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
;
533 for (t
= 0; t
< TXG_SIZE
; t
++)
534 ASSERT(!dmu_objset_is_dirty(os
, t
));
537 if (!dsl_dataset_is_snapshot(ds
)) {
538 VERIFY(0 == dsl_prop_unregister(ds
, "checksum",
539 checksum_changed_cb
, os
));
540 VERIFY(0 == dsl_prop_unregister(ds
, "compression",
541 compression_changed_cb
, os
));
542 VERIFY(0 == dsl_prop_unregister(ds
, "copies",
543 copies_changed_cb
, os
));
544 VERIFY(0 == dsl_prop_unregister(ds
, "dedup",
545 dedup_changed_cb
, os
));
546 VERIFY(0 == dsl_prop_unregister(ds
, "logbias",
547 logbias_changed_cb
, os
));
548 VERIFY(0 == dsl_prop_unregister(ds
, "sync",
549 sync_changed_cb
, os
));
551 VERIFY(0 == dsl_prop_unregister(ds
, "primarycache",
552 primary_cache_changed_cb
, os
));
553 VERIFY(0 == dsl_prop_unregister(ds
, "secondarycache",
554 secondary_cache_changed_cb
, os
));
561 * We should need only a single pass over the dnode list, since
562 * nothing can be added to the list at this point.
564 (void) dmu_objset_evict_dbufs(os
);
566 dnode_special_close(&os
->os_meta_dnode
);
567 if (DMU_USERUSED_DNODE(os
)) {
568 dnode_special_close(&os
->os_userused_dnode
);
569 dnode_special_close(&os
->os_groupused_dnode
);
571 zil_free(os
->os_zil
);
573 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
575 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
, &os
->os_phys_buf
) == 1);
578 * This is a barrier to prevent the objset from going away in
579 * dnode_move() until we can safely ensure that the objset is still in
580 * use. We consider the objset valid before the barrier and invalid
583 rw_enter(&os_lock
, RW_READER
);
586 mutex_destroy(&os
->os_lock
);
587 mutex_destroy(&os
->os_obj_lock
);
588 mutex_destroy(&os
->os_user_ptr_lock
);
589 kmem_free(os
, sizeof (objset_t
));
593 dmu_objset_snap_cmtime(objset_t
*os
)
595 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
598 /* called from dsl for meta-objset */
600 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
601 dmu_objset_type_t type
, dmu_tx_t
*tx
)
606 ASSERT(dmu_tx_is_syncing(tx
));
608 VERIFY(0 == dmu_objset_from_ds(ds
, &os
));
610 VERIFY(0 == dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
612 mdn
= DMU_META_DNODE(os
);
614 dnode_allocate(mdn
, DMU_OT_DNODE
, 1 << DNODE_BLOCK_SHIFT
,
615 DN_MAX_INDBLKSHIFT
, DMU_OT_NONE
, 0, tx
);
618 * We don't want to have to increase the meta-dnode's nlevels
619 * later, because then we could do it in quescing context while
620 * we are also accessing it in open context.
622 * This precaution is not necessary for the MOS (ds == NULL),
623 * because the MOS is only updated in syncing context.
624 * This is most fortunate: the MOS is the only objset that
625 * needs to be synced multiple times as spa_sync() iterates
626 * to convergence, so minimizing its dn_nlevels matters.
632 * Determine the number of levels necessary for the meta-dnode
633 * to contain DN_MAX_OBJECT dnodes.
