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, 2016 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>
55 #include <sys/policy.h>
58 * Needed to close a window in dnode_move() that allows the objset to be freed
59 * before it can be safely accessed.
64 * Tunable to overwrite the maximum number of threads for the parallization
65 * of dmu_objset_find_dp, needed to speed up the import of pools with many
67 * Default is 4 times the number of leaf vdevs.
69 int dmu_find_threads
= 0;
71 static void dmu_objset_find_dp_cb(void *arg
);
76 rw_init(&os_lock
, NULL
, RW_DEFAULT
, NULL
);
86 dmu_objset_spa(objset_t
*os
)
92 dmu_objset_zil(objset_t
*os
)
98 dmu_objset_pool(objset_t
*os
)
102 if ((ds
= os
->os_dsl_dataset
) != NULL
&& ds
->ds_dir
)
103 return (ds
->ds_dir
->dd_pool
);
105 return (spa_get_dsl(os
->os_spa
));
109 dmu_objset_ds(objset_t
*os
)
111 return (os
->os_dsl_dataset
);
115 dmu_objset_type(objset_t
*os
)
117 return (os
->os_phys
->os_type
);
121 dmu_objset_name(objset_t
*os
, char *buf
)
123 dsl_dataset_name(os
->os_dsl_dataset
, buf
);
127 dmu_objset_id(objset_t
*os
)
129 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
131 return (ds
? ds
->ds_object
: 0);
135 dmu_objset_syncprop(objset_t
*os
)
137 return (os
->os_sync
);
141 dmu_objset_logbias(objset_t
*os
)
143 return (os
->os_logbias
);
147 checksum_changed_cb(void *arg
, uint64_t newval
)
152 * Inheritance should have been done by now.
154 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
156 os
->os_checksum
= zio_checksum_select(newval
, ZIO_CHECKSUM_ON_VALUE
);
160 compression_changed_cb(void *arg
, uint64_t newval
)
165 * Inheritance and range checking should have been done by now.
167 ASSERT(newval
!= ZIO_COMPRESS_INHERIT
);
169 os
->os_compress
= zio_compress_select(os
->os_spa
, newval
,
174 copies_changed_cb(void *arg
, uint64_t newval
)
179 * Inheritance and range checking should have been done by now.
182 ASSERT(newval
<= spa_max_replication(os
->os_spa
));
184 os
->os_copies
= newval
;
188 dedup_changed_cb(void *arg
, uint64_t newval
)
191 spa_t
*spa
= os
->os_spa
;
192 enum zio_checksum checksum
;
195 * Inheritance should have been done by now.
197 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
199 checksum
= zio_checksum_dedup_select(spa
, newval
, ZIO_CHECKSUM_OFF
);
201 os
->os_dedup_checksum
= checksum
& ZIO_CHECKSUM_MASK
;
202 os
->os_dedup_verify
= !!(checksum
& ZIO_CHECKSUM_VERIFY
);
206 primary_cache_changed_cb(void *arg
, uint64_t newval
)
211 * Inheritance and range checking should have been done by now.
213 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
214 newval
== ZFS_CACHE_METADATA
);
216 os
->os_primary_cache
= newval
;
220 secondary_cache_changed_cb(void *arg
, uint64_t newval
)
225 * Inheritance and range checking should have been done by now.
227 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
228 newval
== ZFS_CACHE_METADATA
);
230 os
->os_secondary_cache
= newval
;
234 sync_changed_cb(void *arg
, uint64_t newval
)
239 * Inheritance and range checking should have been done by now.
241 ASSERT(newval
== ZFS_SYNC_STANDARD
|| newval
== ZFS_SYNC_ALWAYS
||
242 newval
== ZFS_SYNC_DISABLED
);
244 os
->os_sync
= newval
;
246 zil_set_sync(os
->os_zil
, newval
);
250 redundant_metadata_changed_cb(void *arg
, uint64_t newval
)
255 * Inheritance and range checking should have been done by now.
257 ASSERT(newval
== ZFS_REDUNDANT_METADATA_ALL
||
258 newval
== ZFS_REDUNDANT_METADATA_MOST
);
260 os
->os_redundant_metadata
= newval
;
264 logbias_changed_cb(void *arg
, uint64_t newval
)
268 ASSERT(newval
== ZFS_LOGBIAS_LATENCY
||
269 newval
== ZFS_LOGBIAS_THROUGHPUT
);
270 os
->os_logbias
= newval
;
272 zil_set_logbias(os
->os_zil
, newval
);
276 recordsize_changed_cb(void *arg
, uint64_t newval
)
280 os
->os_recordsize
= newval
;
284 dmu_objset_byteswap(void *buf
, size_t size
)
286 objset_phys_t
*osp
= buf
;
288 ASSERT(size
== OBJSET_OLD_PHYS_SIZE
|| size
== sizeof (objset_phys_t
));
289 dnode_byteswap(&osp
->os_meta_dnode
);
290 byteswap_uint64_array(&osp
->os_zil_header
, sizeof (zil_header_t
));
291 osp
->os_type
= BSWAP_64(osp
->os_type
);
292 osp
->os_flags
= BSWAP_64(osp
->os_flags
);
293 if (size
== sizeof (objset_phys_t
)) {
294 dnode_byteswap(&osp
->os_userused_dnode
);
295 dnode_byteswap(&osp
->os_groupused_dnode
);
300 dmu_objset_open_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
306 ASSERT(ds
== NULL
|| MUTEX_HELD(&ds
->ds_opening_lock
));
308 os
= kmem_zalloc(sizeof (objset_t
), KM_SLEEP
);
309 os
->os_dsl_dataset
= ds
;
312 if (!BP_IS_HOLE(os
->os_rootbp
)) {
313 arc_flags_t aflags
= ARC_FLAG_WAIT
;
315 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
316 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
318 if (DMU_OS_IS_L2CACHEABLE(os
))
319 aflags
|= ARC_FLAG_L2CACHE
;
320 if (DMU_OS_IS_L2COMPRESSIBLE(os
))
321 aflags
|= ARC_FLAG_L2COMPRESS
;
323 dprintf_bp(os
->os_rootbp
, "reading %s", "");
324 err
= arc_read(NULL
, spa
, os
->os_rootbp
,
325 arc_getbuf_func
, &os
->os_phys_buf
,
326 ZIO_PRIORITY_SYNC_READ
, ZIO_FLAG_CANFAIL
, &aflags
, &zb
);
328 kmem_free(os
, sizeof (objset_t
));
329 /* convert checksum errors into IO errors */
331 err
= SET_ERROR(EIO
);
335 /* Increase the blocksize if we are permitted. */
336 if (spa_version(spa
) >= SPA_VERSION_USERSPACE
&&
337 arc_buf_size(os
->os_phys_buf
) < sizeof (objset_phys_t
)) {
338 arc_buf_t
*buf
= arc_buf_alloc(spa
,
339 sizeof (objset_phys_t
), &os
->os_phys_buf
,
341 bzero(buf
->b_data
, sizeof (objset_phys_t
));
342 bcopy(os
->os_phys_buf
->b_data
, buf
->b_data
,
343 arc_buf_size(os
->os_phys_buf
));
344 (void) arc_buf_remove_ref(os
->os_phys_buf
,
346 os
->os_phys_buf
= buf
;
349 os
->os_phys
= os
->os_phys_buf
->b_data
;
350 os
->os_flags
= os
->os_phys
->os_flags
;
352 int size
= spa_version(spa
) >= SPA_VERSION_USERSPACE
?
