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.
31 /* Portions Copyright 2010 Robert Milkowski */
34 #include <sys/zfs_context.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dsl_dir.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/dsl_prop.h>
39 #include <sys/dsl_pool.h>
40 #include <sys/dsl_synctask.h>
41 #include <sys/dsl_deleg.h>
42 #include <sys/dnode.h>
45 #include <sys/dmu_tx.h>
48 #include <sys/dmu_impl.h>
49 #include <sys/zfs_ioctl.h>
51 #include <sys/zfs_onexit.h>
52 #include <sys/dsl_destroy.h>
56 * Needed to close a window in dnode_move() that allows the objset to be freed
57 * before it can be safely accessed.
62 * Tunable to overwrite the maximum number of threads for the parallization
63 * of dmu_objset_find_dp, needed to speed up the import of pools with many
65 * Default is 4 times the number of leaf vdevs.
67 int dmu_find_threads
= 0;
69 static void dmu_objset_find_dp_cb(void *arg
);
74 rw_init(&os_lock
, NULL
, RW_DEFAULT
, NULL
);
84 dmu_objset_spa(objset_t
*os
)
90 dmu_objset_zil(objset_t
*os
)
96 dmu_objset_pool(objset_t
*os
)
100 if ((ds
= os
->os_dsl_dataset
) != NULL
&& ds
->ds_dir
)
101 return (ds
->ds_dir
->dd_pool
);
103 return (spa_get_dsl(os
->os_spa
));
107 dmu_objset_ds(objset_t
*os
)
109 return (os
->os_dsl_dataset
);
113 dmu_objset_type(objset_t
*os
)
115 return (os
->os_phys
->os_type
);
119 dmu_objset_name(objset_t
*os
, char *buf
)
121 dsl_dataset_name(os
->os_dsl_dataset
, buf
);
125 dmu_objset_id(objset_t
*os
)
127 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
129 return (ds
? ds
->ds_object
: 0);
133 dmu_objset_syncprop(objset_t
*os
)
135 return (os
->os_sync
);
139 dmu_objset_logbias(objset_t
*os
)
141 return (os
->os_logbias
);
145 checksum_changed_cb(void *arg
, uint64_t newval
)
150 * Inheritance should have been done by now.
152 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
154 os
->os_checksum
= zio_checksum_select(newval
, ZIO_CHECKSUM_ON_VALUE
);
158 compression_changed_cb(void *arg
, uint64_t newval
)
163 * Inheritance and range checking should have been done by now.
165 ASSERT(newval
!= ZIO_COMPRESS_INHERIT
);
167 os
->os_compress
= zio_compress_select(os
->os_spa
, newval
,
172 copies_changed_cb(void *arg
, uint64_t newval
)
177 * Inheritance and range checking should have been done by now.
180 ASSERT(newval
<= spa_max_replication(os
->os_spa
));
182 os
->os_copies
= newval
;
186 dedup_changed_cb(void *arg
, uint64_t newval
)
189 spa_t
*spa
= os
->os_spa
;
190 enum zio_checksum checksum
;
193 * Inheritance should have been done by now.
195 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
197 checksum
= zio_checksum_dedup_select(spa
, newval
, ZIO_CHECKSUM_OFF
);
199 os
->os_dedup_checksum
= checksum
& ZIO_CHECKSUM_MASK
;
200 os
->os_dedup_verify
= !!(checksum
& ZIO_CHECKSUM_VERIFY
);
204 primary_cache_changed_cb(void *arg
, uint64_t newval
)
209 * Inheritance and range checking should have been done by now.
211 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
212 newval
== ZFS_CACHE_METADATA
);
214 os
->os_primary_cache
= newval
;
218 secondary_cache_changed_cb(void *arg
, uint64_t newval
)
223 * Inheritance and range checking should have been done by now.
225 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
226 newval
== ZFS_CACHE_METADATA
);
228 os
->os_secondary_cache
= newval
;
232 sync_changed_cb(void *arg
, uint64_t newval
)
237 * Inheritance and range checking should have been done by now.
239 ASSERT(newval
== ZFS_SYNC_STANDARD
|| newval
== ZFS_SYNC_ALWAYS
||
240 newval
== ZFS_SYNC_DISABLED
);
242 os
->os_sync
= newval
;
244 zil_set_sync(os
->os_zil
, newval
);
248 redundant_metadata_changed_cb(void *arg
, uint64_t newval
)
253 * Inheritance and range checking should have been done by now.
255 ASSERT(newval
== ZFS_REDUNDANT_METADATA_ALL
||
256 newval
== ZFS_REDUNDANT_METADATA_MOST
);
258 os
->os_redundant_metadata
= newval
;
262 logbias_changed_cb(void *arg
, uint64_t newval
)
266 ASSERT(newval
== ZFS_LOGBIAS_LATENCY
||
267 newval
== ZFS_LOGBIAS_THROUGHPUT
);
268 os
->os_logbias
= newval
;
270 zil_set_logbias(os
->os_zil
, newval
);
274 recordsize_changed_cb(void *arg
, uint64_t newval
)
278 os
->os_recordsize
= newval
;
282 dmu_objset_byteswap(void *buf
, size_t size
)
284 objset_phys_t
*osp
= buf
;
286 ASSERT(size
== OBJSET_OLD_PHYS_SIZE
|| size
== sizeof (objset_phys_t
));
287 dnode_byteswap(&osp
->os_meta_dnode
);
288 byteswap_uint64_array(&osp
->os_zil_header
, sizeof (zil_header_t
));
289 osp
->os_type
= BSWAP_64(osp
->os_type
);
290 osp
->os_flags
= BSWAP_64(osp
->os_flags
);
291 if (size
== sizeof (objset_phys_t
)) {
292 dnode_byteswap(&osp
->os_userused_dnode
);
293 dnode_byteswap(&osp
->os_groupused_dnode
);
298 dmu_objset_open_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
304 ASSERT(ds
== NULL
|| MUTEX_HELD(&ds
->ds_opening_lock
));
306 os
= kmem_zalloc(sizeof (objset_t
), KM_SLEEP
);
307 os
->os_dsl_dataset
= ds
;
310 if (!BP_IS_HOLE(os
->os_rootbp
)) {
311 arc_flags_t aflags
= ARC_FLAG_WAIT
;
313 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
314 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
316 if (DMU_OS_IS_L2CACHEABLE(os
))
317 aflags
|= ARC_FLAG_L2CACHE
;
318 if (DMU_OS_IS_L2COMPRESSIBLE(os
))
319 aflags
|= ARC_FLAG_L2COMPRESS
;
321 dprintf_bp(os
->os_rootbp
, "reading %s", "");
322 err
= arc_read(NULL
, spa
, os
->os_rootbp
,
323 arc_getbuf_func
, &os
->os_phys_buf
,
324 ZIO_PRIORITY_SYNC_READ
, ZIO_FLAG_CANFAIL
, &aflags
, &zb
);
326 kmem_free(os
, sizeof (objset_t
));
327 /* convert checksum errors into IO errors */
329 err
= SET_ERROR(EIO
);
333 /* Increase the blocksize if we are permitted. */
334 if (spa_version(spa
) >= SPA_VERSION_USERSPACE
&&
335 arc_buf_size(os
->os_phys_buf
) < sizeof (objset_phys_t
)) {
336 arc_buf_t
*buf
= arc_buf_alloc(spa
,
337 sizeof (objset_phys_t
), &os
->os_phys_buf
,
339 bzero(buf
->b_data
, sizeof (objset_phys_t
));
340 bcopy(os
->os_phys_buf
->b_data
, buf
->b_data
,
341 arc_buf_size(os
->os_phys_buf
));
342 (void) arc_buf_remove_ref(os
->os_phys_buf
,
344 os
->os_phys_buf
= buf
;
347 os
->os_phys
= os
->os_phys_buf
->b_data
;
348 os
->os_flags
= os
->os_phys
->os_flags
;
350 int size
= spa_version(spa
) >= SPA_VERSION_USERSPACE
?