635 while ((uint64_t)mdn
->dn_nblkptr
<< (mdn
->dn_datablkshift
+
636 (levels
- 1) * (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
637 DN_MAX_OBJECT
* sizeof (dnode_phys_t
))
640 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
641 mdn
->dn_nlevels
= levels
;
644 ASSERT(type
!= DMU_OST_NONE
);
645 ASSERT(type
!= DMU_OST_ANY
);
646 ASSERT(type
< DMU_OST_NUMTYPES
);
647 os
->os_phys
->os_type
= type
;
648 if (dmu_objset_userused_enabled(os
)) {
649 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
650 os
->os_flags
= os
->os_phys
->os_flags
;
653 dsl_dataset_dirty(ds
, tx
);
659 void (*userfunc
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
);
661 dsl_dataset_t
*clone_origin
;
662 const char *lastname
;
663 dmu_objset_type_t type
;
670 dmu_objset_create_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
672 dsl_dir_t
*dd
= arg1
;
673 struct oscarg
*oa
= arg2
;
674 objset_t
*mos
= dd
->dd_pool
->dp_meta_objset
;
678 err
= zap_lookup(mos
, dd
->dd_phys
->dd_child_dir_zapobj
,
679 oa
->lastname
, sizeof (uint64_t), 1, &ddobj
);
681 return (err
? err
: EEXIST
);
683 if (oa
->clone_origin
!= NULL
) {
684 /* You can't clone across pools. */
685 if (oa
->clone_origin
->ds_dir
->dd_pool
!= dd
->dd_pool
)
688 /* You can only clone snapshots, not the head datasets. */
689 if (!dsl_dataset_is_snapshot(oa
->clone_origin
))
697 dmu_objset_create_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
699 dsl_dir_t
*dd
= arg1
;
700 spa_t
*spa
= dd
->dd_pool
->dp_spa
;
701 struct oscarg
*oa
= arg2
;
704 ASSERT(dmu_tx_is_syncing(tx
));
706 obj
= dsl_dataset_create_sync(dd
, oa
->lastname
,
707 oa
->clone_origin
, oa
->flags
, oa
->cr
, tx
);
709 if (oa
->clone_origin
== NULL
) {
710 dsl_pool_t
*dp
= dd
->dd_pool
;
715 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, obj
, FTAG
, &ds
));
716 bp
= dsl_dataset_get_blkptr(ds
);
717 ASSERT(BP_IS_HOLE(bp
));
719 os
= dmu_objset_create_impl(spa
, ds
, bp
, oa
->type
, tx
);
722 oa
->userfunc(os
, oa
->userarg
, oa
->cr
, tx
);
723 dsl_dataset_rele(ds
, FTAG
);
726 spa_history_log_internal(LOG_DS_CREATE
, spa
, tx
, "dataset = %llu", obj
);
730 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
731 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
)
736 struct oscarg oa
= { 0 };
738 ASSERT(strchr(name
, '@') == NULL
);
739 err
= dsl_dir_open(name
, FTAG
, &pdd
, &tail
);
743 dsl_dir_close(pdd
, FTAG
);
754 err
= dsl_sync_task_do(pdd
->dd_pool
, dmu_objset_create_check
,
755 dmu_objset_create_sync
, pdd
, &oa
, 5);
756 dsl_dir_close(pdd
, FTAG
);
761 dmu_objset_clone(const char *name
, dsl_dataset_t
*clone_origin
, uint64_t flags
)
766 struct oscarg oa
= { 0 };
768 ASSERT(strchr(name
, '@') == NULL
);
769 err
= dsl_dir_open(name
, FTAG
, &pdd
, &tail
);
773 dsl_dir_close(pdd
, FTAG
);
778 oa
.clone_origin
= clone_origin
;
782 err
= dsl_sync_task_do(pdd
->dd_pool
, dmu_objset_create_check
,
783 dmu_objset_create_sync
, pdd
, &oa
, 5);
784 dsl_dir_close(pdd
, FTAG
);
789 dmu_objset_destroy(const char *name
, boolean_t defer
)
794 error
= dsl_dataset_own(name
, B_TRUE
, FTAG
, &ds
);
796 error
= dsl_dataset_destroy(ds
, FTAG
, defer
);
797 /* dsl_dataset_destroy() closes the ds. */
804 dsl_sync_task_group_t
*dstg
;
807 char failed
[MAXPATHLEN
];
809 boolean_t needsuspend
;
812 struct dsl_ds_holdarg
*ha
; /* only needed in the temporary case */
813 dsl_dataset_t
*newds
;
817 snapshot_check(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
820 struct snaparg
*sn
= arg2
;
823 /* The props have already been checked by zfs_check_userprops(). */
825 error
= dsl_dataset_snapshot_check(os
->os_dsl_dataset
,
832 * Ideally we would just call
833 * dsl_dataset_user_hold_check() and
834 * dsl_dataset_destroy_check() here. However the
835 * dataset we want to hold and destroy is the snapshot
836 * that we just confirmed we can create, but it won't
837 * exist until after these checks are run. Do any
838 * checks we can here and if more checks are added to
839 * those routines in the future, similar checks may be
842 if (spa_version(os
->os_spa
) < SPA_VERSION_USERREFS
)
845 * Not checking number of tags because the tag will be
846 * unique, as it will be the only tag.
