_lock
);
524 return (dn
!= DMU_META_DNODE(os
));
528 dmu_objset_evict(objset_t
*os
)
530 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
532 for (int t
= 0; t
< TXG_SIZE
; t
++)
533 ASSERT(!dmu_objset_is_dirty(os
, t
));
536 if (!dsl_dataset_is_snapshot(ds
)) {
537 VERIFY(0 == dsl_prop_unregister(ds
, "checksum",
538 checksum_changed_cb
, os
));
539 VERIFY(0 == dsl_prop_unregister(ds
, "compression",
540 compression_changed_cb
, os
));
541 VERIFY(0 == dsl_prop_unregister(ds
, "copies",
542 copies_changed_cb
, os
));
543 VERIFY(0 == dsl_prop_unregister(ds
, "dedup",
544 dedup_changed_cb
, os
));
545 VERIFY(0 == dsl_prop_unregister(ds
, "logbias",
546 logbias_changed_cb
, os
));
547 VERIFY(0 == dsl_prop_unregister(ds
, "sync",
548 sync_changed_cb
, os
));
550 VERIFY(0 == dsl_prop_unregister(ds
, "primarycache",
551 primary_cache_changed_cb
, os
));
552 VERIFY(0 == dsl_prop_unregister(ds
, "secondarycache",
553 secondary_cache_changed_cb
, os
));
560 * We should need only a single pass over the dnode list, since
561 * nothing can be added to the list at this point.
563 (void) dmu_objset_evict_dbufs(os
);
565 dnode_special_close(&os
->os_meta_dnode
);
566 if (DMU_USERUSED_DNODE(os
)) {
567 dnode_special_close(&os
->os_userused_dnode
);
568 dnode_special_close(&os
->os_groupused_dnode
);
570 zil_free(os
->os_zil
);
572 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
574 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
, &os
->os_phys_buf
) == 1);
577 * This is a barrier to prevent the objset from going away in
578 * dnode_move() until we can safely ensure that the objset is still in
579 * use. We consider the objset valid before the barrier and invalid
582 rw_enter(&os_lock
, RW_READER
);
585 mutex_destroy(&os
->os_lock
);
586 mutex_destroy(&os
->os_obj_lock
);
587 mutex_destroy(&os
->os_user_ptr_lock
);
588 kmem_free(os
, sizeof (objset_t
));
592 dmu_objset_snap_cmtime(objset_t
*os
)
594 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
597 /* called from dsl for meta-objset */
599 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
600 dmu_objset_type_t type
, dmu_tx_t
*tx
)
605 ASSERT(dmu_tx_is_syncing(tx
));
607 VERIFY(0 == dmu_objset_from_ds(ds
, &os
));
609 VERIFY(0 == dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
611 mdn
= DMU_META_DNODE(os
);
613 dnode_allocate(mdn
, DMU_OT_DNODE
, 1 << DNODE_BLOCK_SHIFT
,
614 DN_MAX_INDBLKSHIFT
, DMU_OT_NONE
, 0, tx
);
617 * We don't want to have to increase the meta-dnode's nlevels
618 * later, because then we could do it in quescing context while
619 * we are also accessing it in open context.
621 * This precaution is not necessary for the MOS (ds == NULL),
622 * because the MOS is only updated in syncing context.
623 * This is most fortunate: the MOS is the only objset that
624 * needs to be synced multiple times as spa_sync() iterates
625 * to convergence, so minimizing its dn_nlevels matters.
631 * Determine the number of levels necessary for the meta-dnode
632 * to contain DN_MAX_OBJECT dnodes.