351 sizeof (objset_phys_t
) : OBJSET_OLD_PHYS_SIZE
;
352 os
->os_phys_buf
= arc_buf_alloc(spa
, size
,
353 &os
->os_phys_buf
, ARC_BUFC_METADATA
);
354 os
->os_phys
= os
->os_phys_buf
->b_data
;
355 bzero(os
->os_phys
, size
);
359 * Note: the changed_cb will be called once before the register
360 * func returns, thus changing the checksum/compression from the
361 * default (fletcher2/off). Snapshots don't need to know about
362 * checksum/compression/copies.
365 err
= dsl_prop_register(ds
,
366 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
367 primary_cache_changed_cb
, os
);
369 err
= dsl_prop_register(ds
,
370 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
371 secondary_cache_changed_cb
, os
);
373 if (!ds
->ds_is_snapshot
) {
375 err
= dsl_prop_register(ds
,
376 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
377 checksum_changed_cb
, os
);
380 err
= dsl_prop_register(ds
,
381 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
382 compression_changed_cb
, os
);
385 err
= dsl_prop_register(ds
,
386 zfs_prop_to_name(ZFS_PROP_COPIES
),
387 copies_changed_cb
, os
);
390 err
= dsl_prop_register(ds
,
391 zfs_prop_to_name(ZFS_PROP_DEDUP
),
392 dedup_changed_cb
, os
);
395 err
= dsl_prop_register(ds
,
396 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
397 logbias_changed_cb
, os
);
400 err
= dsl_prop_register(ds
,
401 zfs_prop_to_name(ZFS_PROP_SYNC
),
402 sync_changed_cb
, os
);
405 err
= dsl_prop_register(ds
,
407 ZFS_PROP_REDUNDANT_METADATA
),
408 redundant_metadata_changed_cb
, os
);
411 err
= dsl_prop_register(ds
,
412 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
413 recordsize_changed_cb
, os
);
417 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
,
419 kmem_free(os
, sizeof (objset_t
));
423 /* It's the meta-objset. */
424 os
->os_checksum
= ZIO_CHECKSUM_FLETCHER_4
;
425 os
->os_compress
= ZIO_COMPRESS_ON
;
426 os
->os_copies
= spa_max_replication(spa
);
427 os
->os_dedup_checksum
= ZIO_CHECKSUM_OFF
;
428 os
->os_dedup_verify
= B_FALSE
;
429 os
->os_logbias
= ZFS_LOGBIAS_LATENCY
;
430 os
->os_sync
= ZFS_SYNC_STANDARD
;
431 os
->os_primary_cache
= ZFS_CACHE_ALL
;
432 os
->os_secondary_cache
= ZFS_CACHE_ALL
;
435 if (ds
== NULL
|| !ds
->ds_is_snapshot
)
436 os
->os_zil_header
= os
->os_phys
->os_zil_header
;
437 os
->os_zil
= zil_alloc(os
, &os
->os_zil_header
);
439 for (i
= 0; i
< TXG_SIZE
; i
++) {
440 list_create(&os
->os_dirty_dnodes
[i
], sizeof (dnode_t
),
441 offsetof(dnode_t
, dn_dirty_link
[i
]));
442 list_create(&os
->os_free_dnodes
[i
], sizeof (dnode_t
),
443 offsetof(dnode_t
, dn_dirty_link
[i
]));
445 list_create(&os
->os_dnodes
, sizeof (dnode_t
),
446 offsetof(dnode_t
, dn_link
));
447 list_create(&os
->os_downgraded_dbufs
, sizeof (dmu_buf_impl_t
),
448 offsetof(dmu_buf_impl_t
, db_link
));
450 list_link_init(&os
->os_evicting_node
);
452 mutex_init(&os
->os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
453 mutex_init(&os
->os_obj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
454 mutex_init(&os
->os_user_ptr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
456 dnode_special_open(os
, &os
->os_phys
->os_meta_dnode
,
457 DMU_META_DNODE_OBJECT
, &os
->os_meta_dnode
);
458 if (arc_buf_size(os
->os_phys_buf
) >= sizeof (objset_phys_t
)) {
459 dnode_special_open(os
, &os
->os_phys
->os_userused_dnode
,
460 DMU_USERUSED_OBJECT
, &os
->os_userused_dnode
);
461 dnode_special_open(os
, &os
->os_phys
->os_groupused_dnode
,
462 DMU_GROUPUSED_OBJECT
, &os
->os_groupused_dnode
);
470 dmu_objset_from_ds(dsl_dataset_t
*ds
, objset_t
**osp
)
474 mutex_enter(&ds
->ds_opening_lock
);
475 if (ds
->ds_objset
== NULL
) {
477 err
= dmu_objset_open_impl(dsl_dataset_get_spa(ds
),
478 ds
, dsl_dataset_get_blkptr(ds
), &os
);
481 mutex_enter(&ds
->ds_lock
);
482 ASSERT(ds
->ds_objset
== NULL
);
484 mutex_exit(&ds
->ds_lock
);
487 *osp
= ds
->ds_objset
;
488 mutex_exit(&ds
->ds_opening_lock
);
493 * Holds the pool while the objset is held. Therefore only one objset
494 * can be held at a time.
497 dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
)
503 err
= dsl_pool_hold(name
, tag
, &dp
);
506 err
= dsl_dataset_hold(dp
, name
, tag
, &ds
);
508 dsl_pool_rele(dp
, tag
);
512 err
= dmu_objset_from_ds(ds
, osp
);
514 dsl_dataset_rele(ds
, tag
);
515 dsl_pool_rele(dp
, tag
);
522 dmu_objset_own_impl(dsl_dataset_t
*ds
, dmu_objset_type_t type
,
523 boolean_t readonly
, void *tag
, objset_t
**osp
)
527 err
= dmu_objset_from_ds(ds
, osp
);
529 dsl_dataset_disown(ds
, tag
);
530 } else if (type
!= DMU_OST_ANY
&& type
!= (*osp
)->os_phys
->os_type
) {
531 dsl_dataset_disown(ds
, tag
);
532 return (SET_ERROR(EINVAL
));
533 } else if (!readonly
&& dsl_dataset_is_snapshot(ds
)) {
534 dsl_dataset_disown(ds
, tag
);
535 return (SET_ERROR(EROFS
));
541 * dsl_pool must not be held when this is called.
