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.
23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
30 /* Portions Copyright 2010 Robert Milkowski */
33 #include <sys/zfs_context.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_dataset.h>
37 #include <sys/dsl_prop.h>
38 #include <sys/dsl_pool.h>
39 #include <sys/dsl_synctask.h>
40 #include <sys/dsl_deleg.h>
41 #include <sys/dnode.h>
44 #include <sys/dmu_tx.h>
47 #include <sys/dmu_impl.h>
48 #include <sys/zfs_ioctl.h>
50 #include <sys/zfs_onexit.h>
51 #include <sys/dsl_destroy.h>
55 * Needed to close a window in dnode_move() that allows the objset to be freed
56 * before it can be safely accessed.
61 * Tunable to overwrite the maximum number of threads for the parallization
62 * of dmu_objset_find_dp, needed to speed up the import of pools with many
64 * Default is 4 times the number of leaf vdevs.
66 int dmu_find_threads
= 0;
68 static void dmu_objset_find_dp_cb(void *arg
);
73 rw_init(&os_lock
, NULL
, RW_DEFAULT
, NULL
);
83 dmu_objset_spa(objset_t
*os
)
89 dmu_objset_zil(objset_t
*os
)
95 dmu_objset_pool(objset_t
*os
)
99 if ((ds
= os
->os_dsl_dataset
) != NULL
&& ds
->ds_dir
)
100 return (ds
->ds_dir
->dd_pool
);
102 return (spa_get_dsl(os
->os_spa
));
106 dmu_objset_ds(objset_t
*os
)
108 return (os
->os_dsl_dataset
);
112 dmu_objset_type(objset_t
*os
)
114 return (os
->os_phys
->os_type
);
118 dmu_objset_name(objset_t
*os
, char *buf
)
120 dsl_dataset_name(os
->os_dsl_dataset
, buf
);
124 dmu_objset_id(objset_t
*os
)
126 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
128 return (ds
? ds
->ds_object
: 0);
132 dmu_objset_syncprop(objset_t
*os
)
134 return (os
->os_sync
);
138 dmu_objset_logbias(objset_t
*os
)
140 return (os
->os_logbias
);
144 checksum_changed_cb(void *arg
, uint64_t newval
)
149 * Inheritance should have been done by now.
151 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
153 os
->os_checksum
= zio_checksum_select(newval
, ZIO_CHECKSUM_ON_VALUE
);
157 compression_changed_cb(void *arg
, uint64_t newval
)
162 * Inheritance and range checking should have been done by now.
164 ASSERT(newval
!= ZIO_COMPRESS_INHERIT
);
166 os
->os_compress
= zio_compress_select(newval
, ZIO_COMPRESS_ON_VALUE
);
170 copies_changed_cb(void *arg
, uint64_t newval
)
175 * Inheritance and range checking should have been done by now.
178 ASSERT(newval
<= spa_max_replication(os
->os_spa
));
180 os
->os_copies
= newval
;
184 dedup_changed_cb(void *arg
, uint64_t newval
)
187 spa_t
*spa
= os
->os_spa
;
188 enum zio_checksum checksum
;
191 * Inheritance should have been done by now.
193 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
195 checksum
= zio_checksum_dedup_select(spa
, newval
, ZIO_CHECKSUM_OFF
);
197 os
->os_dedup_checksum
= checksum
& ZIO_CHECKSUM_MASK
;
198 os
->os_dedup_verify
= !!(checksum
& ZIO_CHECKSUM_VERIFY
);
202 primary_cache_changed_cb(void *arg
, uint64_t newval
)
207 * Inheritance and range checking should have been done by now.
209 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
210 newval
== ZFS_CACHE_METADATA
);
212 os
->os_primary_cache
= newval
;
216 secondary_cache_changed_cb(void *arg
, uint64_t newval
)
221 * Inheritance and range checking should have been done by now.
223 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
224 newval
== ZFS_CACHE_METADATA
);
226 os
->os_secondary_cache
= newval
;
230 sync_changed_cb(void *arg
, uint64_t newval
)
235 * Inheritance and range checking should have been done by now.
237 ASSERT(newval
== ZFS_SYNC_STANDARD
|| newval
== ZFS_SYNC_ALWAYS
||
238 newval
== ZFS_SYNC_DISABLED
);
240 os
->os_sync
= newval
;
242 zil_set_sync(os
->os_zil
, newval
);
246 redundant_metadata_changed_cb(void *arg
, uint64_t newval
)
251 * Inheritance and range checking should have been done by now.
253 ASSERT(newval
== ZFS_REDUNDANT_METADATA_ALL
||
254 newval
== ZFS_REDUNDANT_METADATA_MOST
);
256 os
->os_redundant_metadata
= newval
;
260 logbias_changed_cb(void *arg
, uint64_t newval
)
264 ASSERT(newval
== ZFS_LOGBIAS_LATENCY
||
265 newval
== ZFS_LOGBIAS_THROUGHPUT
);
266 os
->os_logbias
= newval
;
268 zil_set_logbias(os
->os_zil
, newval
);
272 recordsize_changed_cb(void *arg
, uint64_t newval
)
276 os
->os_recordsize
= newval
;
280 dmu_objset_byteswap(void *buf
, size_t size
)
282 objset_phys_t
*osp
= buf
;
284 ASSERT(size
== OBJSET_OLD_PHYS_SIZE
|| size
== sizeof (objset_phys_t
));
285 dnode_byteswap(&osp
->os_meta_dnode
);
286 byteswap_uint64_array(&osp
->os_zil_header
, sizeof (zil_header_t
));
287 osp
->os_type
= BSWAP_64(osp
->os_type
);
288 osp
->os_flags
= BSWAP_64(osp
->os_flags
);
289 if (size
== sizeof (objset_phys_t
)) {
290 dnode_byteswap(&osp
->os_userused_dnode
);
291 dnode_byteswap(&osp
->os_groupused_dnode
);
296 dmu_objset_open_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
302 ASSERT(ds
== NULL
|| MUTEX_HELD(&ds
->ds_opening_lock
));
304 os
= kmem_zalloc(sizeof (objset_t
), KM_SLEEP
);
305 os
->os_dsl_dataset
= ds
;
308 if (!BP_IS_HOLE(os
->os_rootbp
)) {
309 arc_flags_t aflags
= ARC_FLAG_WAIT
;
311 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
312 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
314 if (DMU_OS_IS_L2CACHEABLE(os
))
315 aflags
|= ARC_FLAG_L2CACHE
;
316 if (DMU_OS_IS_L2COMPRESSIBLE(os
))
317 aflags
|= ARC_FLAG_L2COMPRESS
;
319 dprintf_bp(os
->os_rootbp
, "reading %s", "");
320 err
= arc_read(NULL
, spa
, os
->os_rootbp
,
321 arc_getbuf_func
, &os
->os_phys_buf
,
322 ZIO_PRIORITY_SYNC_READ
, ZIO_FLAG_CANFAIL
, &aflags
, &zb
);
324 kmem_free(os
, sizeof (objset_t
));
325 /* convert checksum errors into IO errors */
327 err
= SET_ERROR(EIO
);
331 /* Increase the blocksize if we are permitted. */
332 if (spa_version(spa
) >= SPA_VERSION_USERSPACE
&&
333 arc_buf_size(os
->os_phys_buf
) < sizeof (objset_phys_t
)) {
334 arc_buf_t
*buf
= arc_buf_alloc(spa
,
335 sizeof (objset_phys_t
), &os
->os_phys_buf
,
337 bzero(buf
->b_data
, sizeof (objset_phys_t
));
338 bcopy(os
->os_phys_buf
->b_data
, buf
->b_data
,
339 arc_buf_size(os
->os_phys_buf
));
340 (void) arc_buf_remove_ref(os
->os_phys_buf
,
342 os
->os_phys_buf
= buf
;
345 os
->os_phys
= os
->os_phys_buf
->b_data
;
346 os
->os_flags
= os
->os_phys
->os_flags
;
348 int size
= spa_version(spa
) >= SPA_VERSION_USERSPACE
?