634 while ((uint64_t)mdn
->dn_nblkptr
<< (mdn
->dn_datablkshift
+
635 (levels
- 1) * (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
636 DN_MAX_OBJECT
* sizeof (dnode_phys_t
))
639 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
640 mdn
->dn_nlevels
= levels
;
643 ASSERT(type
!= DMU_OST_NONE
);
644 ASSERT(type
!= DMU_OST_ANY
);
645 ASSERT(type
< DMU_OST_NUMTYPES
);
646 os
->os_phys
->os_type
= type
;
647 if (dmu_objset_userused_enabled(os
)) {
648 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
649 os
->os_flags
= os
->os_phys
->os_flags
;
652 dsl_dataset_dirty(ds
, tx
);
658 void (*userfunc
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
);
660 dsl_dataset_t
*clone_origin
;
661 const char *lastname
;
662 dmu_objset_type_t type
;
669 dmu_objset_create_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
671 dsl_dir_t
*dd
= arg1
;
672 struct oscarg
*oa
= arg2
;
673 objset_t
*mos
= dd
->dd_pool
->dp_meta_objset
;
677 err
= zap_lookup(mos
, dd
->dd_phys
->dd_child_dir_zapobj
,
678 oa
->lastname
, sizeof (uint64_t), 1, &ddobj
);
680 return (err
? err
: EEXIST
);
682 if (oa
->clone_origin
!= NULL
) {
683 /* You can't clone across pools. */
684 if (oa
->clone_origin
->ds_dir
->dd_pool
!= dd
->dd_pool
)
687 /* You can only clone snapshots, not the head datasets. */
688 if (!dsl_dataset_is_snapshot(oa
->clone_origin
))
696 dmu_objset_create_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
698 dsl_dir_t
*dd
= arg1
;
699 spa_t
*spa
= dd
->dd_pool
->dp_spa
;
700 struct oscarg
*oa
= arg2
;
703 ASSERT(dmu_tx_is_syncing(tx
));
705 obj
= dsl_dataset_create_sync(dd
, oa
->lastname
,
706 oa
->clone_origin
, oa
->flags
, oa
->cr
, tx
);
708 if (oa
->clone_origin
== NULL
) {
709 dsl_pool_t
*dp
= dd
->dd_pool
;
714 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
));
715 bp
= dsl_dataset_get_blkptr(ds
);
716 ASSERT(BP_IS_HOLE(bp
));
718 os
= dmu_objset_create_impl(spa
, ds
, bp
, oa
->type
, tx
);
721 oa
->userfunc(os
, oa
->userarg
, oa
->cr
, tx
);
722 dsl_dataset_rele(ds
, FTAG
);
725 spa_history_log_internal(LOG_DS_CREATE
, spa
, tx
, "dataset = %llu", obj
);
729 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
730 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
)
735 struct oscarg oa
= { 0 };
737 ASSERT(strchr(name
, '@') == NULL
);
738 err
= dsl_dir_open(name
, FTAG
, &pdd
, &tail
);
742 dsl_dir_close(pdd
, FTAG
);
753 err
= dsl_sync_task_do(pdd
->dd_pool
, dmu_objset_create_check
,
754 dmu_objset_create_sync
, pdd
, &oa
, 5);
755 dsl_dir_close(pdd
, FTAG
);
760 dmu_objset_clone(const char *name
, dsl_dataset_t
*clone_origin
, uint64_t flags
)
765 struct oscarg oa
= { 0 };
767 ASSERT(strchr(name
, '@') == NULL
);
768 err
= dsl_dir_open(name
, FTAG
, &pdd
, &tail
);
772 dsl_dir_close(pdd
, FTAG
);
777 oa
.clone_origin
= clone_origin
;
781 err
= dsl_sync_task_do(pdd
->dd_pool
, dmu_objset_create_check
,
782 dmu_objset_create_sync
, pdd
, &oa
, 5);
783 dsl_dir_close(pdd
, FTAG
);
788 dmu_objset_destroy(const char *name
, boolean_t defer
)
793 error
= dsl_dataset_own(name
, B_TRUE
, FTAG
, &ds
);
795 error
= dsl_dataset_destroy(ds
, FTAG
, defer
);
796 /* dsl_dataset_destroy() closes the ds. */
803 dsl_sync_task_group_t
*dstg
;
806 char failed
[MAXPATHLEN
];
808 boolean_t needsuspend
;
811 struct dsl_ds_holdarg
*ha
; /* only needed in the temporary case */
812 dsl_dataset_t
*newds
;
816 snapshot_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
819 struct snaparg
*sn
= arg2
;
822 /* The props have already been checked by zfs_check_userprops(). */
824 error
= dsl_dataset_snapshot_check(os
->os_dsl_dataset
,
831 * Ideally we would just call
832 * dsl_dataset_user_hold_check() and
833 * dsl_dataset_destroy_check() here. However the
834 * dataset we want to hold and destroy is the snapshot
835 * that we just confirmed we can create, but it won't
836 * exist until after these checks are run. Do any
837 * checks we can here and if more checks are added to
838 * those routines in the future, similar checks may be
841 if (spa_version(os
->os_spa
) < SPA_VERSION_USERREFS
)
844 * Not checking number of tags because the tag will be
845 * unique, as it will be the only tag.