542 * Upon successful return, there will be a longhold on the dataset,
543 * and the dsl_pool will not be held.
546 dmu_objset_own(const char *name
, dmu_objset_type_t type
,
547 boolean_t readonly
, void *tag
, objset_t
**osp
)
553 err
= dsl_pool_hold(name
, FTAG
, &dp
);
556 err
= dsl_dataset_own(dp
, name
, tag
, &ds
);
558 dsl_pool_rele(dp
, FTAG
);
561 err
= dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
);
562 dsl_pool_rele(dp
, FTAG
);
568 dmu_objset_own_obj(dsl_pool_t
*dp
, uint64_t obj
, dmu_objset_type_t type
,
569 boolean_t readonly
, void *tag
, objset_t
**osp
)
574 err
= dsl_dataset_own_obj(dp
, obj
, tag
, &ds
);
578 return (dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
));
582 dmu_objset_rele(objset_t
*os
, void *tag
)
584 dsl_pool_t
*dp
= dmu_objset_pool(os
);
585 dsl_dataset_rele(os
->os_dsl_dataset
, tag
);
586 dsl_pool_rele(dp
, tag
);
590 * When we are called, os MUST refer to an objset associated with a dataset
591 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
592 * == tag. We will then release and reacquire ownership of the dataset while
593 * holding the pool config_rwlock to avoid intervening namespace or ownership
596 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
597 * release the hold on its dataset and acquire a new one on the dataset of the
598 * same name so that it can be partially torn down and reconstructed.
601 dmu_objset_refresh_ownership(objset_t
*os
, void *tag
)
604 dsl_dataset_t
*ds
, *newds
;
605 char name
[MAXNAMELEN
];
607 ds
= os
->os_dsl_dataset
;
608 VERIFY3P(ds
, !=, NULL
);
609 VERIFY3P(ds
->ds_owner
, ==, tag
);
610 VERIFY(dsl_dataset_long_held(ds
));
612 dsl_dataset_name(ds
, name
);
613 dp
= dmu_objset_pool(os
);
614 dsl_pool_config_enter(dp
, FTAG
);
615 dmu_objset_disown(os
, tag
);
616 VERIFY0(dsl_dataset_own(dp
, name
, tag
, &newds
));
617 VERIFY3P(newds
, ==, os
->os_dsl_dataset
);
618 dsl_pool_config_exit(dp
, FTAG
);
622 dmu_objset_disown(objset_t
*os
, void *tag
)
624 dsl_dataset_disown(os
->os_dsl_dataset
, tag
);
628 dmu_objset_evict_dbufs(objset_t
*os
)
633 dn_marker
= kmem_alloc(sizeof (dnode_t
), KM_SLEEP
);
635 mutex_enter(&os
->os_lock
);
636 dn
= list_head(&os
->os_dnodes
);
639 * Skip dnodes without holds. We have to do this dance
640 * because dnode_add_ref() only works if there is already a
641 * hold. If the dnode has no holds, then it has no dbufs.
643 if (dnode_add_ref(dn
, FTAG
)) {
644 list_insert_after(&os
->os_dnodes
, dn
, dn_marker
);
645 mutex_exit(&os
->os_lock
);
647 dnode_evict_dbufs(dn
);
648 dnode_rele(dn
, FTAG
);
650 mutex_enter(&os
->os_lock
);
651 dn
= list_next(&os
->os_dnodes
, dn_marker
);
652 list_remove(&os
->os_dnodes
, dn_marker
);
654 dn
= list_next(&os
->os_dnodes
, dn
);
657 mutex_exit(&os
->os_lock
);
659 kmem_free(dn_marker
, sizeof (dnode_t
));
661 if (DMU_USERUSED_DNODE(os
) != NULL
) {
662 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os
));
663 dnode_evict_dbufs(DMU_USERUSED_DNODE(os
));
665 dnode_evict_dbufs(DMU_META_DNODE(os
));
669 * Objset eviction processing is split into into two pieces.
670 * The first marks the objset as evicting, evicts any dbufs that
671 * have a refcount of zero, and then queues up the objset for the
672 * second phase of eviction. Once os->os_dnodes has been cleared by
673 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
674 * The second phase closes the special dnodes, dequeues the objset from
675 * the list of those undergoing eviction, and finally frees the objset.
677 * NOTE: Due to asynchronous eviction processing (invocation of
678 * dnode_buf_pageout()), it is possible for the meta dnode for the
679 * objset to have no holds even though os->os_dnodes is not empty.
682 dmu_objset_evict(objset_t
*os
)
686 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
688 for (t
= 0; t
< TXG_SIZE
; t
++)
689 ASSERT(!dmu_objset_is_dirty(os
, t
));
692 if (!ds
->ds_is_snapshot
) {
693 VERIFY0(dsl_prop_unregister(ds
,
694 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
695 checksum_changed_cb
, os
));
696 VERIFY0(dsl_prop_unregister(ds
,
697 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
698 compression_changed_cb
, os
));
699 VERIFY0(dsl_prop_unregister(ds
,
700 zfs_prop_to_name(ZFS_PROP_COPIES
),
701 copies_changed_cb
, os
));
702 VERIFY0(dsl_prop_unregister(ds
,
703 zfs_prop_to_name(ZFS_PROP_DEDUP
),
704 dedup_changed_cb
, os
));
705 VERIFY0(dsl_prop_unregister(ds
,
706 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
707 logbias_changed_cb
, os
));
708 VERIFY0(dsl_prop_unregister(ds
,
709 zfs_prop_to_name(ZFS_PROP_SYNC
),
710 sync_changed_cb
, os
));
711 VERIFY0(dsl_prop_unregister(ds
,
712 zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA
),
713 redundant_metadata_changed_cb
, os
));
714 VERIFY0(dsl_prop_unregister(ds
,
715 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
716 recordsize_changed_cb
, os
));
718 VERIFY0(dsl_prop_unregister(ds
,
719 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
720 primary_cache_changed_cb
, os
));
721 VERIFY0(dsl_prop_unregister(ds
,
722 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
723 secondary_cache_changed_cb
, os
));
729 dmu_objset_evict_dbufs(os
);
731 mutex_enter(&os
->os_lock
);
732 spa_evicting_os_register(os
->os_spa
, os
);
733 if (list_is_empty(&os
->os_dnodes
)) {
734 mutex_exit(&os
->os_lock
);
735 dmu_objset_evict_done(os
);
737 mutex_exit(&os
->os_lock
);
742 dmu_objset_evict_done(objset_t
*os
)
744 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
746 dnode_special_close(&os
->os_meta_dnode
);
747 if (DMU_USERUSED_DNODE(os
)) {
748 dnode_special_close(&os
->os_userused_dnode
);
749 dnode_special_close(&os
->os_groupused_dnode
);
751 zil_free(os
->os_zil
);
753 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
, &os
->os_phys_buf
));
756 * This is a barrier to prevent the objset from going away in
757 * dnode_move() until we can safely ensure that the objset is still in
758 * use. We consider the objset valid before the barrier and invalid
761 rw_enter(&os_lock
, RW_READER
);
764 mutex_destroy(&os
->os_lock
);
765 mutex_destroy(&os
->os_obj_lock
);
766 mutex_destroy(&os
->os_user_ptr_lock
);
767 spa_evicting_os_deregister(os
->os_spa
, os
);
768 kmem_free(os
, sizeof (objset_t
));
772 dmu_objset_snap_cmtime(objset_t
*os
)
774 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
777 /* called from dsl for meta-objset */
779 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
780 dmu_objset_type_t type
, dmu_tx_t
*tx
)
785 ASSERT(dmu_tx_is_syncing(tx
));
788 VERIFY0(dmu_objset_from_ds(ds
, &os
));
790 VERIFY0(dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
792 mdn
= DMU_META_DNODE(os
);
794 dnode_allocate(mdn
, DMU_OT_DNODE
, 1 << DNODE_BLOCK_SHIFT
,
795 DN_MAX_INDBLKSHIFT
, DMU_OT_NONE
, 0, tx
);
798 * We don't want to have to increase the meta-dnode's nlevels
799 * later, because then we could do it in quescing context while
800 * we are also accessing it in open context.