349 sizeof (objset_phys_t
) : OBJSET_OLD_PHYS_SIZE
;
350 os
->os_phys_buf
= arc_buf_alloc(spa
, size
,
351 &os
->os_phys_buf
, ARC_BUFC_METADATA
);
352 os
->os_phys
= os
->os_phys_buf
->b_data
;
353 bzero(os
->os_phys
, size
);
357 * Note: the changed_cb will be called once before the register
358 * func returns, thus changing the checksum/compression from the
359 * default (fletcher2/off). Snapshots don't need to know about
360 * checksum/compression/copies.
363 err
= dsl_prop_register(ds
,
364 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
365 primary_cache_changed_cb
, os
);
367 err
= dsl_prop_register(ds
,
368 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
369 secondary_cache_changed_cb
, os
);
371 if (!ds
->ds_is_snapshot
) {
373 err
= dsl_prop_register(ds
,
374 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
375 checksum_changed_cb
, os
);
378 err
= dsl_prop_register(ds
,
379 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
380 compression_changed_cb
, os
);
383 err
= dsl_prop_register(ds
,
384 zfs_prop_to_name(ZFS_PROP_COPIES
),
385 copies_changed_cb
, os
);
388 err
= dsl_prop_register(ds
,
389 zfs_prop_to_name(ZFS_PROP_DEDUP
),
390 dedup_changed_cb
, os
);
393 err
= dsl_prop_register(ds
,
394 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
395 logbias_changed_cb
, os
);
398 err
= dsl_prop_register(ds
,
399 zfs_prop_to_name(ZFS_PROP_SYNC
),
400 sync_changed_cb
, os
);
403 err
= dsl_prop_register(ds
,
405 ZFS_PROP_REDUNDANT_METADATA
),
406 redundant_metadata_changed_cb
, os
);
409 err
= dsl_prop_register(ds
,
410 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
411 recordsize_changed_cb
, os
);
415 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
,
417 kmem_free(os
, sizeof (objset_t
));
421 /* It's the meta-objset. */
422 os
->os_checksum
= ZIO_CHECKSUM_FLETCHER_4
;
423 os
->os_compress
= ZIO_COMPRESS_LZJB
;
424 os
->os_copies
= spa_max_replication(spa
);
425 os
->os_dedup_checksum
= ZIO_CHECKSUM_OFF
;
426 os
->os_dedup_verify
= B_FALSE
;
427 os
->os_logbias
= ZFS_LOGBIAS_LATENCY
;
428 os
->os_sync
= ZFS_SYNC_STANDARD
;
429 os
->os_primary_cache
= ZFS_CACHE_ALL
;
430 os
->os_secondary_cache
= ZFS_CACHE_ALL
;
433 if (ds
== NULL
|| !ds
->ds_is_snapshot
)
434 os
->os_zil_header
= os
->os_phys
->os_zil_header
;
435 os
->os_zil
= zil_alloc(os
, &os
->os_zil_header
);
437 for (i
= 0; i
< TXG_SIZE
; i
++) {
438 list_create(&os
->os_dirty_dnodes
[i
], sizeof (dnode_t
),
439 offsetof(dnode_t
, dn_dirty_link
[i
]));
440 list_create(&os
->os_free_dnodes
[i
], sizeof (dnode_t
),
441 offsetof(dnode_t
, dn_dirty_link
[i
]));
443 list_create(&os
->os_dnodes
, sizeof (dnode_t
),
444 offsetof(dnode_t
, dn_link
));
445 list_create(&os
->os_downgraded_dbufs
, sizeof (dmu_buf_impl_t
),
446 offsetof(dmu_buf_impl_t
, db_link
));
448 list_link_init(&os
->os_evicting_node
);
450 mutex_init(&os
->os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
451 mutex_init(&os
->os_obj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
452 mutex_init(&os
->os_user_ptr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
454 dnode_special_open(os
, &os
->os_phys
->os_meta_dnode
,
455 DMU_META_DNODE_OBJECT
, &os
->os_meta_dnode
);
456 if (arc_buf_size(os
->os_phys_buf
) >= sizeof (objset_phys_t
)) {
457 dnode_special_open(os
, &os
->os_phys
->os_userused_dnode
,
458 DMU_USERUSED_OBJECT
, &os
->os_userused_dnode
);
459 dnode_special_open(os
, &os
->os_phys
->os_groupused_dnode
,
460 DMU_GROUPUSED_OBJECT
, &os
->os_groupused_dnode
);
468 dmu_objset_from_ds(dsl_dataset_t
*ds
, objset_t
**osp
)
472 mutex_enter(&ds
->ds_opening_lock
);
473 if (ds
->ds_objset
== NULL
) {
475 err
= dmu_objset_open_impl(dsl_dataset_get_spa(ds
),
476 ds
, dsl_dataset_get_blkptr(ds
), &os
);
479 mutex_enter(&ds
->ds_lock
);
480 ASSERT(ds
->ds_objset
== NULL
);
482 mutex_exit(&ds
->ds_lock
);
485 *osp
= ds
->ds_objset
;
486 mutex_exit(&ds
->ds_opening_lock
);
491 * Holds the pool while the objset is held. Therefore only one objset
492 * can be held at a time.
495 dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
)
501 err
= dsl_pool_hold(name
, tag
, &dp
);
504 err
= dsl_dataset_hold(dp
, name
, tag
, &ds
);
506 dsl_pool_rele(dp
, tag
);
510 err
= dmu_objset_from_ds(ds
, osp
);
512 dsl_dataset_rele(ds
, tag
);
513 dsl_pool_rele(dp
, tag
);
520 dmu_objset_own_impl(dsl_dataset_t
*ds
, dmu_objset_type_t type
,
521 boolean_t readonly
, void *tag
, objset_t
**osp
)
525 err
= dmu_objset_from_ds(ds
, osp
);
527 dsl_dataset_disown(ds
, tag
);
528 } else if (type
!= DMU_OST_ANY
&& type
!= (*osp
)->os_phys
->os_type
) {
529 dsl_dataset_disown(ds
, tag
);
530 return (SET_ERROR(EINVAL
));
531 } else if (!readonly
&& dsl_dataset_is_snapshot(ds
)) {
532 dsl_dataset_disown(ds
, tag
);
533 return (SET_ERROR(EROFS
));
539 * dsl_pool must not be held when this is called.