848 if (strlen(sn
->htag
) + MAX_TAG_PREFIX_LEN
>= MAXNAMELEN
)
851 sn
->ha
= kmem_alloc(sizeof (struct dsl_ds_holdarg
), KM_SLEEP
);
852 sn
->ha
->temphold
= B_TRUE
;
853 sn
->ha
->htag
= sn
->htag
;
859 snapshot_sync(void *arg1
, void *arg2
, dmu_tx_t
*tx
)
862 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
863 struct snaparg
*sn
= arg2
;
865 dsl_dataset_snapshot_sync(ds
, sn
->snapname
, tx
);
869 pa
.pa_props
= sn
->props
;
870 pa
.pa_source
= ZPROP_SRC_LOCAL
;
871 dsl_props_set_sync(ds
->ds_prev
, &pa
, tx
);
875 struct dsl_ds_destroyarg da
;
877 dsl_dataset_user_hold_sync(ds
->ds_prev
, sn
->ha
, tx
);
878 kmem_free(sn
->ha
, sizeof (struct dsl_ds_holdarg
));
880 sn
->newds
= ds
->ds_prev
;
884 dsl_dataset_destroy_sync(&da
, FTAG
, tx
);
889 dmu_objset_snapshot_one(const char *name
, void *arg
)
891 struct snaparg
*sn
= arg
;
897 * If the objset starts with a '%', then ignore it unless it was
898 * explicitly named (ie, not recursive). These hidden datasets
899 * are always inconsistent, and by not opening them here, we can
900 * avoid a race with dsl_dir_destroy_check().
902 cp
= strrchr(name
, '/');
903 if (cp
&& cp
[1] == '%' && sn
->recursive
)
906 (void) strcpy(sn
->failed
, name
);
909 * Check permissions if we are doing a recursive snapshot. The
910 * permission checks for the starting dataset have already been
911 * performed in zfs_secpolicy_snapshot()
913 if (sn
->recursive
&& (err
= zfs_secpolicy_snapshot_perms(name
, CRED())))
916 err
= dmu_objset_hold(name
, sn
, &os
);
921 * If the objset is in an inconsistent state (eg, in the process
922 * of being destroyed), don't snapshot it. As with %hidden
923 * datasets, we return EBUSY if this name was explicitly
924 * requested (ie, not recursive), and otherwise ignore it.
926 if (os
->os_dsl_dataset
->ds_phys
->ds_flags
& DS_FLAG_INCONSISTENT
) {
927 dmu_objset_rele(os
, sn
);
928 return (sn
->recursive
? 0 : EBUSY
);
931 if (sn
->needsuspend
) {
932 err
= zil_suspend(dmu_objset_zil(os
));
934 dmu_objset_rele(os
, sn
);
938 dsl_sync_task_create(sn
->dstg
, snapshot_check
, snapshot_sync
,
945 dmu_objset_snapshot(char *fsname
, char *snapname
, char *tag
,
946 nvlist_t
*props
, boolean_t recursive
, boolean_t temporary
, int cleanup_fd
)
948 dsl_sync_task_t
*dst
;
954 sn
= kmem_alloc(sizeof (struct snaparg
), KM_SLEEP
);
955 (void) strcpy(sn
->failed
, fsname
);
957 err
= spa_open(fsname
, &spa
, FTAG
);
959 kmem_free(sn
, sizeof (struct snaparg
));
964 if (cleanup_fd
< 0) {
965 spa_close(spa
, FTAG
);
968 if ((err
= zfs_onexit_fd_hold(cleanup_fd
, &minor
)) != 0) {
969 spa_close(spa
, FTAG
);
974 sn
->dstg
= dsl_sync_task_group_create(spa_get_dsl(spa
));
975 sn
->snapname
= snapname
;
978 sn
->recursive
= recursive
;
979 sn
->needsuspend
= (spa_version(spa
) < SPA_VERSION_FAST_SNAP
);
980 sn
->temporary
= temporary
;
985 err
= dmu_objset_find(fsname
,
986 dmu_objset_snapshot_one
, sn
, DS_FIND_CHILDREN
);
988 err
= dmu_objset_snapshot_one(fsname
, sn
);
992 err
= dsl_sync_task_group_wait(sn
->dstg
);
994 for (dst
= list_head(&sn
->dstg
->dstg_tasks
); dst
;
995 dst
= list_next(&sn
->dstg
->dstg_tasks
, dst
)) {
996 objset_t
*os
= dst
->dst_arg1
;
997 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
999 dsl_dataset_name(ds
, sn
->failed
);
1000 } else if (temporary
) {
1001 dsl_register_onexit_hold_cleanup(sn
->newds
, tag
, minor
);
1003 if (sn
->needsuspend
)
1004 zil_resume(dmu_objset_zil(os
));
1005 dmu_objset_rele(os
, sn
);
1009 (void) strcpy(fsname
, sn
->failed
);
1011 zfs_onexit_fd_rele(cleanup_fd
);
1012 dsl_sync_task_group_destroy(sn
->dstg
);
1013 spa_close(spa
, FTAG
);
1014 kmem_free(sn
, sizeof (struct snaparg
));
1019 dmu_objset_sync_dnodes(list_t
*list
, list_t
*newlist
, dmu_tx_t
*tx
)
1023 while ((dn
= list_head(list
))) {
1024 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1025 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1027 * Initialize dn_zio outside dnode_sync() because the
1028 * meta-dnode needs to set it ouside dnode_sync().
