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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
26 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
29 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
30 * Copyright 2017 Nexenta Systems, Inc.
33 /* Portions Copyright 2010 Robert Milkowski */
35 #include <sys/zfeature.h>
37 #include <sys/zfs_context.h>
38 #include <sys/dmu_objset.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/dsl_dataset.h>
41 #include <sys/dsl_prop.h>
42 #include <sys/dsl_pool.h>
43 #include <sys/dsl_synctask.h>
44 #include <sys/dsl_deleg.h>
45 #include <sys/dnode.h>
48 #include <sys/dmu_tx.h>
51 #include <sys/dmu_impl.h>
52 #include <sys/zfs_ioctl.h>
54 #include <sys/zfs_onexit.h>
55 #include <sys/dsl_destroy.h>
57 #include <sys/policy.h>
58 #include <sys/spa_impl.h>
61 * Needed to close a window in dnode_move() that allows the objset to be freed
62 * before it can be safely accessed.
67 * Tunable to overwrite the maximum number of threads for the parallelization
68 * of dmu_objset_find_dp, needed to speed up the import of pools with many
70 * Default is 4 times the number of leaf vdevs.
72 int dmu_find_threads
= 0;
75 * Backfill lower metadnode objects after this many have been freed.
76 * Backfilling negatively impacts object creation rates, so only do it
77 * if there are enough holes to fill.
79 int dmu_rescan_dnode_threshold
= 1 << DN_MAX_INDBLKSHIFT
;
81 static void dmu_objset_find_dp_cb(void *arg
);
83 static void dmu_objset_upgrade(objset_t
*os
, dmu_objset_upgrade_cb_t cb
);
84 static void dmu_objset_upgrade_stop(objset_t
*os
);
89 rw_init(&os_lock
, NULL
, RW_DEFAULT
, NULL
);
99 dmu_objset_spa(objset_t
*os
)
105 dmu_objset_zil(objset_t
*os
)
111 dmu_objset_pool(objset_t
*os
)
115 if ((ds
= os
->os_dsl_dataset
) != NULL
&& ds
->ds_dir
)
116 return (ds
->ds_dir
->dd_pool
);
118 return (spa_get_dsl(os
->os_spa
));
122 dmu_objset_ds(objset_t
*os
)
124 return (os
->os_dsl_dataset
);
128 dmu_objset_type(objset_t
*os
)
130 return (os
->os_phys
->os_type
);
134 dmu_objset_name(objset_t
*os
, char *buf
)
136 dsl_dataset_name(os
->os_dsl_dataset
, buf
);
140 dmu_objset_id(objset_t
*os
)
142 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
144 return (ds
? ds
->ds_object
: 0);
148 dmu_objset_dnodesize(objset_t
*os
)
150 return (os
->os_dnodesize
);
154 dmu_objset_syncprop(objset_t
*os
)
156 return (os
->os_sync
);
160 dmu_objset_logbias(objset_t
*os
)
162 return (os
->os_logbias
);
166 checksum_changed_cb(void *arg
, uint64_t newval
)
171 * Inheritance should have been done by now.
173 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
175 os
->os_checksum
= zio_checksum_select(newval
, ZIO_CHECKSUM_ON_VALUE
);
179 compression_changed_cb(void *arg
, uint64_t newval
)
184 * Inheritance and range checking should have been done by now.
186 ASSERT(newval
!= ZIO_COMPRESS_INHERIT
);
188 os
->os_compress
= zio_compress_select(os
->os_spa
, newval
,
193 copies_changed_cb(void *arg
, uint64_t newval
)
198 * Inheritance and range checking should have been done by now.
201 ASSERT(newval
<= spa_max_replication(os
->os_spa
));
203 os
->os_copies
= newval
;
207 dedup_changed_cb(void *arg
, uint64_t newval
)
210 spa_t
*spa
= os
->os_spa
;
211 enum zio_checksum checksum
;
214 * Inheritance should have been done by now.
