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 Nexenta Systems, Inc. All rights reserved.
28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
29 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
32 /* Portions Copyright 2010 Robert Milkowski */
35 #include <sys/zfs_context.h>
36 #include <sys/dmu_objset.h>
37 #include <sys/dsl_dir.h>
38 #include <sys/dsl_dataset.h>
39 #include <sys/dsl_prop.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_synctask.h>
42 #include <sys/dsl_deleg.h>
43 #include <sys/dnode.h>
46 #include <sys/dmu_tx.h>
49 #include <sys/dmu_impl.h>
50 #include <sys/zfs_ioctl.h>
52 #include <sys/zfs_onexit.h>
53 #include <sys/dsl_destroy.h>
57 * Needed to close a window in dnode_move() that allows the objset to be freed
58 * before it can be safely accessed.
63 * Tunable to overwrite the maximum number of threads for the parallization
64 * of dmu_objset_find_dp, needed to speed up the import of pools with many
66 * Default is 4 times the number of leaf vdevs.
68 int dmu_find_threads
= 0;
70 static void dmu_objset_find_dp_cb(void *arg
);
75 rw_init(&os_lock
, NULL
, RW_DEFAULT
, NULL
);
85 dmu_objset_spa(objset_t
*os
)
91 dmu_objset_zil(objset_t
*os
)
97 dmu_objset_pool(objset_t
*os
)
101 if ((ds
= os
->os_dsl_dataset
) != NULL
&& ds
->ds_dir
)
102 return (ds
->ds_dir
->dd_pool
);
104 return (spa_get_dsl(os
->os_spa
));
108 dmu_objset_ds(objset_t
*os
)
110 return (os
->os_dsl_dataset
);
114 dmu_objset_type(objset_t
*os
)
116 return (os
->os_phys
->os_type
);
120 dmu_objset_name(objset_t
*os
, char *buf
)
122 dsl_dataset_name(os
->os_dsl_dataset
, buf
);
126 dmu_objset_id(objset_t
*os
)
128 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
130 return (ds
? ds
->ds_object
: 0);
134 dmu_objset_syncprop(objset_t
*os
)
136 return (os
->os_sync
);
140 dmu_objset_logbias(objset_t
*os
)
142 return (os
->os_logbias
);
146 checksum_changed_cb(void *arg
, uint64_t newval
)
151 * Inheritance should have been done by now.
153 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
155 os
->os_checksum
= zio_checksum_select(newval
, ZIO_CHECKSUM_ON_VALUE
);
159 compression_changed_cb(void *arg
, uint64_t newval
)
164 * Inheritance and range checking should have been done by now.
166 ASSERT(newval
!= ZIO_COMPRESS_INHERIT
);
168 os
->os_compress
= zio_compress_select(os
->os_spa
, newval
,
173 copies_changed_cb(void *arg
, uint64_t newval
)
178 * Inheritance and range checking should have been done by now.
181 ASSERT(newval
<= spa_max_replication(os
->os_spa
));
183 os
->os_copies
= newval
;
187 dedup_changed_cb(void *arg
, uint64_t newval
)
190 spa_t
*spa
= os
->os_spa
;
191 enum zio_checksum checksum
;
194 * Inheritance should have been done by now.
196 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
198 checksum
= zio_checksum_dedup_select(spa
, newval
, ZIO_CHECKSUM_OFF
);
200 os
->os_dedup_checksum
= checksum
& ZIO_CHECKSUM_MASK
;
201 os
->os_dedup_verify
= !!(checksum
& ZIO_CHECKSUM_VERIFY
);
205 primary_cache_changed_cb(void *arg
, uint64_t newval
)
210 * Inheritance and range checking should have been done by now.
212 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
213 newval
== ZFS_CACHE_METADATA
);
215 os
->os_primary_cache
= newval
;
219 secondary_cache_changed_cb(void *arg
, uint64_t newval
)
224 * Inheritance and range checking should have been done by now.
226 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
227 newval
== ZFS_CACHE_METADATA
);
229 os
->os_secondary_cache
= newval
;
233 sync_changed_cb(void *arg
, uint64_t newval
)
238 * Inheritance and range checking should have been done by now.
240 ASSERT(newval
== ZFS_SYNC_STANDARD
|| newval
== ZFS_SYNC_ALWAYS
||
241 newval
== ZFS_SYNC_DISABLED
);
243 os
->os_sync
= newval
;
245 zil_set_sync(os
->os_zil
, newval
);
249 redundant_metadata_changed_cb(void *arg
, uint64_t newval
)
254 * Inheritance and range checking should have been done by now.
256 ASSERT(newval
== ZFS_REDUNDANT_METADATA_ALL
||
257 newval
== ZFS_REDUNDANT_METADATA_MOST
);
259 os
->os_redundant_metadata
= newval
;
263 logbias_changed_cb(void *arg
, uint64_t newval
)
267 ASSERT(newval
== ZFS_LOGBIAS_LATENCY
||
268 newval
== ZFS_LOGBIAS_THROUGHPUT
);
269 os
->os_logbias
= newval
;
271 zil_set_logbias(os
->os_zil
, newval
);
275 recordsize_changed_cb(void *arg
, uint64_t newval
)
279 os
->os_recordsize
= newval
;
283 dmu_objset_byteswap(void *buf
, size_t size
)
285 objset_phys_t
*osp
= buf
;
287 ASSERT(size
== OBJSET_OLD_PHYS_SIZE
|| size
== sizeof (objset_phys_t
));
288 dnode_byteswap(&osp
->os_meta_dnode
);
289 byteswap_uint64_array(&osp
->os_zil_header
, sizeof (zil_header_t
));
290 osp
->os_type
= BSWAP_64(osp
->os_type
);
291 osp
->os_flags
= BSWAP_64(osp
->os_flags
);
292 if (size
== sizeof (objset_phys_t
)) {
293 dnode_byteswap(&osp
->os_userused_dnode
);
294 dnode_byteswap(&osp
->os_groupused_dnode
);
299 dmu_objset_open_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
305 ASSERT(ds
== NULL
|| MUTEX_HELD(&ds
->ds_opening_lock
));
307 os
= kmem_zalloc(sizeof (objset_t
), KM_SLEEP
);
308 os
->os_dsl_dataset
= ds
;
311 if (!BP_IS_HOLE(os
->os_rootbp
)) {
312 arc_flags_t aflags
= ARC_FLAG_WAIT
;
314 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
315 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
317 if (DMU_OS_IS_L2CACHEABLE(os
))
318 aflags
|= ARC_FLAG_L2CACHE
;
319 if (DMU_OS_IS_L2COMPRESSIBLE(os
))
320 aflags
|= ARC_FLAG_L2COMPRESS
;
322 dprintf_bp(os
->os_rootbp
, "reading %s", "");
323 err
= arc_read(NULL
, spa
, os
->os_rootbp
,
324 arc_getbuf_func
, &os
->os_phys_buf
,
325 ZIO_PRIORITY_SYNC_READ
, ZIO_FLAG_CANFAIL
, &aflags
, &zb
);
327 kmem_free(os
, sizeof (objset_t
));
328 /* convert checksum errors into IO errors */
330 err
= SET_ERROR(EIO
);
334 /* Increase the blocksize if we are permitted. */
335 if (spa_version(spa
) >= SPA_VERSION_USERSPACE
&&
336 arc_buf_size(os
->os_phys_buf
) < sizeof (objset_phys_t
)) {
337 arc_buf_t
*buf
= arc_buf_alloc(spa
,
338 sizeof (objset_phys_t
), &os
->os_phys_buf
,
340 bzero(buf
->b_data
, sizeof (objset_phys_t
));
341 bcopy(os
->os_phys_buf
->b_data
, buf
->b_data
,
342 arc_buf_size(os
->os_phys_buf
));
343 (void) arc_buf_remove_ref(os
->os_phys_buf
,
345 os
->os_phys_buf
= buf
;
348 os
->os_phys
= os
->os_phys_buf
->b_data
;
349 os
->os_flags
= os
->os_phys
->os_flags
;
351 int size
= spa_version(spa
) >= SPA_VERSION_USERSPACE
?
