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 2015 Nexenta Systems, Inc. All rights reserved.
28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
32 /* Portions Copyright 2010 Robert Milkowski */
35 #include <sys/zfs_context.h>
36 #include <sys/dmu_objset.h>
37 #include <sys/dsl_dir.h>
38 #include <sys/dsl_dataset.h>
39 #include <sys/dsl_prop.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_synctask.h>
42 #include <sys/dsl_deleg.h>
43 #include <sys/dnode.h>
46 #include <sys/dmu_tx.h>
49 #include <sys/dmu_impl.h>
50 #include <sys/zfs_ioctl.h>
52 #include <sys/zfs_onexit.h>
53 #include <sys/dsl_destroy.h>
57 * Needed to close a window in dnode_move() that allows the objset to be freed
58 * before it can be safely accessed.
63 * Tunable to overwrite the maximum number of threads for the parallization
64 * of dmu_objset_find_dp, needed to speed up the import of pools with many
66 * Default is 4 times the number of leaf vdevs.
68 int dmu_find_threads
= 0;
70 static void dmu_objset_find_dp_cb(void *arg
);
75 rw_init(&os_lock
, NULL
, RW_DEFAULT
, NULL
);
85 dmu_objset_spa(objset_t
*os
)
91 dmu_objset_zil(objset_t
*os
)
97 dmu_objset_pool(objset_t
*os
)
101 if ((ds
= os
->os_dsl_dataset
) != NULL
&& ds
->ds_dir
)
102 return (ds
->ds_dir
->dd_pool
);
104 return (spa_get_dsl(os
->os_spa
));
108 dmu_objset_ds(objset_t
*os
)
110 return (os
->os_dsl_dataset
);
114 dmu_objset_type(objset_t
*os
)
116 return (os
->os_phys
->os_type
);
120 dmu_objset_name(objset_t
*os
, char *buf
)
122 dsl_dataset_name(os
->os_dsl_dataset
, buf
);
126 dmu_objset_id(objset_t
*os
)
128 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
130 return (ds
? ds
->ds_object
: 0);
134 dmu_objset_syncprop(objset_t
*os
)
136 return (os
->os_sync
);
140 dmu_objset_logbias(objset_t
*os
)
142 return (os
->os_logbias
);
146 checksum_changed_cb(void *arg
, uint64_t newval
)
151 * Inheritance should have been done by now.
153 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
155 os
->os_checksum
= zio_checksum_select(newval
, ZIO_CHECKSUM_ON_VALUE
);
159 compression_changed_cb(void *arg
, uint64_t newval
)
164 * Inheritance and range checking should have been done by now.
166 ASSERT(newval
!= ZIO_COMPRESS_INHERIT
);
168 os
->os_compress
= zio_compress_select(os
->os_spa
, newval
,
173 copies_changed_cb(void *arg
, uint64_t newval
)
178 * Inheritance and range checking should have been done by now.
181 ASSERT(newval
<= spa_max_replication(os
->os_spa
));
183 os
->os_copies
= newval
;
187 dedup_changed_cb(void *arg
, uint64_t newval
)
190 spa_t
*spa
= os
->os_spa
;
191 enum zio_checksum checksum
;
194 * Inheritance should have been done by now.
196 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
198 checksum
= zio_checksum_dedup_select(spa
, newval
, ZIO_CHECKSUM_OFF
);
200 os
->os_dedup_checksum
= checksum
& ZIO_CHECKSUM_MASK
;
201 os
->os_dedup_verify
= !!(checksum
& ZIO_CHECKSUM_VERIFY
);
205 primary_cache_changed_cb(void *arg
, uint64_t newval
)
210 * Inheritance and range checking should have been done by now.
212 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
213 newval
== ZFS_CACHE_METADATA
);
215 os
->os_primary_cache
= newval
;
219 secondary_cache_changed_cb(void *arg
, uint64_t newval
)
224 * Inheritance and range checking should have been done by now.
226 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
227 newval
== ZFS_CACHE_METADATA
);
229 os
->os_secondary_cache
= newval
;
233 sync_changed_cb(void *arg
, uint64_t newval
)
238 * Inheritance and range checking should have been done by now.
240 ASSERT(newval
== ZFS_SYNC_STANDARD
|| newval
== ZFS_SYNC_ALWAYS
||
241 newval
== ZFS_SYNC_DISABLED
);
243 os
->os_sync
= newval
;
245 zil_set_sync(os
->os_zil
, newval
);
249 redundant_metadata_changed_cb(void *arg
, uint64_t newval
)
254 * Inheritance and range checking should have been done by now.
256 ASSERT(newval
== ZFS_REDUNDANT_METADATA_ALL
||
257 newval
== ZFS_REDUNDANT_METADATA_MOST
);
259 os
->os_redundant_metadata
= newval
;
263 logbias_changed_cb(void *arg
, uint64_t newval
)
267 ASSERT(newval
== ZFS_LOGBIAS_LATENCY
||
268 newval
== ZFS_LOGBIAS_THROUGHPUT
);
269 os
->os_logbias
= newval
;
271 zil_set_logbias(os
->os_zil
, newval
);
275 recordsize_changed_cb(void *arg
, uint64_t newval
)
279 os
->os_recordsize
= newval
;
283 dmu_objset_byteswap(void *buf
, size_t size
)
285 objset_phys_t
*osp
= buf
;
287 ASSERT(size
== OBJSET_OLD_PHYS_SIZE
|| size
== sizeof (objset_phys_t
));
288 dnode_byteswap(&osp
->os_meta_dnode
);
289 byteswap_uint64_array(&osp
->os_zil_header
, sizeof (zil_header_t
));
290 osp
->os_type
= BSWAP_64(osp
->os_type
);
291 osp
->os_flags
= BSWAP_64(osp
->os_flags
);
292 if (size
== sizeof (objset_phys_t
)) {
293 dnode_byteswap(&osp
->os_userused_dnode
);
294 dnode_byteswap(&osp
->os_groupused_dnode
);
299 dmu_objset_open_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
305 ASSERT(ds
== NULL
|| MUTEX_HELD(&ds
->ds_opening_lock
));
307 os
= kmem_zalloc(sizeof (objset_t
), KM_SLEEP
);
308 os
->os_dsl_dataset
= ds
;
311 if (!BP_IS_HOLE(os
->os_rootbp
)) {
312 arc_flags_t aflags
= ARC_FLAG_WAIT
;
314 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
315 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
317 if (DMU_OS_IS_L2CACHEABLE(os
))
318 aflags
|= ARC_FLAG_L2CACHE
;
320 dprintf_bp(os
->os_rootbp
, "reading %s", "");
321 err
= arc_read(NULL
, spa
, os
->os_rootbp
,
322 arc_getbuf_func
, &os
->os_phys_buf
,
323 ZIO_PRIORITY_SYNC_READ
, ZIO_FLAG_CANFAIL
, &aflags
, &zb
);
325 kmem_free(os
, sizeof (objset_t
));
326 /* convert checksum errors into IO errors */
328 err
= SET_ERROR(EIO
);
332 /* Increase the blocksize if we are permitted. */
333 if (spa_version(spa
) >= SPA_VERSION_USERSPACE
&&
334 arc_buf_size(os
->os_phys_buf
) < sizeof (objset_phys_t
)) {
335 arc_buf_t
*buf
= arc_alloc_buf(spa
,
336 sizeof (objset_phys_t
), &os
->os_phys_buf
,
338 bzero(buf
->b_data
, sizeof (objset_phys_t
));
339 bcopy(os
->os_phys_buf
->b_data
, buf
->b_data
,
340 arc_buf_size(os
->os_phys_buf
));
341 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
342 os
->os_phys_buf
= buf
;
345 os
->os_phys
= os
->os_phys_buf
->b_data
;
346 os
->os_flags
= os
->os_phys
->os_flags
;
348 int size
= spa_version(spa
) >= SPA_VERSION_USERSPACE
?