802 * This precaution is not necessary for the MOS (ds == NULL),
803 * because the MOS is only updated in syncing context.
804 * This is most fortunate: the MOS is the only objset that
805 * needs to be synced multiple times as spa_sync() iterates
806 * to convergence, so minimizing its dn_nlevels matters.
812 * Determine the number of levels necessary for the meta-dnode
813 * to contain DN_MAX_OBJECT dnodes.
815 while ((uint64_t)mdn
->dn_nblkptr
<< (mdn
->dn_datablkshift
+
816 (levels
- 1) * (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
817 DN_MAX_OBJECT
* sizeof (dnode_phys_t
))
820 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
821 mdn
->dn_nlevels
= levels
;
824 ASSERT(type
!= DMU_OST_NONE
);
825 ASSERT(type
!= DMU_OST_ANY
);
826 ASSERT(type
< DMU_OST_NUMTYPES
);
827 os
->os_phys
->os_type
= type
;
828 if (dmu_objset_userused_enabled(os
)) {
829 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
830 os
->os_flags
= os
->os_phys
->os_flags
;
833 dsl_dataset_dirty(ds
, tx
);
838 typedef struct dmu_objset_create_arg
{
839 const char *doca_name
;
841 void (*doca_userfunc
)(objset_t
*os
, void *arg
,
842 cred_t
*cr
, dmu_tx_t
*tx
);
844 dmu_objset_type_t doca_type
;
846 } dmu_objset_create_arg_t
;
850 dmu_objset_create_check(void *arg
, dmu_tx_t
*tx
)
852 dmu_objset_create_arg_t
*doca
= arg
;
853 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
858 if (strchr(doca
->doca_name
, '@') != NULL
)
859 return (SET_ERROR(EINVAL
));
861 error
= dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
);
865 dsl_dir_rele(pdd
, FTAG
);
866 return (SET_ERROR(EEXIST
));
868 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
870 dsl_dir_rele(pdd
, FTAG
);
876 dmu_objset_create_sync(void *arg
, dmu_tx_t
*tx
)
878 dmu_objset_create_arg_t
*doca
= arg
;
879 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
887 VERIFY0(dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
));
889 obj
= dsl_dataset_create_sync(pdd
, tail
, NULL
, doca
->doca_flags
,
890 doca
->doca_cred
, tx
);
892 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
893 bp
= dsl_dataset_get_blkptr(ds
);
894 os
= dmu_objset_create_impl(pdd
->dd_pool
->dp_spa
,
895 ds
, bp
, doca
->doca_type
, tx
);
897 if (doca
->doca_userfunc
!= NULL
) {
898 doca
->doca_userfunc(os
, doca
->doca_userarg
,
899 doca
->doca_cred
, tx
);
902 spa_history_log_internal_ds(ds
, "create", tx
, "");
903 dsl_dataset_rele(ds
, FTAG
);
904 dsl_dir_rele(pdd
, FTAG
);
908 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
909 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
)
911 dmu_objset_create_arg_t doca
;
913 doca
.doca_name
= name
;
914 doca
.doca_cred
= CRED();
915 doca
.doca_flags
= flags
;
916 doca
.doca_userfunc
= func
;
917 doca
.doca_userarg
= arg
;
918 doca
.doca_type
= type
;
920 return (dsl_sync_task(name
,
921 dmu_objset_create_check
, dmu_objset_create_sync
, &doca
,
922 5, ZFS_SPACE_CHECK_NORMAL
));
925 typedef struct dmu_objset_clone_arg
{
926 const char *doca_clone
;
927 const char *doca_origin
;
929 } dmu_objset_clone_arg_t
;
933 dmu_objset_clone_check(void *arg
, dmu_tx_t
*tx
)
935 dmu_objset_clone_arg_t
*doca
= arg
;
939 dsl_dataset_t
*origin
;
940 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
942 if (strchr(doca
->doca_clone
, '@') != NULL
)
943 return (SET_ERROR(EINVAL
));
945 error
= dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
);
949 dsl_dir_rele(pdd
, FTAG
);
950 return (SET_ERROR(EEXIST
));
953 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
956 dsl_dir_rele(pdd
, FTAG
);
957 return (SET_ERROR(EDQUOT
));
959 dsl_dir_rele(pdd
, FTAG
);
961 error
= dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
);
965 /* You can only clone snapshots, not the head datasets. */
966 if (!origin
->ds_is_snapshot
) {
967 dsl_dataset_rele(origin
, FTAG
);
968 return (SET_ERROR(EINVAL
));
970 dsl_dataset_rele(origin
, FTAG
);
976 dmu_objset_clone_sync(void *arg
, dmu_tx_t
*tx
)
978 dmu_objset_clone_arg_t
*doca
= arg
;
979 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
982 dsl_dataset_t
*origin
, *ds
;
984 char namebuf
[MAXNAMELEN
];
986 VERIFY0(dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
));
987 VERIFY0(dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
));
989 obj
= dsl_dataset_create_sync(pdd
, tail
, origin
, 0,
990 doca
->doca_cred
, tx
);
992 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
993 dsl_dataset_name(origin
, namebuf
);
994 spa_history_log_internal_ds(ds
, "clone", tx
,
995 "origin=%s (%llu)", namebuf
, origin
->ds_object
);
996 dsl_dataset_rele(ds
, FTAG
);
997 dsl_dataset_rele(origin
, FTAG
);
998 dsl_dir_rele(pdd
, FTAG
);
1002 dmu_objset_clone(const char *clone
, const char *origin
)
1004 dmu_objset_clone_arg_t doca
;
1006 doca
.doca_clone
= clone
;
1007 doca
.doca_origin
= origin
;
1008 doca
.doca_cred
= CRED();
1010 return (dsl_sync_task(clone
,
1011 dmu_objset_clone_check
, dmu_objset_clone_sync
, &doca
,
1012 5, ZFS_SPACE_CHECK_NORMAL
));
1016 dmu_objset_snapshot_one(const char *fsname
, const char *snapname
)
1019 char *longsnap
= kmem_asprintf("%s@%s", fsname
, snapname
);
1020 nvlist_t
*snaps
= fnvlist_alloc();
1022 fnvlist_add_boolean(snaps
, longsnap
);
1024 err
= dsl_dataset_snapshot(snaps
, NULL
, NULL
);
1025 fnvlist_free(snaps
);
1030 dmu_objset_sync_dnodes(list_t
*list
, list_t
*newlist
, dmu_tx_t
*tx
)
1034 while ((dn
= list_head(list
))) {
1035 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1036 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1038 * Initialize dn_zio outside dnode_sync() because the
1039 * meta-dnode needs to set it ouside dnode_sync().