540 * Upon successful return, there will be a longhold on the dataset,
541 * and the dsl_pool will not be held.
544 dmu_objset_own(const char *name
, dmu_objset_type_t type
,
545 boolean_t readonly
, void *tag
, objset_t
**osp
)
551 err
= dsl_pool_hold(name
, FTAG
, &dp
);
554 err
= dsl_dataset_own(dp
, name
, tag
, &ds
);
556 dsl_pool_rele(dp
, FTAG
);
559 err
= dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
);
560 dsl_pool_rele(dp
, FTAG
);
566 dmu_objset_own_obj(dsl_pool_t
*dp
, uint64_t obj
, dmu_objset_type_t type
,
567 boolean_t readonly
, void *tag
, objset_t
**osp
)
572 err
= dsl_dataset_own_obj(dp
, obj
, tag
, &ds
);
576 return (dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
));
580 dmu_objset_rele(objset_t
*os
, void *tag
)
582 dsl_pool_t
*dp
= dmu_objset_pool(os
);
583 dsl_dataset_rele(os
->os_dsl_dataset
, tag
);
584 dsl_pool_rele(dp
, tag
);
588 * When we are called, os MUST refer to an objset associated with a dataset
589 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
590 * == tag. We will then release and reacquire ownership of the dataset while
591 * holding the pool config_rwlock to avoid intervening namespace or ownership
594 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
595 * release the hold on its dataset and acquire a new one on the dataset of the
596 * same name so that it can be partially torn down and reconstructed.
599 dmu_objset_refresh_ownership(objset_t
*os
, void *tag
)
602 dsl_dataset_t
*ds
, *newds
;
603 char name
[MAXNAMELEN
];
605 ds
= os
->os_dsl_dataset
;
606 VERIFY3P(ds
, !=, NULL
);
607 VERIFY3P(ds
->ds_owner
, ==, tag
);
608 VERIFY(dsl_dataset_long_held(ds
));
610 dsl_dataset_name(ds
, name
);
611 dp
= dmu_objset_pool(os
);
612 dsl_pool_config_enter(dp
, FTAG
);
613 dmu_objset_disown(os
, tag
);
614 VERIFY0(dsl_dataset_own(dp
, name
, tag
, &newds
));
615 VERIFY3P(newds
, ==, os
->os_dsl_dataset
);
616 dsl_pool_config_exit(dp
, FTAG
);
620 dmu_objset_disown(objset_t
*os
, void *tag
)
622 dsl_dataset_disown(os
->os_dsl_dataset
, tag
);
626 dmu_objset_evict_dbufs(objset_t
*os
)
631 dn_marker
= kmem_alloc(sizeof (dnode_t
), KM_SLEEP
);
633 mutex_enter(&os
->os_lock
);
634 dn
= list_head(&os
->os_dnodes
);
637 * Skip dnodes without holds. We have to do this dance
638 * because dnode_add_ref() only works if there is already a
639 * hold. If the dnode has no holds, then it has no dbufs.
641 if (dnode_add_ref(dn
, FTAG
)) {
642 list_insert_after(&os
->os_dnodes
, dn
, dn_marker
);
643 mutex_exit(&os
->os_lock
);
645 dnode_evict_dbufs(dn
);
646 dnode_rele(dn
, FTAG
);
648 mutex_enter(&os
->os_lock
);
649 dn
= list_next(&os
->os_dnodes
, dn_marker
);
650 list_remove(&os
->os_dnodes
, dn_marker
);
652 dn
= list_next(&os
->os_dnodes
, dn
);
655 mutex_exit(&os
->os_lock
);
657 kmem_free(dn_marker
, sizeof (dnode_t
));
659 if (DMU_USERUSED_DNODE(os
) != NULL
) {
660 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os
));
661 dnode_evict_dbufs(DMU_USERUSED_DNODE(os
));
663 dnode_evict_dbufs(DMU_META_DNODE(os
));
667 * Objset eviction processing is split into into two pieces.
668 * The first marks the objset as evicting, evicts any dbufs that
669 * have a refcount of zero, and then queues up the objset for the
670 * second phase of eviction. Once os->os_dnodes has been cleared by
671 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
672 * The second phase closes the special dnodes, dequeues the objset from
673 * the list of those undergoing eviction, and finally frees the objset.
675 * NOTE: Due to asynchronous eviction processing (invocation of
676 * dnode_buf_pageout()), it is possible for the meta dnode for the
677 * objset to have no holds even though os->os_dnodes is not empty.
680 dmu_objset_evict(objset_t
*os
)
684 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
686 for (t
= 0; t
< TXG_SIZE
; t
++)
687 ASSERT(!dmu_objset_is_dirty(os
, t
));
690 if (!ds
->ds_is_snapshot
) {
691 VERIFY0(dsl_prop_unregister(ds
,
692 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
693 checksum_changed_cb
, os
));
694 VERIFY0(dsl_prop_unregister(ds
,
695 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
696 compression_changed_cb
, os
));
697 VERIFY0(dsl_prop_unregister(ds
,
698 zfs_prop_to_name(ZFS_PROP_COPIES
),
699 copies_changed_cb
, os
));
700 VERIFY0(dsl_prop_unregister(ds
,
701 zfs_prop_to_name(ZFS_PROP_DEDUP
),
702 dedup_changed_cb
, os
));
703 VERIFY0(dsl_prop_unregister(ds
,
704 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
705 logbias_changed_cb
, os
));
706 VERIFY0(dsl_prop_unregister(ds
,
707 zfs_prop_to_name(ZFS_PROP_SYNC
),
708 sync_changed_cb
, os
));
709 VERIFY0(dsl_prop_unregister(ds
,
710 zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA
),
711 redundant_metadata_changed_cb
, os
));
712 VERIFY0(dsl_prop_unregister(ds
,
713 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
714 recordsize_changed_cb
, os
));
716 VERIFY0(dsl_prop_unregister(ds
,
717 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
718 primary_cache_changed_cb
, os
));
719 VERIFY0(dsl_prop_unregister(ds
,
720 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
721 secondary_cache_changed_cb
, os
));
727 os
->os_evicting
= B_TRUE
;
728 dmu_objset_evict_dbufs(os
);
730 mutex_enter(&os
->os_lock
);
731 spa_evicting_os_register(os
->os_spa
, os
);
732 if (list_is_empty(&os
->os_dnodes
)) {
733 mutex_exit(&os
->os_lock
);
734 dmu_objset_evict_done(os
);
736 mutex_exit(&os
->os_lock
);
741 dmu_objset_evict_done(objset_t
*os
)
743 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
745 dnode_special_close(&os
->os_meta_dnode
);
746 if (DMU_USERUSED_DNODE(os
)) {
747 dnode_special_close(&os
->os_userused_dnode
);
748 dnode_special_close(&os
->os_groupused_dnode
);
750 zil_free(os
->os_zil
);
752 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
, &os
->os_phys_buf
));
755 * This is a barrier to prevent the objset from going away in
756 * dnode_move() until we can safely ensure that the objset is still in
757 * use. We consider the objset valid before the barrier and invalid
760 rw_enter(&os_lock
, RW_READER
);
763 mutex_destroy(&os
->os_lock
);
764 mutex_destroy(&os
->os_obj_lock
);
765 mutex_destroy(&os
->os_user_ptr_lock
);
766 spa_evicting_os_deregister(os
->os_spa
, os
);
767 kmem_free(os
, sizeof (objset_t
));
771 dmu_objset_snap_cmtime(objset_t
*os
)
773 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
776 /* called from dsl for meta-objset */
778 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
779 dmu_objset_type_t type
, dmu_tx_t
*tx
)
784 ASSERT(dmu_tx_is_syncing(tx
));
787 VERIFY0(dmu_objset_from_ds(ds
, &os
));
789 VERIFY0(dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
791 mdn
= DMU_META_DNODE(os
);
793 dnode_allocate(mdn
, DMU_OT_DNODE
, 1 << DNODE_BLOCK_SHIFT
,
794 DN_MAX_INDBLKSHIFT
, DMU_OT_NONE
, 0, tx
);
797 * We don't want to have to increase the meta-dnode's nlevels
798 * later, because then we could do it in quescing context while
799 * we are also accessing it in open context.