353 sizeof (objset_phys_t
) : OBJSET_OLD_PHYS_SIZE
;
354 os
->os_phys_buf
= arc_buf_alloc(spa
, size
,
355 &os
->os_phys_buf
, ARC_BUFC_METADATA
);
356 os
->os_phys
= os
->os_phys_buf
->b_data
;
357 bzero(os
->os_phys
, size
);
361 * Note: the changed_cb will be called once before the register
362 * func returns, thus changing the checksum/compression from the
363 * default (fletcher2/off). Snapshots don't need to know about
364 * checksum/compression/copies.
367 err
= dsl_prop_register(ds
,
368 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
369 primary_cache_changed_cb
, os
);
371 err
= dsl_prop_register(ds
,
372 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
373 secondary_cache_changed_cb
, os
);
375 if (!ds
->ds_is_snapshot
) {
377 err
= dsl_prop_register(ds
,
378 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
379 checksum_changed_cb
, os
);
382 err
= dsl_prop_register(ds
,
383 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
384 compression_changed_cb
, os
);
387 err
= dsl_prop_register(ds
,
388 zfs_prop_to_name(ZFS_PROP_COPIES
),
389 copies_changed_cb
, os
);
392 err
= dsl_prop_register(ds
,
393 zfs_prop_to_name(ZFS_PROP_DEDUP
),
394 dedup_changed_cb
, os
);
397 err
= dsl_prop_register(ds
,
398 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
399 logbias_changed_cb
, os
);
402 err
= dsl_prop_register(ds
,
403 zfs_prop_to_name(ZFS_PROP_SYNC
),
404 sync_changed_cb
, os
);
407 err
= dsl_prop_register(ds
,
409 ZFS_PROP_REDUNDANT_METADATA
),
410 redundant_metadata_changed_cb
, os
);
413 err
= dsl_prop_register(ds
,
414 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
415 recordsize_changed_cb
, os
);
419 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
,
421 kmem_free(os
, sizeof (objset_t
));
425 /* It's the meta-objset. */
426 os
->os_checksum
= ZIO_CHECKSUM_FLETCHER_4
;
427 os
->os_compress
= ZIO_COMPRESS_ON
;
428 os
->os_copies
= spa_max_replication(spa
);
429 os
->os_dedup_checksum
= ZIO_CHECKSUM_OFF
;
430 os
->os_dedup_verify
= B_FALSE
;
431 os
->os_logbias
= ZFS_LOGBIAS_LATENCY
;
432 os
->os_sync
= ZFS_SYNC_STANDARD
;
433 os
->os_primary_cache
= ZFS_CACHE_ALL
;
434 os
->os_secondary_cache
= ZFS_CACHE_ALL
;
437 if (ds
== NULL
|| !ds
->ds_is_snapshot
)
438 os
->os_zil_header
= os
->os_phys
->os_zil_header
;
439 os
->os_zil
= zil_alloc(os
, &os
->os_zil_header
);
441 for (i
= 0; i
< TXG_SIZE
; i
++) {
442 list_create(&os
->os_dirty_dnodes
[i
], sizeof (dnode_t
),
443 offsetof(dnode_t
, dn_dirty_link
[i
]));
444 list_create(&os
->os_free_dnodes
[i
], sizeof (dnode_t
),
445 offsetof(dnode_t
, dn_dirty_link
[i
]));
447 list_create(&os
->os_dnodes
, sizeof (dnode_t
),
448 offsetof(dnode_t
, dn_link
));
449 list_create(&os
->os_downgraded_dbufs
, sizeof (dmu_buf_impl_t
),
450 offsetof(dmu_buf_impl_t
, db_link
));
452 list_link_init(&os
->os_evicting_node
);
454 mutex_init(&os
->os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
455 mutex_init(&os
->os_obj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
456 mutex_init(&os
->os_user_ptr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
458 dnode_special_open(os
, &os
->os_phys
->os_meta_dnode
,
459 DMU_META_DNODE_OBJECT
, &os
->os_meta_dnode
);
460 if (arc_buf_size(os
->os_phys_buf
) >= sizeof (objset_phys_t
)) {
461 dnode_special_open(os
, &os
->os_phys
->os_userused_dnode
,
462 DMU_USERUSED_OBJECT
, &os
->os_userused_dnode
);
463 dnode_special_open(os
, &os
->os_phys
->os_groupused_dnode
,
464 DMU_GROUPUSED_OBJECT
, &os
->os_groupused_dnode
);
472 dmu_objset_from_ds(dsl_dataset_t
*ds
, objset_t
**osp
)
476 mutex_enter(&ds
->ds_opening_lock
);
477 if (ds
->ds_objset
== NULL
) {
479 err
= dmu_objset_open_impl(dsl_dataset_get_spa(ds
),
480 ds
, dsl_dataset_get_blkptr(ds
), &os
);
483 mutex_enter(&ds
->ds_lock
);
484 ASSERT(ds
->ds_objset
== NULL
);
486 mutex_exit(&ds
->ds_lock
);
489 *osp
= ds
->ds_objset
;
490 mutex_exit(&ds
->ds_opening_lock
);
495 * Holds the pool while the objset is held. Therefore only one objset
496 * can be held at a time.
499 dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
)
505 err
= dsl_pool_hold(name
, tag
, &dp
);
508 err
= dsl_dataset_hold(dp
, name
, tag
, &ds
);
510 dsl_pool_rele(dp
, tag
);
514 err
= dmu_objset_from_ds(ds
, osp
);
516 dsl_dataset_rele(ds
, tag
);
517 dsl_pool_rele(dp
, tag
);
524 dmu_objset_own_impl(dsl_dataset_t
*ds
, dmu_objset_type_t type
,
525 boolean_t readonly
, void *tag
, objset_t
**osp
)
529 err
= dmu_objset_from_ds(ds
, osp
);
531 dsl_dataset_disown(ds
, tag
);
532 } else if (type
!= DMU_OST_ANY
&& type
!= (*osp
)->os_phys
->os_type
) {
533 dsl_dataset_disown(ds
, tag
);
534 return (SET_ERROR(EINVAL
));
535 } else if (!readonly
&& dsl_dataset_is_snapshot(ds
)) {
536 dsl_dataset_disown(ds
, tag
);
537 return (SET_ERROR(EROFS
));
543 * dsl_pool must not be held when this is called.