349 sizeof (objset_phys_t
) : OBJSET_OLD_PHYS_SIZE
;
350 os
->os_phys_buf
= arc_alloc_buf(spa
, size
,
351 &os
->os_phys_buf
, ARC_BUFC_METADATA
);
352 os
->os_phys
= os
->os_phys_buf
->b_data
;
353 bzero(os
->os_phys
, size
);
357 * Note: the changed_cb will be called once before the register
358 * func returns, thus changing the checksum/compression from the
359 * default (fletcher2/off). Snapshots don't need to know about
360 * checksum/compression/copies.
363 boolean_t needlock
= B_FALSE
;
366 * Note: it's valid to open the objset if the dataset is
367 * long-held, in which case the pool_config lock will not
370 if (!dsl_pool_config_held(dmu_objset_pool(os
))) {
372 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
374 err
= dsl_prop_register(ds
,
375 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
376 primary_cache_changed_cb
, os
);
378 err
= dsl_prop_register(ds
,
379 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
380 secondary_cache_changed_cb
, os
);
382 if (!ds
->ds_is_snapshot
) {
384 err
= dsl_prop_register(ds
,
385 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
386 checksum_changed_cb
, os
);
389 err
= dsl_prop_register(ds
,
390 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
391 compression_changed_cb
, os
);
394 err
= dsl_prop_register(ds
,
395 zfs_prop_to_name(ZFS_PROP_COPIES
),
396 copies_changed_cb
, os
);
399 err
= dsl_prop_register(ds
,
400 zfs_prop_to_name(ZFS_PROP_DEDUP
),
401 dedup_changed_cb
, os
);
404 err
= dsl_prop_register(ds
,
405 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
406 logbias_changed_cb
, os
);
409 err
= dsl_prop_register(ds
,
410 zfs_prop_to_name(ZFS_PROP_SYNC
),
411 sync_changed_cb
, os
);
414 err
= dsl_prop_register(ds
,
416 ZFS_PROP_REDUNDANT_METADATA
),
417 redundant_metadata_changed_cb
, os
);
420 err
= dsl_prop_register(ds
,
421 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
422 recordsize_changed_cb
, os
);
426 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
428 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
429 kmem_free(os
, sizeof (objset_t
));
433 /* It's the meta-objset. */
434 os
->os_checksum
= ZIO_CHECKSUM_FLETCHER_4
;
435 os
->os_compress
= ZIO_COMPRESS_ON
;
436 os
->os_copies
= spa_max_replication(spa
);
437 os
->os_dedup_checksum
= ZIO_CHECKSUM_OFF
;
438 os
->os_dedup_verify
= B_FALSE
;
439 os
->os_logbias
= ZFS_LOGBIAS_LATENCY
;
440 os
->os_sync
= ZFS_SYNC_STANDARD
;
441 os
->os_primary_cache
= ZFS_CACHE_ALL
;
442 os
->os_secondary_cache
= ZFS_CACHE_ALL
;
445 if (ds
== NULL
|| !ds
->ds_is_snapshot
)
446 os
->os_zil_header
= os
->os_phys
->os_zil_header
;
447 os
->os_zil
= zil_alloc(os
, &os
->os_zil_header
);
449 for (i
= 0; i
< TXG_SIZE
; i
++) {
450 list_create(&os
->os_dirty_dnodes
[i
], sizeof (dnode_t
),
451 offsetof(dnode_t
, dn_dirty_link
[i
]));
452 list_create(&os
->os_free_dnodes
[i
], sizeof (dnode_t
),
453 offsetof(dnode_t
, dn_dirty_link
[i
]));
455 list_create(&os
->os_dnodes
, sizeof (dnode_t
),
456 offsetof(dnode_t
, dn_link
));
457 list_create(&os
->os_downgraded_dbufs
, sizeof (dmu_buf_impl_t
),
458 offsetof(dmu_buf_impl_t
, db_link
));
460 mutex_init(&os
->os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
461 mutex_init(&os
->os_obj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
462 mutex_init(&os
->os_user_ptr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
464 dnode_special_open(os
, &os
->os_phys
->os_meta_dnode
,
465 DMU_META_DNODE_OBJECT
, &os
->os_meta_dnode
);
466 if (arc_buf_size(os
->os_phys_buf
) >= sizeof (objset_phys_t
)) {
467 dnode_special_open(os
, &os
->os_phys
->os_userused_dnode
,
468 DMU_USERUSED_OBJECT
, &os
->os_userused_dnode
);
469 dnode_special_open(os
, &os
->os_phys
->os_groupused_dnode
,
470 DMU_GROUPUSED_OBJECT
, &os
->os_groupused_dnode
);
478 dmu_objset_from_ds(dsl_dataset_t
*ds
, objset_t
**osp
)
483 * We shouldn't be doing anything with dsl_dataset_t's unless the
484 * pool_config lock is held, or the dataset is long-held.
486 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
) ||
487 dsl_dataset_long_held(ds
));
489 mutex_enter(&ds
->ds_opening_lock
);
490 if (ds
->ds_objset
== NULL
) {
492 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
493 err
= dmu_objset_open_impl(dsl_dataset_get_spa(ds
),
494 ds
, dsl_dataset_get_blkptr(ds
), &os
);
495 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
498 mutex_enter(&ds
->ds_lock
);
499 ASSERT(ds
->ds_objset
== NULL
);
501 mutex_exit(&ds
->ds_lock
);
504 *osp
= ds
->ds_objset
;
505 mutex_exit(&ds
->ds_opening_lock
);
510 * Holds the pool while the objset is held. Therefore only one objset
511 * can be held at a time.
514 dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
)
520 err
= dsl_pool_hold(name
, tag
, &dp
);
523 err
= dsl_dataset_hold(dp
, name
, tag
, &ds
);
525 dsl_pool_rele(dp
, tag
);
529 err
= dmu_objset_from_ds(ds
, osp
);
531 dsl_dataset_rele(ds
, tag
);
532 dsl_pool_rele(dp
, tag
);
539 dmu_objset_own_impl(dsl_dataset_t
*ds
, dmu_objset_type_t type
,
540 boolean_t readonly
, void *tag
, objset_t
**osp
)
544 err
= dmu_objset_from_ds(ds
, osp
);
546 dsl_dataset_disown(ds
, tag
);
547 } else if (type
!= DMU_OST_ANY
&& type
!= (*osp
)->os_phys
->os_type
) {
548 dsl_dataset_disown(ds
, tag
);
549 return (SET_ERROR(EINVAL
));
550 } else if (!readonly
&& dsl_dataset_is_snapshot(ds
)) {
551 dsl_dataset_disown(ds
, tag
);
552 return (SET_ERROR(EROFS
));
558 * dsl_pool must not be held when this is called.