847 if (strlen(sn
->htag
) + MAX_TAG_PREFIX_LEN
>= MAXNAMELEN
)
850 sn
->ha
= kmem_alloc(sizeof (struct dsl_ds_holdarg
), KM_SLEEP
);
851 sn
->ha
->temphold
= B_TRUE
;
852 sn
->ha
->htag
= sn
->htag
;
858 snapshot_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
861 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
862 struct snaparg
*sn
= arg2
;
864 dsl_dataset_snapshot_sync(ds
, sn
->snapname
, tx
);
868 pa
.pa_props
= sn
->props
;
869 pa
.pa_source
= ZPROP_SRC_LOCAL
;
870 dsl_props_set_sync(ds
->ds_prev
, &pa
, tx
);
874 struct dsl_ds_destroyarg da
;
876 dsl_dataset_user_hold_sync(ds
->ds_prev
, sn
->ha
, tx
);
877 kmem_free(sn
->ha
, sizeof (struct dsl_ds_holdarg
));
879 sn
->newds
= ds
->ds_prev
;
883 dsl_dataset_destroy_sync(&da
, FTAG
, tx
);
888 dmu_objset_snapshot_one(const char *name
, void *arg
)
890 struct snaparg
*sn
= arg
;
896 * If the objset starts with a '%', then ignore it unless it was
897 * explicitly named (ie, not recursive). These hidden datasets
898 * are always inconsistent, and by not opening them here, we can
899 * avoid a race with dsl_dir_destroy_check().
901 cp
= strrchr(name
, '/');
902 if (cp
&& cp
[1] == '%' && sn
->recursive
)
905 (void) strcpy(sn
->failed
, name
);
908 * Check permissions if we are doing a recursive snapshot. The
909 * permission checks for the starting dataset have already been
910 * performed in zfs_secpolicy_snapshot()
912 if (sn
->recursive
&& (err
= zfs_secpolicy_snapshot_perms(name
, CRED())))
915 err
= dmu_objset_hold(name
, sn
, &os
);
920 * If the objset is in an inconsistent state (eg, in the process
921 * of being destroyed), don't snapshot it. As with %hidden
922 * datasets, we return EBUSY if this name was explicitly
923 * requested (ie, not recursive), and otherwise ignore it.
925 if (os
->os_dsl_dataset
->ds_phys
->ds_flags
& DS_FLAG_INCONSISTENT
) {
926 dmu_objset_rele(os
, sn
);
927 return (sn
->recursive
? 0 : EBUSY
);
930 if (sn
->needsuspend
) {
931 err
= zil_suspend(dmu_objset_zil(os
));
933 dmu_objset_rele(os
, sn
);
937 dsl_sync_task_create(sn
->dstg
, snapshot_check
, snapshot_sync
,
944 dmu_objset_snapshot(char *fsname
, char *snapname
, char *tag
,
945 nvlist_t
*props
, boolean_t recursive
, boolean_t temporary
, int cleanup_fd
)
947 dsl_sync_task_t
*dst
;
953 (void) strcpy(sn
.failed
, fsname
);
955 err
= spa_open(fsname
, &spa
, FTAG
);
960 if (cleanup_fd
< 0) {
961 spa_close(spa
, FTAG
);
964 if ((err
= zfs_onexit_fd_hold(cleanup_fd
, &minor
)) != 0) {
965 spa_close(spa
, FTAG
);
970 sn
.dstg
= dsl_sync_task_group_create(spa_get_dsl(spa
));
971 sn
.snapname
= snapname
;
974 sn
.recursive
= recursive
;
975 sn
.needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
976 sn
.temporary
= temporary
;
981 err
= dmu_objset_find(fsname
,
982 dmu_objset_snapshot_one
, &sn
, DS_FIND_CHILDREN
);
984 err
= dmu_objset_snapshot_one(fsname
, &sn
);
988 err
= dsl_sync_task_group_wait(sn
.dstg
);
990 for (dst
= list_head(&sn
.dstg
->dstg_tasks
); dst
;
991 dst
= list_next(&sn
.dstg
->dstg_tasks
, dst
)) {
992 objset_t
*os
= dst
->dst_arg1
;
993 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
995 dsl_dataset_name(ds
, sn
.failed
);
996 } else if (temporary
) {
997 dsl_register_onexit_hold_cleanup(sn
.newds
, tag
, minor
);
1000 zil_resume(dmu_objset_zil(os
));
1001 dmu_objset_rele(os
, &sn
);
1005 (void) strcpy(fsname
, sn
.failed
);
1007 zfs_onexit_fd_rele(cleanup_fd
);
1008 dsl_sync_task_group_destroy(sn
.dstg
);
1009 spa_close(spa
, FTAG
);
1014 dmu_objset_sync_dnodes(list_t