544 * Upon successful return, there will be a longhold on the dataset,
545 * and the dsl_pool will not be held.
548 dmu_objset_own(const char *name
, dmu_objset_type_t type
,
549 boolean_t readonly
, void *tag
, objset_t
**osp
)
555 err
= dsl_pool_hold(name
, FTAG
, &dp
);
558 err
= dsl_dataset_own(dp
, name
, tag
, &ds
);
560 dsl_pool_rele(dp
, FTAG
);
563 err
= dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
);
564 dsl_pool_rele(dp
, FTAG
);
570 dmu_objset_own_obj(dsl_pool_t
*dp
, uint64_t obj
, dmu_objset_type_t type
,
571 boolean_t readonly
, void *tag
, objset_t
**osp
)
576 err
= dsl_dataset_own_obj(dp
, obj
, tag
, &ds
);
580 return (dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
));
584 dmu_objset_rele(objset_t
*os
, void *tag
)
586 dsl_pool_t
*dp
= dmu_objset_pool(os
);
587 dsl_dataset_rele(os
->os_dsl_dataset
, tag
);
588 dsl_pool_rele(dp
, tag
);
592 * When we are called, os MUST refer to an objset associated with a dataset
593 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
594 * == tag. We will then release and reacquire ownership of the dataset while
595 * holding the pool config_rwlock to avoid intervening namespace or ownership
598 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
599 * release the hold on its dataset and acquire a new one on the dataset of the
600 * same name so that it can be partially torn down and reconstructed.
603 dmu_objset_refresh_ownership(objset_t
*os
, void *tag
)
606 dsl_dataset_t
*ds
, *newds
;
607 char name
[MAXNAMELEN
];
609 ds
= os
->os_dsl_dataset
;
610 VERIFY3P(ds
, !=, NULL
);
611 VERIFY3P(ds
->ds_owner
, ==, tag
);
612 VERIFY(dsl_dataset_long_held(ds
));
614 dsl_dataset_name(ds
, name
);
615 dp
= dmu_objset_pool(os
);
616 dsl_pool_config_enter(dp
, FTAG
);
617 dmu_objset_disown(os
, tag
);
618 VERIFY0(dsl_dataset_own(dp
, name
, tag
, &newds
));
619 VERIFY3P(newds
, ==, os
->os_dsl_dataset
);
620 dsl_pool_config_exit(dp
, FTAG
);
624 dmu_objset_disown(objset_t
*os
, void *tag
)
626 dsl_dataset_disown(os
->os_dsl_dataset
, tag
);
630 dmu_objset_evict_dbufs(objset_t
*os
)
635 dn_marker
= kmem_alloc(sizeof (dnode_t
), KM_SLEEP
);
637 mutex_enter(&os
->os_lock
);
638 dn
= list_head(&os
->os_dnodes
);
641 * Skip dnodes without holds. We have to do this dance
642 * because dnode_add_ref() only works if there is already a
643 * hold. If the dnode has no holds, then it has no dbufs.
645 if (dnode_add_ref(dn
, FTAG
)) {
646 list_insert_after(&os
->os_dnodes
, dn
, dn_marker
);
647 mutex_exit(&os
->os_lock
);
649 dnode_evict_dbufs(dn
);
650 dnode_rele(dn
, FTAG
);
652 mutex_enter(&os
->os_lock
);
653 dn
= list_next(&os
->os_dnodes
, dn_marker
);
654 list_remove(&os
->os_dnodes
, dn_marker
);
656 dn
= list_next(&os
->os_dnodes
, dn
);
659 mutex_exit(&os
->os_lock
);
661 kmem_free(dn_marker
, sizeof (dnode_t
));
663 if (DMU_USERUSED_DNODE(os
) != NULL
) {
664 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os
));
665 dnode_evict_dbufs(DMU_USERUSED_DNODE(os
));
667 dnode_evict_dbufs(DMU_META_DNODE(os
));
671 * Objset eviction processing is split into into two pieces.
672 * The first marks the objset as evicting, evicts any dbufs that
673 * have a refcount of zero, and then queues up the objset for the
674 * second phase of eviction. Once os->os_dnodes has been cleared by
675 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
676 * The second phase closes the special dnodes, dequeues the objset from
677 * the list of those undergoing eviction, and finally frees the objset.
679 * NOTE: Due to asynchronous eviction processing (invocation of
680 * dnode_buf_pageout()), it is possible for the meta dnode for the
681 * objset to have no holds even though os->os_dnodes is not empty.
684 dmu_objset_evict(objset_t
*os
)
688 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
690 for (t
= 0; t
< TXG_SIZE
; t
++)
691 ASSERT(!dmu_objset_is_dirty(os
, t
));
694 dsl_prop_unregister_all(ds
, os
);
699 dmu_objset_evict_dbufs(os
);
701 mutex_enter(&os
->os_lock
);
702 spa_evicting_os_register(os
->os_spa
, os
);
703 if (list_is_empty(&os
->os_dnodes
)) {
704 mutex_exit(&os
->os_lock
);
705 dmu_objset_evict_done(os
);
707 mutex_exit(&os
->os_lock
);
712 dmu_objset_evict_done(objset_t
*os
)
714 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
716 dnode_special_close(&os
->os_meta_dnode
);
717 if (DMU_USERUSED_DNODE(os
)) {
718 dnode_special_close(&os
->os_userused_dnode
);
719 dnode_special_close(&os
->os_groupused_dnode
);
721 zil_free(os
->os_zil
);
723 VERIFY(arc_buf_remove_ref(os
->os_phys_buf
, &os
->os_phys_buf
));
726 * This is a barrier to prevent the objset from going away in
727 * dnode_move() until we can safely ensure that the objset is still in
728 * use. We consider the objset valid before the barrier and invalid
731 rw_enter(&os_lock
, RW_READER
);
734 mutex_destroy(&os
->os_lock
);
735 mutex_destroy(&os
->os_obj_lock
);
736 mutex_destroy(&os
->os_user_ptr_lock
);
737 spa_evicting_os_deregister(os
->os_spa
, os
);
738 kmem_free(os
, sizeof (objset_t
));
742 dmu_objset_snap_cmtime(objset_t
*os
)
744 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
747 /* called from dsl for meta-objset */
749 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
750 dmu_objset_type_t type
, dmu_tx_t
*tx
)
755 ASSERT(dmu_tx_is_syncing(tx
));
758 VERIFY0(dmu_objset_from_ds(ds
, &os
));
760 VERIFY0(dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
762 mdn
= DMU_META_DNODE(os
);
764 dnode_allocate(mdn
, DMU_OT_DNODE
, 1 << DNODE_BLOCK_SHIFT
,
765 DN_MAX_INDBLKSHIFT
, DMU_OT_NONE
, 0, tx
);
768 * We don't want to have to increase the meta-dnode's nlevels
769 * later, because then we could do it in quescing context while
770 * we are also accessing it in open context.