559 * Upon successful return, there will be a longhold on the dataset,
560 * and the dsl_pool will not be held.
563 dmu_objset_own(const char *name
, dmu_objset_type_t type
,
564 boolean_t readonly
, void *tag
, objset_t
**osp
)
570 err
= dsl_pool_hold(name
, FTAG
, &dp
);
573 err
= dsl_dataset_own(dp
, name
, tag
, &ds
);
575 dsl_pool_rele(dp
, FTAG
);
578 err
= dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
);
579 dsl_pool_rele(dp
, FTAG
);
585 dmu_objset_own_obj(dsl_pool_t
*dp
, uint64_t obj
, dmu_objset_type_t type
,
586 boolean_t readonly
, void *tag
, objset_t
**osp
)
591 err
= dsl_dataset_own_obj(dp
, obj
, tag
, &ds
);
595 return (dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
));
599 dmu_objset_rele(objset_t
*os
, void *tag
)
601 dsl_pool_t
*dp
= dmu_objset_pool(os
);
602 dsl_dataset_rele(os
->os_dsl_dataset
, tag
);
603 dsl_pool_rele(dp
, tag
);
607 * When we are called, os MUST refer to an objset associated with a dataset
608 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
609 * == tag. We will then release and reacquire ownership of the dataset while
610 * holding the pool config_rwlock to avoid intervening namespace or ownership
613 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
614 * release the hold on its dataset and acquire a new one on the dataset of the
615 * same name so that it can be partially torn down and reconstructed.
618 dmu_objset_refresh_ownership(objset_t
*os
, void *tag
)
621 dsl_dataset_t
*ds
, *newds
;
622 char name
[ZFS_MAX_DATASET_NAME_LEN
];
624 ds
= os
->os_dsl_dataset
;
625 VERIFY3P(ds
, !=, NULL
);
626 VERIFY3P(ds
->ds_owner
, ==, tag
);
627 VERIFY(dsl_dataset_long_held(ds
));
629 dsl_dataset_name(ds
, name
);
630 dp
= dmu_objset_pool(os
);
631 dsl_pool_config_enter(dp
, FTAG
);
632 dmu_objset_disown(os
, tag
);
633 VERIFY0(dsl_dataset_own(dp
, name
, tag
, &newds
));
634 VERIFY3P(newds
, ==, os
->os_dsl_dataset
);
635 dsl_pool_config_exit(dp
, FTAG
);
639 dmu_objset_disown(objset_t
*os
, void *tag
)
641 dsl_dataset_disown(os
->os_dsl_dataset
, tag
);
645 dmu_objset_evict_dbufs(objset_t
*os
)
650 mutex_enter(&os
->os_lock
);
651 dn
= list_head(&os
->os_dnodes
);
654 * Skip dnodes without holds. We have to do this dance
655 * because dnode_add_ref() only works if there is already a
656 * hold. If the dnode has no holds, then it has no dbufs.
658 if (dnode_add_ref(dn
, FTAG
)) {
659 list_insert_after(&os
->os_dnodes
, dn
, &dn_marker
);
660 mutex_exit(&os
->os_lock
);
662 dnode_evict_dbufs(dn
);
663 dnode_rele(dn
, FTAG
);
665 mutex_enter(&os
->os_lock
);
666 dn
= list_next(&os
->os_dnodes
, &dn_marker
);
667 list_remove(&os
->os_dnodes
, &dn_marker
);
669 dn
= list_next(&os
->os_dnodes
, dn
);
672 mutex_exit(&os
->os_lock
);
674 if (DMU_USERUSED_DNODE(os
) != NULL
) {
675 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os
));
676 dnode_evict_dbufs(DMU_USERUSED_DNODE(os
));
678 dnode_evict_dbufs(DMU_META_DNODE(os
));
682 * Objset eviction processing is split into into two pieces.
683 * The first marks the objset as evicting, evicts any dbufs that
684 * have a refcount of zero, and then queues up the objset for the
685 * second phase of eviction. Once os->os_dnodes has been cleared by
686 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
687 * The second phase closes the special dnodes, dequeues the objset from
688 * the list of those undergoing eviction, and finally frees the objset.
690 * NOTE: Due to asynchronous eviction processing (invocation of
691 * dnode_buf_pageout()), it is possible for the meta dnode for the
692 * objset to have no holds even though os->os_dnodes is not empty.
695 dmu_objset_evict(objset_t
*os
)
697 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
699 for (int t
= 0; t
< TXG_SIZE
; t
++)
700 ASSERT(!dmu_objset_is_dirty(os
, t
));
703 dsl_prop_unregister_all(ds
, os
);
708 dmu_objset_evict_dbufs(os
);
710 mutex_enter(&os
->os_lock
);
711 spa_evicting_os_register(os
->os_spa
, os
);
712 if (list_is_empty(&os
->os_dnodes
)) {
713 mutex_exit(&os
->os_lock
);
714 dmu_objset_evict_done(os
);
716 mutex_exit(&os
->os_lock
);
721 dmu_objset_evict_done(objset_t
*os
)
723 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
725 dnode_special_close(&os
->os_meta_dnode
);
726 if (DMU_USERUSED_DNODE(os
)) {
727 dnode_special_close(&os
->os_userused_dnode
);
728 dnode_special_close(&os
->os_groupused_dnode
);
730 zil_free(os
->os_zil
);
732 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
735 * This is a barrier to prevent the objset from going away in
736 * dnode_move() until we can safely ensure that the objset is still in
737 * use. We consider the objset valid before the barrier and invalid
740 rw_enter(&os_lock
, RW_READER
);
743 mutex_destroy(&os
->os_lock
);
744 mutex_destroy(&os
->os_obj_lock
);
745 mutex_destroy(&os
->os_user_ptr_lock
);
746 spa_evicting_os_deregister(os
->os_spa
, os
);
747 kmem_free(os
, sizeof (objset_t
));
751 dmu_objset_snap_cmtime(objset_t
*os
)
753 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
756 /* called from dsl for meta-objset */
758 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
759 dmu_objset_type_t type
, dmu_tx_t
*tx
)
764 ASSERT(dmu_tx_is_syncing(tx
));
767 VERIFY0(dmu_objset_from_ds(ds
, &os
));
769 VERIFY0(dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
771 mdn
= DMU_META_DNODE(os
);
773 dnode_allocate(mdn
, DMU_OT_DNODE
, 1 << DNODE_BLOCK_SHIFT
,
774 DN_MAX_INDBLKSHIFT
, DMU_OT_NONE
, 0, tx
);
777 * We don't want to have to increase the meta-dnode's nlevels
778 * later, because then we could do it in quescing context while
779 * we are also accessing it in open context.