352 sizeof (objset_phys_t
) : OBJSET_OLD_PHYS_SIZE
;
353 os
->os_phys_buf
= arc_buf_alloc(spa
, size
,
354 &os
->os_phys_buf
, ARC_BUFC_METADATA
);
355 os
->os_phys
= os
->os_phys_buf
->b_data
;
356 bzero(os
->os_phys
, size
);
360 * Note: the changed_cb will be called once before the register
361 * func returns, thus changing the checksum/compression from the
362 * default (fletcher2/off). Snapshots don't need to know about
363 * checksum/compression/copies.
366 err
= dsl_prop_register(ds
,
367 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
368 primary_cache_changed_cb
, os
);
370 err
= dsl_prop_register(ds
,
371 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
372 secondary_cache_changed_cb
, os
);
374 if (!ds
->ds_is_snapshot
) {
376 err
= dsl_prop_register(ds
,
377 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
378 checksum_changed_cb
, os
);
381 err
= dsl_prop_register(ds
,
382 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
383 compression_changed_cb
, os
);
386 err
= dsl_prop_register(ds
,
387 zfs_prop_to_name(ZFS_PROP_COPIES
),
388 copies_changed_cb
, os
);
391 err
= dsl_prop_register(ds
,
392 zfs_prop_to_name(ZFS_PROP_DEDUP
),
393 dedup_changed_cb
, os
);
396 err
= dsl_prop_register(ds
,
397 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
398 logbias_changed_cb
, os
);
401 err
= dsl_prop_register(ds
,
402 zfs_prop_to_name(ZFS_PROP_SYNC
),
403 sync_changed_cb
, os
);
406 err
= dsl_prop_register(ds
,
408 ZFS_PROP_REDUNDANT_METADATA
),
409 redundant_metadata_changed_cb
, os
);
412 err
= dsl_prop_register(ds
,
413 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
414 recordsize_changed_cb
, os
);
418 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
,
420 kmem_free(os
, sizeof (objset_t
));
424 /* It's the meta-objset. */
425 os
->os_checksum
= ZIO_CHECKSUM_FLETCHER_4
;
426 os
->os_compress
= ZIO_COMPRESS_ON
;
427 os
->os_copies
= spa_max_replication(spa
);
428 os
->os_dedup_checksum
= ZIO_CHECKSUM_OFF
;
429 os
->os_dedup_verify
= B_FALSE
;
430 os
->os_logbias
= ZFS_LOGBIAS_LATENCY
;
431 os
->os_sync
= ZFS_SYNC_STANDARD
;
432 os
->os_primary_cache
= ZFS_CACHE_ALL
;
433 os
->os_secondary_cache
= ZFS_CACHE_ALL
;
436 if (ds
== NULL
|| !ds
->ds_is_snapshot
)
437 os
->os_zil_header
= os
->os_phys
->os_zil_header
;
438 os
->os_zil
= zil_alloc(os
, &os
->os_zil_header
);
440 for (i
= 0; i
< TXG_SIZE
; i
++) {
441 list_create(&os
->os_dirty_dnodes
[i
], sizeof (dnode_t
),
442 offsetof(dnode_t
, dn_dirty_link
[i
]));
443 list_create(&os
->os_free_dnodes
[i
], sizeof (dnode_t
),
444 offsetof(dnode_t
, dn_dirty_link
[i
]));
446 list_create(&os
->os_dnodes
, sizeof (dnode_t
),
447 offsetof(dnode_t
, dn_link
));
448 list_create(&os
->os_downgraded_dbufs
, sizeof (dmu_buf_impl_t
),
449 offsetof(dmu_buf_impl_t
, db_link
));
451 list_link_init(&os
->os_evicting_node
);
453 mutex_init(&os
->os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
454 mutex_init(&os
->os_obj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
455 mutex_init(&os
->os_user_ptr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
457 dnode_special_open(os
, &os
->os_phys
->os_meta_dnode
,
458 DMU_META_DNODE_OBJECT
, &os
->os_meta_dnode
);
459 if (arc_buf_size(os
->os_phys_buf
) >= sizeof (objset_phys_t
)) {
460 dnode_special_open(os
, &os
->os_phys
->os_userused_dnode
,
461 DMU_USERUSED_OBJECT
, &os
->os_userused_dnode
);
462 dnode_special_open(os
, &os
->os_phys
->os_groupused_dnode
,
463 DMU_GROUPUSED_OBJECT
, &os
->os_groupused_dnode
);
471 dmu_objset_from_ds(dsl_dataset_t
*ds
, objset_t
**osp
)
475 mutex_enter(&ds
->ds_opening_lock
);
476 if (ds
->ds_objset
== NULL
) {
478 err
= dmu_objset_open_impl(dsl_dataset_get_spa(ds
),
479 ds
, dsl_dataset_get_blkptr(ds
), &os
);
482 mutex_enter(&ds
->ds_lock
);
483 ASSERT(ds
->ds_objset
== NULL
);
485 mutex_exit(&ds
->ds_lock
);
488 *osp
= ds
->ds_objset
;
489 mutex_exit(&ds
->ds_opening_lock
);
494 * Holds the pool while the objset is held. Therefore only one objset
495 * can be held at a time.
498 dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
)
504 err
= dsl_pool_hold(name
, tag
, &dp
);
507 err
= dsl_dataset_hold(dp
, name
, tag
, &ds
);
509 dsl_pool_rele(dp
, tag
);
513 err
= dmu_objset_from_ds(ds
, osp
);
515 dsl_dataset_rele(ds
, tag
);
516 dsl_pool_rele(dp
, tag
);
523 dmu_objset_own_impl(dsl_dataset_t
*ds
, dmu_objset_type_t type
,
524 boolean_t readonly
, void *tag
, objset_t
**osp
)
528 err
= dmu_objset_from_ds(ds
, osp
);
530 dsl_dataset_disown(ds
, tag
);
531 } else if (type
!= DMU_OST_ANY
&& type
!= (*osp
)->os_phys
->os_type
) {
532 dsl_dataset_disown(ds
, tag
);
533 return (SET_ERROR(EINVAL
));
534 } else if (!readonly
&& dsl_dataset_is_snapshot(ds
)) {
535 dsl_dataset_disown(ds
, tag
);
536 return (SET_ERROR(EROFS
));
542 * dsl_pool must not be held when this is called.