216 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
218 checksum
= zio_checksum_dedup_select(spa
, newval
, ZIO_CHECKSUM_OFF
);
220 os
->os_dedup_checksum
= checksum
& ZIO_CHECKSUM_MASK
;
221 os
->os_dedup_verify
= !!(checksum
& ZIO_CHECKSUM_VERIFY
);
225 primary_cache_changed_cb(void *arg
, uint64_t newval
)
230 * Inheritance and range checking should have been done by now.
232 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
233 newval
== ZFS_CACHE_METADATA
);
235 os
->os_primary_cache
= newval
;
239 secondary_cache_changed_cb(void *arg
, uint64_t newval
)
244 * Inheritance and range checking should have been done by now.
246 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
247 newval
== ZFS_CACHE_METADATA
);
249 os
->os_secondary_cache
= newval
;
253 sync_changed_cb(void *arg
, uint64_t newval
)
258 * Inheritance and range checking should have been done by now.
260 ASSERT(newval
== ZFS_SYNC_STANDARD
|| newval
== ZFS_SYNC_ALWAYS
||
261 newval
== ZFS_SYNC_DISABLED
);
263 os
->os_sync
= newval
;
265 zil_set_sync(os
->os_zil
, newval
);
269 redundant_metadata_changed_cb(void *arg
, uint64_t newval
)
274 * Inheritance and range checking should have been done by now.
276 ASSERT(newval
== ZFS_REDUNDANT_METADATA_ALL
||
277 newval
== ZFS_REDUNDANT_METADATA_MOST
);
279 os
->os_redundant_metadata
= newval
;
283 dnodesize_changed_cb(void *arg
, uint64_t newval
)
288 case ZFS_DNSIZE_LEGACY
:
289 os
->os_dnodesize
= DNODE_MIN_SIZE
;
291 case ZFS_DNSIZE_AUTO
:
293 * Choose a dnode size that will work well for most
294 * workloads if the user specified "auto". Future code
295 * improvements could dynamically select a dnode size
296 * based on observed workload patterns.
298 os
->os_dnodesize
= DNODE_MIN_SIZE
* 2;
305 os
->os_dnodesize
= newval
;
311 logbias_changed_cb(void *arg
, uint64_t newval
)
315 ASSERT(newval
== ZFS_LOGBIAS_LATENCY
||
316 newval
== ZFS_LOGBIAS_THROUGHPUT
);
317 os
->os_logbias
= newval
;
319 zil_set_logbias(os
->os_zil
, newval
);
323 recordsize_changed_cb(void *arg
, uint64_t newval
)
327 os
->os_recordsize
= newval
;
331 dmu_objset_byteswap(void *buf
, size_t size
)
333 objset_phys_t
*osp
= buf
;
335 ASSERT(size
== OBJSET_OLD_PHYS_SIZE
|| size
== sizeof (objset_phys_t
));
336 dnode_byteswap(&osp
->os_meta_dnode
);
337 byteswap_uint64_array(&osp
->os_zil_header
, sizeof (zil_header_t
));
338 osp
->os_type
= BSWAP_64(osp
->os_type
);
339 osp
->os_flags
= BSWAP_64(osp
->os_flags
);
340 if (size
== sizeof (objset_phys_t
)) {
341 dnode_byteswap(&osp
->os_userused_dnode
);
342 dnode_byteswap(&osp
->os_groupused_dnode
);
347 * The hash is a CRC-based hash of the objset_t pointer and the object number.
350 dnode_hash(const objset_t
*os
, uint64_t obj
)
352 uintptr_t osv
= (uintptr_t)os
;
353 uint64_t crc
= -1ULL;
355 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
357 * The low 6 bits of the pointer don't have much entropy, because
358 * the objset_t is larger than 2^6 bytes long.