1030 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1033 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1034 list_remove(list
, dn
);
1037 (void) dnode_add_ref(dn
, newlist
);
1038 list_insert_tail(newlist
, dn
);
1047 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1051 blkptr_t
*bp
= zio
->io_bp
;
1053 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1055 ASSERT(bp
== os
->os_rootbp
);
1056 ASSERT(BP_GET_TYPE(bp
) == DMU_OT_OBJSET
);
1057 ASSERT(BP_GET_LEVEL(bp
) == 0);
1060 * Update rootbp fill count: it should be the number of objects
1061 * allocated in the object set (not counting the "special"
1062 * objects that are stored in the objset_phys_t -- the meta
1063 * dnode and user/group accounting objects).
1066 for (i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1067 bp
->blk_fill
+= dnp
->dn_blkptr
[i
].blk_fill
;
1072 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1074 blkptr_t
*bp
= zio
->io_bp
;
1075 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1078 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1079 ASSERT(BP_EQUAL(bp
, bp_orig
));
1081 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1082 dmu_tx_t
*tx
= os
->os_synctx
;
1084 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1085 dsl_dataset_block_born(ds
, bp
, tx
);
1089 /* called from dsl */
1091 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1098 list_t
*newlist
= NULL
;
1099 dbuf_dirty_record_t
*dr
;
1101 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", tx
->tx_txg
);
1103 ASSERT(dmu_tx_is_syncing(tx
));
1104 /* XXX the write_done callback should really give us the tx... */
1107 if (os
->os_dsl_dataset
== NULL
) {
1109 * This is the MOS. If we have upgraded,
1110 * spa_max_replication() could change, so reset
1113 os
->os_copies
= spa_max_replication(os
->os_spa
);
1117 * Create the root block IO
1119 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1120 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1121 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1122 VERIFY3U(0, ==, arc_release_bp(os
->os_phys_buf
, &os
->os_phys_buf
,
1123 os
->os_rootbp
, os
->os_spa
, &zb
));
1125 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1127 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1128 os
->os_rootbp
, os
->os_phys_buf
, DMU_OS_IS_L2CACHEABLE(os
), &zp
,
1129 dmu_objset_write_ready
, dmu_objset_write_done
, os
,
1130 ZIO_PRIORITY_ASYNC_WRITE
, ZIO_FLAG_MUSTSUCCEED
, &zb
);
1133 * Sync special dnodes - the parent IO for the sync is the root block
1135 DMU_META_DNODE(os
)->dn_zio
= zio
;
1136 dnode_sync(DMU_META_DNODE(os
), tx
);
1138 os
->os_phys
->os_flags
= os
->os_flags
;
1140 if (DMU_USERUSED_DNODE(os
) &&
1141 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1142 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1143 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1144 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1145 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1148 txgoff
= tx
->tx_txg
& TXG_MASK
;
1150 if (dmu_objset_userused_enabled(os
)) {
1151 newlist
= &os
->os_synced_dnodes
;
1153 * We must create the list here because it uses the
1154 * dn_dirty_link[] of this txg.
1156 list_create(newlist
, sizeof (dnode_t
),
1157 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1160 dmu_objset_sync_dnodes(&os
->os_free_dnodes
[txgoff
], newlist
, tx
);
1161 dmu_objset_sync_dnodes(&os
->os_dirty_dnodes
[txgoff
], newlist
, tx
);
1163 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1164 while ((dr
= list_head(list
)) != NULL
) {
1165 ASSERT(dr
->dr_dbuf
->db_level
== 0);
1166 list_remove(list
, dr
);
1168 zio_nowait(dr
->dr_zio
);
1171 * Free intent log blocks up to this tx.