781 * This precaution is not necessary for the MOS (ds == NULL),
782 * because the MOS is only updated in syncing context.
783 * This is most fortunate: the MOS is the only objset that
784 * needs to be synced multiple times as spa_sync() iterates
785 * to convergence, so minimizing its dn_nlevels matters.
791 * Determine the number of levels necessary for the meta-dnode
792 * to contain DN_MAX_OBJECT dnodes. Note that in order to
793 * ensure that we do not overflow 64 bits, there has to be
794 * a nlevels that gives us a number of blocks > DN_MAX_OBJECT
795 * but < 2^64. Therefore,
796 * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be
797 * less than (64 - log2(DN_MAX_OBJECT)) (16).
799 while ((uint64_t)mdn
->dn_nblkptr
<<
800 (mdn
->dn_datablkshift
- DNODE_SHIFT
+
801 (levels
- 1) * (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
805 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
806 mdn
->dn_nlevels
= levels
;
809 ASSERT(type
!= DMU_OST_NONE
);
810 ASSERT(type
!= DMU_OST_ANY
);
811 ASSERT(type
< DMU_OST_NUMTYPES
);
812 os
->os_phys
->os_type
= type
;
813 if (dmu_objset_userused_enabled(os
)) {
814 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
815 os
->os_flags
= os
->os_phys
->os_flags
;
818 dsl_dataset_dirty(ds
, tx
);
823 typedef struct dmu_objset_create_arg
{
824 const char *doca_name
;
826 void (*doca_userfunc
)(objset_t
*os
, void *arg
,
827 cred_t
*cr
, dmu_tx_t
*tx
);
829 dmu_objset_type_t doca_type
;
831 } dmu_objset_create_arg_t
;
835 dmu_objset_create_check(void *arg
, dmu_tx_t
*tx
)
837 dmu_objset_create_arg_t
*doca
= arg
;
838 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
843 if (strchr(doca
->doca_name
, '@') != NULL
)
844 return (SET_ERROR(EINVAL
));
846 if (strlen(doca
->doca_name
) >= ZFS_MAX_DATASET_NAME_LEN
)
847 return (SET_ERROR(ENAMETOOLONG
));
849 error
= dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
);
853 dsl_dir_rele(pdd
, FTAG
);
854 return (SET_ERROR(EEXIST
));
856 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
858 dsl_dir_rele(pdd
, FTAG
);
864 dmu_objset_create_sync(void *arg
, dmu_tx_t
*tx
)
866 dmu_objset_create_arg_t
*doca
= arg
;
867 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
875 VERIFY0(dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
));
877 obj
= dsl_dataset_create_sync(pdd
, tail
, NULL
, doca
->doca_flags
,
878 doca
->doca_cred
, tx
);
880 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
881 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
882 bp
= dsl_dataset_get_blkptr(ds
);
883 os
= dmu_objset_create_impl(pdd
->dd_pool
->dp_spa
,
884 ds
, bp
, doca
->doca_type
, tx
);
885 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
887 if (doca
->doca_userfunc
!= NULL
) {
888 doca
->doca_userfunc(os
, doca
->doca_userarg
,
889 doca
->doca_cred
, tx
);
892 spa_history_log_internal_ds(ds
, "create", tx
, "");
893 dsl_dataset_rele(ds
, FTAG
);
894 dsl_dir_rele(pdd
, FTAG
);
898 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
899 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
)
901 dmu_objset_create_arg_t doca
;
903 doca
.doca_name
= name
;
904 doca
.doca_cred
= CRED();
905 doca
.doca_flags
= flags
;
906 doca
.doca_userfunc
= func
;
907 doca
.doca_userarg
= arg
;
908 doca
.doca_type
= type
;
910 return (dsl_sync_task(name
,
911 dmu_objset_create_check
, dmu_objset_create_sync
, &doca
,
912 5, ZFS_SPACE_CHECK_NORMAL
));
915 typedef struct dmu_objset_clone_arg
{
916 const char *doca_clone
;
917 const char *doca_origin
;
919 } dmu_objset_clone_arg_t
;
923 dmu_objset_clone_check(void *arg
, dmu_tx_t
*tx
)
925 dmu_objset_clone_arg_t
*doca
= arg
;
929 dsl_dataset_t
*origin
;
930 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
932 if (strchr(doca
->doca_clone
, '@') != NULL
)
933 return (SET_ERROR(EINVAL
));
935 if (strlen(doca
->doca_clone
) >= ZFS_MAX_DATASET_NAME_LEN
)
936 return (SET_ERROR(ENAMETOOLONG
));
938 error
= dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
);
942 dsl_dir_rele(pdd
, FTAG
);
943 return (SET_ERROR(EEXIST
));
946 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
949 dsl_dir_rele(pdd
, FTAG
);
950 return (SET_ERROR(EDQUOT
));
952 dsl_dir_rele(pdd
, FTAG
);
954 error
= dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
);
958 /* You can only clone snapshots, not the head datasets. */
959 if (!origin
->ds_is_snapshot
) {
960 dsl_dataset_rele(origin
, FTAG
);
961 return (SET_ERROR(EINVAL
));
963 dsl_dataset_rele(origin
, FTAG
);
969 dmu_objset_clone_sync(void *arg
, dmu_tx_t
*tx
)
971 dmu_objset_clone_arg_t
*doca
= arg
;
972 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
975 dsl_dataset_t
*origin
, *ds
;
977 char namebuf
[ZFS_MAX_DATASET_NAME_LEN
];
979 VERIFY0(dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
));
980 VERIFY0(dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
));
982 obj
= dsl_dataset_create_sync(pdd
, tail
, origin
, 0,
983 doca
->doca_cred
, tx
);
985 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
986 dsl_dataset_name(origin
, namebuf
);
987 spa_history_log_internal_ds(ds
, "clone", tx
,
988 "origin=%s (%llu)", namebuf
, origin
->ds_object
);
989 dsl_dataset_rele(ds
, FTAG
);
990 dsl_dataset_rele(origin
, FTAG
);
991 dsl_dir_rele(pdd
, FTAG
);
995 dmu_objset_clone(const char *clone
, const char *origin
)
997 dmu_objset_clone_arg_t doca
;
999 doca
.doca_clone
= clone
;
1000 doca
.doca_origin
= origin
;
1001 doca
.doca_cred
= CRED();
1003 return (dsl_sync_task(clone
,
1004 dmu_objset_clone_check
, dmu_objset_clone_sync
, &doca
,
1005 5, ZFS_SPACE_CHECK_NORMAL
));
1009 dmu_objset_snapshot_one(const char *fsname
, const char *snapname
)
1012 char *longsnap
= kmem_asprintf("%s@%s", fsname
, snapname
);
1013 nvlist_t
*snaps
= fnvlist_alloc();
1015 fnvlist_add_boolean(snaps
, longsnap
);
1017 err
= dsl_dataset_snapshot(snaps
, NULL
, NULL
);
1018 fnvlist_free(snaps
);
1023 dmu_objset_sync_dnodes(list_t
*list
, list_t
*newlist
, dmu_tx_t
*tx
)
1027 while (dn
= list_head(list
)) {
1028 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1029 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1031 * Initialize dn_zio outside dnode_sync() because the
1032 * meta-dnode needs to set it ouside dnode_sync().