1041 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1044 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1045 list_remove(list
, dn
);
1048 (void) dnode_add_ref(dn
, newlist
);
1049 list_insert_tail(newlist
, dn
);
1058 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1062 blkptr_t
*bp
= zio
->io_bp
;
1064 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1066 ASSERT(!BP_IS_EMBEDDED(bp
));
1067 ASSERT3P(bp
, ==, os
->os_rootbp
);
1068 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_OBJSET
);
1069 ASSERT0(BP_GET_LEVEL(bp
));
1072 * Update rootbp fill count: it should be the number of objects
1073 * allocated in the object set (not counting the "special"
1074 * objects that are stored in the objset_phys_t -- the meta
1075 * dnode and user/group accounting objects).
1078 for (i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1079 bp
->blk_fill
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
1084 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1086 blkptr_t
*bp
= zio
->io_bp
;
1087 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1090 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1091 ASSERT(BP_EQUAL(bp
, bp_orig
));
1093 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1094 dmu_tx_t
*tx
= os
->os_synctx
;
1096 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1097 dsl_dataset_block_born(ds
, bp
, tx
);
1101 /* called from dsl */
1103 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1106 zbookmark_phys_t zb
;
1110 list_t
*newlist
= NULL
;
1111 dbuf_dirty_record_t
*dr
;
1113 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", tx
->tx_txg
);
1115 ASSERT(dmu_tx_is_syncing(tx
));
1116 /* XXX the write_done callback should really give us the tx... */
1119 if (os
->os_dsl_dataset
== NULL
) {
1121 * This is the MOS. If we have upgraded,
1122 * spa_max_replication() could change, so reset
1125 os
->os_copies
= spa_max_replication(os
->os_spa
);
1129 * Create the root block IO
1131 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1132 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1133 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1134 arc_release(os
->os_phys_buf
, &os
->os_phys_buf
);
1136 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1138 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1139 os
->os_rootbp
, os
->os_phys_buf
, DMU_OS_IS_L2CACHEABLE(os
),
1140 DMU_OS_IS_L2COMPRESSIBLE(os
), &zp
, dmu_objset_write_ready
,
1141 NULL
, dmu_objset_write_done
, os
, ZIO_PRIORITY_ASYNC_WRITE
,
1142 ZIO_FLAG_MUSTSUCCEED
, &zb
);
1145 * Sync special dnodes - the parent IO for the sync is the root block
1147 DMU_META_DNODE(os
)->dn_zio
= zio
;
1148 dnode_sync(DMU_META_DNODE(os
), tx
);
1150 os
->os_phys
->os_flags
= os
->os_flags
;
1152 if (DMU_USERUSED_DNODE(os
) &&
1153 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1154 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1155 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1156 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1157 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1160 txgoff
= tx
->tx_txg
& TXG_MASK
;
1162 if (dmu_objset_userused_enabled(os
)) {
1163 newlist
= &os
->os_synced_dnodes
;
1165 * We must create the list here because it uses the
1166 * dn_dirty_link[] of this txg.
1168 list_create(newlist
, sizeof (dnode_t
),
1169 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1172 dmu_objset_sync_dnodes(&os
->os_free_dnodes
[txgoff
], newlist
, tx
);
1173 dmu_objset_sync_dnodes(&os
->os_dirty_dnodes
[txgoff
], newlist
, tx
);
1175 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1176 while ((dr
= list_head(list
))) {
1177 ASSERT0(dr
->dr_dbuf
->db_level
);
1178 list_remove(list
, dr
);
1180 zio_nowait(dr
->dr_zio
);
1183 * Free intent log blocks up to this tx.
1185 zil_sync(os
->os_zil
, tx
);
1186 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1191 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1193 return (!list_is_empty(&os
->os_dirty_dnodes
[txg
& TXG_MASK
]) ||
1194 !list_is_empty(&os
->os_free_dnodes
[txg
& TXG_MASK
]));
1197 static objset_used_cb_t
*used_cbs
[DMU_OST_NUMTYPES
];
1200 dmu_objset_register_type(dmu_objset_type_t ost
, objset_used_cb_t
*cb
)
1206 dmu_objset_userused_enabled(objset_t
*os
)
1208 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1209 used_cbs
[os
->os_phys
->os_type
] != NULL
&&
1210 DMU_USERUSED_DNODE(os
) != NULL
);
1214 do_userquota_update(objset_t
*os
, uint64_t used
, uint64_t flags
,
1215 uint64_t user
, uint64_t group
, boolean_t subtract
, dmu_tx_t
*tx
)
1217 if ((flags
& DNODE_FLAG_USERUSED_ACCOUNTED
)) {
1218 int64_t delta
= DNODE_SIZE
+ used
;
1221 VERIFY3U(0, ==, zap_increment_int(os
, DMU_USERUSED_OBJECT
,
1223 VERIFY3U(0, ==, zap_increment_int(os
, DMU_GROUPUSED_OBJECT
,
1229 dmu_objset_do_userquota_updates(objset_t
*os
, dmu_tx_t
*tx
)
1232 list_t
*list
= &os
->os_synced_dnodes
;
1234 ASSERT(list_head(list
) == NULL
|| dmu_objset_userused_enabled(os
));
1236 while ((dn
= list_head(list
))) {
1238 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
1239 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
1240 dn
->dn_phys
->dn_flags
&
1241 DNODE_FLAG_USERUSED_ACCOUNTED
);
1243 /* Allocate the user/groupused objects if necessary. */
1244 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
1245 VERIFY(0 == zap_create_claim(os
,
1246 DMU_USERUSED_OBJECT
,
1247 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1248 VERIFY(0 == zap_create_claim(os
,
1249 DMU_GROUPUSED_OBJECT
,
1250 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1254 * We intentionally modify the zap object even if the
1255 * net delta is zero. Otherwise
1256 * the block of the zap obj could be shared between
1257 * datasets but need to be different between them after
1261 flags
= dn
->dn_id_flags
;
1263 if (flags
& DN_ID_OLD_EXIST
) {
1264 do_userquota_update(os
, dn
->dn_oldused
, dn
->dn_oldflags
,
1265 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
, tx
);
1267 if (flags
& DN_ID_NEW_EXIST
) {
1268 do_userquota_update(os
, DN_USED_BYTES(dn
->dn_phys
),
1269 dn
->dn_phys
->dn_flags
, dn
->dn_newuid
,
1270 dn
->dn_newgid
, B_FALSE
, tx
);
1273 mutex_enter(&dn
->dn_mtx
);
1275 dn
->dn_oldflags
= 0;
1276 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
1277 dn
->dn_olduid
= dn
->dn_newuid
;
1278 dn
->dn_oldgid
= dn
->dn_newgid
;
1279 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1280 if (dn
->dn_bonuslen
== 0)
1281 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1283 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1285 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
1286 mutex_exit(&dn
->dn_mtx
);
1288 list_remove(list
, dn
);
1289 dnode_rele(dn
, list
);
1294 * Returns a pointer to data to find uid/gid from
1296 * If a dirty record for transaction group that is syncing can't
1297 * be found then NULL is returned. In the NULL case it is assumed
1298 * the uid/gid aren't changing.