1173 zil_sync(os
->os_zil
, tx
);
1174 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1179 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1181 return (!list_is_empty(&os
->os_dirty_dnodes
[txg
& TXG_MASK
]) ||
1182 !list_is_empty(&os
->os_free_dnodes
[txg
& TXG_MASK
]));
1186 dmu_objset_is_dirty_anywhere(objset_t
*os
)
1190 for (t
= 0; t
< TXG_SIZE
; t
++)
1191 if (dmu_objset_is_dirty(os
, t
))
1196 static objset_used_cb_t
*used_cbs
[DMU_OST_NUMTYPES
];
1199 dmu_objset_register_type(dmu_objset_type_t ost
, objset_used_cb_t
*cb
)
1205 dmu_objset_userused_enabled(objset_t
*os
)
1207 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1208 used_cbs
[os
->os_phys
->os_type
] != NULL
&&
1209 DMU_USERUSED_DNODE(os
) != NULL
);
1213 do_userquota_update(objset_t
*os
, uint64_t used
, uint64_t flags
,
1214 uint64_t user
, uint64_t group
, boolean_t subtract
, dmu_tx_t
*tx
)
1216 if ((flags
& DNODE_FLAG_USERUSED_ACCOUNTED
)) {
1217 int64_t delta
= DNODE_SIZE
+ used
;
1220 VERIFY3U(0, ==, zap_increment_int(os
, DMU_USERUSED_OBJECT
,
1222 VERIFY3U(0, ==, zap_increment_int(os
, DMU_GROUPUSED_OBJECT
,
1228 dmu_objset_do_userquota_updates(objset_t
*os
, dmu_tx_t
*tx
)
1231 list_t
*list
= &os
->os_synced_dnodes
;
1233 ASSERT(list_head(list
) == NULL
|| dmu_objset_userused_enabled(os
));
1235 while ((dn
= list_head(list
)) != NULL
) {
1237 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
1238 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
1239 dn
->dn_phys
->dn_flags
&
1240 DNODE_FLAG_USERUSED_ACCOUNTED
);
1242 /* Allocate the user/groupused objects if necessary. */
1243 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
1244 VERIFY(0 == zap_create_claim(os
,
1245 DMU_USERUSED_OBJECT
,
1246 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1247 VERIFY(0 == zap_create_claim(os
,
1248 DMU_GROUPUSED_OBJECT
,
1249 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1253 * We intentionally modify the zap object even if the
1254 * net delta is zero. Otherwise
1255 * the block of the zap obj could be shared between
1256 * datasets but need to be different between them after
1260 flags
= dn
->dn_id_flags
;
1262 if (flags
& DN_ID_OLD_EXIST
) {
1263 do_userquota_update(os
, dn
->dn_oldused
, dn
->dn_oldflags
,
1264 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
, tx
);
1266 if (flags
& DN_ID_NEW_EXIST
) {
1267 do_userquota_update(os
, DN_USED_BYTES(dn
->dn_phys
),
1268 dn
->dn_phys
->dn_flags
, dn
->dn_newuid
,
1269 dn
->dn_newgid
, B_FALSE
, tx
);
1272 mutex_enter(&dn
->dn_mtx
);
1274 dn
->dn_oldflags
= 0;
1275 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
1276 dn
->dn_olduid
= dn
->dn_newuid
;
1277 dn
->dn_oldgid
= dn
->dn_newgid
;
1278 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1279 if (dn
->dn_bonuslen
== 0)
1280 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1282 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1284 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
1285 mutex_exit(&dn
->dn_mtx
);
1287 list_remove(list
, dn
);
1288 dnode_rele(dn
, list
);
1293 * Returns a pointer to data to find uid/gid from
1295 * If a dirty record for transaction group that is syncing can't
1296 * be found then NULL is returned. In the NULL case it is assumed
1297 * the uid/gid aren't changing.