543 * Upon successful return, there will be a longhold on the dataset,
544 * and the dsl_pool will not be held.
547 dmu_objset_own(const char *name
, dmu_objset_type_t type
,
548 boolean_t readonly
, void *tag
, objset_t
**osp
)
554 err
= dsl_pool_hold(name
, FTAG
, &dp
);
557 err
= dsl_dataset_own(dp
, name
, tag
, &ds
);
559 dsl_pool_rele(dp
, FTAG
);
562 err
= dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
);
563 dsl_pool_rele(dp
, FTAG
);
569 dmu_objset_own_obj(dsl_pool_t
*dp
, uint64_t obj
, dmu_objset_type_t type
,
570 boolean_t readonly
, void *tag
, objset_t
**osp
)
575 err
= dsl_dataset_own_obj(dp
, obj
, tag
, &ds
);
579 return (dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
));
583 dmu_objset_rele(objset_t
*os
, void *tag
)
585 dsl_pool_t
*dp
= dmu_objset_pool(os
);
586 dsl_dataset_rele(os
->os_dsl_dataset
, tag
);
587 dsl_pool_rele(dp
, tag
);
591 * When we are called, os MUST refer to an objset associated with a dataset
592 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
593 * == tag. We will then release and reacquire ownership of the dataset while
594 * holding the pool config_rwlock to avoid intervening namespace or ownership
597 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
598 * release the hold on its dataset and acquire a new one on the dataset of the
599 * same name so that it can be partially torn down and reconstructed.
602 dmu_objset_refresh_ownership(objset_t
*os
, void *tag
)
605 dsl_dataset_t
*ds
, *newds
;
606 char name
[MAXNAMELEN
];
608 ds
= os
->os_dsl_dataset
;
609 VERIFY3P(ds
, !=, NULL
);
610 VERIFY3P(ds
->ds_owner
, ==, tag
);
611 VERIFY(dsl_dataset_long_held(ds
));
613 dsl_dataset_name(ds
, name
);
614 dp
= dmu_objset_pool(os
);
615 dsl_pool_config_enter(dp
, FTAG
);
616 dmu_objset_disown(os
, tag
);
617 VERIFY0(dsl_dataset_own(dp
, name
, tag
, &newds
));
618 VERIFY3P(newds
, ==, os
->os_dsl_dataset
);
619 dsl_pool_config_exit(dp
, FTAG
);
623 dmu_objset_disown(objset_t
*os
, void *tag
)
625 dsl_dataset_disown(os
->os_dsl_dataset
, tag
);
629 dmu_objset_evict_dbufs(objset_t
*os
)
634 dn_marker
= kmem_alloc(sizeof (dnode_t
), KM_SLEEP
);
636 mutex_enter(&os
->os_lock
);
637 dn
= list_head(&os
->os_dnodes
);
640 * Skip dnodes without holds. We have to do this dance
641 * because dnode_add_ref() only works if there is already a
642 * hold. If the dnode has no holds, then it has no dbufs.
644 if (dnode_add_ref(dn
, FTAG
)) {
645 list_insert_after(&os
->os_dnodes
, dn
, dn_marker
);
646 mutex_exit(&os
->os_lock
);
648 dnode_evict_dbufs(dn
);
649 dnode_rele(dn
, FTAG
);
651 mutex_enter(&os
->os_lock
);
652 dn
= list_next(&os
->os_dnodes
, dn_marker
);
653 list_remove(&os
->os_dnodes
, dn_marker
);
655 dn
= list_next(&os
->os_dnodes
, dn
);
658 mutex_exit(&os
->os_lock
);
660 kmem_free(dn_marker
, sizeof (dnode_t
));
662 if (DMU_USERUSED_DNODE(os
) != NULL
) {
663 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os
));
664 dnode_evict_dbufs(DMU_USERUSED_DNODE(os
));
666 dnode_evict_dbufs(DMU_META_DNODE(os
));
670 * Objset eviction processing is split into into two pieces.
671 * The first marks the objset as evicting, evicts any dbufs that
672 * have a refcount of zero, and then queues up the objset for the
673 * second phase of eviction. Once os->os_dnodes has been cleared by
674 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
675 * The second phase closes the special dnodes, dequeues the objset from
676 * the list of those undergoing eviction, and finally frees the objset.
678 * NOTE: Due to asynchronous eviction processing (invocation of
679 * dnode_buf_pageout()), it is possible for the meta dnode for the
680 * objset to have no holds even though os->os_dnodes is not empty.
683 dmu_objset_evict(objset_t
*os
)
687 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
689 for (t
= 0; t
< TXG_SIZE
; t
++)
690 ASSERT(!dmu_objset_is_dirty(os
, t
));
693 dsl_prop_unregister_all(ds
, os
);
698 dmu_objset_evict_dbufs(os
);
700 mutex_enter(&os
->os_lock
);
701 spa_evicting_os_register(os
->os_spa
, os
);
702 if (list_is_empty(&os
->os_dnodes
)) {
703 mutex_exit(&os
->os_lock
);
704 dmu_objset_evict_done(os
);
706 mutex_exit(&os
->os_lock
);
711 dmu_objset_evict_done(objset_t
*os
)
713 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
715 dnode_special_close(&os
->os_meta_dnode
);
716 if (DMU_USERUSED_DNODE(os
)) {
717 dnode_special_close(&os
->os_userused_dnode
);
718 dnode_special_close(&os
->os_groupused_dnode
);
720 zil_free(os
->os_zil
);
722 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
, &os
->os_phys_buf
));
725 * This is a barrier to prevent the objset from going away in
726 * dnode_move() until we can safely ensure that the objset is still in
727 * use. We consider the objset valid before the barrier and invalid
730 rw_enter(&os_lock
, RW_READER
);
733 mutex_destroy(&os
->os_lock
);
734 mutex_destroy(&os
->os_obj_lock
);
735 mutex_destroy(&os
->os_user_ptr_lock
);
736 spa_evicting_os_deregister(os
->os_spa
, os
);
737 kmem_free(os
, sizeof (objset_t
));
741 dmu_objset_snap_cmtime(objset_t
*os
)
743 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
746 /* called from dsl for meta-objset */
748 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
749 dmu_objset_type_t type
, dmu_tx_t
*tx
)
754 ASSERT(dmu_tx_is_syncing(tx
));
757 VERIFY0(dmu_objset_from_ds(ds
, &os
));
759 VERIFY0(dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
761 mdn
= DMU_META_DNODE(os
);
763 dnode_allocate(mdn
, DMU_OT_DNODE
, 1 << DNODE_BLOCK_SHIFT
,
764 DN_MAX_INDBLKSHIFT
, DMU_OT_NONE
, 0, tx
);
767 * We don't want to have to increase the meta-dnode's nlevels
768 * later, because then we could do it in quescing context while
769 * we are also accessing it in open context.