1301 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
1303 dbuf_dirty_record_t
*dr
, **drp
;
1306 if (db
->db_dirtycnt
== 0)
1307 return (db
->db
.db_data
); /* Nothing is changing */
1309 for (drp
= &db
->db_last_dirty
; (dr
= *drp
) != NULL
; drp
= &dr
->dr_next
)
1310 if (dr
->dr_txg
== tx
->tx_txg
)
1318 DB_DNODE_ENTER(dr
->dr_dbuf
);
1319 dn
= DB_DNODE(dr
->dr_dbuf
);
1321 if (dn
->dn_bonuslen
== 0 &&
1322 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
1323 data
= dr
->dt
.dl
.dr_data
->b_data
;
1325 data
= dr
->dt
.dl
.dr_data
;
1327 DB_DNODE_EXIT(dr
->dr_dbuf
);
1334 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
1336 objset_t
*os
= dn
->dn_objset
;
1338 dmu_buf_impl_t
*db
= NULL
;
1339 uint64_t *user
= NULL
;
1340 uint64_t *group
= NULL
;
1341 int flags
= dn
->dn_id_flags
;
1343 boolean_t have_spill
= B_FALSE
;
1345 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
1348 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
1349 DN_ID_CHKED_SPILL
)))
1352 if (before
&& dn
->dn_bonuslen
!= 0)
1353 data
= DN_BONUS(dn
->dn_phys
);
1354 else if (!before
&& dn
->dn_bonuslen
!= 0) {
1357 mutex_enter(&db
->db_mtx
);
1358 data
= dmu_objset_userquota_find_data(db
, tx
);
1360 data
= DN_BONUS(dn
->dn_phys
);
1362 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
1365 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
1366 rf
|= DB_RF_HAVESTRUCT
;
1367 error
= dmu_spill_hold_by_dnode(dn
,
1368 rf
| DB_RF_MUST_SUCCEED
,
1369 FTAG
, (dmu_buf_t
**)&db
);
1371 mutex_enter(&db
->db_mtx
);
1372 data
= (before
) ? db
->db
.db_data
:
1373 dmu_objset_userquota_find_data(db
, tx
);
1374 have_spill
= B_TRUE
;
1376 mutex_enter(&dn
->dn_mtx
);
1377 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1378 mutex_exit(&dn
->dn_mtx
);
1384 user
= &dn
->dn_olduid
;
1385 group
= &dn
->dn_oldgid
;
1387 user
= &dn
->dn_newuid
;
1388 group
= &dn
->dn_newgid
;
1392 * Must always call the callback in case the object
1393 * type has changed and that type isn't an object type to track
1395 error
= used_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
,
1399 * Preserve existing uid/gid when the callback can't determine
1400 * what the new uid/gid are and the callback returned EEXIST.
1401 * The EEXIST error tells us to just use the existing uid/gid.
1402 * If we don't know what the old values are then just assign
1403 * them to 0, since that is a new file being created.
1405 if (!before
&& data
== NULL
&& error
== EEXIST
) {
1406 if (flags
& DN_ID_OLD_EXIST
) {
1407 dn
->dn_newuid
= dn
->dn_olduid
;
1408 dn
->dn_newgid
= dn
->dn_oldgid
;
1417 mutex_exit(&db
->db_mtx
);
1419 mutex_enter(&dn
->dn_mtx
);
1420 if (error
== 0 && before
)
1421 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1422 if (error
== 0 && !before
)
1423 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
1426 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1428 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1430 mutex_exit(&dn
->dn_mtx
);
1432 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
1436 dmu_objset_userspace_present(objset_t
*os
)
1438 return (os
->os_phys
->os_flags
&
1439 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
1443 dmu_objset_userspace_upgrade(objset_t
*os
)
1448 if (dmu_objset_userspace_present(os
))
1450 if (!dmu_objset_userused_enabled(os
))
1451 return (SET_ERROR(ENOTSUP
));
1452 if (dmu_objset_is_snapshot(os
))
1453 return (SET_ERROR(EINVAL
));
1456 * We simply need to mark every object dirty, so that it will be
1457 * synced out and now accounted. If this is called
1458 * concurrently, or if we already did some work before crashing,
1459 * that's fine, since we track each object's accounted state
1463 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
1468 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1469 return (SET_ERROR(EINTR
));
1471 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
1474 tx
= dmu_tx_create(os
);
1475 dmu_tx_hold_bonus(tx
, obj
);
1476 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
1481 dmu_buf_will_dirty(db
, tx
);
1482 dmu_buf_rele(db
, FTAG
);
1486 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
1487 txg_wait_synced(dmu_objset_pool(os
), 0);
1492 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
1493 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1495 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
1496 usedobjsp
, availobjsp
);
1500 dmu_objset_fsid_guid(objset_t
*os
)
1502 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
1506 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
1508 stat
->dds_type
= os
->os_phys
->os_type
;
1509 if (os
->os_dsl_dataset
)
1510 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
1514 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
1516 ASSERT(os
->os_dsl_dataset
||
1517 os
->os_phys
->os_type
== DMU_OST_META
);
1519 if (os
->os_dsl_dataset
!= NULL
)
1520 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
1522 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
1523 os
->os_phys
->os_type
);
1524 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
1525 dmu_objset_userspace_present(os
));
1529 dmu_objset_is_snapshot(objset_t
*os
)
1531 if (os
->os_dsl_dataset
!= NULL
)
1532 return (os
->os_dsl_dataset
->ds_is_snapshot
);
1538 dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
, int maxlen
,
1539 boolean_t
*conflict
)
1541 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1544 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1545 return (SET_ERROR(ENOENT
));
1547 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1548 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, name
, 8, 1, &ignored
,
1549 MT_FIRST
, real
, maxlen
, conflict
));
1553 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
1554 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
1556 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1557 zap_cursor_t cursor
;
1558 zap_attribute_t attr
;
1560 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
1562 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1563 return (SET_ERROR(ENOENT
));
1565 zap_cursor_init_serialized(&cursor
,