772 * This precaution is not necessary for the MOS (ds == NULL),
773 * because the MOS is only updated in syncing context.
774 * This is most fortunate: the MOS is the only objset that
775 * needs to be synced multiple times as spa_sync() iterates
776 * to convergence, so minimizing its dn_nlevels matters.
782 * Determine the number of levels necessary for the meta-dnode
783 * to contain DN_MAX_OBJECT dnodes.
785 while ((uint64_t)mdn
->dn_nblkptr
<< (mdn
->dn_datablkshift
+
786 (levels
- 1) * (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
787 DN_MAX_OBJECT
* sizeof (dnode_phys_t
))
790 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
791 mdn
->dn_nlevels
= levels
;
794 ASSERT(type
!= DMU_OST_NONE
);
795 ASSERT(type
!= DMU_OST_ANY
);
796 ASSERT(type
< DMU_OST_NUMTYPES
);
797 os
->os_phys
->os_type
= type
;
798 if (dmu_objset_userused_enabled(os
)) {
799 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
800 os
->os_flags
= os
->os_phys
->os_flags
;
803 dsl_dataset_dirty(ds
, tx
);
808 typedef struct dmu_objset_create_arg
{
809 const char *doca_name
;
811 void (*doca_userfunc
)(objset_t
*os
, void *arg
,
812 cred_t
*cr
, dmu_tx_t
*tx
);
814 dmu_objset_type_t doca_type
;
816 } dmu_objset_create_arg_t
;
820 dmu_objset_create_check(void *arg
, dmu_tx_t
*tx
)
822 dmu_objset_create_arg_t
*doca
= arg
;
823 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
828 if (strchr(doca
->doca_name
, '@') != NULL
)
829 return (SET_ERROR(EINVAL
));
831 error
= dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
);
835 dsl_dir_rele(pdd
, FTAG
);
836 return (SET_ERROR(EEXIST
));
838 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
840 dsl_dir_rele(pdd
, FTAG
);
846 dmu_objset_create_sync(void *arg
, dmu_tx_t
*tx
)
848 dmu_objset_create_arg_t
*doca
= arg
;
849 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
857 VERIFY0(dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
));
859 obj
= dsl_dataset_create_sync(pdd
, tail
, NULL
, doca
->doca_flags
,
860 doca
->doca_cred
, tx
);
862 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
863 bp
= dsl_dataset_get_blkptr(ds
);
864 os
= dmu_objset_create_impl(pdd
->dd_pool
->dp_spa
,
865 ds
, bp
, doca
->doca_type
, tx
);
867 if (doca
->doca_userfunc
!= NULL
) {
868 doca
->doca_userfunc(os
, doca
->doca_userarg
,
869 doca
->doca_cred
, tx
);
872 spa_history_log_internal_ds(ds
, "create", tx
, "");
873 zvol_create_minors(dp
->dp_spa
, doca
->doca_name
, B_TRUE
);
875 dsl_dataset_rele(ds
, FTAG
);
876 dsl_dir_rele(pdd
, FTAG
);
880 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
881 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
)
883 dmu_objset_create_arg_t doca
;
885 doca
.doca_name
= name
;
886 doca
.doca_cred
= CRED();
887 doca
.doca_flags
= flags
;
888 doca
.doca_userfunc
= func
;
889 doca
.doca_userarg
= arg
;
890 doca
.doca_type
= type
;
892 return (dsl_sync_task(name
,
893 dmu_objset_create_check
, dmu_objset_create_sync
, &doca
,
894 5, ZFS_SPACE_CHECK_NORMAL
));
897 typedef struct dmu_objset_clone_arg
{
898 const char *doca_clone
;
899 const char *doca_origin
;
901 } dmu_objset_clone_arg_t
;
905 dmu_objset_clone_check(void *arg
, dmu_tx_t
*tx
)
907 dmu_objset_clone_arg_t
*doca
= arg
;
911 dsl_dataset_t
*origin
;
912 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
914 if (strchr(doca
->doca_clone
, '@') != NULL
)
915 return (SET_ERROR(EINVAL
));
917 error
= dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
);
921 dsl_dir_rele(pdd
, FTAG
);
922 return (SET_ERROR(EEXIST
));
925 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
928 dsl_dir_rele(pdd
, FTAG
);
929 return (SET_ERROR(EDQUOT
));
931 dsl_dir_rele(pdd
, FTAG
);
933 error
= dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
);
937 /* You can only clone snapshots, not the head datasets. */
938 if (!origin
->ds_is_snapshot
) {
939 dsl_dataset_rele(origin
, FTAG
);
940 return (SET_ERROR(EINVAL
));
942 dsl_dataset_rele(origin
, FTAG
);
948 dmu_objset_clone_sync(void *arg
, dmu_tx_t
*tx
)
950 dmu_objset_clone_arg_t
*doca
= arg
;
951 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
954 dsl_dataset_t
*origin
, *ds
;
956 char namebuf
[MAXNAMELEN
];
958 VERIFY0(dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
));
959 VERIFY0(dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
));
961 obj
= dsl_dataset_create_sync(pdd
, tail
, origin
, 0,
962 doca
->doca_cred
, tx
);
964 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
965 dsl_dataset_name(origin
, namebuf
);
966 spa_history_log_internal_ds(ds
, "clone", tx
,
967 "origin=%s (%llu)", namebuf
, origin
->ds_object
);
968 zvol_create_minors(dp
->dp_spa
, doca
->doca_clone
, B_TRUE
);
969 dsl_dataset_rele(ds
, FTAG
);
970 dsl_dataset_rele(origin
, FTAG
);
971 dsl_dir_rele(pdd
, FTAG
);
975 dmu_objset_clone(const char *clone
, const char *origin
)
977 dmu_objset_clone_arg_t doca
;
979 doca
.doca_clone
= clone
;
980 doca
.doca_origin
= origin
;
981 doca
.doca_cred
= CRED();
983 return (dsl_sync_task(clone
,
984 dmu_objset_clone_check
, dmu_objset_clone_sync
, &doca
,
985 5, ZFS_SPACE_CHECK_NORMAL
));
989 dmu_objset_snapshot_one(const char *fsname
, const char *snapname
)
992 char *longsnap
= kmem_asprintf("%s@%s", fsname
, snapname
);
993 nvlist_t
*snaps
= fnvlist_alloc();
995 fnvlist_add_boolean(snaps
, longsnap
);
997 err
= dsl_dataset_snapshot(snaps
, NULL
, NULL
);
1003 dmu_objset_sync_dnodes(list_t
*list
, list_t
*newlist
, dmu_tx_t
*tx
)
1007 while ((dn
= list_head(list
))) {
1008 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1009 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1011 * Initialize dn_zio outside dnode_sync() because the
1012 * meta-dnode needs to set it ouside dnode_sync().