1300 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
1302 dbuf_dirty_record_t
*dr
, **drp
;
1305 if (db
->db_dirtycnt
== 0)
1306 return (db
->db
.db_data
); /* Nothing is changing */
1308 for (drp
= &db
->db_last_dirty
; (dr
= *drp
) != NULL
; drp
= &dr
->dr_next
)
1309 if (dr
->dr_txg
== tx
->tx_txg
)
1317 DB_DNODE_ENTER(dr
->dr_dbuf
);
1318 dn
= DB_DNODE(dr
->dr_dbuf
);
1320 if (dn
->dn_bonuslen
== 0 &&
1321 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
1322 data
= dr
->dt
.dl
.dr_data
->b_data
;
1324 data
= dr
->dt
.dl
.dr_data
;
1326 DB_DNODE_EXIT(dr
->dr_dbuf
);
1333 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
1335 objset_t
*os
= dn
->dn_objset
;
1337 dmu_buf_impl_t
*db
= NULL
;
1338 uint64_t *user
= NULL
, *group
= NULL
;
1339 int flags
= dn
->dn_id_flags
;
1341 boolean_t have_spill
= B_FALSE
;
1343 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
1346 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
1347 DN_ID_CHKED_SPILL
)))
1350 if (before
&& dn
->dn_bonuslen
!= 0)
1351 data
= DN_BONUS(dn
->dn_phys
);
1352 else if (!before
&& dn
->dn_bonuslen
!= 0) {
1355 mutex_enter(&db
->db_mtx
);
1356 data
= dmu_objset_userquota_find_data(db
, tx
);
1358 data
= DN_BONUS(dn
->dn_phys
);
1360 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
1363 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
1364 rf
|= DB_RF_HAVESTRUCT
;
1365 error
= dmu_spill_hold_by_dnode(dn
,
1366 rf
| DB_RF_MUST_SUCCEED
,
1367 FTAG
, (dmu_buf_t
**)&db
);
1369 mutex_enter(&db
->db_mtx
);
1370 data
= (before
) ? db
->db
.db_data
:
1371 dmu_objset_userquota_find_data(db
, tx
);
1372 have_spill
= B_TRUE
;
1374 mutex_enter(&dn
->dn_mtx
);
1375 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1376 mutex_exit(&dn
->dn_mtx
);
1382 user
= &dn
->dn_olduid
;
1383 group
= &dn
->dn_oldgid
;
1385 user
= &dn
->dn_newuid
;
1386 group
= &dn
->dn_newgid
;
1390 * Must always call the callback in case the object
1391 * type has changed and that type isn't an object type to track
1393 error
= used_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
,
1397 * Preserve existing uid/gid when the callback can't determine
1398 * what the new uid/gid are and the callback returned EEXIST.
1399 * The EEXIST error tells us to just use the existing uid/gid.
1400 * If we don't know what the old values are then just assign
1401 * them to 0, since that is a new file being created.
1403 if (!before
&& data
== NULL
&& error
== EEXIST
) {
1404 if (flags
& DN_ID_OLD_EXIST
) {
1405 dn
->dn_newuid
= dn
->dn_olduid
;
1406 dn
->dn_newgid
= dn
->dn_oldgid
;
1415 mutex_exit(&db
->db_mtx
);
1417 mutex_enter(&dn
->dn_mtx
);
1418 if (error
== 0 && before
)
1419 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1420 if (error
== 0 && !before
)
1421 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
1424 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1426 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1428 mutex_exit(&dn
->dn_mtx
);
1430 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
1434 dmu_objset_userspace_present(objset_t
*os
)
1436 return (os
->os_phys
->os_flags
&
1437 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
1441 dmu_objset_userspace_upgrade(objset_t
*os
)
1446 if (dmu_objset_userspace_present(os
))
1448 if (!dmu_objset_userused_enabled(os
))
1450 if (dmu_objset_is_snapshot(os
))
1454 * We simply need to mark every object dirty, so that it will be
1455 * synced out and now accounted. If this is called
1456 * concurrently, or if we already did some work before crashing,
1457 * that's fine, since we track each object's accounted state
1461 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
1466 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1469 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
1472 tx
= dmu_tx_create(os
);
1473 dmu_tx_hold_bonus(tx
, obj
);
1474 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
1479 dmu_buf_will_dirty(db
, tx
);
1480 dmu_buf_rele(db
, FTAG
);
1484 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
1485 txg_wait_synced(dmu_objset_pool(os
), 0);
1490 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
1491 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1493 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
1494 usedobjsp
, availobjsp
);
1498 dmu_objset_fsid_guid(objset_t
*os
)
1500 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
1504 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
1506 stat