771 * This precaution is not necessary for the MOS (ds == NULL),
772 * because the MOS is only updated in syncing context.
773 * This is most fortunate: the MOS is the only objset that
774 * needs to be synced multiple times as spa_sync() iterates
775 * to convergence, so minimizing its dn_nlevels matters.
781 * Determine the number of levels necessary for the meta-dnode
782 * to contain DN_MAX_OBJECT dnodes.
784 while ((uint64_t)mdn
->dn_nblkptr
<< (mdn
->dn_datablkshift
+
785 (levels
- 1) * (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
786 DN_MAX_OBJECT
* sizeof (dnode_phys_t
))
789 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
790 mdn
->dn_nlevels
= levels
;
793 ASSERT(type
!= DMU_OST_NONE
);
794 ASSERT(type
!= DMU_OST_ANY
);
795 ASSERT(type
< DMU_OST_NUMTYPES
);
796 os
->os_phys
->os_type
= type
;
797 if (dmu_objset_userused_enabled(os
)) {
798 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
799 os
->os_flags
= os
->os_phys
->os_flags
;
802 dsl_dataset_dirty(ds
, tx
);
807 typedef struct dmu_objset_create_arg
{
808 const char *doca_name
;
810 void (*doca_userfunc
)(objset_t
*os
, void *arg
,
811 cred_t
*cr
, dmu_tx_t
*tx
);
813 dmu_objset_type_t doca_type
;
815 } dmu_objset_create_arg_t
;
819 dmu_objset_create_check(void *arg
, dmu_tx_t
*tx
)
821 dmu_objset_create_arg_t
*doca
= arg
;
822 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
827 if (strchr(doca
->doca_name
, '@') != NULL
)
828 return (SET_ERROR(EINVAL
));
830 error
= dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
);
834 dsl_dir_rele(pdd
, FTAG
);
835 return (SET_ERROR(EEXIST
));
837 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
839 dsl_dir_rele(pdd
, FTAG
);
845 dmu_objset_create_sync(void *arg
, dmu_tx_t
*tx
)
847 dmu_objset_create_arg_t
*doca
= arg
;
848 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
856 VERIFY0(dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
));
858 obj
= dsl_dataset_create_sync(pdd
, tail
, NULL
, doca
->doca_flags
,
859 doca
->doca_cred
, tx
);
861 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
862 bp
= dsl_dataset_get_blkptr(ds
);
863 os
= dmu_objset_create_impl(pdd
->dd_pool
->dp_spa
,
864 ds
, bp
, doca
->doca_type
, tx
);
866 if (doca
->doca_userfunc
!= NULL
) {
867 doca
->doca_userfunc(os
, doca
->doca_userarg
,
868 doca
->doca_cred
, tx
);
871 spa_history_log_internal_ds(ds
, "create", tx
, "");
872 zvol_create_minors(dp
->dp_spa
, doca
->doca_name
, B_TRUE
);
874 dsl_dataset_rele(ds
, FTAG
);
875 dsl_dir_rele(pdd
, FTAG
);
879 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
880 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
)
882 dmu_objset_create_arg_t doca
;
884 doca
.doca_name
= name
;
885 doca
.doca_cred
= CRED();
886 doca
.doca_flags
= flags
;
887 doca
.doca_userfunc
= func
;
888 doca
.doca_userarg
= arg
;
889 doca
.doca_type
= type
;
891 return (dsl_sync_task(name
,
892 dmu_objset_create_check
, dmu_objset_create_sync
, &doca
,
893 5, ZFS_SPACE_CHECK_NORMAL
));
896 typedef struct dmu_objset_clone_arg
{
897 const char *doca_clone
;
898 const char *doca_origin
;
900 } dmu_objset_clone_arg_t
;
904 dmu_objset_clone_check(void *arg
, dmu_tx_t
*tx
)
906 dmu_objset_clone_arg_t
*doca
= arg
;
910 dsl_dataset_t
*origin
;
911 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
913 if (strchr(doca
->doca_clone
, '@') != NULL
)
914 return (SET_ERROR(EINVAL
));
916 error
= dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
);
920 dsl_dir_rele(pdd
, FTAG
);
921 return (SET_ERROR(EEXIST
));
924 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
927 dsl_dir_rele(pdd
, FTAG
);
928 return (SET_ERROR(EDQUOT
));
930 dsl_dir_rele(pdd
, FTAG
);
932 error
= dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
);
936 /* You can only clone snapshots, not the head datasets. */
937 if (!origin
->ds_is_snapshot
) {
938 dsl_dataset_rele(origin
, FTAG
);
939 return (SET_ERROR(EINVAL
));
941 dsl_dataset_rele(origin
, FTAG
);
947 dmu_objset_clone_sync(void *arg
, dmu_tx_t
*tx
)
949 dmu_objset_clone_arg_t
*doca
= arg
;
950 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
953 dsl_dataset_t
*origin
, *ds
;
955 char namebuf
[MAXNAMELEN
];
957 VERIFY0(dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
));
958 VERIFY0(dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
));
960 obj
= dsl_dataset_create_sync(pdd
, tail
, origin
, 0,
961 doca
->doca_cred
, tx
);
963 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
964 dsl_dataset_name(origin
, namebuf
);
965 spa_history_log_internal_ds(ds
, "clone", tx
,
966 "origin=%s (%llu)", namebuf
, origin
->ds_object
);
967 zvol_create_minors(dp
->dp_spa
, doca
->doca_clone
, B_TRUE
);
968 dsl_dataset_rele(ds
, FTAG
);
969 dsl_dataset_rele(origin
, FTAG
);
970 dsl_dir_rele(pdd
, FTAG
);
974 dmu_objset_clone(const char *clone
, const char *origin
)
976 dmu_objset_clone_arg_t doca
;
978 doca
.doca_clone
= clone
;
979 doca
.doca_origin
= origin
;
980 doca
.doca_cred
= CRED();
982 return (dsl_sync_task(clone
,
983 dmu_objset_clone_check
, dmu_objset_clone_sync
, &doca
,
984 5, ZFS_SPACE_CHECK_NORMAL
));
988 dmu_objset_snapshot_one(const char *fsname
, const char *snapname
)
991 char *longsnap
= kmem_asprintf("%s@%s", fsname
, snapname
);
992 nvlist_t
*snaps
= fnvlist_alloc();
994 fnvlist_add_boolean(snaps
, longsnap
);
996 err
= dsl_dataset_snapshot(snaps
, NULL
, NULL
);
1002 dmu_objset_sync_dnodes(list_t
*list
, list_t
*newlist
, dmu_tx_t
*tx
)
1006 while ((dn
= list_head(list
))) {
1007 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1008 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1010 * Initialize dn_zio outside dnode_sync() because the
1011 * meta-dnode needs to set it ouside dnode_sync().