*list
, list_t
*newlist
, dmu_tx_t
*tx
)
1018 while (dn
= list_head(list
)) {
1019 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1020 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1022 * Initialize dn_zio outside dnode_sync() because the
1023 * meta-dnode needs to set it ouside dnode_sync().
1025 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1028 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1029 list_remove(list
, dn
);
1032 (void) dnode_add_ref(dn
, newlist
);
1033 list_insert_tail(newlist
, dn
);
1042 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1044 blkptr_t
*bp
= zio
->io_bp
;
1046 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1048 ASSERT(bp
== os
->os_rootbp
);
1049 ASSERT(BP_GET_TYPE(bp
) == DMU_OT_OBJSET
);
1050 ASSERT(BP_GET_LEVEL(bp
) == 0);
1053 * Update rootbp fill count: it should be the number of objects
1054 * allocated in the object set (not counting the "special"
1055 * objects that are stored in the objset_phys_t -- the meta
1056 * dnode and user/group accounting objects).
1059 for (int i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1060 bp
->blk_fill
+= dnp
->dn_blkptr
[i
].blk_fill
;
1065 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1067 blkptr_t
*bp
= zio
->io_bp
;
1068 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1071 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1072 ASSERT(BP_EQUAL(bp
, bp_orig
));
1074 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1075 dmu_tx_t
*tx
= os
->os_synctx
;
1077 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1078 dsl_dataset_block_born(ds
, bp
, tx
);
1082 /* called from dsl */
1084 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1091 list_t
*newlist
= NULL
;
1092 dbuf_dirty_record_t
*dr
;
1094 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", tx
->tx_txg
);
1096 ASSERT(dmu_tx_is_syncing(tx
));
1097 /* XXX the write_done callback should really give us the tx... */
1100 if (os
->os_dsl_dataset
== NULL
) {
1102 * This is the MOS. If we have upgraded,
1103 * spa_max_replication() could change, so reset
1106 os
->os_copies
= spa_max_replication(os
->os_spa
);
1110 * Create the root block IO
1112 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1113 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1114 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1115 VERIFY3U(0, ==, arc_release_bp(os
->os_phys_buf
, &os
->os_phys_buf
,
1116 os
->os_rootbp
, os
->os_spa
, &zb
));
1118 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1120 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1121 os
->os_rootbp
, os
->os_phys_buf
, DMU_OS_IS_L2CACHEABLE(os
), &zp
,
1122 dmu_objset_write_ready
, dmu_objset_write_done
, os
,
1123 ZIO_PRIORITY_ASYNC_WRITE
, ZIO_FLAG_MUSTSUCCEED
, &zb
);
1126 * Sync special dnodes - the parent IO for the sync is the root block
1128 DMU_META_DNODE(os
)->dn_zio
= zio
;
1129 dnode_sync(DMU_META_DNODE(os
), tx
);
1131 os
->os_phys
->os_flags
= os
->os_flags
;
1133 if (DMU_USERUSED_DNODE(os
) &&
1134 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1135 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1136 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1137 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1138 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1141 txgoff
= tx
->tx_txg
& TXG_MASK
;
1143 if (dmu_objset_userused_enabled(os
)) {
1144 newlist
= &os
->os_synced_dnodes
;
1146 * We must create the list here because it uses the
1147 * dn_dirty_link[] of this txg.