->dds_type
= os
->os_phys
->os_type
;
1507 if (os
->os_dsl_dataset
)
1508 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
1512 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
1514 ASSERT(os
->os_dsl_dataset
||
1515 os
->os_phys
->os_type
== DMU_OST_META
);
1517 if (os
->os_dsl_dataset
!= NULL
)
1518 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
1520 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
1521 os
->os_phys
->os_type
);
1522 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
1523 dmu_objset_userspace_present(os
));
1527 dmu_objset_is_snapshot(objset_t
*os
)
1529 if (os
->os_dsl_dataset
!= NULL
)
1530 return (dsl_dataset_is_snapshot(os
->os_dsl_dataset
));
1536 dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
, int maxlen
,
1537 boolean_t
*conflict
)
1539 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1542 if (ds
->ds_phys
->ds_snapnames_zapobj
== 0)
1545 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1546 ds
->ds_phys
->ds_snapnames_zapobj
, name
, 8, 1, &ignored
, MT_FIRST
,
1547 real
, maxlen
, conflict
));
1551 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
1552 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
1554 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1555 zap_cursor_t cursor
;
1556 zap_attribute_t attr
;
1558 if (ds
->ds_phys
->ds_snapnames_zapobj
== 0)
1561 zap_cursor_init_serialized(&cursor
,
1562 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1563 ds
->ds_phys
->ds_snapnames_zapobj
, *offp
);
1565 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1566 zap_cursor_fini(&cursor
);
1570 if (strlen(attr
.za_name
) + 1 > namelen
) {
1571 zap_cursor_fini(&cursor
);
1572 return (ENAMETOOLONG
);
1575 (void) strcpy(name
, attr
.za_name
);
1577 *idp
= attr
.za_first_integer
;
1579 *case_conflict
= attr
.za_normalization_conflict
;
1580 zap_cursor_advance(&cursor
);
1581 *offp
= zap_cursor_serialize(&cursor
);
1582 zap_cursor_fini(&cursor
);
1588 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
1589 uint64_t *idp
, uint64_t *offp
)
1591 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
1592 zap_cursor_t cursor
;
1593 zap_attribute_t attr
;
1595 /* there is no next dir on a snapshot! */
1596 if (os
->os_dsl_dataset
->ds_object
!=
1597 dd
->dd_phys
->dd_head_dataset_obj
)
1600 zap_cursor_init_serialized(&cursor
,
1601 dd
->dd_pool
->dp_meta_objset
,
1602 dd
->dd_phys
->dd_child_dir_zapobj
, *offp
);
1604 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1605 zap_cursor_fini(&cursor
);
1609 if (strlen(attr
.za_name
) + 1 > namelen
) {
1610 zap_cursor_fini(&cursor
);
1611 return (ENAMETOOLONG
);
1614 (void) strcpy(name
, attr
.za_name
);
1616 *idp
= attr
.za_first_integer
;
1617 zap_cursor_advance(&cursor
);
1618 *offp
= zap_cursor_serialize(&cursor
);
1619 zap_cursor_fini(&cursor
);
1625 int (*func
)(const char *, void *);
1631 findfunc(spa_t
*spa
, uint64_t dsobj
, const char *dsname
, void *arg
)
1633 struct findarg
*fa
= arg
;
1634 return (fa
->func(dsname
, fa
->arg
));
1638 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1639 * Perhaps change all callers to use dmu_objset_find_spa()?
1642 dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
1648 return (dmu_objset_find_spa(NULL
, name
, findfunc
, &fa
, flags
));
1652 * Find all objsets under name, call func on each
1655 dmu_objset_find_spa(spa_t
*spa
, const char *name
,
1656 int func(spa_t
*, uint64_t, const char *, void *), void *arg
, int flags
)
1662 zap_attribute_t
*attr
;
1668 name
= spa_name(spa
);
1669 err
= dsl_dir_open_spa(spa
, name
, FTAG
, &dd
, NULL
);
1673 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1674 if (dd
->dd_myname
[0] == '$') {
1675 dsl_dir_close(dd
, FTAG
);
1679 thisobj
= dd
->dd_phys
->dd_head_dataset_obj
;
1680 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1684 * Iterate over all children.
1686 if (flags
& DS_FIND_CHILDREN
) {
1687 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1688 dd
->dd_phys
->dd_child_dir_zapobj
);
1689 zap_cursor_retrieve(&zc
, attr
) == 0;
1690 (void) zap_cursor_advance(&zc
)) {
1691 ASSERT(attr
->za_integer_length
== sizeof (uint64_t));
1692 ASSERT(attr
->za_num_integers
== 1);
1694 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
1695 err
= dmu_objset_find_spa(spa
, child
, func
, arg
, flags
);
1700 zap_cursor_fini(&zc
);
1703 dsl_dir_close(dd
, FTAG
);
1704 kmem_free(attr
, sizeof (zap_attribute_t
));
1710 * Iterate over all snapshots.