1034 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1037 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1038 list_remove(list
, dn
);
1041 (void) dnode_add_ref(dn
, newlist
);
1042 list_insert_tail(newlist
, dn
);
1051 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1053 blkptr_t
*bp
= zio
->io_bp
;
1055 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1057 ASSERT(!BP_IS_EMBEDDED(bp
));
1058 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_OBJSET
);
1059 ASSERT0(BP_GET_LEVEL(bp
));
1062 * Update rootbp fill count: it should be the number of objects
1063 * allocated in the object set (not counting the "special"
1064 * objects that are stored in the objset_phys_t -- the meta
1065 * dnode and user/group accounting objects).
1068 for (int i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1069 bp
->blk_fill
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
1070 if (os
->os_dsl_dataset
!= NULL
)
1071 rrw_enter(&os
->os_dsl_dataset
->ds_bp_rwlock
, RW_WRITER
, FTAG
);
1072 *os
->os_rootbp
= *bp
;
1073 if (os
->os_dsl_dataset
!= NULL
)
1074 rrw_exit(&os
->os_dsl_dataset
->ds_bp_rwlock
, FTAG
);
1079 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1081 blkptr_t
*bp
= zio
->io_bp
;
1082 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1085 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1086 ASSERT(BP_EQUAL(bp
, bp_orig
));
1088 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1089 dmu_tx_t
*tx
= os
->os_synctx
;
1091 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1092 dsl_dataset_block_born(ds
, bp
, tx
);
1094 kmem_free(bp
, sizeof (*bp
));
1097 /* called from dsl */
1099 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1102 zbookmark_phys_t zb
;
1106 list_t
*newlist
= NULL
;
1107 dbuf_dirty_record_t
*dr
;
1108 blkptr_t
*blkptr_copy
= kmem_alloc(sizeof (*os
->os_rootbp
), KM_SLEEP
);
1109 *blkptr_copy
= *os
->os_rootbp
;
1111 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", tx
->tx_txg
);
1113 ASSERT(dmu_tx_is_syncing(tx
));
1114 /* XXX the write_done callback should really give us the tx... */
1117 if (os
->os_dsl_dataset
== NULL
) {
1119 * This is the MOS. If we have upgraded,
1120 * spa_max_replication() could change, so reset
1123 os
->os_copies
= spa_max_replication(os
->os_spa
);
1127 * Create the root block IO
1129 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1130 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1131 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1132 arc_release(os
->os_phys_buf
, &os
->os_phys_buf
);
1134 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1136 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1137 blkptr_copy
, os
->os_phys_buf
, DMU_OS_IS_L2CACHEABLE(os
),
1138 &zp
, dmu_objset_write_ready
, NULL
, NULL
, dmu_objset_write_done
,
1139 os
, ZIO_PRIORITY_ASYNC_WRITE
, ZIO_FLAG_MUSTSUCCEED
, &zb
);
1142 * Sync special dnodes - the parent IO for the sync is the root block
1144 DMU_META_DNODE(os
)->dn_zio
= zio
;
1145 dnode_sync(DMU_META_DNODE(os
), tx
);
1147 os
->os_phys
->os_flags
= os
->os_flags
;
1149 if (DMU_USERUSED_DNODE(os
) &&
1150 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1151 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1152 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1153 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1154 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1157 txgoff
= tx
->tx_txg
& TXG_MASK
;
1159 if (dmu_objset_userused_enabled(os
)) {
1160 newlist
= &os
->os_synced_dnodes
;
1162 * We must create the list here because it uses the
1163 * dn_dirty_link[] of this txg.
1165 list_create(newlist
, sizeof (dnode_t
),
1166 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1169 dmu_objset_sync_dnodes(&os
->os_free_dnodes
[txgoff
], newlist
, tx
);
1170 dmu_objset_sync_dnodes(&os
->os_dirty_dnodes
[txgoff
], newlist
, tx
);
1172 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1173 while (dr
= list_head(list
)) {
1174 ASSERT0(dr
->dr_dbuf
->db_level
);
1175 list_remove(list
, dr
);
1177 zio_nowait(dr
->dr_zio
);
1180 * Free intent log blocks up to this tx.
1182 zil_sync(os
->os_zil
, tx
);
1183 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1188 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1190 return (!list_is_empty(&os
->os_dirty_dnodes
[txg
& TXG_MASK
]) ||
1191 !list_is_empty(&os
->os_free_dnodes
[txg
& TXG_MASK
]));
1194 static objset_used_cb_t
*used_cbs
[DMU_OST_NUMTYPES
];
1197 dmu_objset_register_type(dmu_objset_type_t ost
, objset_used_cb_t
*cb
)
1203 dmu_objset_userused_enabled(objset_t
*os
)
1205 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1206 used_cbs
[os
->os_phys
->os_type
] != NULL
&&
1207 DMU_USERUSED_DNODE(os
) != NULL
);
1210 typedef struct userquota_node
{
1213 avl_node_t uqn_node
;
1216 typedef struct userquota_cache
{
1217 avl_tree_t uqc_user_deltas
;
1218 avl_tree_t uqc_group_deltas
;
1219 } userquota_cache_t
;
1222 userquota_compare(const void *l
, const void *r
)
1224 const userquota_node_t
*luqn
= l
;
1225 const userquota_node_t
*ruqn
= r
;
1227 if (luqn
->uqn_id
< ruqn
->uqn_id
)
1229 if (luqn
->uqn_id
> ruqn
->uqn_id
)
1235 do_userquota_cacheflush(objset_t
*os
, userquota_cache_t
*cache
, dmu_tx_t
*tx
)
1238 userquota_node_t
*uqn
;
1240 ASSERT(dmu_tx_is_syncing(tx
));
1243 while ((uqn
= avl_destroy_nodes(&cache
->uqc_user_deltas
,
1244 &cookie
)) != NULL
) {
1245 VERIFY0(zap_increment_int(os
, DMU_USERUSED_OBJECT
,
1246 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1247 kmem_free(uqn
, sizeof (*uqn
));
1249 avl_destroy(&cache
->uqc_user_deltas
);
1252 while ((uqn
= avl_destroy_nodes(&cache
->uqc_group_deltas