1149 list_create(newlist
, sizeof (dnode_t
),
1150 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1153 dmu_objset_sync_dnodes(&os
->os_free_dnodes
[txgoff
], newlist
, tx
);
1154 dmu_objset_sync_dnodes(&os
->os_dirty_dnodes
[txgoff
], newlist
, tx
);
1156 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1157 while (dr
= list_head(list
)) {
1158 ASSERT(dr
->dr_dbuf
->db_level
== 0);
1159 list_remove(list
, dr
);
1161 zio_nowait(dr
->dr_zio
);
1164 * Free intent log blocks up to this tx.
1166 zil_sync(os
->os_zil
, tx
);
1167 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1172 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1174 return (!list_is_empty(&os
->os_dirty_dnodes
[txg
& TXG_MASK
]) ||
1175 !list_is_empty(&os
->os_free_dnodes
[txg
& TXG_MASK
]));
1179 dmu_objset_is_dirty_anywhere(objset_t
*os
)
1181 for (int t
= 0; t
< TXG_SIZE
; t
++)
1182 if (dmu_objset_is_dirty(os
, t
))
1187 static objset_used_cb_t
*used_cbs
[DMU_OST_NUMTYPES
];
1190 dmu_objset_register_type(dmu_objset_type_t ost
, objset_used_cb_t
*cb
)
1196 dmu_objset_userused_enabled(objset_t
*os
)
1198 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1199 used_cbs
[os
->os_phys
->os_type
] != NULL
&&
1200 DMU_USERUSED_DNODE(os
) != NULL
);
1204 do_userquota_update(objset_t
*os
, uint64_t used
, uint64_t flags
,
1205 uint64_t user
, uint64_t group
, boolean_t subtract
, dmu_tx_t
*tx
)
1207 if ((flags
& DNODE_FLAG_USERUSED_ACCOUNTED
)) {
1208 int64_t delta
= DNODE_SIZE
+ used
;
1211 VERIFY3U(0, ==, zap_increment_int(os
, DMU_USERUSED_OBJECT
,
1213 VERIFY3U(0, ==, zap_increment_int(os
, DMU_GROUPUSED_OBJECT
,
1219 dmu_objset_do_userquota_updates(objset_t
*os
, dmu_tx_t
*tx
)
1222 list_t
*list
= &os
->os_synced_dnodes
;
1224 ASSERT(list_head(list
) == NULL
|| dmu_objset_userused_enabled(os
));
1226 while (dn
= list_head(list
)) {
1228 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
1229 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
1230 dn
->dn_phys
->dn_flags
&
1231 DNODE_FLAG_USERUSED_ACCOUNTED
);
1233 /* Allocate the user/groupused objects if necessary. */
1234 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
1235 VERIFY(0 == zap_create_claim(os
,
1236 DMU_USERUSED_OBJECT
,
1237 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1238 VERIFY(0 == zap_create_claim(os
,
1239 DMU_GROUPUSED_OBJECT
,
1240 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1244 * We intentionally modify the zap object even if the
1245 * net delta is zero. Otherwise
1246 * the block of the zap obj could be shared between
1247 * datasets but need to be different between them after
1251 flags
= dn
->dn_id_flags
;
1253 if (flags
& DN_ID_OLD_EXIST
) {
1254 do_userquota_update(os
, dn
->dn_oldused
, dn
->dn_oldflags
,
1255 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
, tx
);
1257 if (flags
& DN_ID_NEW_EXIST
) {
1258 do_userquota_update(os
, DN_USED_BYTES(dn
->dn_phys
),
1259 dn
->dn_phys
->dn_flags
, dn
->dn_newuid
,
1260 dn
->dn_newgid
, B_FALSE
, tx
);
1263 mutex_enter(&dn
->dn_mtx
);
1265 dn
->dn_oldflags
= 0;
1266 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
1267 dn
->dn_olduid
= dn
->dn_newuid
;
1268 dn
->dn_oldgid
= dn
->dn_newgid
;
1269 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1270 if (dn
->dn_bonuslen
== 0)
1271 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1273 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1275 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
1276 mutex_exit(&dn
->dn_mtx
);
1278 list_remove(list
, dn
);
1279 dnode_rele(dn
, list
);
1284 * Returns a pointer to data to find uid/gid from
1286 * If a dirty record for transaction group that is syncing can't
1287 * be found then NULL is returned. In the NULL case it is assumed
1288 * the uid/gid aren't changing.