1712 if (flags
& DS_FIND_SNAPSHOTS
) {
1713 if (!dsl_pool_sync_context(dp
))
1714 rw_enter(&dp
->dp_config_rwlock
, RW_READER
);
1715 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1716 if (!dsl_pool_sync_context(dp
))
1717 rw_exit(&dp
->dp_config_rwlock
);
1720 uint64_t snapobj
= ds
->ds_phys
->ds_snapnames_zapobj
;
1721 dsl_dataset_rele(ds
, FTAG
);
1723 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1724 zap_cursor_retrieve(&zc
, attr
) == 0;
1725 (void) zap_cursor_advance(&zc
)) {
1726 ASSERT(attr
->za_integer_length
==
1728 ASSERT(attr
->za_num_integers
== 1);
1730 child
= kmem_asprintf("%s@%s",
1731 name
, attr
->za_name
);
1732 err
= func(spa
, attr
->za_first_integer
,
1738 zap_cursor_fini(&zc
);
1742 dsl_dir_close(dd
, FTAG
);
1743 kmem_free(attr
, sizeof (zap_attribute_t
));
1749 * Apply to self if appropriate.
1751 err
= func(spa
, thisobj
, name
, arg
);
1757 dmu_objset_prefetch(const char *name
, void *arg
)
1761 if (dsl_dataset_hold(name
, FTAG
, &ds
))
1764 if (!BP_IS_HOLE(&ds
->ds_phys
->ds_bp
)) {
1765 mutex_enter(&ds
->ds_opening_lock
);
1766 if (ds
->ds_objset
== NULL
) {
1767 uint32_t aflags
= ARC_NOWAIT
| ARC_PREFETCH
;
1770 SET_BOOKMARK(&zb
, ds
->ds_object
, ZB_ROOT_OBJECT
,
1771 ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1773 (void) dsl_read_nolock(NULL
, dsl_dataset_get_spa(ds
),
1774 &ds
->ds_phys
->ds_bp
, NULL
, NULL
,
1775 ZIO_PRIORITY_ASYNC_READ
,
1776 ZIO_FLAG_CANFAIL
| ZIO_FLAG_SPECULATIVE
,
1779 mutex_exit(&ds
->ds_opening_lock
);
1782 dsl_dataset_rele(ds
, FTAG
);
1787 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
1789 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
1790 os
->os_user_ptr
= user_ptr
;
1794 dmu_objset_get_user(objset_t
*os
)
1796 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
1797 return (os
->os_user_ptr
);
1800 #if defined(_KERNEL) && defined(HAVE_SPL)
1801 EXPORT_SYMBOL(dmu_objset_pool
);
1802 EXPORT_SYMBOL(dmu_objset_name
);
1803 EXPORT_SYMBOL(dmu_objset_hold
);
1804 EXPORT_SYMBOL(dmu_objset_own
);
1805 EXPORT_SYMBOL(dmu_objset_rele
);
1806 EXPORT_SYMBOL(dmu_objset_disown
);
1807 EXPORT_SYMBOL(dmu_objset_from_ds
);
1808 EXPORT_SYMBOL(dmu_objset_create
);
1809 EXPORT_SYMBOL(dmu_objset_clone
);
1810 EXPORT_SYMBOL(dmu_objset_destroy
);
1811 EXPORT_SYMBOL(dmu_objset_snapshot
);
1812 EXPORT_SYMBOL(dmu_objset_stats
);
1813 EXPORT_SYMBOL(dmu_objset_fast_stat
);
1814 EXPORT_SYMBOL(dmu_objset_spa
);
1815 EXPORT_SYMBOL(dmu_objset_space
);
1816 EXPORT_SYMBOL(dmu_objset_fsid_guid
);
1817 EXPORT_SYMBOL(dmu_objset_find
);
1818 EXPORT_SYMBOL(dmu_objset_find_spa
);
1819 EXPORT_SYMBOL(dmu_objset_prefetch
);
1820 EXPORT_SYMBOL(dmu_objset_byteswap
);
1821 EXPORT_SYMBOL(dmu_objset_evict_dbufs
);
1822 EXPORT_SYMBOL(dmu_objset_snap_cmtime
);
1824 EXPORT_SYMBOL(dmu_objset_sync
);
1825 EXPORT_SYMBOL(dmu_objset_is_dirty
);
1826 EXPORT_SYMBOL(dmu_objset_create_impl
);
1827 EXPORT_SYMBOL(dmu_objset_open_impl
);
1828 EXPORT_SYMBOL(dmu_objset_evict
);
1829 EXPORT_SYMBOL(dmu_objset_register_type
);
1830 EXPORT_SYMBOL(dmu_objset_do_userquota_updates
);
1831 EXPORT_SYMBOL(dmu_objset_userquota_get_ids
);
1832 EXPORT_SYMBOL(dmu_objset_userused_enabled
);
1833 EXPORT_SYMBOL(dmu_objset_userspace_upgrade
);
1834 EXPORT_SYMBOL(dmu_objset_userspace_present
);