1013 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1016 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1017 list_remove(list
, dn
);
1020 (void) dnode_add_ref(dn
, newlist
);
1021 list_insert_tail(newlist
, dn
);
1030 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1034 blkptr_t
*bp
= zio
->io_bp
;
1036 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1038 ASSERT(!BP_IS_EMBEDDED(bp
));
1039 ASSERT3P(bp
, ==, os
->os_rootbp
);
1040 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_OBJSET
);
1041 ASSERT0(BP_GET_LEVEL(bp
));
1044 * Update rootbp fill count: it should be the number of objects
1045 * allocated in the object set (not counting the "special"
1046 * objects that are stored in the objset_phys_t -- the meta
1047 * dnode and user/group accounting objects).
1050 for (i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1051 bp
->blk_fill
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
1056 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1058 blkptr_t
*bp
= zio
->io_bp
;
1059 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1062 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1063 ASSERT(BP_EQUAL(bp
, bp_orig
));
1065 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1066 dmu_tx_t
*tx
= os
->os_synctx
;
1068 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1069 dsl_dataset_block_born(ds
, bp
, tx
);
1073 /* called from dsl */
1075 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1078 zbookmark_phys_t zb
;
1082 list_t
*newlist
= NULL
;
1083 dbuf_dirty_record_t
*dr
;
1085 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", tx
->tx_txg
);
1087 ASSERT(dmu_tx_is_syncing(tx
));
1088 /* XXX the write_done callback should really give us the tx... */
1091 if (os
->os_dsl_dataset
== NULL
) {
1093 * This is the MOS. If we have upgraded,
1094 * spa_max_replication() could change, so reset
1097 os
->os_copies
= spa_max_replication(os
->os_spa
);
1101 * Create the root block IO
1103 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1104 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1105 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1106 arc_release(os
->os_phys_buf
, &os
->os_phys_buf
);
1108 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1110 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1111 os
->os_rootbp
, os
->os_phys_buf
, DMU_OS_IS_L2CACHEABLE(os
),
1112 DMU_OS_IS_L2COMPRESSIBLE(os
), &zp
, dmu_objset_write_ready
,
1113 NULL
, dmu_objset_write_done
, os
, ZIO_PRIORITY_ASYNC_WRITE
,
1114 ZIO_FLAG_MUSTSUCCEED
, &zb
);
1117 * Sync special dnodes - the parent IO for the sync is the root block
1119 DMU_META_DNODE(os
)->dn_zio
= zio
;
1120 dnode_sync(DMU_META_DNODE(os
), tx
);
1122 os
->os_phys
->os_flags
= os
->os_flags
;
1124 if (DMU_USERUSED_DNODE(os
) &&
1125 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1126 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1127 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1128 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1129 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1132 txgoff
= tx
->tx_txg
& TXG_MASK
;
1134 if (dmu_objset_userused_enabled(os
)) {
1135 newlist
= &os
->os_synced_dnodes
;
1137 * We must create the list here because it uses the
1138 * dn_dirty_link[] of this txg.
1140 list_create(newlist
, sizeof (dnode_t
),
1141 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1144 dmu_objset_sync_dnodes(&os
->os_free_dnodes
[txgoff
], newlist
, tx
);
1145 dmu_objset_sync_dnodes(&os
->os_dirty_dnodes
[txgoff
], newlist
, tx
);
1147 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1148 while ((dr
= list_head(list
))) {
1149 ASSERT0(dr
->dr_dbuf
->db_level
);
1150 list_remove(list
, dr
);
1152 zio_nowait(dr
->dr_zio
);
1155 * Free intent log blocks up to this tx.
1157 zil_sync(os
->os_zil
, tx
);
1158 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1163 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1165 return (!list_is_empty(&os
->os_dirty_dnodes
[txg
& TXG_MASK
]) ||
1166 !list_is_empty(&os
->os_free_dnodes
[txg
& TXG_MASK
]));
1169 static objset_used_cb_t
*used_cbs
[DMU_OST_NUMTYPES
];
1172 dmu_objset_register_type(dmu_objset_type_t ost
, objset_used_cb_t
*cb
)
1178 dmu_objset_userused_enabled(objset_t
*os
)
1180 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1181 used_cbs
[os
->os_phys
->os_type
] != NULL
&&
1182 DMU_USERUSED_DNODE(os
) != NULL
);
1186 do_userquota_update(objset_t
*os
, uint64_t used
, uint64_t flags
,
1187 uint64_t user
, uint64_t group
, boolean_t subtract
, dmu_tx_t
*tx
)
1189 if ((flags
& DNODE_FLAG_USERUSED_ACCOUNTED
)) {
1190 int64_t delta
= DNODE_SIZE
+ used
;
1193 VERIFY3U(0, ==, zap_increment_int(os
, DMU_USERUSED_OBJECT
,
1195 VERIFY3U(0, ==, zap_increment_int(os
, DMU_GROUPUSED_OBJECT
,
1201 dmu_objset_do_userquota_updates(objset_t
*os
, dmu_tx_t
*tx
)
1204 list_t
*list
= &os
->os_synced_dnodes
;
1206 ASSERT(list_head(list
) == NULL
|| dmu_objset_userused_enabled(os
));
1208 while ((dn
= list_head(list
))) {
1210 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
1211 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
1212 dn
->dn_phys
->dn_flags
&
1213 DNODE_FLAG_USERUSED_ACCOUNTED
);
1215 /* Allocate the user/groupused objects if necessary. */
1216 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
1217 VERIFY(0 == zap_create_claim(os
,
1218 DMU_USERUSED_OBJECT
,
1219 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1220 VERIFY(0 == zap_create_claim(os
,
1221 DMU_GROUPUSED_OBJECT
,
1222 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1226 * We intentionally modify the zap object even if the
1227 * net delta is zero. Otherwise
1228 * the block of the zap obj could be shared between
1229 * datasets but need to be different between them after
1233 flags
= dn
->dn_id_flags
;
1235 if (flags
& DN_ID_OLD_EXIST
) {
1236 do_userquota_update(os
, dn
->dn_oldused
, dn
->dn_oldflags
,
1237 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
, tx
);
1239 if (flags
& DN_ID_NEW_EXIST
) {
1240 do_userquota_update(os
, DN_USED_BYTES(dn
->dn_phys
),
1241 dn
->dn_phys
->dn_flags
, dn
->dn_newuid
,
1242 dn
->dn_newgid
, B_FALSE
, tx
);
1245 mutex_enter(&dn
->dn_mtx
);
1247 dn
->dn_oldflags
= 0;
1248 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
1249 dn
->dn_olduid
= dn
->dn_newuid
;
1250 dn
->dn_oldgid
= dn
->dn_newgid
;
1251 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1252 if (dn
->dn_bonuslen
== 0)
1253 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1255 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1257 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
1258 mutex_exit(&dn
->dn_mtx
);
1260 list_remove(list
, dn
);
1261 dnode_rele(dn
, list
);
1266 * Returns a pointer to data to find uid/gid from
1268 * If a dirty record for transaction group that is syncing can't
1269 * be found then NULL is returned. In the NULL case it is assumed
1270 * the uid/gid aren't changing.