801 * This precaution is not necessary for the MOS (ds == NULL),
802 * because the MOS is only updated in syncing context.
803 * This is most fortunate: the MOS is the only objset that
804 * needs to be synced multiple times as spa_sync() iterates
805 * to convergence, so minimizing its dn_nlevels matters.
811 * Determine the number of levels necessary for the meta-dnode
812 * to contain DN_MAX_OBJECT dnodes.
814 while ((uint64_t)mdn
->dn_nblkptr
<< (mdn
->dn_datablkshift
+
815 (levels
- 1) * (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
816 DN_MAX_OBJECT
* sizeof (dnode_phys_t
))
819 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
820 mdn
->dn_nlevels
= levels
;
823 ASSERT(type
!= DMU_OST_NONE
);
824 ASSERT(type
!= DMU_OST_ANY
);
825 ASSERT(type
< DMU_OST_NUMTYPES
);
826 os
->os_phys
->os_type
= type
;
827 if (dmu_objset_userused_enabled(os
)) {
828 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
829 os
->os_flags
= os
->os_phys
->os_flags
;
832 dsl_dataset_dirty(ds
, tx
);
837 typedef struct dmu_objset_create_arg
{
838 const char *doca_name
;
840 void (*doca_userfunc
)(objset_t
*os
, void *arg
,
841 cred_t
*cr
, dmu_tx_t
*tx
);
843 dmu_objset_type_t doca_type
;
845 } dmu_objset_create_arg_t
;
849 dmu_objset_create_check(void *arg
, dmu_tx_t
*tx
)
851 dmu_objset_create_arg_t
*doca
= arg
;
852 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
857 if (strchr(doca
->doca_name
, '@') != NULL
)
858 return (SET_ERROR(EINVAL
));
860 error
= dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
);
864 dsl_dir_rele(pdd
, FTAG
);
865 return (SET_ERROR(EEXIST
));
867 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
869 dsl_dir_rele(pdd
, FTAG
);
875 dmu_objset_create_sync(void *arg
, dmu_tx_t
*tx
)
877 dmu_objset_create_arg_t
*doca
= arg
;
878 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
886 VERIFY0(dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
));
888 obj
= dsl_dataset_create_sync(pdd
, tail
, NULL
, doca
->doca_flags
,
889 doca
->doca_cred
, tx
);
891 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
892 bp
= dsl_dataset_get_blkptr(ds
);
893 os
= dmu_objset_create_impl(pdd
->dd_pool
->dp_spa
,
894 ds
, bp
, doca
->doca_type
, tx
);
896 if (doca
->doca_userfunc
!= NULL
) {
897 doca
->doca_userfunc(os
, doca
->doca_userarg
,
898 doca
->doca_cred
, tx
);
901 spa_history_log_internal_ds(ds
, "create", tx
, "");
902 dsl_dataset_rele(ds
, FTAG
);
903 dsl_dir_rele(pdd
, FTAG
);
907 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
908 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
)
910 dmu_objset_create_arg_t doca
;
912 doca
.doca_name
= name
;
913 doca
.doca_cred
= CRED();
914 doca
.doca_flags
= flags
;
915 doca
.doca_userfunc
= func
;
916 doca
.doca_userarg
= arg
;
917 doca
.doca_type
= type
;
919 return (dsl_sync_task(name
,
920 dmu_objset_create_check
, dmu_objset_create_sync
, &doca
,
921 5, ZFS_SPACE_CHECK_NORMAL
));
924 typedef struct dmu_objset_clone_arg
{
925 const char *doca_clone
;
926 const char *doca_origin
;
928 } dmu_objset_clone_arg_t
;
932 dmu_objset_clone_check(void *arg
, dmu_tx_t
*tx
)
934 dmu_objset_clone_arg_t
*doca
= arg
;
938 dsl_dataset_t
*origin
;
939 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
941 if (strchr(doca
->doca_clone
, '@') != NULL
)
942 return (SET_ERROR(EINVAL
));
944 error
= dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
);
948 dsl_dir_rele(pdd
, FTAG
);
949 return (SET_ERROR(EEXIST
));
951 /* You can't clone across pools. */
952 if (pdd
->dd_pool
!= dp
) {
953 dsl_dir_rele(pdd
, FTAG
);
954 return (SET_ERROR(EXDEV
));
956 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
959 dsl_dir_rele(pdd
, FTAG
);
960 return (SET_ERROR(EDQUOT
));
962 dsl_dir_rele(pdd
, FTAG
);
964 error
= dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
);
968 /* You can't clone across pools. */
969 if (origin
->ds_dir
->dd_pool
!= dp
) {
970 dsl_dataset_rele(origin
, FTAG
);
971 return (SET_ERROR(EXDEV
));
974 /* You can only clone snapshots, not the head datasets. */
975 if (!origin
->ds_is_snapshot
) {
976 dsl_dataset_rele(origin
, FTAG
);
977 return (SET_ERROR(EINVAL
));
979 dsl_dataset_rele(origin
, FTAG
);
985 dmu_objset_clone_sync(void *arg
, dmu_tx_t
*tx
)
987 dmu_objset_clone_arg_t
*doca
= arg
;
988 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
991 dsl_dataset_t
*origin
, *ds
;
993 char namebuf
[MAXNAMELEN
];
995 VERIFY0(dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
));
996 VERIFY0(dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
));
998 obj
= dsl_dataset_create_sync(pdd
, tail
, origin
, 0,
999 doca
->doca_cred
, tx
);
1001 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
1002 dsl_dataset_name(origin
, namebuf
);
1003 spa_history_log_internal_ds(ds
, "clone", tx
,
1004 "origin=%s (%llu)", namebuf
, origin
->ds_object
);
1005 dsl_dataset_rele(ds
, FTAG
);
1006 dsl_dataset_rele(origin
, FTAG
);
1007 dsl_dir_rele(pdd
, FTAG
);
1011 dmu_objset_clone(const char *clone
, const char *origin
)
1013 dmu_objset_clone_arg_t doca
;
1015 doca
.doca_clone
= clone
;
1016 doca
.doca_origin
= origin
;
1017 doca
.doca_cred
= CRED();
1019 return (dsl_sync_task(clone
,
1020 dmu_objset_clone_check
, dmu_objset_clone_sync
, &doca
,
1021 5, ZFS_SPACE_CHECK_NORMAL
));
1025 dmu_objset_snapshot_one(const char *fsname
, const char *snapname
)
1028 char *longsnap
= kmem_asprintf("%s@%s", fsname
, snapname
);
1029 nvlist_t
*snaps
= fnvlist_alloc();
1031 fnvlist_add_boolean(snaps
, longsnap
);
1033 err
= dsl_dataset_snapshot(snaps
, NULL
, NULL
);
1034 fnvlist_free(snaps
);
1039 dmu_objset_sync_dnodes(list_t
*list
, list_t
*newlist
, dmu_tx_t
*tx
)
1043 while ((dn
= list_head(list
))) {
1044 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1045 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1047 * Initialize dn_zio outside dnode_sync() because the
1048 * meta-dnode needs to set it ouside dnode_sync().