1014 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1017 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1018 list_remove(list
, dn
);
1021 (void) dnode_add_ref(dn
, newlist
);
1022 list_insert_tail(newlist
, dn
);
1031 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1035 blkptr_t
*bp
= zio
->io_bp
;
1037 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1039 ASSERT(!BP_IS_EMBEDDED(bp
));
1040 ASSERT3P(bp
, ==, os
->os_rootbp
);
1041 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_OBJSET
);
1042 ASSERT0(BP_GET_LEVEL(bp
));
1045 * Update rootbp fill count: it should be the number of objects
1046 * allocated in the object set (not counting the "special"
1047 * objects that are stored in the objset_phys_t -- the meta
1048 * dnode and user/group accounting objects).
1051 for (i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1052 bp
->blk_fill
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
1057 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1059 blkptr_t
*bp
= zio
->io_bp
;
1060 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1063 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1064 ASSERT(BP_EQUAL(bp
, bp_orig
));
1066 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1067 dmu_tx_t
*tx
= os
->os_synctx
;
1069 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1070 dsl_dataset_block_born(ds
, bp
, tx
);
1074 /* called from dsl */
1076 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1079 zbookmark_phys_t zb
;
1083 list_t
*newlist
= NULL
;
1084 dbuf_dirty_record_t
*dr
;
1086 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", tx
->tx_txg
);
1088 ASSERT(dmu_tx_is_syncing(tx
));
1089 /* XXX the write_done callback should really give us the tx... */
1092 if (os
->os_dsl_dataset
== NULL
) {
1094 * This is the MOS. If we have upgraded,
1095 * spa_max_replication() could change, so reset
1098 os
->os_copies
= spa_max_replication(os
->os_spa
);
1102 * Create the root block IO
1104 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1105 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1106 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1107 arc_release(os
->os_phys_buf
, &os
->os_phys_buf
);
1109 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1111 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1112 os
->os_rootbp
, os
->os_phys_buf
, DMU_OS_IS_L2CACHEABLE(os
),
1113 DMU_OS_IS_L2COMPRESSIBLE(os
),
1114 &zp
, dmu_objset_write_ready
, NULL
, NULL
, dmu_objset_write_done
,
1115 os
, ZIO_PRIORITY_ASYNC_WRITE
, ZIO_FLAG_MUSTSUCCEED
, &zb
);
1118 * Sync special dnodes - the parent IO for the sync is the root block
1120 DMU_META_DNODE(os
)->dn_zio
= zio
;
1121 dnode_sync(DMU_META_DNODE(os
), tx
);
1123 os
->os_phys
->os_flags
= os
->os_flags
;
1125 if (DMU_USERUSED_DNODE(os
) &&
1126 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1127 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1128 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1129 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1130 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1133 txgoff
= tx
->tx_txg
& TXG_MASK
;
1135 if (dmu_objset_userused_enabled(os
)) {
1136 newlist
= &os
->os_synced_dnodes
;
1138 * We must create the list here because it uses the
1139 * dn_dirty_link[] of this txg.
1141 list_create(newlist
, sizeof (dnode_t
),
1142 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1145 dmu_objset_sync_dnodes(&os
->os_free_dnodes
[txgoff
], newlist
, tx
);
1146 dmu_objset_sync_dnodes(&os
->os_dirty_dnodes
[txgoff
], newlist
, tx
);
1148 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1149 while ((dr
= list_head(list
))) {
1150 ASSERT0(dr
->dr_dbuf
->db_level
);
1151 list_remove(list
, dr
);
1153 zio_nowait(dr
->dr_zio
);
1156 * Free intent log blocks up to this tx.
1158 zil_sync(os
->os_zil
, tx
);
1159 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1164 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1166 return (!list_is_empty(&os
->os_dirty_dnodes
[txg
& TXG_MASK
]) ||
1167 !list_is_empty(&os
->os_free_dnodes
[txg
& TXG_MASK
]));
1170 static objset_used_cb_t
*used_cbs
[DMU_OST_NUMTYPES
];
1173 dmu_objset_register_type(dmu_objset_type_t ost
, objset_used_cb_t
*cb
)
1179 dmu_objset_userused_enabled(objset_t
*os
)
1181 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1182 used_cbs
[os
->os_phys
->os_type
] != NULL
&&
1183 DMU_USERUSED_DNODE(os
) != NULL
);
1187 do_userquota_update(objset_t
*os
, uint64_t used
, uint64_t flags
,
1188 uint64_t user
, uint64_t group
, boolean_t subtract
, dmu_tx_t
*tx
)
1190 if ((flags
& DNODE_FLAG_USERUSED_ACCOUNTED
)) {
1191 int64_t delta
= DNODE_SIZE
+ used
;
1194 VERIFY3U(0, ==, zap_increment_int(os
, DMU_USERUSED_OBJECT
,
1196 VERIFY3U(0, ==, zap_increment_int(os
, DMU_GROUPUSED_OBJECT
,
1202 dmu_objset_do_userquota_updates(objset_t
*os
, dmu_tx_t
*tx
)
1205 list_t
*list
= &os
->os_synced_dnodes
;
1207 ASSERT(list_head(list
) == NULL
|| dmu_objset_userused_enabled(os
));
1209 while ((dn
= list_head(list
))) {
1211 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
1212 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
1213 dn
->dn_phys
->dn_flags
&
1214 DNODE_FLAG_USERUSED_ACCOUNTED
);
1216 /* Allocate the user/groupused objects if necessary. */
1217 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
1218 VERIFY(0 == zap_create_claim(os
,
1219 DMU_USERUSED_OBJECT
,
1220 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1221 VERIFY(0 == zap_create_claim(os
,
1222 DMU_GROUPUSED_OBJECT
,
1223 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1227 * We intentionally modify the zap object even if the
1228 * net delta is zero. Otherwise
1229 * the block of the zap obj could be shared between
1230 * datasets but need to be different between them after
1234 flags
= dn
->dn_id_flags
;
1236 if (flags
& DN_ID_OLD_EXIST
) {
1237 do_userquota_update(os
, dn
->dn_oldused
, dn
->dn_oldflags
,
1238 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
, tx
);
1240 if (flags
& DN_ID_NEW_EXIST
) {
1241 do_userquota_update(os
, DN_USED_BYTES(dn
->dn_phys
),
1242 dn
->dn_phys
->dn_flags
, dn
->dn_newuid
,
1243 dn
->dn_newgid
, B_FALSE
, tx
);
1246 mutex_enter(&dn
->dn_mtx
);
1248 dn
->dn_oldflags
= 0;
1249 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
1250 dn
->dn_olduid
= dn
->dn_newuid
;
1251 dn
->dn_oldgid
= dn
->dn_newgid
;
1252 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1253 if (dn
->dn_bonuslen
== 0)
1254 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1256 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1258 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
1259 mutex_exit(&dn
->dn_mtx
);
1261 list_remove(list
, dn
);
1262 dnode_rele(dn
, list
);
1267 * Returns a pointer to data to find uid/gid from
1269 * If a dirty record for transaction group that is syncing can't
1270 * be found then NULL is returned. In the NULL case it is assumed
1271 * the uid/gid aren't changing.