1291 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
1293 dbuf_dirty_record_t
*dr
, **drp
;
1296 if (db
->db_dirtycnt
== 0)
1297 return (db
->db
.db_data
); /* Nothing is changing */
1299 for (drp
= &db
->db_last_dirty
; (dr
= *drp
) != NULL
; drp
= &dr
->dr_next
)
1300 if (dr
->dr_txg
== tx
->tx_txg
)
1308 DB_DNODE_ENTER(dr
->dr_dbuf
);
1309 dn
= DB_DNODE(dr
->dr_dbuf
);
1311 if (dn
->dn_bonuslen
== 0 &&
1312 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
1313 data
= dr
->dt
.dl
.dr_data
->b_data
;
1315 data
= dr
->dt
.dl
.dr_data
;
1317 DB_DNODE_EXIT(dr
->dr_dbuf
);
1324 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
1326 objset_t
*os
= dn
->dn_objset
;
1328 dmu_buf_impl_t
*db
= NULL
;
1329 uint64_t *user
, *group
;
1330 int flags
= dn
->dn_id_flags
;
1332 boolean_t have_spill
= B_FALSE
;
1334 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
1337 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
1338 DN_ID_CHKED_SPILL
)))
1341 if (before
&& dn
->dn_bonuslen
!= 0)
1342 data
= DN_BONUS(dn
->dn_phys
);
1343 else if (!before
&& dn
->dn_bonuslen
!= 0) {
1346 mutex_enter(&db
->db_mtx
);
1347 data
= dmu_objset_userquota_find_data(db
, tx
);
1349 data
= DN_BONUS(dn
->dn_phys
);
1351 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
1354 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
1355 rf
|= DB_RF_HAVESTRUCT
;
1356 error
= dmu_spill_hold_by_dnode(dn
,
1357 rf
| DB_RF_MUST_SUCCEED
,
1358 FTAG
, (dmu_buf_t
**)&db
);
1360 mutex_enter(&db
->db_mtx
);
1361 data
= (before
) ? db
->db
.db_data
:
1362 dmu_objset_userquota_find_data(db
, tx
);
1363 have_spill
= B_TRUE
;
1365 mutex_enter(&dn
->dn_mtx
);
1366 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1367 mutex_exit(&dn
->dn_mtx
);
1373 user
= &dn
->dn_olduid
;
1374 group
= &dn
->dn_oldgid
;
1376 user
= &dn
->dn_newuid
;
1377 group
= &dn
->dn_newgid
;
1381 * Must always call the callback in case the object
1382 * type has changed and that type isn't an object type to track
1384 error
= used_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
,
1388 * Preserve existing uid/gid when the callback can't determine
1389 * what the new uid/gid are and the callback returned EEXIST.
1390 * The EEXIST error tells us to just use the existing uid/gid.
1391 * If we don't know what the old values are then just assign
1392 * them to 0, since that is a new file being created.