,
1253 &cookie
)) != NULL
) {
1254 VERIFY0(zap_increment_int(os
, DMU_GROUPUSED_OBJECT
,
1255 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1256 kmem_free(uqn
, sizeof (*uqn
));
1258 avl_destroy(&cache
->uqc_group_deltas
);
1262 userquota_update_cache(avl_tree_t
*avl
, uint64_t id
, int64_t delta
)
1264 userquota_node_t search
= { .uqn_id
= id
};
1267 userquota_node_t
*uqn
= avl_find(avl
, &search
, &idx
);
1269 uqn
= kmem_zalloc(sizeof (*uqn
), KM_SLEEP
);
1271 avl_insert(avl
, uqn
, idx
);
1273 uqn
->uqn_delta
+= delta
;
1277 do_userquota_update(userquota_cache_t
*cache
, uint64_t used
, uint64_t flags
,
1278 uint64_t user
, uint64_t group
, boolean_t subtract
)
1280 if ((flags
& DNODE_FLAG_USERUSED_ACCOUNTED
)) {
1281 int64_t delta
= DNODE_SIZE
+ used
;
1285 userquota_update_cache(&cache
->uqc_user_deltas
, user
, delta
);
1286 userquota_update_cache(&cache
->uqc_group_deltas
, group
, delta
);
1291 dmu_objset_do_userquota_updates(objset_t
*os
, dmu_tx_t
*tx
)
1294 list_t
*list
= &os
->os_synced_dnodes
;
1295 userquota_cache_t cache
= { 0 };
1297 ASSERT(list_head(list
) == NULL
|| dmu_objset_userused_enabled(os
));
1299 avl_create(&cache
.uqc_user_deltas
, userquota_compare
,
1300 sizeof (userquota_node_t
), offsetof(userquota_node_t
, uqn_node
));
1301 avl_create(&cache
.uqc_group_deltas
, userquota_compare
,
1302 sizeof (userquota_node_t
), offsetof(userquota_node_t
, uqn_node
));
1304 while (dn
= list_head(list
)) {
1306 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
1307 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
1308 dn
->dn_phys
->dn_flags
&
1309 DNODE_FLAG_USERUSED_ACCOUNTED
);
1311 /* Allocate the user/groupused objects if necessary. */
1312 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
1313 VERIFY0(zap_create_claim(os
,
1314 DMU_USERUSED_OBJECT
,
1315 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1316 VERIFY0(zap_create_claim(os
,
1317 DMU_GROUPUSED_OBJECT
,
1318 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1321 flags
= dn
->dn_id_flags
;
1323 if (flags
& DN_ID_OLD_EXIST
) {
1324 do_userquota_update(&cache
,
1325 dn
->dn_oldused
, dn
->dn_oldflags
,
1326 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
);
1328 if (flags
& DN_ID_NEW_EXIST
) {
1329 do_userquota_update(&cache
,
1330 DN_USED_BYTES(dn
->dn_phys
),
1331 dn
->dn_phys
->dn_flags
, dn
->dn_newuid
,
1332 dn
->dn_newgid
, B_FALSE
);
1335 mutex_enter(&dn
->dn_mtx
);
1337 dn
->dn_oldflags
= 0;
1338 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
1339 dn
->dn_olduid
= dn
->dn_newuid
;
1340 dn
->dn_oldgid
= dn
->dn_newgid
;
1341 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1342 if (dn
->dn_bonuslen
== 0)
1343 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1345 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1347 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
1348 mutex_exit(&dn
->dn_mtx
);
1350 list_remove(list
, dn
);
1351 dnode_rele(dn
, list
);
1353 do_userquota_cacheflush(os
, &cache
, tx
);
1357 * Returns a pointer to data to find uid/gid from
1359 * If a dirty record for transaction group that is syncing can't
1360 * be found then NULL is returned. In the NULL case it is assumed
1361 * the uid/gid aren't changing.
1364 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
1366 dbuf_dirty_record_t
*dr
, **drp
;
1369 if (db
->db_dirtycnt
== 0)
1370 return (db
->db
.db_data
); /* Nothing is changing */
1372 for (drp
= &db
->db_last_dirty
; (dr
= *drp
) != NULL
; drp
= &dr
->dr_next
)
1373 if (dr
->dr_txg
== tx
->tx_txg
)
1381 DB_DNODE_ENTER(dr
->dr_dbuf
);
1382 dn
= DB_DNODE(dr
->dr_dbuf
);
1384 if (dn
->dn_bonuslen
== 0 &&
1385 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
1386 data
= dr
->dt
.dl
.dr_data
->b_data
;
1388 data
= dr
->dt
.dl
.dr_data
;
1390 DB_DNODE_EXIT(dr
->dr_dbuf
);
1397 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
1399 objset_t
*os
= dn
->dn_objset
;
1401 dmu_buf_impl_t
*db
= NULL
;
1402 uint64_t *user
= NULL
;
1403 uint64_t *group
= NULL
;
1404 int flags
= dn
->dn_id_flags
;
1406 boolean_t have_spill
= B_FALSE
;
1408 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
1411 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
1412 DN_ID_CHKED_SPILL
)))
1415 if (before
&& dn
->dn_bonuslen
!= 0)
1416 data
= DN_BONUS(dn
->dn_phys
);
1417 else if (!before
&& dn
->dn_bonuslen
!= 0) {
1420 mutex_enter(&db
->db_mtx
);
1421 data
= dmu_objset_userquota_find_data(db
, tx
);
1423 data
= DN_BONUS(dn
->dn_phys
);
1425 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
1428 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
1429 rf
|= DB_RF_HAVESTRUCT
;
1430 error
= dmu_spill_hold_by_dnode(dn
,
1431 rf
| DB_RF_MUST_SUCCEED
,
1432 FTAG
, (dmu_buf_t
**)&db
);
1434 mutex_enter(&db
->db_mtx
);
1435 data
= (before
) ? db
->db
.db_data
:
1436 dmu_objset_userquota_find_data(db
, tx
);
1437 have_spill
= B_TRUE
;
1439 mutex_enter(&dn
->dn_mtx
);
1440 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1441 mutex_exit(&dn
->dn_mtx
);
1447 user
= &dn
->dn_olduid
;
1448 group
= &dn
->dn_oldgid
;
1450 user
= &dn
->dn_newuid
;
1451 group
= &dn
->dn_newgid
;
1455 * Must always call the callback in case the object
1456 * type has changed and that type isn't an object type to track
1458 error
= used_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
,
1462 * Preserve existing uid/gid when the callback can't determine
1463 * what the new uid/gid are and the callback returned EEXIST.
1464 * The EEXIST error tells us to just use the existing uid/gid.
1465 * If we don't know what the old values are then just assign
1466 * them to 0, since that is a new file being created.