1050 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1053 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1054 list_remove(list
, dn
);
1057 (void) dnode_add_ref(dn
, newlist
);
1058 list_insert_tail(newlist
, dn
);
1067 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1071 blkptr_t
*bp
= zio
->io_bp
;
1073 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1075 ASSERT(!BP_IS_EMBEDDED(bp
));
1076 ASSERT3P(bp
, ==, os
->os_rootbp
);
1077 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_OBJSET
);
1078 ASSERT0(BP_GET_LEVEL(bp
));
1081 * Update rootbp fill count: it should be the number of objects
1082 * allocated in the object set (not counting the "special"
1083 * objects that are stored in the objset_phys_t -- the meta
1084 * dnode and user/group accounting objects).
1087 for (i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1088 bp
->blk_fill
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
1093 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1095 blkptr_t
*bp
= zio
->io_bp
;
1096 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1099 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1100 ASSERT(BP_EQUAL(bp
, bp_orig
));
1102 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1103 dmu_tx_t
*tx
= os
->os_synctx
;
1105 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1106 dsl_dataset_block_born(ds
, bp
, tx
);
1110 /* called from dsl */
1112 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1115 zbookmark_phys_t zb
;
1119 list_t
*newlist
= NULL
;
1120 dbuf_dirty_record_t
*dr
;
1122 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", tx
->tx_txg
);
1124 ASSERT(dmu_tx_is_syncing(tx
));
1125 /* XXX the write_done callback should really give us the tx... */
1128 if (os
->os_dsl_dataset
== NULL
) {
1130 * This is the MOS. If we have upgraded,
1131 * spa_max_replication() could change, so reset
1134 os
->os_copies
= spa_max_replication(os
->os_spa
);
1138 * Create the root block IO
1140 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1141 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1142 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1143 arc_release(os
->os_phys_buf
, &os
->os_phys_buf
);
1145 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1147 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1148 os
->os_rootbp
, os
->os_phys_buf
, DMU_OS_IS_L2CACHEABLE(os
),
1149 DMU_OS_IS_L2COMPRESSIBLE(os
), &zp
, dmu_objset_write_ready
,
1150 NULL
, dmu_objset_write_done
, os
, ZIO_PRIORITY_ASYNC_WRITE
,
1151 ZIO_FLAG_MUSTSUCCEED
, &zb
);
1154 * Sync special dnodes - the parent IO for the sync is the root block
1156 DMU_META_DNODE(os
)->dn_zio
= zio
;
1157 dnode_sync(DMU_META_DNODE(os
), tx
);
1159 os
->os_phys
->os_flags
= os
->os_flags
;
1161 if (DMU_USERUSED_DNODE(os
) &&
1162 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1163 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1164 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1165 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1166 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1169 txgoff
= tx
->tx_txg
& TXG_MASK
;
1171 if (dmu_objset_userused_enabled(os
)) {
1172 newlist
= &os
->os_synced_dnodes
;
1174 * We must create the list here because it uses the
1175 * dn_dirty_link[] of this txg.
1177 list_create(newlist
, sizeof (dnode_t
),
1178 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1181 dmu_objset_sync_dnodes(&os
->os_free_dnodes
[txgoff
], newlist
, tx
);
1182 dmu_objset_sync_dnodes(&os
->os_dirty_dnodes
[txgoff
], newlist
, tx
);
1184 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1185 while ((dr
= list_head(list
))) {
1186 ASSERT0(dr
->dr_dbuf
->db_level
);
1187 list_remove(list
, dr
);
1189 zio_nowait(dr
->dr_zio
);
1192 * Free intent log blocks up to this tx.
1194 zil_sync(os
->os_zil
, tx
);
1195 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1200 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1202 return (!list_is_empty(&os
->os_dirty_dnodes
[txg
& TXG_MASK
]) ||
1203 !list_is_empty(&os
->os_free_dnodes
[txg
& TXG_MASK
]));
1206 static objset_used_cb_t
*used_cbs
[DMU_OST_NUMTYPES
];
1209 dmu_objset_register_type(dmu_objset_type_t ost
, objset_used_cb_t
*cb
)
1215 dmu_objset_userused_enabled(objset_t
*os
)
1217 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1218 used_cbs
[os
->os_phys
->os_type
] != NULL
&&
1219 DMU_USERUSED_DNODE(os
) != NULL
);
1223 do_userquota_update(objset_t
*os
, uint64_t used
, uint64_t flags
,
1224 uint64_t user
, uint64_t group
, boolean_t subtract
, dmu_tx_t
*tx
)
1226 if ((flags
& DNODE_FLAG_USERUSED_ACCOUNTED
)) {
1227 int64_t delta
= DNODE_SIZE
+ used
;
1230 VERIFY3U(0, ==, zap_increment_int(os
, DMU_USERUSED_OBJECT
,
1232 VERIFY3U(0, ==, zap_increment_int(os
, DMU_GROUPUSED_OBJECT
,
1238 dmu_objset_do_userquota_updates(objset_t
*os
, dmu_tx_t
*tx
)
1241 list_t
*list
= &os
->os_synced_dnodes
;
1243 ASSERT(list_head(list
) == NULL
|| dmu_objset_userused_enabled(os
));
1245 while ((dn
= list_head(list
))) {
1247 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
1248 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
1249 dn
->dn_phys
->dn_flags
&
1250 DNODE_FLAG_USERUSED_ACCOUNTED
);
1252 /* Allocate the user/groupused objects if necessary. */
1253 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
1254 VERIFY(0 == zap_create_claim(os
,
1255 DMU_USERUSED_OBJECT
,
1256 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1257 VERIFY(0 == zap_create_claim(os
,
1258 DMU_GROUPUSED_OBJECT
,
1259 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1263 * We intentionally modify the zap object even if the
1264 * net delta is zero. Otherwise
1265 * the block of the zap obj could be shared between
1266 * datasets but need to be different between them after
1270 flags
= dn
->dn_id_flags
;
1272 if (flags
& DN_ID_OLD_EXIST
) {
1273 do_userquota_update(os
, dn
->dn_oldused
, dn
->dn_oldflags
,
1274 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
, tx
);
1276 if (flags
& DN_ID_NEW_EXIST
) {
1277 do_userquota_update(os
, DN_USED_BYTES(dn
->dn_phys
),
1278 dn
->dn_phys
->dn_flags
, dn
->dn_newuid
,
1279 dn
->dn_newgid
, B_FALSE
, tx
);
1282 mutex_enter(&dn
->dn_mtx
);
1284 dn
->dn_oldflags
= 0;
1285 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
1286 dn
->dn_olduid
= dn
->dn_newuid
;
1287 dn
->dn_oldgid
= dn
->dn_newgid
;
1288 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1289 if (dn
->dn_bonuslen
== 0)
1290 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1292 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1294 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
1295 mutex_exit(&dn
->dn_mtx
);
1297 list_remove(list
, dn
);
1298 dnode_rele(dn
, list
);
1303 * Returns a pointer to data to find uid/gid from
1305 * If a dirty record for transaction group that is syncing can't
1306 * be found then NULL is returned. In the NULL case it is assumed
1307 * the uid/gid aren't changing.