1394 if (!before
&& data
== NULL
&& error
== EEXIST
) {
1395 if (flags
& DN_ID_OLD_EXIST
) {
1396 dn
->dn_newuid
= dn
->dn_olduid
;
1397 dn
->dn_newgid
= dn
->dn_oldgid
;
1406 mutex_exit(&db
->db_mtx
);
1408 mutex_enter(&dn
->dn_mtx
);
1409 if (error
== 0 && before
)
1410 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1411 if (error
== 0 && !before
)
1412 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
1415 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1417 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1419 mutex_exit(&dn
->dn_mtx
);
1421 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
1425 dmu_objset_userspace_present(objset_t
*os
)
1427 return (os
->os_phys
->os_flags
&
1428 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
1432 dmu_objset_userspace_upgrade(objset_t
*os
)
1437 if (dmu_objset_userspace_present(os
))
1439 if (!dmu_objset_userused_enabled(os
))
1441 if (dmu_objset_is_snapshot(os
))
1445 * We simply need to mark every object dirty, so that it will be
1446 * synced out and now accounted. If this is called
1447 * concurrently, or if we already did some work before crashing,
1448 * that's fine, since we track each object's accounted state
1452 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
1457 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1460 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
1463 tx
= dmu_tx_create(os
);
1464 dmu_tx_hold_bonus(tx
, obj
);
1465 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
1470 dmu_buf_will_dirty(db
, tx
);
1471 dmu_buf_rele(db
, FTAG
);
1475 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
1476 txg_wait_synced(dmu_objset_pool(os
), 0);
1481 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
1482 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1484 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
1485 usedobjsp
, availobjsp
);
1489 dmu_objset_fsid_guid(objset_t
*os
)
1491 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
1495 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
1497 stat
->dds_type
= os
->os_phys
->os_type
;
1498 if (os
->os_dsl_dataset
)
1499 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
1503 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
1505 ASSERT(os
->os_dsl_dataset
||
1506 os
->os_phys
->os_type
== DMU_OST_META
);
1508 if (os
->os_dsl_dataset
!= NULL
)
1509 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
1511 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
1512 os
->os_phys
->os_type
);
1513 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
1514 dmu_objset_userspace_present(os
));
1518 dmu_objset_is_snapshot(objset_t
*os
)
1520 if (os
->os_dsl_dataset
!= NULL
)
1521 return (dsl_dataset_is_snapshot(os
->os_dsl_dataset
));
1527 dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
, int maxlen
,
1528 boolean_t
*conflict
)
1530 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1533 if (ds
->ds_phys
->ds_snapnames_zapobj
== 0)
1536 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1537 ds
->ds_phys
->ds_snapnames_zapobj
, name
, 8, 1, &ignored
, MT_FIRST
,
1538 real
, maxlen
, conflict
));
1542 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
1543 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
1545 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1546 zap_cursor_t cursor
;
1547 zap_attribute_t attr
;
1549 if (ds
->ds_phys
->ds_snapnames_zapobj
== 0)
1552 zap_cursor_init_serialized(&cursor
,
1553 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1554 ds
->ds_phys
->ds_snapnames_zapobj
, *offp
);
1556 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1557 zap_cursor_fini(&cursor
);
1561 if (strlen(attr
.za_name
) + 1 > namelen
) {
1562 zap_cursor_fini(&cursor
);
1563 return (ENAMETOOLONG
);
1566 (void) strcpy(name
, attr
.za_name
);
1568 *idp
= attr
.za_first_integer
;
1570 *case_conflict
= attr
.za_normalization_conflict
;
1571 zap_cursor_advance(&cursor
);
1572 *offp
= zap_cursor_serialize(&cursor
);
1573 zap_cursor_fini(&cursor
);
1579 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
1580 uint64_t *idp
, uint64_t *offp
)
1582 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
1583 zap_cursor_t cursor
;
1584 zap_attribute_t attr
;
1586 /* there is no next dir on a snapshot! */
1587 if (os
->os_dsl_dataset
->ds_object
!=
1588 dd
->dd_phys
->dd_head_dataset_obj
)
1591 zap_cursor_init_serialized(&cursor
,
1592 dd
->dd_pool
->dp_meta_objset
,
1593 dd
->dd_phys
->dd_child_dir_zapobj
, *offp
);
1595 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1596 zap_cursor_fini(&cursor
);
1600 if (strlen(attr
.za_name
) + 1 > namelen
) {
1601 zap_cursor_fini(&cursor
);
1602 return (ENAMETOOLONG
);
1605 (void) strcpy(name
, attr
.za_name
);
1607 *idp
= attr
.za_first_integer
;
1608 zap_cursor_advance(&cursor
);
1609 *offp
= zap_cursor_serialize(&cursor
);
1610 zap_cursor_fini(&cursor
);
1616 int (*func
)(const char *, void *);
1622 findfunc(spa_t
*spa
, uint64_t dsobj
, const char *dsname
, void *arg
)
1624 struct findarg
*fa
= arg
;
1625 return (fa
->func(dsname
, fa
->arg
));
1629 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1630 * Perhaps change all callers to use dmu_objset_find_spa()?