1468 if (!before
&& data
== NULL
&& error
== EEXIST
) {
1469 if (flags
& DN_ID_OLD_EXIST
) {
1470 dn
->dn_newuid
= dn
->dn_olduid
;
1471 dn
->dn_newgid
= dn
->dn_oldgid
;
1480 mutex_exit(&db
->db_mtx
);
1482 mutex_enter(&dn
->dn_mtx
);
1483 if (error
== 0 && before
)
1484 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1485 if (error
== 0 && !before
)
1486 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
1489 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1491 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1493 mutex_exit(&dn
->dn_mtx
);
1495 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
1499 dmu_objset_userspace_present(objset_t
*os
)
1501 return (os
->os_phys
->os_flags
&
1502 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
1506 dmu_objset_userspace_upgrade(objset_t
*os
)
1511 if (dmu_objset_userspace_present(os
))
1513 if (!dmu_objset_userused_enabled(os
))
1514 return (SET_ERROR(ENOTSUP
));
1515 if (dmu_objset_is_snapshot(os
))
1516 return (SET_ERROR(EINVAL
));
1519 * We simply need to mark every object dirty, so that it will be
1520 * synced out and now accounted. If this is called
1521 * concurrently, or if we already did some work before crashing,
1522 * that's fine, since we track each object's accounted state
1526 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
1531 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1532 return (SET_ERROR(EINTR
));
1534 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
1537 tx
= dmu_tx_create(os
);
1538 dmu_tx_hold_bonus(tx
, obj
);
1539 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
1544 dmu_buf_will_dirty(db
, tx
);
1545 dmu_buf_rele(db
, FTAG
);
1549 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
1550 txg_wait_synced(dmu_objset_pool(os
), 0);
1555 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
1556 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1558 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
1559 usedobjsp
, availobjsp
);
1563 dmu_objset_fsid_guid(objset_t
*os
)
1565 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
1569 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
1571 stat
->dds_type
= os
->os_phys
->os_type
;
1572 if (os
->os_dsl_dataset
)
1573 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
1577 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
1579 ASSERT(os
->os_dsl_dataset
||
1580 os
->os_phys
->os_type
== DMU_OST_META
);
1582 if (os
->os_dsl_dataset
!= NULL
)
1583 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
1585 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
1586 os
->os_phys
->os_type
);
1587 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
1588 dmu_objset_userspace_present(os
));
1592 dmu_objset_is_snapshot(objset_t
*os
)
1594 if (os
->os_dsl_dataset
!= NULL
)
1595 return (os
->os_dsl_dataset
->ds_is_snapshot
);
1601 dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
, int maxlen
,
1602 boolean_t
*conflict
)
1604 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1607 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1608 return (SET_ERROR(ENOENT
));
1610 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1611 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, name
, 8, 1, &ignored
,
1612 MT_FIRST
, real
, maxlen
, conflict
));
1616 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
1617 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
1619 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1620 zap_cursor_t cursor
;
1621 zap_attribute_t attr
;
1623 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
1625 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1626 return (SET_ERROR(ENOENT
));
1628 zap_cursor_init_serialized(&cursor
,
1629 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1630 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, *offp
);
1632 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1633 zap_cursor_fini(&cursor
);
1634 return (SET_ERROR(ENOENT
));
1637 if (strlen(attr
.za_name
) + 1 > namelen
) {
1638 zap_cursor_fini(&cursor
);
1639 return (SET_ERROR(ENAMETOOLONG
));
1642 (void) strcpy(name
, attr
.za_name
);
1644 *idp
= attr
.za_first_integer
;
1646 *case_conflict
= attr
.za_normalization_conflict
;
1647 zap_cursor_advance(&cursor
);
1648 *offp
= zap_cursor_serialize(&cursor
);
1649 zap_cursor_fini(&cursor
);
1655 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
1656 uint64_t *idp
, uint64_t *offp
)
1658 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
1659 zap_cursor_t cursor
;
1660 zap_attribute_t attr
;
1662 /* there is no next dir on a snapshot! */
1663 if (os
->os_dsl_dataset
->ds_object
!=
1664 dsl_dir_phys(dd
)->dd_head_dataset_obj
)
1665 return (SET_ERROR(ENOENT
));
1667 zap_cursor_init_serialized(&cursor
,
1668 dd
->dd_pool
->dp_meta_objset
,
1669 dsl_dir_phys(dd
)->dd_child_dir_zapobj
, *offp
);
1671 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1672 zap_cursor_fini(&cursor
);
1673 return (SET_ERROR(ENOENT
));
1676 if (strlen(attr
.za_name
) + 1 > namelen
) {
1677 zap_cursor_fini(&cursor
);
1678 return (SET_ERROR(ENAMETOOLONG
));
1681 (void) strcpy(name
, attr
.za_name
);
1683 *idp
= attr
.za_first_integer
;
1684 zap_cursor_advance(&cursor
);
1685 *offp
= zap_cursor_serialize(&cursor
);
1686 zap_cursor_fini(&cursor
);
1691 typedef struct dmu_objset_find_ctx
{
1695 int (*dc_func
)(dsl_pool_t
*, dsl_dataset_t
*, void *);
1698 kmutex_t
*dc_error_lock
;
1700 } dmu_objset_find_ctx_t
;
1703 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t
*dcp
)
1705 dsl_pool_t
*dp
= dcp
->dc_dp
;
1706 dmu_objset_find_ctx_t
*child_dcp
;
1710 zap_attribute_t
*attr
;
1714 /* don't process if there already was an error */
1715 if (*dcp
->dc_error
!= 0)
1718 err
= dsl_dir_hold_obj(dp
, dcp
->dc_ddobj
, NULL
, FTAG
, &dd
);
1722 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1723 if (dd
->dd_myname
[0] == '$') {
1724 dsl_dir_rele(dd
, FTAG
);
1728 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
1729 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1732 * Iterate over all children.
1734 if (dcp
->dc_flags
& DS_FIND_CHILDREN
) {
1735 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1736 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
1737 zap_cursor_retrieve(&zc
, attr
) == 0;
1738 (void) zap_cursor_advance(&zc
)) {
1739 ASSERT3U(attr
->za_integer_length
, ==,
1741 ASSERT3U(attr
->za_num_integers
, ==, 1);
1743 child_dcp
= kmem_alloc(sizeof (*child_dcp
), KM_SLEEP
);
1745 child_dcp
->dc_ddobj
= attr
->za_first_integer
;
1746 if (dcp
->dc_tq
!= NULL
)
1747 (void) taskq_dispatch(dcp
->dc_tq
,
1748 dmu_objset_find_dp_cb
, child_dcp
, TQ_SLEEP
);
1750 dmu_objset_find_dp_impl(child_dcp
);
1752 zap_cursor_fini(&zc
);
1756 * Iterate over all snapshots.
1758 if (dcp
->dc_flags
& DS_FIND_SNAPSHOTS
) {
1760 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1765 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
1766 dsl_dataset_rele(ds
, FTAG
);
1768 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1769 zap_cursor_retrieve(&zc
, attr
) == 0;
1770 (void) zap_cursor_advance(&zc
)) {
1771 ASSERT3U(attr
->za_integer_length
, ==,
1773 ASSERT3U(attr
->za_num_integers
, ==, 1);
1775 err
= dsl_dataset_hold_obj(dp
,
1776 attr
->za_first_integer
, FTAG
, &ds
);
1779 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
1780 dsl_dataset_rele(ds
, FTAG
);
1784 zap_cursor_fini(&zc
);
1788 dsl_dir_rele(dd
, FTAG
);
1789 kmem_free(attr
, sizeof (zap_attribute_t
));
1797 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1800 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
1801 dsl_dataset_rele(ds
, FTAG
);
1805 mutex_enter(dcp
->dc_error_lock
);
1806 /* only keep first error */
1807 if (*dcp
->dc_error
== 0)
1808 *dcp
->dc_error
= err
;
1809 mutex_exit(dcp
->dc_error_lock
);
1812 kmem_free(dcp
, sizeof (*dcp
));
1816 dmu_objset_find_dp_cb(void *arg
)
1818 dmu_objset_find_ctx_t
*dcp
= arg
;
1819 dsl_pool_t
*dp
= dcp
->dc_dp
;
1822 * We need to get a pool_config_lock here, as there are several
1823 * asssert(pool_config_held) down the stack. Getting a lock via
1824 * dsl_pool_config_enter is risky, as it might be stalled by a
1825 * pending writer. This would deadlock, as the write lock can
1826 * only be granted when our parent thread gives up the lock.