1310 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
1312 dbuf_dirty_record_t
*dr
, **drp
;
1315 if (db
->db_dirtycnt
== 0)
1316 return (db
->db
.db_data
); /* Nothing is changing */
1318 for (drp
= &db
->db_last_dirty
; (dr
= *drp
) != NULL
; drp
= &dr
->dr_next
)
1319 if (dr
->dr_txg
== tx
->tx_txg
)
1327 DB_DNODE_ENTER(dr
->dr_dbuf
);
1328 dn
= DB_DNODE(dr
->dr_dbuf
);
1330 if (dn
->dn_bonuslen
== 0 &&
1331 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
1332 data
= dr
->dt
.dl
.dr_data
->b_data
;
1334 data
= dr
->dt
.dl
.dr_data
;
1336 DB_DNODE_EXIT(dr
->dr_dbuf
);
1343 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
1345 objset_t
*os
= dn
->dn_objset
;
1347 dmu_buf_impl_t
*db
= NULL
;
1348 uint64_t *user
= NULL
;
1349 uint64_t *group
= NULL
;
1350 int flags
= dn
->dn_id_flags
;
1352 boolean_t have_spill
= B_FALSE
;
1353 boolean_t have_bonus
= B_FALSE
;
1355 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
1358 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
1359 DN_ID_CHKED_SPILL
)))
1362 if (before
&& dn
->dn_bonuslen
!= 0)
1363 data
= DN_BONUS(dn
->dn_phys
);
1364 else if (!before
&& dn
->dn_bonuslen
!= 0) {
1367 if (!RW_WRITE_HELD(&dn
->dn_struct_rwlock
)) {
1368 have_bonus
= dbuf_try_add_ref((dmu_buf_t
*)db
,
1369 dn
->dn_objset
, dn
->dn_object
,
1370 DMU_BONUS_BLKID
, FTAG
);
1373 * The hold will fail if the buffer is
1374 * being evicted due to unlink, in which
1375 * case nothing needs to be done.
1380 mutex_enter(&db
->db_mtx
);
1381 data
= dmu_objset_userquota_find_data(db
, tx
);
1383 data
= DN_BONUS(dn
->dn_phys
);
1385 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
1388 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
1389 rf
|= DB_RF_HAVESTRUCT
;
1390 error
= dmu_spill_hold_by_dnode(dn
,
1391 rf
| DB_RF_MUST_SUCCEED
,
1392 FTAG
, (dmu_buf_t
**)&db
);
1394 mutex_enter(&db
->db_mtx
);
1395 data
= (before
) ? db
->db
.db_data
:
1396 dmu_objset_userquota_find_data(db
, tx
);
1397 have_spill
= B_TRUE
;
1399 mutex_enter(&dn
->dn_mtx
);
1400 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1401 mutex_exit(&dn
->dn_mtx
);
1407 user
= &dn
->dn_olduid
;
1408 group
= &dn
->dn_oldgid
;
1410 user
= &dn
->dn_newuid
;
1411 group
= &dn
->dn_newgid
;
1415 * Must always call the callback in case the object
1416 * type has changed and that type isn't an object type to track
1418 error
= used_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
,
1422 * Preserve existing uid/gid when the callback can't determine
1423 * what the new uid/gid are and the callback returned EEXIST.
1424 * The EEXIST error tells us to just use the existing uid/gid.
1425 * If we don't know what the old values are then just assign
1426 * them to 0, since that is a new file being created.
1428 if (!before
&& data
== NULL
&& error
== EEXIST
) {
1429 if (flags
& DN_ID_OLD_EXIST
) {
1430 dn
->dn_newuid
= dn
->dn_olduid
;
1431 dn
->dn_newgid
= dn
->dn_oldgid
;
1440 mutex_exit(&db
->db_mtx
);
1442 mutex_enter(&dn
->dn_mtx
);
1443 if (error
== 0 && before
)
1444 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1445 if (error
== 0 && !before
)
1446 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
1449 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1451 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1453 mutex_exit(&dn
->dn_mtx
);
1454 if (have_spill
|| have_bonus
)
1455 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
1459 dmu_objset_userspace_present(objset_t
*os
)
1461 return (os
->os_phys
->os_flags
&
1462 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
1466 dmu_objset_userspace_upgrade(objset_t
*os
)
1471 if (dmu_objset_userspace_present(os
))
1473 if (!dmu_objset_userused_enabled(os
))
1474 return (SET_ERROR(ENOTSUP
));
1475 if (dmu_objset_is_snapshot(os
))
1476 return (SET_ERROR(EINVAL
));
1479 * We simply need to mark every object dirty, so that it will be
1480 * synced out and now accounted. If this is called
1481 * concurrently, or if we already did some work before crashing,
1482 * that's fine, since we track each object's accounted state
1486 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
1491 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1492 return (SET_ERROR(EINTR
));
1494 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
1497 tx
= dmu_tx_create(os
);
1498 dmu_tx_hold_bonus(tx
, obj
);
1499 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
1504 dmu_buf_will_dirty(db
, tx
);
1505 dmu_buf_rele(db
, FTAG
);
1509 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
1510 txg_wait_synced(dmu_objset_pool(os
), 0);
1515 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
1516 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1518 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
1519 usedobjsp
, availobjsp
);
1523 dmu_objset_fsid_guid(objset_t
*os
)
1525 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
1529 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
1531 stat
->dds_type
= os
->os_phys
->os_type
;
1532 if (os
->os_dsl_dataset
)
1533 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
1537 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
1539 ASSERT(os
->os_dsl_dataset
||
1540 os
->os_phys
->os_type
== DMU_OST_META
);
1542 if (os
->os_dsl_dataset
!= NULL
)
1543 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
1545 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
1546 os
->os_phys
->os_type
);
1547 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
1548 dmu_objset_userspace_present(os
));
1552 dmu_objset_is_snapshot(objset_t
*os
)
1554 if (os
->os_dsl_dataset
!= NULL
)
1555 return (os
->os_dsl_dataset
->ds_is_snapshot
);
1561 dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
, int maxlen
,
1562 boolean_t
*conflict
)
1564 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1567 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1568 return (SET_ERROR(ENOENT
));
1570 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1571 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, name
, 8, 1, &ignored
,
1572 MT_FIRST
, real
, maxlen
, conflict
));
1576 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