1274 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
1276 dbuf_dirty_record_t
*dr
, **drp
;
1279 if (db
->db_dirtycnt
== 0)
1280 return (db
->db
.db_data
); /* Nothing is changing */
1282 for (drp
= &db
->db_last_dirty
; (dr
= *drp
) != NULL
; drp
= &dr
->dr_next
)
1283 if (dr
->dr_txg
== tx
->tx_txg
)
1291 DB_DNODE_ENTER(dr
->dr_dbuf
);
1292 dn
= DB_DNODE(dr
->dr_dbuf
);
1294 if (dn
->dn_bonuslen
== 0 &&
1295 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
1296 data
= dr
->dt
.dl
.dr_data
->b_data
;
1298 data
= dr
->dt
.dl
.dr_data
;
1300 DB_DNODE_EXIT(dr
->dr_dbuf
);
1307 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
1309 objset_t
*os
= dn
->dn_objset
;
1311 dmu_buf_impl_t
*db
= NULL
;
1312 uint64_t *user
= NULL
;
1313 uint64_t *group
= NULL
;
1314 int flags
= dn
->dn_id_flags
;
1316 boolean_t have_spill
= B_FALSE
;
1318 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
1321 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
1322 DN_ID_CHKED_SPILL
)))
1325 if (before
&& dn
->dn_bonuslen
!= 0)
1326 data
= DN_BONUS(dn
->dn_phys
);
1327 else if (!before
&& dn
->dn_bonuslen
!= 0) {
1330 mutex_enter(&db
->db_mtx
);
1331 data
= dmu_objset_userquota_find_data(db
, tx
);
1333 data
= DN_BONUS(dn
->dn_phys
);
1335 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
1338 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
1339 rf
|= DB_RF_HAVESTRUCT
;
1340 error
= dmu_spill_hold_by_dnode(dn
,
1341 rf
| DB_RF_MUST_SUCCEED
,
1342 FTAG
, (dmu_buf_t
**)&db
);
1344 mutex_enter(&db
->db_mtx
);
1345 data
= (before
) ? db
->db
.db_data
:
1346 dmu_objset_userquota_find_data(db
, tx
);
1347 have_spill
= B_TRUE
;
1349 mutex_enter(&dn
->dn_mtx
);
1350 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1351 mutex_exit(&dn
->dn_mtx
);
1357 user
= &dn
->dn_olduid
;
1358 group
= &dn
->dn_oldgid
;
1360 user
= &dn
->dn_newuid
;
1361 group
= &dn
->dn_newgid
;
1365 * Must always call the callback in case the object
1366 * type has changed and that type isn't an object type to track
1368 error
= used_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
,
1372 * Preserve existing uid/gid when the callback can't determine
1373 * what the new uid/gid are and the callback returned EEXIST.
1374 * The EEXIST error tells us to just use the existing uid/gid.
1375 * If we don't know what the old values are then just assign
1376 * them to 0, since that is a new file being created.
1378 if (!before
&& data
== NULL
&& error
== EEXIST
) {
1379 if (flags
& DN_ID_OLD_EXIST
) {
1380 dn
->dn_newuid
= dn
->dn_olduid
;
1381 dn
->dn_newgid
= dn
->dn_oldgid
;
1390 mutex_exit(&db
->db_mtx
);
1392 mutex_enter(&dn
->dn_mtx
);
1393 if (error
== 0 && before
)
1394 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1395 if (error
== 0 && !before
)
1396 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
1399 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1401 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1403 mutex_exit(&dn
->dn_mtx
);
1405 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
1409 dmu_objset_userspace_present(objset_t
*os
)
1411 return (os
->os_phys
->os_flags
&
1412 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
1416 dmu_objset_userspace_upgrade(objset_t
*os
)
1421 if (dmu_objset_userspace_present(os
))
1423 if (!dmu_objset_userused_enabled(os
))
1424 return (SET_ERROR(ENOTSUP
));
1425 if (dmu_objset_is_snapshot(os
))
1426 return (SET_ERROR(EINVAL
));
1429 * We simply need to mark every object dirty, so that it will be
1430 * synced out and now accounted. If this is called
1431 * concurrently, or if we already did some work before crashing,
1432 * that's fine, since we track each object's accounted state
1436 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
1441 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1442 return (SET_ERROR(EINTR
));
1444 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
1447 tx
= dmu_tx_create(os
);
1448 dmu_tx_hold_bonus(tx
, obj
);
1449 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
1454 dmu_buf_will_dirty(db
, tx
);
1455 dmu_buf_rele(db
, FTAG
);
1459 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
1460 txg_wait_synced(dmu_objset_pool(os
), 0);
1465 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
1466 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1468 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
1469 usedobjsp
, availobjsp
);
1473 dmu_objset_fsid_guid(objset_t
*os
)
1475 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
1479 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
1481 stat
->dds_type
= os
->os_phys
->os_type
;
1482 if (os
->os_dsl_dataset
)
1483 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
1487 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
1489 ASSERT(os
->os_dsl_dataset
||
1490 os
->os_phys
->os_type
== DMU_OST_META
);
1492 if (os
->os_dsl_dataset
!= NULL
)
1493 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
1495 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
1496 os
->os_phys
->os_type
);
1497 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
1498 dmu_objset_userspace_present(os
));
1502 dmu_objset_is_snapshot(objset_t
*os
)
1504 if (os
->os_dsl_dataset
!= NULL
)
1505 return (os
->os_dsl_dataset
->ds_is_snapshot
);
1511 dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
, int maxlen
,
1512 boolean_t
*conflict
)
1514 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1517 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1518 return (SET_ERROR(ENOENT
));
1520 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1521 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, name
, 8, 1, &ignored
,
1522 MT_FIRST
, real
, maxlen
, conflict
));
1526 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
1527 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
1529 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1530 zap_cursor_t cursor
;
1531 zap_attribute_t attr
;
1533 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
1535 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1536 return (SET_ERROR(ENOENT
));
1538 zap_cursor_init_serialized(&cursor
,