1273 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
1275 dbuf_dirty_record_t
*dr
, **drp
;
1278 if (db
->db_dirtycnt
== 0)
1279 return (db
->db
.db_data
); /* Nothing is changing */
1281 for (drp
= &db
->db_last_dirty
; (dr
= *drp
) != NULL
; drp
= &dr
->dr_next
)
1282 if (dr
->dr_txg
== tx
->tx_txg
)
1290 DB_DNODE_ENTER(dr
->dr_dbuf
);
1291 dn
= DB_DNODE(dr
->dr_dbuf
);
1293 if (dn
->dn_bonuslen
== 0 &&
1294 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
1295 data
= dr
->dt
.dl
.dr_data
->b_data
;
1297 data
= dr
->dt
.dl
.dr_data
;
1299 DB_DNODE_EXIT(dr
->dr_dbuf
);
1306 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
1308 objset_t
*os
= dn
->dn_objset
;
1310 dmu_buf_impl_t
*db
= NULL
;
1311 uint64_t *user
= NULL
;
1312 uint64_t *group
= NULL
;
1313 int flags
= dn
->dn_id_flags
;
1315 boolean_t have_spill
= B_FALSE
;
1317 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
1320 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
1321 DN_ID_CHKED_SPILL
)))
1324 if (before
&& dn
->dn_bonuslen
!= 0)
1325 data
= DN_BONUS(dn
->dn_phys
);
1326 else if (!before
&& dn
->dn_bonuslen
!= 0) {
1329 mutex_enter(&db
->db_mtx
);
1330 data
= dmu_objset_userquota_find_data(db
, tx
);
1332 data
= DN_BONUS(dn
->dn_phys
);
1334 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
1337 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
1338 rf
|= DB_RF_HAVESTRUCT
;
1339 error
= dmu_spill_hold_by_dnode(dn
,
1340 rf
| DB_RF_MUST_SUCCEED
,
1341 FTAG
, (dmu_buf_t
**)&db
);
1343 mutex_enter(&db
->db_mtx
);
1344 data
= (before
) ? db
->db
.db_data
:
1345 dmu_objset_userquota_find_data(db
, tx
);
1346 have_spill
= B_TRUE
;
1348 mutex_enter(&dn
->dn_mtx
);
1349 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1350 mutex_exit(&dn
->dn_mtx
);
1356 user
= &dn
->dn_olduid
;
1357 group
= &dn
->dn_oldgid
;
1359 user
= &dn
->dn_newuid
;
1360 group
= &dn
->dn_newgid
;
1364 * Must always call the callback in case the object
1365 * type has changed and that type isn't an object type to track
1367 error
= used_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
,
1371 * Preserve existing uid/gid when the callback can't determine
1372 * what the new uid/gid are and the callback returned EEXIST.
1373 * The EEXIST error tells us to just use the existing uid/gid.
1374 * If we don't know what the old values are then just assign
1375 * them to 0, since that is a new file being created.
1377 if (!before
&& data
== NULL
&& error
== EEXIST
) {
1378 if (flags
& DN_ID_OLD_EXIST
) {
1379 dn
->dn_newuid
= dn
->dn_olduid
;
1380 dn
->dn_newgid
= dn
->dn_oldgid
;
1389 mutex_exit(&db
->db_mtx
);
1391 mutex_enter(&dn
->dn_mtx
);
1392 if (error
== 0 && before
)
1393 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1394 if (error
== 0 && !before
)
1395 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
1398 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1400 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1402 mutex_exit(&dn
->dn_mtx
);
1404 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
1408 dmu_objset_userspace_present(objset_t
*os
)
1410 return (os
->os_phys
->os_flags
&
1411 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
1415 dmu_objset_userspace_upgrade(objset_t
*os
)
1420 if (dmu_objset_userspace_present(os
))
1422 if (!dmu_objset_userused_enabled(os
))
1423 return (SET_ERROR(ENOTSUP
));
1424 if (dmu_objset_is_snapshot(os
))
1425 return (SET_ERROR(EINVAL
));
1428 * We simply need to mark every object dirty, so that it will be
1429 * synced out and now accounted. If this is called
1430 * concurrently, or if we already did some work before crashing,
1431 * that's fine, since we track each object's accounted state
1435 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
1440 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1441 return (SET_ERROR(EINTR
));
1443 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
1446 tx
= dmu_tx_create(os
);
1447 dmu_tx_hold_bonus(tx
, obj
);
1448 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
1453 dmu_buf_will_dirty(db
, tx
);
1454 dmu_buf_rele(db
, FTAG
);
1458 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
1459 txg_wait_synced(dmu_objset_pool(os
), 0);
1464 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
1465 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1467 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
1468 usedobjsp
, availobjsp
);
1472 dmu_objset_fsid_guid(objset_t
*os
)
1474 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
1478 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
1480 stat
->dds_type
= os
->os_phys
->os_type
;
1481 if (os
->os_dsl_dataset
)
1482 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
1486 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
1488 ASSERT(os
->os_dsl_dataset
||
1489 os
->os_phys
->os_type
== DMU_OST_META
);
1491 if (os
->os_dsl_dataset
!= NULL
)
1492 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
1494 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
1495 os
->os_phys
->os_type
);
1496 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
1497 dmu_objset_userspace_present(os
));
1501 dmu_objset_is_snapshot(objset_t
*os
)
1503 if (os
->os_dsl_dataset
!= NULL
)
1504 return (os
->os_dsl_dataset
->ds_is_snapshot
);
1510 dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
, int maxlen
,
1511 boolean_t
*conflict
)
1513 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1516 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1517 return (SET_ERROR(ENOENT
));
1519 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1520 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, name
, 8, 1, &ignored
,
1521 MT_FIRST
, real
, maxlen
, conflict
));
1525 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
1526 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
1528 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1529 zap_cursor_t cursor
;
1530 zap_attribute_t attr
;
1532 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
1534 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1535 return (SET_ERROR(ENOENT
));
1537 zap_cursor_init_serialized(&cursor
,