1577 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
1579 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1580 zap_cursor_t cursor
;
1581 zap_attribute_t attr
;
1583 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
1585 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1586 return (SET_ERROR(ENOENT
));
1588 zap_cursor_init_serialized(&cursor
,
1589 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1590 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, *offp
);
1592 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1593 zap_cursor_fini(&cursor
);
1594 return (SET_ERROR(ENOENT
));
1597 if (strlen(attr
.za_name
) + 1 > namelen
) {
1598 zap_cursor_fini(&cursor
);
1599 return (SET_ERROR(ENAMETOOLONG
));
1602 (void) strcpy(name
, attr
.za_name
);
1604 *idp
= attr
.za_first_integer
;
1606 *case_conflict
= attr
.za_normalization_conflict
;
1607 zap_cursor_advance(&cursor
);
1608 *offp
= zap_cursor_serialize(&cursor
);
1609 zap_cursor_fini(&cursor
);
1615 dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *value
)
1617 return (dsl_dataset_snap_lookup(os
->os_dsl_dataset
, name
, value
));
1621 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
1622 uint64_t *idp
, uint64_t *offp
)
1624 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
1625 zap_cursor_t cursor
;
1626 zap_attribute_t attr
;
1628 /* there is no next dir on a snapshot! */
1629 if (os
->os_dsl_dataset
->ds_object
!=
1630 dsl_dir_phys(dd
)->dd_head_dataset_obj
)
1631 return (SET_ERROR(ENOENT
));
1633 zap_cursor_init_serialized(&cursor
,
1634 dd
->dd_pool
->dp_meta_objset
,
1635 dsl_dir_phys(dd
)->dd_child_dir_zapobj
, *offp
);
1637 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1638 zap_cursor_fini(&cursor
);
1639 return (SET_ERROR(ENOENT
));
1642 if (strlen(attr
.za_name
) + 1 > namelen
) {
1643 zap_cursor_fini(&cursor
);
1644 return (SET_ERROR(ENAMETOOLONG
));
1647 (void) strcpy(name
, attr
.za_name
);
1649 *idp
= attr
.za_first_integer
;
1650 zap_cursor_advance(&cursor
);
1651 *offp
= zap_cursor_serialize(&cursor
);
1652 zap_cursor_fini(&cursor
);
1657 typedef struct dmu_objset_find_ctx
{
1661 int (*dc_func
)(dsl_pool_t
*, dsl_dataset_t
*, void *);
1664 kmutex_t
*dc_error_lock
;
1666 } dmu_objset_find_ctx_t
;
1669 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t
*dcp
)
1671 dsl_pool_t
*dp
= dcp
->dc_dp
;
1672 dmu_objset_find_ctx_t
*child_dcp
;
1676 zap_attribute_t
*attr
;
1680 /* don't process if there already was an error */
1681 if (*dcp
->dc_error
!= 0)
1684 err
= dsl_dir_hold_obj(dp
, dcp
->dc_ddobj
, NULL
, FTAG
, &dd
);
1688 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1689 if (dd
->dd_myname
[0] == '$') {
1690 dsl_dir_rele(dd
, FTAG
);
1694 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
1695 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1698 * Iterate over all children.
1700 if (dcp
->dc_flags
& DS_FIND_CHILDREN
) {
1701 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1702 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
1703 zap_cursor_retrieve(&zc
, attr
) == 0;
1704 (void) zap_cursor_advance(&zc
)) {
1705 ASSERT3U(attr
->za_integer_length
, ==,
1707 ASSERT3U(attr
->za_num_integers
, ==, 1);
1709 child_dcp
= kmem_alloc(sizeof (*child_dcp
), KM_SLEEP
);
1711 child_dcp
->dc_ddobj
= attr
->za_first_integer
;
1712 if (dcp
->dc_tq
!= NULL
)
1713 (void) taskq_dispatch(dcp
->dc_tq
,
1714 dmu_objset_find_dp_cb
, child_dcp
, TQ_SLEEP
);
1716 dmu_objset_find_dp_impl(child_dcp
);
1718 zap_cursor_fini(&zc
);
1722 * Iterate over all snapshots.
1724 if (dcp
->dc_flags
& DS_FIND_SNAPSHOTS
) {
1726 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1731 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
1732 dsl_dataset_rele(ds
, FTAG
);
1734 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1735 zap_cursor_retrieve(&zc
, attr
) == 0;
1736 (void) zap_cursor_advance(&zc
)) {
1737 ASSERT3U(attr
->za_integer_length
, ==,
1739 ASSERT3U(attr
->za_num_integers
, ==, 1);
1741 err
= dsl_dataset_hold_obj(dp
,
1742 attr
->za_first_integer
, FTAG
, &ds
);
1745 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
1746 dsl_dataset_rele(ds
, FTAG
);
1750 zap_cursor_fini(&zc
);
1754 dsl_dir_rele(dd
, FTAG
);
1755 kmem_free(attr
, sizeof (zap_attribute_t
));
1763 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1766 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
1767 dsl_dataset_rele(ds
, FTAG
);
1771 mutex_enter(dcp
->dc_error_lock
);
1772 /* only keep first error */
1773 if (*dcp
->dc_error
== 0)
1774 *dcp
->dc_error
= err
;
1775 mutex_exit(dcp
->dc_error_lock
);
1778 kmem_free(dcp
, sizeof (*dcp
));
1782 dmu_objset_find_dp_cb(void *arg
)
1784 dmu_objset_find_ctx_t
*dcp
= arg
;
1785 dsl_pool_t
*dp
= dcp
->dc_dp
;
1787 dsl_pool_config_enter(dp
, FTAG
);
1789 dmu_objset_find_dp_impl(dcp
);
1791 dsl_pool_config_exit(dp
, FTAG
);
1795 * Find objsets under and including ddobj, call func(ds) on each.
1796 * The order for the enumeration is completely undefined.
1797 * func is called with dsl_pool_config held.
1800 dmu_objset_find_dp(dsl_pool_t
*dp
, uint64_t ddobj
,
1801 int func(dsl_pool_t
*, dsl_dataset_t
*, void *), void *arg
, int flags
)
1806 dmu_objset_find_ctx_t
*dcp
;
1809 mutex_init(&err_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1810 dcp
= kmem_alloc(sizeof (*dcp
), KM_SLEEP
);
1813 dcp
->dc_ddobj
= ddobj
;
1814 dcp
->dc_func
= func
;
1816 dcp
->dc_flags
= flags
;
1817 dcp
->dc_error_lock
= &err_lock
;
1818 dcp
->dc_error
= &error
;
1820 if ((flags
& DS_FIND_SERIALIZE
) || dsl_pool_config_held_writer(dp
)) {
1822 * In case a write lock is held we can't make use of
1823 * parallelism, as down the stack of the worker threads
1824 * the lock is asserted via dsl_pool_config_held.
1825 * In case of a read lock this is solved by getting a read
1826 * lock in each worker thread, which isn't possible in case
1827 * of a writer lock. So we fall back to the synchronous path
1829 * In the future it might be possible to get some magic into
1830 * dsl_pool_config_held in a way that it returns true for
1831 * the worker threads so that a single lock held from this
1832 * thread suffices. For now, stay single threaded.