360 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (osv
>> 6)) & 0xFF];
361 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 0)) & 0xFF];
362 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 8)) & 0xFF];
363 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 16)) & 0xFF];
365 crc
^= (osv
>>14) ^ (obj
>>24);
371 dnode_multilist_index_func(multilist_t
*ml
, void *obj
)
374 return (dnode_hash(dn
->dn_objset
, dn
->dn_object
) %
375 multilist_get_num_sublists(ml
));
379 dmu_objset_open_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
385 ASSERT(ds
== NULL
|| MUTEX_HELD(&ds
->ds_opening_lock
));
387 os
= kmem_zalloc(sizeof (objset_t
), KM_SLEEP
);
388 os
->os_dsl_dataset
= ds
;
391 if (!BP_IS_HOLE(os
->os_rootbp
)) {
392 arc_flags_t aflags
= ARC_FLAG_WAIT
;
394 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
395 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
397 if (DMU_OS_IS_L2CACHEABLE(os
))
398 aflags
|= ARC_FLAG_L2CACHE
;
400 dprintf_bp(os
->os_rootbp
, "reading %s", "");
401 err
= arc_read(NULL
, spa
, os
->os_rootbp
,
402 arc_getbuf_func
, &os
->os_phys_buf
,
403 ZIO_PRIORITY_SYNC_READ
, ZIO_FLAG_CANFAIL
, &aflags
, &zb
);
405 kmem_free(os
, sizeof (objset_t
));
406 /* convert checksum errors into IO errors */
408 err
= SET_ERROR(EIO
);
412 /* Increase the blocksize if we are permitted. */
413 if (spa_version(spa
) >= SPA_VERSION_USERSPACE
&&
414 arc_buf_size(os
->os_phys_buf
) < sizeof (objset_phys_t
)) {
415 arc_buf_t
*buf
= arc_alloc_buf(spa
, &os
->os_phys_buf
,
416 ARC_BUFC_METADATA
, sizeof (objset_phys_t
));
417 bzero(buf
->b_data
, sizeof (objset_phys_t
));
418 bcopy(os
->os_phys_buf
->b_data
, buf
->b_data
,
419 arc_buf_size(os
->os_phys_buf
));
420 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
421 os
->os_phys_buf
= buf
;
424 os
->os_phys
= os
->os_phys_buf
->b_data
;
425 os
->os_flags
= os
->os_phys
->os_flags
;
427 int size
= spa_version(spa
) >= SPA_VERSION_USERSPACE
?
428 sizeof (objset_phys_t
) : OBJSET_OLD_PHYS_SIZE
;
429 os
->os_phys_buf
= arc_alloc_buf(spa
, &os
->os_phys_buf
,
430 ARC_BUFC_METADATA
, size
);
431 os
->os_phys
= os
->os_phys_buf
->b_data
;
432 bzero(os
->os_phys
, size
);
436 * Note: the changed_cb will be called once before the register
437 * func returns, thus changing the checksum/compression from the
438 * default (fletcher2/off). Snapshots don't need to know about
439 * checksum/compression/copies.
442 boolean_t needlock
= B_FALSE
;
445 * Note: it's valid to open the objset if the dataset is
446 * long-held, in which case the pool_config lock will not
449 if (!dsl_pool_config_held(dmu_objset_pool(os
))) {
451 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
453 err
= dsl_prop_register(ds
,
454 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
455 primary_cache_changed_cb
, os
);
457 err
= dsl_prop_register(ds
,
458 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
459 secondary_cache_changed_cb
, os
);
461 if (!ds
->ds_is_snapshot
) {
463 err
= dsl_prop_register(ds
,
464 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
465 checksum_changed_cb
, os
);
468 err
= dsl_prop_register(ds
,
469 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
470 compression_changed_cb
, os
);
473 err
= dsl_prop_register(ds
,
474 zfs_prop_to_name(ZFS_PROP_COPIES
),
475 copies_changed_cb
, os
);
478 err
= dsl_prop_register(ds
,
479 zfs_prop_to_name(ZFS_PROP_DEDUP
),
480 dedup_changed_cb
, os
);
483 err
= dsl_prop_register(ds
,
484 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
485 logbias_changed_cb
, os
);
488 err