1566 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1567 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, *offp
);
1569 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1570 zap_cursor_fini(&cursor
);
1571 return (SET_ERROR(ENOENT
));
1574 if (strlen(attr
.za_name
) + 1 > namelen
) {
1575 zap_cursor_fini(&cursor
);
1576 return (SET_ERROR(ENAMETOOLONG
));
1579 (void) strcpy(name
, attr
.za_name
);
1581 *idp
= attr
.za_first_integer
;
1583 *case_conflict
= attr
.za_normalization_conflict
;
1584 zap_cursor_advance(&cursor
);
1585 *offp
= zap_cursor_serialize(&cursor
);
1586 zap_cursor_fini(&cursor
);
1592 dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *value
)
1594 return (dsl_dataset_snap_lookup(os
->os_dsl_dataset
, name
, value
));
1598 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
1599 uint64_t *idp
, uint64_t *offp
)
1601 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
1602 zap_cursor_t cursor
;
1603 zap_attribute_t attr
;
1605 /* there is no next dir on a snapshot! */
1606 if (os
->os_dsl_dataset
->ds_object
!=
1607 dsl_dir_phys(dd
)->dd_head_dataset_obj
)
1608 return (SET_ERROR(ENOENT
));
1610 zap_cursor_init_serialized(&cursor
,
1611 dd
->dd_pool
->dp_meta_objset
,
1612 dsl_dir_phys(dd
)->dd_child_dir_zapobj
, *offp
);
1614 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1615 zap_cursor_fini(&cursor
);
1616 return (SET_ERROR(ENOENT
));
1619 if (strlen(attr
.za_name
) + 1 > namelen
) {
1620 zap_cursor_fini(&cursor
);
1621 return (SET_ERROR(ENAMETOOLONG
));
1624 (void) strcpy(name
, attr
.za_name
);
1626 *idp
= attr
.za_first_integer
;
1627 zap_cursor_advance(&cursor
);
1628 *offp
= zap_cursor_serialize(&cursor
);
1629 zap_cursor_fini(&cursor
);
1634 typedef struct dmu_objset_find_ctx
{
1638 int (*dc_func
)(dsl_pool_t
*, dsl_dataset_t
*, void *);
1641 kmutex_t
*dc_error_lock
;
1643 } dmu_objset_find_ctx_t
;
1646 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t
*dcp
)
1648 dsl_pool_t
*dp
= dcp
->dc_dp
;
1649 dmu_objset_find_ctx_t
*child_dcp
;
1653 zap_attribute_t
*attr
;
1657 /* don't process if there already was an error */
1658 if (*dcp
->dc_error
!= 0)
1661 err
= dsl_dir_hold_obj(dp
, dcp
->dc_ddobj
, NULL
, FTAG
, &dd
);
1665 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1666 if (dd
->dd_myname
[0] == '$') {
1667 dsl_dir_rele(dd
, FTAG
);
1671 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
1672 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1675 * Iterate over all children.
1677 if (dcp
->dc_flags
& DS_FIND_CHILDREN
) {
1678 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1679 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
1680 zap_cursor_retrieve(&zc
, attr
) == 0;
1681 (void) zap_cursor_advance(&zc
)) {
1682 ASSERT3U(attr
->za_integer_length
, ==,
1684 ASSERT3U(attr
->za_num_integers
, ==, 1);
1686 child_dcp
= kmem_alloc(sizeof (*child_dcp
), KM_SLEEP
);
1688 child_dcp
->dc_ddobj
= attr
->za_first_integer
;
1689 if (dcp
->dc_tq
!= NULL
)
1690 (void) taskq_dispatch(dcp
->dc_tq
,
1691 dmu_objset_find_dp_cb
, child_dcp
, TQ_SLEEP
);
1693 dmu_objset_find_dp_impl(child_dcp
);
1695 zap_cursor_fini(&zc
);
1699 * Iterate over all snapshots.
1701 if (dcp
->dc_flags
& DS_FIND_SNAPSHOTS
) {
1703 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1708 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
1709 dsl_dataset_rele(ds
, FTAG
);
1711 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1712 zap_cursor_retrieve(&zc
, attr
) == 0;
1713 (void) zap_cursor_advance(&zc
)) {
1714 ASSERT3U(attr
->za_integer_length
, ==,
1716 ASSERT3U(attr
->za_num_integers
, ==, 1);
1718 err
= dsl_dataset_hold_obj(dp
,
1719 attr
->za_first_integer
, FTAG
, &ds
);
1722 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
1723 dsl_dataset_rele(ds
, FTAG
);
1727 zap_cursor_fini(&zc
);
1731 dsl_dir_rele(dd
, FTAG
);
1732 kmem_free(attr
, sizeof (zap_attribute_t
));
1740 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1743 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
1744 dsl_dataset_rele(ds
, FTAG
);
1748 mutex_enter(dcp
->dc_error_lock
);
1749 /* only keep first error */
1750 if (*dcp
->dc_error
== 0)
1751 *dcp
->dc_error
= err
;
1752 mutex_exit(dcp
->dc_error_lock
);
1755 kmem_free(dcp
, sizeof (*dcp
));
1759 dmu_objset_find_dp_cb(void *arg
)
1761 dmu_objset_find_ctx_t
*dcp
= arg
;
1762 dsl_pool_t
*dp
= dcp
->dc_dp
;
1765 * We need to get a pool_config_lock here, as there are several
1766 * asssert(pool_config_held) down the stack. Getting a lock via
1767 * dsl_pool_config_enter is risky, as it might be stalled by a
1768 * pending writer. This would deadlock, as the write lock can
1769 * only be granted when our parent thread gives up the lock.
1770 * The _prio interface gives us priority over a pending writer.
1772 dsl_pool_config_enter_prio(dp
, FTAG
);
1774 dmu_objset_find_dp_impl(dcp
);
1776 dsl_pool_config_exit(dp
, FTAG
);
1780 * Find objsets under and including ddobj, call func(ds) on each.
1781 * The order for the enumeration is completely undefined.
1782 * func is called with dsl_pool_config held.
1785 dmu_objset_find_dp(dsl_pool_t
*dp
, uint64_t ddobj
,
1786 int func(dsl_pool_t
*, dsl_dataset_t
*, void *), void *arg
, int flags
)
1791 dmu_objset_find_ctx_t
*dcp
;
1794 mutex_init(&err_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1795 dcp
= kmem_alloc(sizeof (*dcp
), KM_SLEEP
);
1798 dcp
->dc_ddobj
= ddobj
;
1799 dcp
->dc_func
= func
;
1801 dcp
->dc_flags
= flags
;
1802 dcp
->dc_error_lock
= &err_lock
;
1803 dcp
->dc_error
= &error
;
1805 if ((flags
& DS_FIND_SERIALIZE
) || dsl_pool_config_held_writer(dp
)) {
1807 * In case a write lock is held we can't make use of
1808 * parallelism, as down the stack of the worker threads
1809 * the lock is asserted via dsl_pool_config_held.
1810 * In case of a read lock this is solved by getting a read
1811 * lock in each worker thread, which isn't possible in case
1812 * of a writer lock. So we fall back to the synchronous path
1814 * In the future it might be possible to get some magic into
1815 * dsl_pool_config_held in a way that it returns true for
1816 * the worker threads so that a single lock held from this
1817 * thread suffices. For now, stay single threaded.