1834 dmu_objset_find_dp_impl(dcp
);
1839 ntasks
= dmu_find_threads
;
1841 ntasks
= vdev_count_leaves(dp
->dp_spa
) * 4;
1842 tq
= taskq_create("dmu_objset_find", ntasks
, minclsyspri
, ntasks
,
1845 kmem_free(dcp
, sizeof (*dcp
));
1846 return (SET_ERROR(ENOMEM
));
1850 /* dcp will be freed by task */
1851 (void) taskq_dispatch(tq
, dmu_objset_find_dp_cb
, dcp
, TQ_SLEEP
);
1854 * PORTING: this code relies on the property of taskq_wait to wait
1855 * until no more tasks are queued and no more tasks are active. As
1856 * we always queue new tasks from within other tasks, task_wait
1857 * reliably waits for the full recursion to finish, even though we
1858 * enqueue new tasks after taskq_wait has been called.
1859 * On platforms other than illumos, taskq_wait may not have this
1864 mutex_destroy(&err_lock
);
1870 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1871 * The dp_config_rwlock must not be held when this is called, and it
1872 * will not be held when the callback is called.
1873 * Therefore this function should only be used when the pool is not changing
1874 * (e.g. in syncing context), or the callback can deal with the possible races.
1877 dmu_objset_find_impl(spa_t
*spa
, const char *name
,
1878 int func(const char *, void *), void *arg
, int flags
)
1881 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1884 zap_attribute_t
*attr
;
1889 dsl_pool_config_enter(dp
, FTAG
);
1891 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, NULL
);
1893 dsl_pool_config_exit(dp
, FTAG
);
1897 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1898 if (dd
->dd_myname
[0] == '$') {
1899 dsl_dir_rele(dd
, FTAG
);
1900 dsl_pool_config_exit(dp
, FTAG
);
1904 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
1905 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1908 * Iterate over all children.
1910 if (flags
& DS_FIND_CHILDREN
) {
1911 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1912 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
1913 zap_cursor_retrieve(&zc
, attr
) == 0;
1914 (void) zap_cursor_advance(&zc
)) {
1915 ASSERT3U(attr
->za_integer_length
, ==,
1917 ASSERT3U(attr
->za_num_integers
, ==, 1);
1919 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
1920 dsl_pool_config_exit(dp
, FTAG
);
1921 err
= dmu_objset_find_impl(spa
, child
,
1923 dsl_pool_config_enter(dp
, FTAG
);
1928 zap_cursor_fini(&zc
);
1931 dsl_dir_rele(dd
, FTAG
);
1932 dsl_pool_config_exit(dp
, FTAG
);
1933 kmem_free(attr
, sizeof (zap_attribute_t
));
1939 * Iterate over all snapshots.
1941 if (flags
& DS_FIND_SNAPSHOTS
) {
1942 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1947 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
1948 dsl_dataset_rele(ds
, FTAG
);
1950 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1951 zap_cursor_retrieve(&zc
, attr
) == 0;
1952 (void) zap_cursor_advance(&zc
)) {
1953 ASSERT3U(attr
->za_integer_length
, ==,
1955 ASSERT3U(attr
->za_num_integers
, ==, 1);
1957 child
= kmem_asprintf("%s@%s",
1958 name
, attr
->za_name
);
1959 dsl_pool_config_exit(dp
, FTAG
);
1960 err
= func(child
, arg
);
1961 dsl_pool_config_enter(dp
, FTAG
);
1966 zap_cursor_fini(&zc
);
1970 dsl_dir_rele(dd
, FTAG
);
1971 kmem_free(attr
, sizeof (zap_attribute_t
));
1972 dsl_pool_config_exit(dp
, FTAG
);
1977 /* Apply to self. */
1978 return (func(name
, arg
));
1982 * See comment above dmu_objset_find_impl().
1985 dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
1991 error
= spa_open(name
, &spa
, FTAG
);
1994 error
= dmu_objset_find_impl(spa
, name
, func
, arg
, flags
);
1995 spa_close(spa
, FTAG
);
2000 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
2002 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2003 os
->os_user_ptr
= user_ptr
;
2007 dmu_objset_get_user(objset_t
*os
)
2009 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2010 return (os
->os_user_ptr
);
2014 * Determine name of filesystem, given name of snapshot.
2015 * buf must be at least MAXNAMELEN bytes
2018 dmu_fsname(const char *snapname
, char *buf
)
2020 char *atp
= strchr(snapname
, '@');
2022 return (SET_ERROR(EINVAL
));
2023 if (atp
- snapname
>= MAXNAMELEN
)
2024 return (SET_ERROR(ENAMETOOLONG
));
2025 (void) strlcpy(buf
, snapname
, atp
- snapname
+ 1);
2029 #if defined(_KERNEL) && defined(HAVE_SPL)
2030 EXPORT_SYMBOL(dmu_objset_zil
);
2031 EXPORT_SYMBOL(dmu_objset_pool
);
2032 EXPORT_SYMBOL(dmu_objset_ds
);
2033 EXPORT_SYMBOL(dmu_objset_type
);
2034 EXPORT_SYMBOL(dmu_objset_name
);
2035 EXPORT_SYMBOL(dmu_objset_hold
);
2036 EXPORT_SYMBOL(dmu_objset_own
);
2037 EXPORT_SYMBOL(dmu_objset_rele
);
2038 EXPORT_SYMBOL(dmu_objset_disown
);
2039 EXPORT_SYMBOL(dmu_objset_from_ds
);
2040 EXPORT_SYMBOL(dmu_objset_create
);
2041 EXPORT_SYMBOL(dmu_objset_clone
);
2042 EXPORT_SYMBOL(dmu_objset_stats
);
2043 EXPORT_SYMBOL(dmu_objset_fast_stat
);
2044 EXPORT_SYMBOL(dmu_objset_spa
);
2045 EXPORT_SYMBOL(dmu_objset_space
);
2046 EXPORT_SYMBOL(dmu_objset_fsid_guid
);
2047 EXPORT_SYMBOL(dmu_objset_find
);
2048 EXPORT_SYMBOL(dmu_objset_byteswap
);
2049 EXPORT_SYMBOL(dmu_objset_evict_dbufs
);
2050 EXPORT_SYMBOL(dmu_objset_snap_cmtime
);
2052 EXPORT_SYMBOL(dmu_objset_sync
);
2053 EXPORT_SYMBOL(dmu_objset_is_dirty
);
2054 EXPORT_SYMBOL(dmu_objset_create_impl
);
2055 EXPORT_SYMBOL(dmu_objset_open_impl
);
2056 EXPORT_SYMBOL(dmu_objset_evict
);
2057 EXPORT_SYMBOL(dmu_objset_register_type
);
2058 EXPORT_SYMBOL(dmu_objset_do_userquota_updates
);
2059 EXPORT_SYMBOL(dmu_objset_userquota_get_ids
);
2060 EXPORT_SYMBOL(dmu_objset_userused_enabled
);
2061 EXPORT_SYMBOL(dmu_objset_userspace_upgrade
);
2062 EXPORT_SYMBOL(dmu_objset_userspace_present
);