1539 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1540 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, *offp
);
1542 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1543 zap_cursor_fini(&cursor
);
1544 return (SET_ERROR(ENOENT
));
1547 if (strlen(attr
.za_name
) + 1 > namelen
) {
1548 zap_cursor_fini(&cursor
);
1549 return (SET_ERROR(ENAMETOOLONG
));
1552 (void) strcpy(name
, attr
.za_name
);
1554 *idp
= attr
.za_first_integer
;
1556 *case_conflict
= attr
.za_normalization_conflict
;
1557 zap_cursor_advance(&cursor
);
1558 *offp
= zap_cursor_serialize(&cursor
);
1559 zap_cursor_fini(&cursor
);
1565 dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *value
)
1567 return (dsl_dataset_snap_lookup(os
->os_dsl_dataset
, name
, value
));
1571 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
1572 uint64_t *idp
, uint64_t *offp
)
1574 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
1575 zap_cursor_t cursor
;
1576 zap_attribute_t attr
;
1578 /* there is no next dir on a snapshot! */
1579 if (os
->os_dsl_dataset
->ds_object
!=
1580 dsl_dir_phys(dd
)->dd_head_dataset_obj
)
1581 return (SET_ERROR(ENOENT
));
1583 zap_cursor_init_serialized(&cursor
,
1584 dd
->dd_pool
->dp_meta_objset
,
1585 dsl_dir_phys(dd
)->dd_child_dir_zapobj
, *offp
);
1587 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1588 zap_cursor_fini(&cursor
);
1589 return (SET_ERROR(ENOENT
));
1592 if (strlen(attr
.za_name
) + 1 > namelen
) {
1593 zap_cursor_fini(&cursor
);
1594 return (SET_ERROR(ENAMETOOLONG
));
1597 (void) strcpy(name
, attr
.za_name
);
1599 *idp
= attr
.za_first_integer
;
1600 zap_cursor_advance(&cursor
);
1601 *offp
= zap_cursor_serialize(&cursor
);
1602 zap_cursor_fini(&cursor
);
1607 typedef struct dmu_objset_find_ctx
{
1611 int (*dc_func
)(dsl_pool_t
*, dsl_dataset_t
*, void *);
1614 kmutex_t
*dc_error_lock
;
1616 } dmu_objset_find_ctx_t
;
1619 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t
*dcp
)
1621 dsl_pool_t
*dp
= dcp
->dc_dp
;
1622 dmu_objset_find_ctx_t
*child_dcp
;
1626 zap_attribute_t
*attr
;
1630 /* don't process if there already was an error */
1631 if (*dcp
->dc_error
!= 0)
1634 err
= dsl_dir_hold_obj(dp
, dcp
->dc_ddobj
, NULL
, FTAG
, &dd
);
1638 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1639 if (dd
->dd_myname
[0] == '$') {
1640 dsl_dir_rele(dd
, FTAG
);
1644 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
1645 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1648 * Iterate over all children.
1650 if (dcp
->dc_flags
& DS_FIND_CHILDREN
) {
1651 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1652 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
1653 zap_cursor_retrieve(&zc
, attr
) == 0;
1654 (void) zap_cursor_advance(&zc
)) {
1655 ASSERT3U(attr
->za_integer_length
, ==,
1657 ASSERT3U(attr
->za_num_integers
, ==, 1);
1659 child_dcp
= kmem_alloc(sizeof (*child_dcp
), KM_SLEEP
);
1661 child_dcp
->dc_ddobj
= attr
->za_first_integer
;
1662 if (dcp
->dc_tq
!= NULL
)
1663 (void) taskq_dispatch(dcp
->dc_tq
,
1664 dmu_objset_find_dp_cb
, child_dcp
, TQ_SLEEP
);
1666 dmu_objset_find_dp_impl(child_dcp
);
1668 zap_cursor_fini(&zc
);
1672 * Iterate over all snapshots.
1674 if (dcp
->dc_flags
& DS_FIND_SNAPSHOTS
) {
1676 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1681 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
1682 dsl_dataset_rele(ds
, FTAG
);
1684 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1685 zap_cursor_retrieve(&zc
, attr
) == 0;
1686 (void) zap_cursor_advance(&zc
)) {
1687 ASSERT3U(attr
->za_integer_length
, ==,
1689 ASSERT3U(attr
->za_num_integers
, ==, 1);
1691 err
= dsl_dataset_hold_obj(dp
,
1692 attr
->za_first_integer
, FTAG
, &ds
);
1695 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
1696 dsl_dataset_rele(ds
, FTAG
);
1700 zap_cursor_fini(&zc
);
1704 dsl_dir_rele(dd
, FTAG
);
1705 kmem_free(attr
, sizeof (zap_attribute_t
));
1713 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1716 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
1717 dsl_dataset_rele(ds
, FTAG
);
1721 mutex_enter(dcp
->dc_error_lock
);
1722 /* only keep first error */
1723 if (*dcp
->dc_error
== 0)
1724 *dcp
->dc_error
= err
;
1725 mutex_exit(dcp
->dc_error_lock
);
1728 kmem_free(dcp
, sizeof (*dcp
));
1732 dmu_objset_find_dp_cb(void *arg
)
1734 dmu_objset_find_ctx_t
*dcp
= arg
;
1735 dsl_pool_t
*dp
= dcp
->dc_dp
;
1738 * We need to get a pool_config_lock here, as there are several
1739 * asssert(pool_config_held) down the stack. Getting a lock via
1740 * dsl_pool_config_enter is risky, as it might be stalled by a
1741 * pending writer. This would deadlock, as the write lock can
1742 * only be granted when our parent thread gives up the lock.
1743 * The _prio interface gives us priority over a pending writer.
1745 dsl_pool_config_enter_prio(dp
, FTAG
);
1747 dmu_objset_find_dp_impl(dcp
);
1749 dsl_pool_config_exit(dp
, FTAG
);
1753 * Find objsets under and including ddobj, call func(ds) on each.
1754 * The order for the enumeration is completely undefined.
1755 * func is called with dsl_pool_config held.
1758 dmu_objset_find_dp(dsl_pool_t
*dp
, uint64_t ddobj
,
1759 int func(dsl_pool_t
*, dsl_dataset_t
*, void *), void *arg
, int flags
)
1764 dmu_objset_find_ctx_t
*dcp
;
1767 mutex_init(&err_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1768 dcp
= kmem_alloc(sizeof (*dcp
), KM_SLEEP
);
1771 dcp
->dc_ddobj
= ddobj
;
1772 dcp
->dc_func
= func
;
1774 dcp
->dc_flags
= flags
;
1775 dcp
->dc_error_lock
= &err_lock
;
1776 dcp
->dc_error
= &error
;
1778 if ((flags
& DS_FIND_SERIALIZE
) || dsl_pool_config_held_writer(dp
)) {
1780 * In case a write lock is held we can't make use of
1781 * parallelism, as down the stack of the worker threads
1782 * the lock is asserted via dsl_pool_config_held.
1783 * In case of a read lock this is solved by getting a read
1784 * lock in each worker thread, which isn't possible in case
1785 * of a writer lock. So we fall back to the synchronous path
1787 * In the future it might be possible to get some magic into
1788 * dsl_pool_config_held in a way that it returns true for
1789 * the worker threads so that a single lock held from this
1790 * thread suffices. For now, stay single threaded.