1633 dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
1639 return (dmu_objset_find_spa(NULL
, name
, findfunc
, &fa
, flags
));
1643 * Find all objsets under name, call func on each
1646 dmu_objset_find_spa(spa_t
*spa
, const char *name
,
1647 int func(spa_t
*, uint64_t, const char *, void *), void *arg
, int flags
)
1653 zap_attribute_t
*attr
;
1659 name
= spa_name(spa
);
1660 err
= dsl_dir_open_spa(spa
, name
, FTAG
, &dd
, NULL
);
1664 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1665 if (dd
->dd_myname
[0] == '$') {
1666 dsl_dir_close(dd
, FTAG
);
1670 thisobj
= dd
->dd_phys
->dd_head_dataset_obj
;
1671 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1675 * Iterate over all children.
1677 if (flags
& DS_FIND_CHILDREN
) {
1678 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1679 dd
->dd_phys
->dd_child_dir_zapobj
);
1680 zap_cursor_retrieve(&zc
, attr
) == 0;
1681 (void) zap_cursor_advance(&zc
)) {
1682 ASSERT(attr
->za_integer_length
== sizeof (uint64_t));
1683 ASSERT(attr
->za_num_integers
== 1);
1685 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
1686 err
= dmu_objset_find_spa(spa
, child
, func
, arg
, flags
);
1691 zap_cursor_fini(&zc
);
1694 dsl_dir_close(dd
, FTAG
);
1695 kmem_free(attr
, sizeof (zap_attribute_t
));
1701 * Iterate over all snapshots.
1703 if (flags
& DS_FIND_SNAPSHOTS
) {
1704 if (!dsl_pool_sync_context(dp
))
1705 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
1706 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1707 if (!dsl_pool_sync_context(dp
))
1708 rw_exit(&dp
->dp_config_rwlock
);
1711 uint64_t snapobj
= ds
->ds_phys
->ds_snapnames_zapobj
;
1712 dsl_dataset_rele(ds
, FTAG
);
1714 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1715 zap_cursor_retrieve(&zc
, attr
) == 0;
1716 (void) zap_cursor_advance(&zc
)) {
1717 ASSERT(attr
->za_integer_length
==
1719 ASSERT(attr
->za_num_integers
== 1);
1721 child
= kmem_asprintf("%s@%s",
1722 name
, attr
->za_name
);
1723 err
= func(spa
, attr
->za_first_integer
,
1729 zap_cursor_fini(&zc
);
1733 dsl_dir_close(dd
, FTAG
);
1734 kmem_free(attr
, sizeof (zap_attribute_t
));
1740 * Apply to self if appropriate.
1742 err
= func(spa
, thisobj
, name
, arg
);
1748 dmu_objset_prefetch(const char *name
, void *arg
)
1752 if (dsl_dataset_hold(name
, FTAG
, &ds
))
1755 if (!BP_IS_HOLE(&ds
->ds_phys
->ds_bp
)) {
1756 mutex_enter(&ds
->ds_opening_lock
);
1757 if (ds
->ds_objset
== NULL
) {
1758 uint32_t aflags
= ARC_NOWAIT
| ARC_PREFETCH
;
1761 SET_BOOKMARK(&zb
, ds
->ds_object
, ZB_ROOT_OBJECT
,
1762 ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1764 (void) dsl_read_nolock(NULL
, dsl_dataset_get_spa(ds
),
1765 &ds
->ds_phys
->ds_bp
, NULL
, NULL
,
1766 ZIO_PRIORITY_ASYNC_READ
,
1767 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
,
1770 mutex_exit(&ds
->ds_opening_lock
);
1773 dsl_dataset_rele(ds
, FTAG
);
1778 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
1780 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
1781 os
->os_user_ptr
= user_ptr
;
1785 dmu_objset_get_user(objset_t
*os
)
1787 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
1788 return (os
->os_user_ptr
);