1538 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1539 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, *offp
);
1541 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1542 zap_cursor_fini(&cursor
);
1543 return (SET_ERROR(ENOENT
));
1546 if (strlen(attr
.za_name
) + 1 > namelen
) {
1547 zap_cursor_fini(&cursor
);
1548 return (SET_ERROR(ENAMETOOLONG
));
1551 (void) strcpy(name
, attr
.za_name
);
1553 *idp
= attr
.za_first_integer
;
1555 *case_conflict
= attr
.za_normalization_conflict
;
1556 zap_cursor_advance(&cursor
);
1557 *offp
= zap_cursor_serialize(&cursor
);
1558 zap_cursor_fini(&cursor
);
1564 dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *value
)
1566 return (dsl_dataset_snap_lookup(os
->os_dsl_dataset
, name
, value
));
1570 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
1571 uint64_t *idp
, uint64_t *offp
)
1573 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
1574 zap_cursor_t cursor
;
1575 zap_attribute_t attr
;
1577 /* there is no next dir on a snapshot! */
1578 if (os
->os_dsl_dataset
->ds_object
!=
1579 dsl_dir_phys(dd
)->dd_head_dataset_obj
)
1580 return (SET_ERROR(ENOENT
));
1582 zap_cursor_init_serialized(&cursor
,
1583 dd
->dd_pool
->dp_meta_objset
,
1584 dsl_dir_phys(dd
)->dd_child_dir_zapobj
, *offp
);
1586 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1587 zap_cursor_fini(&cursor
);
1588 return (SET_ERROR(ENOENT
));
1591 if (strlen(attr
.za_name
) + 1 > namelen
) {
1592 zap_cursor_fini(&cursor
);
1593 return (SET_ERROR(ENAMETOOLONG
));
1596 (void) strcpy(name
, attr
.za_name
);
1598 *idp
= attr
.za_first_integer
;
1599 zap_cursor_advance(&cursor
);
1600 *offp
= zap_cursor_serialize(&cursor
);
1601 zap_cursor_fini(&cursor
);
1606 typedef struct dmu_objset_find_ctx
{
1610 int (*dc_func
)(dsl_pool_t
*, dsl_dataset_t
*, void *);
1613 kmutex_t
*dc_error_lock
;
1615 } dmu_objset_find_ctx_t
;
1618 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t
*dcp
)
1620 dsl_pool_t
*dp
= dcp
->dc_dp
;
1621 dmu_objset_find_ctx_t
*child_dcp
;
1625 zap_attribute_t
*attr
;
1629 /* don't process if there already was an error */
1630 if (*dcp
->dc_error
!= 0)
1633 err
= dsl_dir_hold_obj(dp
, dcp
->dc_ddobj
, NULL
, FTAG
, &dd
);
1637 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1638 if (dd
->dd_myname
[0] == '$') {
1639 dsl_dir_rele(dd
, FTAG
);
1643 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
1644 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1647 * Iterate over all children.
1649 if (dcp
->dc_flags
& DS_FIND_CHILDREN
) {
1650 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1651 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
1652 zap_cursor_retrieve(&zc
, attr
) == 0;
1653 (void) zap_cursor_advance(&zc
)) {
1654 ASSERT3U(attr
->za_integer_length
, ==,
1656 ASSERT3U(attr
->za_num_integers
, ==, 1);
1658 child_dcp
= kmem_alloc(sizeof (*child_dcp
), KM_SLEEP
);
1660 child_dcp
->dc_ddobj
= attr
->za_first_integer
;
1661 if (dcp
->dc_tq
!= NULL
)
1662 (void) taskq_dispatch(dcp
->dc_tq
,
1663 dmu_objset_find_dp_cb
, child_dcp
, TQ_SLEEP
);
1665 dmu_objset_find_dp_impl(child_dcp
);
1667 zap_cursor_fini(&zc
);
1671 * Iterate over all snapshots.
1673 if (dcp
->dc_flags
& DS_FIND_SNAPSHOTS
) {
1675 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1680 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
1681 dsl_dataset_rele(ds
, FTAG
);
1683 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1684 zap_cursor_retrieve(&zc
, attr
) == 0;
1685 (void) zap_cursor_advance(&zc
)) {
1686 ASSERT3U(attr
->za_integer_length
, ==,
1688 ASSERT3U(attr
->za_num_integers
, ==, 1);
1690 err
= dsl_dataset_hold_obj(dp
,
1691 attr
->za_first_integer
, FTAG
, &ds
);
1694 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
1695 dsl_dataset_rele(ds
, FTAG
);
1699 zap_cursor_fini(&zc
);
1703 dsl_dir_rele(dd
, FTAG
);
1704 kmem_free(attr
, sizeof (zap_attribute_t
));
1712 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1715 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
1716 dsl_dataset_rele(ds
, FTAG
);
1720 mutex_enter(dcp
->dc_error_lock
);
1721 /* only keep first error */
1722 if (*dcp
->dc_error
== 0)
1723 *dcp
->dc_error
= err
;
1724 mutex_exit(dcp
->dc_error_lock
);
1727 kmem_free(dcp
, sizeof (*dcp
));
1731 dmu_objset_find_dp_cb(void *arg
)
1733 dmu_objset_find_ctx_t
*dcp
= arg
;
1734 dsl_pool_t
*dp
= dcp
->dc_dp
;
1737 * We need to get a pool_config_lock here, as there are several
1738 * asssert(pool_config_held) down the stack. Getting a lock via
1739 * dsl_pool_config_enter is risky, as it might be stalled by a
1740 * pending writer. This would deadlock, as the write lock can
1741 * only be granted when our parent thread gives up the lock.
1742 * The _prio interface gives us priority over a pending writer.
1744 dsl_pool_config_enter_prio(dp
, FTAG
);
1746 dmu_objset_find_dp_impl(dcp
);
1748 dsl_pool_config_exit(dp
, FTAG
);
1752 * Find objsets under and including ddobj, call func(ds) on each.
1753 * The order for the enumeration is completely undefined.
1754 * func is called with dsl_pool_config held.
1757 dmu_objset_find_dp(dsl_pool_t
*dp
, uint64_t ddobj
,
1758 int func(dsl_pool_t
*, dsl_dataset_t
*, void *), void *arg
, int flags
)
1763 dmu_objset_find_ctx_t
*dcp
;
1766 mutex_init(&err_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1767 dcp
= kmem_alloc(sizeof (*dcp
), KM_SLEEP
);
1770 dcp
->dc_ddobj
= ddobj
;
1771 dcp
->dc_func
= func
;
1773 dcp
->dc_flags
= flags
;
1774 dcp
->dc_error_lock
= &err_lock
;
1775 dcp
->dc_error
= &error
;
1777 if ((flags
& DS_FIND_SERIALIZE
) || dsl_pool_config_held_writer(dp
)) {
1779 * In case a write lock is held we can't make use of
1780 * parallelism, as down the stack of the worker threads
1781 * the lock is asserted via dsl_pool_config_held.
1782 * In case of a read lock this is solved by getting a read
1783 * lock in each worker thread, which isn't possible in case
1784 * of a writer lock. So we fall back to the synchronous path
1786 * In the future it might be possible to get some magic into
1787 * dsl_pool_config_held in a way that it returns true for
1788 * the worker threads so that a single lock held from this
1789 * thread suffices. For now, stay single threaded.