1819 dmu_objset_find_dp_impl(dcp
);
1824 ntasks
= dmu_find_threads
;
1826 ntasks
= vdev_count_leaves(dp
->dp_spa
) * 4;
1827 tq
= taskq_create("dmu_objset_find", ntasks
, maxclsyspri
, ntasks
,
1830 kmem_free(dcp
, sizeof (*dcp
));
1831 return (SET_ERROR(ENOMEM
));
1835 /* dcp will be freed by task */
1836 (void) taskq_dispatch(tq
, dmu_objset_find_dp_cb
, dcp
, TQ_SLEEP
);
1839 * PORTING: this code relies on the property of taskq_wait to wait
1840 * until no more tasks are queued and no more tasks are active. As
1841 * we always queue new tasks from within other tasks, task_wait
1842 * reliably waits for the full recursion to finish, even though we
1843 * enqueue new tasks after taskq_wait has been called.
1844 * On platforms other than illumos, taskq_wait may not have this
1849 mutex_destroy(&err_lock
);
1855 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1856 * The dp_config_rwlock must not be held when this is called, and it
1857 * will not be held when the callback is called.
1858 * Therefore this function should only be used when the pool is not changing
1859 * (e.g. in syncing context), or the callback can deal with the possible races.
1862 dmu_objset_find_impl(spa_t
*spa
, const char *name
,
1863 int func(const char *, void *), void *arg
, int flags
)
1866 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1869 zap_attribute_t
*attr
;
1874 dsl_pool_config_enter(dp
, FTAG
);
1876 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, NULL
);
1878 dsl_pool_config_exit(dp
, FTAG
);
1882 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1883 if (dd
->dd_myname
[0] == '$') {
1884 dsl_dir_rele(dd
, FTAG
);
1885 dsl_pool_config_exit(dp
, FTAG
);
1889 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
1890 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1893 * Iterate over all children.
1895 if (flags
& DS_FIND_CHILDREN
) {
1896 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1897 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
1898 zap_cursor_retrieve(&zc
, attr
) == 0;
1899 (void) zap_cursor_advance(&zc
)) {
1900 ASSERT3U(attr
->za_integer_length
, ==,
1902 ASSERT3U(attr
->za_num_integers
, ==, 1);
1904 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
1905 dsl_pool_config_exit(dp
, FTAG
);
1906 err
= dmu_objset_find_impl(spa
, child
,
1908 dsl_pool_config_enter(dp
, FTAG
);
1913 zap_cursor_fini(&zc
);
1916 dsl_dir_rele(dd
, FTAG
);
1917 dsl_pool_config_exit(dp
, FTAG
);
1918 kmem_free(attr
, sizeof (zap_attribute_t
));
1924 * Iterate over all snapshots.
1926 if (flags
& DS_FIND_SNAPSHOTS
) {
1927 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1932 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
1933 dsl_dataset_rele(ds
, FTAG
);
1935 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1936 zap_cursor_retrieve(&zc
, attr
) == 0;
1937 (void) zap_cursor_advance(&zc
)) {
1938 ASSERT3U(attr
->za_integer_length
, ==,
1940 ASSERT3U(attr
->za_num_integers
, ==, 1);
1942 child
= kmem_asprintf("%s@%s",
1943 name
, attr
->za_name
);
1944 dsl_pool_config_exit(dp
, FTAG
);
1945 err
= func(child
, arg
);
1946 dsl_pool_config_enter(dp
, FTAG
);
1951 zap_cursor_fini(&zc
);
1955 dsl_dir_rele(dd
, FTAG
);
1956 kmem_free(attr
, sizeof (zap_attribute_t
));
1957 dsl_pool_config_exit(dp
, FTAG
);
1962 /* Apply to self. */
1963 return (func(name
, arg
));
1967 * See comment above dmu_objset_find_impl().
1970 dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
1976 error
= spa_open(name
, &spa
, FTAG
);
1979 error
= dmu_objset_find_impl(spa
, name
, func
, arg
, flags
);
1980 spa_close(spa
, FTAG
);
1985 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
1987 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
1988 os
->os_user_ptr
= user_ptr
;
1992 dmu_objset_get_user(objset_t
*os
)
1994 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
1995 return (os
->os_user_ptr
);
1999 * Determine name of filesystem, given name of snapshot.
2000 * buf must be at least MAXNAMELEN bytes
2003 dmu_fsname(const char *snapname
, char *buf
)
2005 char *atp
= strchr(snapname
, '@');
2007 return (SET_ERROR(EINVAL
));
2008 if (atp
- snapname
>= MAXNAMELEN
)
2009 return (SET_ERROR(ENAMETOOLONG
));
2010 (void) strlcpy(buf
, snapname
, atp
- snapname
+ 1);
2014 #if defined(_KERNEL) && defined(HAVE_SPL)
2015 EXPORT_SYMBOL(dmu_objset_zil
);
2016 EXPORT_SYMBOL(dmu_objset_pool
);
2017 EXPORT_SYMBOL(dmu_objset_ds
);
2018 EXPORT_SYMBOL(dmu_objset_type
);
2019 EXPORT_SYMBOL(dmu_objset_name
);
2020 EXPORT_SYMBOL(dmu_objset_hold
);
2021 EXPORT_SYMBOL(dmu_objset_own
);
2022 EXPORT_SYMBOL(dmu_objset_rele
);
2023 EXPORT_SYMBOL(dmu_objset_disown
);
2024 EXPORT_SYMBOL(dmu_objset_from_ds
);
2025 EXPORT_SYMBOL(dmu_objset_create
);
2026 EXPORT_SYMBOL(dmu_objset_clone
);
2027 EXPORT_SYMBOL(dmu_objset_stats
);
2028 EXPORT_SYMBOL(dmu_objset_fast_stat
);
2029 EXPORT_SYMBOL(dmu_objset_spa
);
2030 EXPORT_SYMBOL(dmu_objset_space
);
2031 EXPORT_SYMBOL(dmu_objset_fsid_guid
);
2032 EXPORT_SYMBOL(dmu_objset_find
);
2033 EXPORT_SYMBOL(dmu_objset_byteswap
);
2034 EXPORT_SYMBOL(dmu_objset_evict_dbufs
);
2035 EXPORT_SYMBOL(dmu_objset_snap_cmtime
);
2037 EXPORT_SYMBOL(dmu_objset_sync
);
2038 EXPORT_SYMBOL(dmu_objset_is_dirty
);
2039 EXPORT_SYMBOL(dmu_objset_create_impl
);
2040 EXPORT_SYMBOL(dmu_objset_open_impl
);
2041 EXPORT_SYMBOL(dmu_objset_evict
);
2042 EXPORT_SYMBOL(dmu_objset_register_type
);
2043 EXPORT_SYMBOL(dmu_objset_do_userquota_updates
);
2044 EXPORT_SYMBOL(dmu_objset_userquota_get_ids
);
2045 EXPORT_SYMBOL(dmu_objset_userused_enabled
);
2046 EXPORT_SYMBOL(dmu_objset_userspace_upgrade
);
2047 EXPORT_SYMBOL(dmu_objset_userspace_present
);