1827 * The _prio interface gives us priority over a pending writer.
1829 dsl_pool_config_enter_prio(dp
, FTAG
);
1831 dmu_objset_find_dp_impl(dcp
);
1833 dsl_pool_config_exit(dp
, FTAG
);
1837 * Find objsets under and including ddobj, call func(ds) on each.
1838 * The order for the enumeration is completely undefined.
1839 * func is called with dsl_pool_config held.
1842 dmu_objset_find_dp(dsl_pool_t
*dp
, uint64_t ddobj
,
1843 int func(dsl_pool_t
*, dsl_dataset_t
*, void *), void *arg
, int flags
)
1848 dmu_objset_find_ctx_t
*dcp
;
1851 mutex_init(&err_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1852 dcp
= kmem_alloc(sizeof (*dcp
), KM_SLEEP
);
1855 dcp
->dc_ddobj
= ddobj
;
1856 dcp
->dc_func
= func
;
1858 dcp
->dc_flags
= flags
;
1859 dcp
->dc_error_lock
= &err_lock
;
1860 dcp
->dc_error
= &error
;
1862 if ((flags
& DS_FIND_SERIALIZE
) || dsl_pool_config_held_writer(dp
)) {
1864 * In case a write lock is held we can't make use of
1865 * parallelism, as down the stack of the worker threads
1866 * the lock is asserted via dsl_pool_config_held.
1867 * In case of a read lock this is solved by getting a read
1868 * lock in each worker thread, which isn't possible in case
1869 * of a writer lock. So we fall back to the synchronous path
1871 * In the future it might be possible to get some magic into
1872 * dsl_pool_config_held in a way that it returns true for
1873 * the worker threads so that a single lock held from this
1874 * thread suffices. For now, stay single threaded.
1876 dmu_objset_find_dp_impl(dcp
);
1877 mutex_destroy(&err_lock
);
1882 ntasks
= dmu_find_threads
;
1884 ntasks
= vdev_count_leaves(dp
->dp_spa
) * 4;
1885 tq
= taskq_create("dmu_objset_find", ntasks
, minclsyspri
, ntasks
,
1888 kmem_free(dcp
, sizeof (*dcp
));
1889 mutex_destroy(&err_lock
);
1891 return (SET_ERROR(ENOMEM
));
1895 /* dcp will be freed by task */
1896 (void) taskq_dispatch(tq
, dmu_objset_find_dp_cb
, dcp
, TQ_SLEEP
);
1899 * PORTING: this code relies on the property of taskq_wait to wait
1900 * until no more tasks are queued and no more tasks are active. As
1901 * we always queue new tasks from within other tasks, task_wait
1902 * reliably waits for the full recursion to finish, even though we
1903 * enqueue new tasks after taskq_wait has been called.
1904 * On platforms other than illumos, taskq_wait may not have this
1909 mutex_destroy(&err_lock
);
1915 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1916 * The dp_config_rwlock must not be held when this is called, and it
1917 * will not be held when the callback is called.
1918 * Therefore this function should only be used when the pool is not changing
1919 * (e.g. in syncing context), or the callback can deal with the possible races.
1922 dmu_objset_find_impl(spa_t
*spa
, const char *name
,
1923 int func(const char *, void *), void *arg
, int flags
)
1926 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1929 zap_attribute_t
*attr
;
1934 dsl_pool_config_enter(dp
, FTAG
);
1936 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, NULL
);
1938 dsl_pool_config_exit(dp
, FTAG
);
1942 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1943 if (dd
->dd_myname
[0] == '$') {
1944 dsl_dir_rele(dd
, FTAG
);
1945 dsl_pool_config_exit(dp
, FTAG
);
1949 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
1950 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1953 * Iterate over all children.
1955 if (flags
& DS_FIND_CHILDREN
) {
1956 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
1957 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
1958 zap_cursor_retrieve(&zc
, attr
) == 0;
1959 (void) zap_cursor_advance(&zc
)) {
1960 ASSERT3U(attr
->za_integer_length
, ==,
1962 ASSERT3U(attr
->za_num_integers
, ==, 1);
1964 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
1965 dsl_pool_config_exit(dp
, FTAG
);
1966 err
= dmu_objset_find_impl(spa
, child
,
1968 dsl_pool_config_enter(dp
, FTAG
);
1973 zap_cursor_fini(&zc
);
1976 dsl_dir_rele(dd
, FTAG
);
1977 dsl_pool_config_exit(dp
, FTAG
);
1978 kmem_free(attr
, sizeof (zap_attribute_t
));
1984 * Iterate over all snapshots.
1986 if (flags
& DS_FIND_SNAPSHOTS
) {
1987 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
1992 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
1993 dsl_dataset_rele(ds
, FTAG
);
1995 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
1996 zap_cursor_retrieve(&zc
, attr
) == 0;
1997 (void) zap_cursor_advance(&zc
)) {
1998 ASSERT3U(attr
->za_integer_length
, ==,
2000 ASSERT3U(attr
->za_num_integers
, ==, 1);
2002 child
= kmem_asprintf("%s@%s",
2003 name
, attr
->za_name
);
2004 dsl_pool_config_exit(dp
, FTAG
);
2005 err
= func(child
, arg
);
2006 dsl_pool_config_enter(dp
, FTAG
);
2011 zap_cursor_fini(&zc
);
2015 dsl_dir_rele(dd
, FTAG
);
2016 kmem_free(attr
, sizeof (zap_attribute_t
));
2017 dsl_pool_config_exit(dp
, FTAG
);
2022 /* Apply to self. */
2023 return (func(name
, arg
));
2027 * See comment above dmu_objset_find_impl().
2030 dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
2036 error
= spa_open(name
, &spa
, FTAG
);
2039 error
= dmu_objset_find_impl(spa
, name
, func
, arg
, flags
);
2040 spa_close(spa
, FTAG
);
2045 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
2047 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2048 os
->os_user_ptr
= user_ptr
;
2052 dmu_objset_get_user(objset_t
*os
)
2054 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2055 return (os
->os_user_ptr
);
2059 * Determine name of filesystem, given name of snapshot.
2060 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2063 dmu_fsname(const char *snapname
, char *buf
)
2065 char *atp
= strchr(snapname
, '@');
2067 return (SET_ERROR(EINVAL
));
2068 if (atp
- snapname
>= ZFS_MAX_DATASET_NAME_LEN
)
2069 return (SET_ERROR(ENAMETOOLONG
));
2070 (void) strlcpy(buf
, snapname
, atp
- snapname
+ 1);