1791 dmu_objset_find_dp_impl(dcp
);
1792 mutex_destroy(&err_lock
);
1797 ntasks
= dmu_find_threads
;
1799 ntasks
= vdev_count_leaves(dp
->dp_spa
) * 4;
1800 tq
= taskq_create("dmu_objset_find", ntasks
, maxclsyspri
, ntasks
,
1803 kmem_free(dcp
, sizeof (*dcp
));
1804 mutex_destroy(&err_lock
);
1806 return (SET_ERROR(ENOMEM
));
1810 /* dcp will be freed by task */
1811 (void) taskq_dispatch(tq
, dmu_objset_find_dp_cb
, dcp
, TQ_SLEEP
);
1814 * PORTING: this code relies on the property of taskq_wait to wait
1815 * until no more tasks are queued and no more tasks are active. As
1816 * we always queue new tasks from within other tasks, task_wait
1817 * reliably waits for the full recursion to finish, even though we
1818 * enqueue new tasks after taskq_wait has been called.
1819 * On platforms other than illumos, taskq_wait may not have this
1824 mutex_destroy(&err_lock
);
1830 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1831 * The dp_config_rwlock must not be held when this is called, and it
1832 * will not be held when the callback is called.
1833 * Therefore this function should only be used when the pool is not changing
1834 * (e.g. in syncing context), or the callback can deal with the possible races.
1837 dmu_objset_find_impl(spa_t
*spa
, const char *name
,
1838 int func(const char *, void *), void *arg
, int flags
)
1841 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1844 zap_attribute_t
*attr
;
1849 dsl_pool_config_enter(dp
, FTAG
);
1851 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, NULL
);
1853 dsl_pool_config_exit(dp
, FTAG
);
1857 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1858 if (dd
->dd_myname
[0] == '$') {
1859 dsl_dir_rele(dd
, FTAG
);
1860 dsl_pool_config_exit(dp
, FTAG
);
1864 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
1865 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1868 * Iterate over all children.
1870 if (flags
& DS_FIND_CHILDREN
) {
1871 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1872 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
1873 zap_cursor_retrieve(&zc
, attr
) == 0;
1874 (void) zap_cursor_advance(&zc
)) {
1875 ASSERT3U(attr
->za_integer_length
, ==,
1877 ASSERT3U(attr
->za_num_integers
, ==, 1);
1879 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
1880 dsl_pool_config_exit(dp
, FTAG
);
1881 err
= dmu_objset_find_impl(spa
, child
,
1883 dsl_pool_config_enter(dp
, FTAG
);
1888 zap_cursor_fini(&zc
);
1891 dsl_dir_rele(dd
, FTAG
);
1892 dsl_pool_config_exit(dp
, FTAG
);
1893 kmem_free(attr
, sizeof (zap_attribute_t
));
1899 * Iterate over all snapshots.
1901 if (flags
& DS_FIND_SNAPSHOTS
) {
1902 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1907 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
1908 dsl_dataset_rele(ds
, FTAG
);
1910 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1911 zap_cursor_retrieve(&zc
, attr
) == 0;
1912 (void) zap_cursor_advance(&zc
)) {
1913 ASSERT3U(attr
->za_integer_length
, ==,
1915 ASSERT3U(attr
->za_num_integers
, ==, 1);
1917 child
= kmem_asprintf("%s@%s",
1918 name
, attr
->za_name
);
1919 dsl_pool_config_exit(dp
, FTAG
);
1920 err
= func(child
, arg
);
1921 dsl_pool_config_enter(dp
, FTAG
);
1926 zap_cursor_fini(&zc
);
1930 dsl_dir_rele(dd
, FTAG
);
1931 kmem_free(attr
, sizeof (zap_attribute_t
));
1932 dsl_pool_config_exit(dp
, FTAG
);
1937 /* Apply to self. */
1938 return (func(name
, arg
));
1942 * See comment above dmu_objset_find_impl().
1945 dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
1951 error
= spa_open(name
, &spa
, FTAG
);
1954 error
= dmu_objset_find_impl(spa
, name
, func
, arg
, flags
);
1955 spa_close(spa
, FTAG
);
1960 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
1962 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
1963 os
->os_user_ptr
= user_ptr
;
1967 dmu_objset_get_user(objset_t
*os
)
1969 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
1970 return (os
->os_user_ptr
);
1974 * Determine name of filesystem, given name of snapshot.
1975 * buf must be at least MAXNAMELEN bytes
1978 dmu_fsname(const char *snapname
, char *buf
)
1980 char *atp
= strchr(snapname
, '@');
1982 return (SET_ERROR(EINVAL
));
1983 if (atp
- snapname
>= MAXNAMELEN
)
1984 return (SET_ERROR(ENAMETOOLONG
));
1985 (void) strlcpy(buf
, snapname
, atp
- snapname
+ 1);
1989 #if defined(_KERNEL) && defined(HAVE_SPL)
1990 EXPORT_SYMBOL(dmu_objset_zil
);
1991 EXPORT_SYMBOL(dmu_objset_pool
);
1992 EXPORT_SYMBOL(dmu_objset_ds
);
1993 EXPORT_SYMBOL(dmu_objset_type
);
1994 EXPORT_SYMBOL(dmu_objset_name
);
1995 EXPORT_SYMBOL(dmu_objset_hold
);
1996 EXPORT_SYMBOL(dmu_objset_own
);
1997 EXPORT_SYMBOL(dmu_objset_rele
);
1998 EXPORT_SYMBOL(dmu_objset_disown
);
1999 EXPORT_SYMBOL(dmu_objset_from_ds
);
2000 EXPORT_SYMBOL(dmu_objset_create
);
2001 EXPORT_SYMBOL(dmu_objset_clone
);
2002 EXPORT_SYMBOL(dmu_objset_stats
);
2003 EXPORT_SYMBOL(dmu_objset_fast_stat
);
2004 EXPORT_SYMBOL(dmu_objset_spa
);
2005 EXPORT_SYMBOL(dmu_objset_space
);
2006 EXPORT_SYMBOL(dmu_objset_fsid_guid
);
2007 EXPORT_SYMBOL(dmu_objset_find
);
2008 EXPORT_SYMBOL(dmu_objset_byteswap
);
2009 EXPORT_SYMBOL(dmu_objset_evict_dbufs
);
2010 EXPORT_SYMBOL(dmu_objset_snap_cmtime
);
2012 EXPORT_SYMBOL(dmu_objset_sync
);
2013 EXPORT_SYMBOL(dmu_objset_is_dirty
);
2014 EXPORT_SYMBOL(dmu_objset_create_impl
);
2015 EXPORT_SYMBOL(dmu_objset_open_impl
);
2016 EXPORT_SYMBOL(dmu_objset_evict
);
2017 EXPORT_SYMBOL(dmu_objset_register_type
);
2018 EXPORT_SYMBOL(dmu_objset_do_userquota_updates
);
2019 EXPORT_SYMBOL(dmu_objset_userquota_get_ids
);
2020 EXPORT_SYMBOL(dmu_objset_userused_enabled
);
2021 EXPORT_SYMBOL(dmu_objset_userspace_upgrade
);
2022 EXPORT_SYMBOL(dmu_objset_userspace_present
);