1792 dmu_objset_find_dp_impl(dcp
);
1793 mutex_destroy(&err_lock
);
1798 ntasks
= dmu_find_threads
;
1800 ntasks
= vdev_count_leaves(dp
->dp_spa
) * 4;
1801 tq
= taskq_create("dmu_objset_find", ntasks
, maxclsyspri
, ntasks
,
1804 kmem_free(dcp
, sizeof (*dcp
));
1805 mutex_destroy(&err_lock
);
1807 return (SET_ERROR(ENOMEM
));
1811 /* dcp will be freed by task */
1812 (void) taskq_dispatch(tq
, dmu_objset_find_dp_cb
, dcp
, TQ_SLEEP
);
1815 * PORTING: this code relies on the property of taskq_wait to wait
1816 * until no more tasks are queued and no more tasks are active. As
1817 * we always queue new tasks from within other tasks, task_wait
1818 * reliably waits for the full recursion to finish, even though we
1819 * enqueue new tasks after taskq_wait has been called.
1820 * On platforms other than illumos, taskq_wait may not have this
1825 mutex_destroy(&err_lock
);
1831 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1832 * The dp_config_rwlock must not be held when this is called, and it
1833 * will not be held when the callback is called.
1834 * Therefore this function should only be used when the pool is not changing
1835 * (e.g. in syncing context), or the callback can deal with the possible races.
1838 dmu_objset_find_impl(spa_t
*spa
, const char *name
,
1839 int func(const char *, void *), void *arg
, int flags
)
1842 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1845 zap_attribute_t
*attr
;
1850 dsl_pool_config_enter(dp
, FTAG
);
1852 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, NULL
);
1854 dsl_pool_config_exit(dp
, FTAG
);
1858 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1859 if (dd
->dd_myname
[0] == '$') {
1860 dsl_dir_rele(dd
, FTAG
);
1861 dsl_pool_config_exit(dp
, FTAG
);
1865 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
1866 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1869 * Iterate over all children.
1871 if (flags
& DS_FIND_CHILDREN
) {
1872 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1873 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
1874 zap_cursor_retrieve(&zc
, attr
) == 0;
1875 (void) zap_cursor_advance(&zc
)) {
1876 ASSERT3U(attr
->za_integer_length
, ==,
1878 ASSERT3U(attr
->za_num_integers
, ==, 1);
1880 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
1881 dsl_pool_config_exit(dp
, FTAG
);
1882 err
= dmu_objset_find_impl(spa
, child
,
1884 dsl_pool_config_enter(dp
, FTAG
);
1889 zap_cursor_fini(&zc
);
1892 dsl_dir_rele(dd
, FTAG
);
1893 dsl_pool_config_exit(dp
, FTAG
);
1894 kmem_free(attr
, sizeof (zap_attribute_t
));
1900 * Iterate over all snapshots.
1902 if (flags
& DS_FIND_SNAPSHOTS
) {
1903 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1908 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
1909 dsl_dataset_rele(ds
, FTAG
);
1911 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1912 zap_cursor_retrieve(&zc
, attr
) == 0;
1913 (void) zap_cursor_advance(&zc
)) {
1914 ASSERT3U(attr
->za_integer_length
, ==,
1916 ASSERT3U(attr
->za_num_integers
, ==, 1);
1918 child
= kmem_asprintf("%s@%s",
1919 name
, attr
->za_name
);
1920 dsl_pool_config_exit(dp
, FTAG
);
1921 err
= func(child
, arg
);
1922 dsl_pool_config_enter(dp
, FTAG
);
1927 zap_cursor_fini(&zc
);
1931 dsl_dir_rele(dd
, FTAG
);
1932 kmem_free(attr
, sizeof (zap_attribute_t
));
1933 dsl_pool_config_exit(dp
, FTAG
);
1938 /* Apply to self. */
1939 return (func(name
, arg
));
1943 * See comment above dmu_objset_find_impl().
1946 dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
1952 error
= spa_open(name
, &spa
, FTAG
);
1955 error
= dmu_objset_find_impl(spa
, name
, func
, arg
, flags
);
1956 spa_close(spa
, FTAG
);
1961 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
1963 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
1964 os
->os_user_ptr
= user_ptr
;
1968 dmu_objset_get_user(objset_t
*os
)
1970 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
1971 return (os
->os_user_ptr
);
1975 * Determine name of filesystem, given name of snapshot.
1976 * buf must be at least MAXNAMELEN bytes
1979 dmu_fsname(const char *snapname
, char *buf
)
1981 char *atp
= strchr(snapname
, '@');
1983 return (SET_ERROR(EINVAL
));
1984 if (atp
- snapname
>= MAXNAMELEN
)
1985 return (SET_ERROR(ENAMETOOLONG
));
1986 (void) strlcpy(buf
, snapname
, atp
- snapname
+ 1);
1990 #if defined(_KERNEL) && defined(HAVE_SPL)
1991 EXPORT_SYMBOL(dmu_objset_zil
);
1992 EXPORT_SYMBOL(dmu_objset_pool
);
1993 EXPORT_SYMBOL(dmu_objset_ds
);
1994 EXPORT_SYMBOL(dmu_objset_type
);
1995 EXPORT_SYMBOL(dmu_objset_name
);
1996 EXPORT_SYMBOL(dmu_objset_hold
);
1997 EXPORT_SYMBOL(dmu_objset_own
);
1998 EXPORT_SYMBOL(dmu_objset_rele
);
1999 EXPORT_SYMBOL(dmu_objset_disown
);
2000 EXPORT_SYMBOL(dmu_objset_from_ds
);
2001 EXPORT_SYMBOL(dmu_objset_create
);
2002 EXPORT_SYMBOL(dmu_objset_clone
);
2003 EXPORT_SYMBOL(dmu_objset_stats
);
2004 EXPORT_SYMBOL(dmu_objset_fast_stat
);
2005 EXPORT_SYMBOL(dmu_objset_spa
);
2006 EXPORT_SYMBOL(dmu_objset_space
);
2007 EXPORT_SYMBOL(dmu_objset_fsid_guid
);
2008 EXPORT_SYMBOL(dmu_objset_find
);
2009 EXPORT_SYMBOL(dmu_objset_byteswap
);
2010 EXPORT_SYMBOL(dmu_objset_evict_dbufs
);
2011 EXPORT_SYMBOL(dmu_objset_snap_cmtime
);
2013 EXPORT_SYMBOL(dmu_objset_sync
);
2014 EXPORT_SYMBOL(dmu_objset_is_dirty
);
2015 EXPORT_SYMBOL(dmu_objset_create_impl
);
2016 EXPORT_SYMBOL(dmu_objset_open_impl
);
2017 EXPORT_SYMBOL(dmu_objset_evict
);
2018 EXPORT_SYMBOL(dmu_objset_register_type
);
2019 EXPORT_SYMBOL(dmu_objset_do_userquota_updates
);
2020 EXPORT_SYMBOL(dmu_objset_userquota_get_ids
);
2021 EXPORT_SYMBOL(dmu_objset_userused_enabled
);
2022 EXPORT_SYMBOL(dmu_objset_userspace_upgrade
);
2023 EXPORT_SYMBOL(dmu_objset_userspace_present
);