= dsl_prop_register(ds
,
489 zfs_prop_to_name(ZFS_PROP_SYNC
),
490 sync_changed_cb
, os
);
493 err
= dsl_prop_register(ds
,
495 ZFS_PROP_REDUNDANT_METADATA
),
496 redundant_metadata_changed_cb
, os
);
499 err
= dsl_prop_register(ds
,
500 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
501 recordsize_changed_cb
, os
);
504 err
= dsl_prop_register(ds
,
505 zfs_prop_to_name(ZFS_PROP_DNODESIZE
),
506 dnodesize_changed_cb
, os
);
510 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
512 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
513 kmem_free(os
, sizeof (objset_t
));
517 /* It's the meta-objset. */
518 os
->os_checksum
= ZIO_CHECKSUM_FLETCHER_4
;
519 os
->os_compress
= ZIO_COMPRESS_ON
;
520 os
->os_copies
= spa_max_replication(spa
);
521 os
->os_dedup_checksum
= ZIO_CHECKSUM_OFF
;
522 os
->os_dedup_verify
= B_FALSE
;
523 os
->os_logbias
= ZFS_LOGBIAS_LATENCY
;
524 os
->os_sync
= ZFS_SYNC_STANDARD
;
525 os
->os_primary_cache
= ZFS_CACHE_ALL
;
526 os
->os_secondary_cache
= ZFS_CACHE_ALL
;
527 os
->os_dnodesize
= DNODE_MIN_SIZE
;
530 if (ds
== NULL
|| !ds
->ds_is_snapshot
)
531 os
->os_zil_header
= os
->os_phys
->os_zil_header
;
532 os
->os_zil
= zil_alloc(os
, &os
->os_zil_header
);
534 for (i
= 0; i
< TXG_SIZE
; i
++) {
535 os
->os_dirty_dnodes
[i
] = multilist_create(sizeof (dnode_t
),
536 offsetof(dnode_t
, dn_dirty_link
[i
]),
537 dnode_multilist_index_func
);
539 list_create(&os
->os_dnodes
, sizeof (dnode_t
),
540 offsetof(dnode_t
, dn_link
));
541 list_create(&os
->os_downgraded_dbufs
, sizeof (dmu_buf_impl_t
),
542 offsetof(dmu_buf_impl_t
, db_link
));
544 list_link_init(&os
->os_evicting_node
);
546 mutex_init(&os
->os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
547 mutex_init(&os
->os_userused_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
548 mutex_init(&os
->os_obj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
549 mutex_init(&os
->os_user_ptr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
551 dnode_special_open(os
, &os
->os_phys
->os_meta_dnode
,
552 DMU_META_DNODE_OBJECT
, &os
->os_meta_dnode
);
553 if (arc_buf_size(os
->os_phys_buf
) >= sizeof (objset_phys_t
)) {
554 dnode_special_open(os
, &os
->os_phys
->os_userused_dnode
,
555 DMU_USERUSED_OBJECT
, &os
->os_userused_dnode
);
556 dnode_special_open(os
, &os
->os_phys
->os_groupused_dnode
,
557 DMU_GROUPUSED_OBJECT
, &os
->os_groupused_dnode
);
560 mutex_init(&os
->os_upgrade_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
567 dmu_objset_from_ds(dsl_dataset_t
*ds
, objset_t
**osp
)
572 * We shouldn't be doing anything with dsl_dataset_t's unless the
573 * pool_config lock is held, or the dataset is long-held.
575 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
) ||
576 dsl_dataset_long_held(ds
));
578 mutex_enter(&ds
->ds_opening_lock
);
579 if (ds
->ds_objset
== NULL
) {
581 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
582 err
= dmu_objset_open_impl(dsl_dataset_get_spa(ds
),
583 ds
, dsl_dataset_get_blkptr(ds
), &os
);
584 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
587 mutex_enter(&ds
->ds_lock
);
588 ASSERT(ds
->ds_objset
== NULL
);
590 mutex_exit(&ds
->ds_lock
);
593 *osp
= ds
->ds_objset
;
594 mutex_exit(&ds
->ds_opening_lock
);
599 * Holds the pool while the objset is held. Therefore only one objset
600 * can be held at a time.
603 dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
)
609 err
= dsl_pool_hold(name
, tag
, &dp
);
612 err
= dsl_dataset_hold(dp
, name
, tag
, &ds
);
614 dsl_pool_rele(dp
, tag
);
618 err
= dmu_objset_from_ds(ds
, osp
);
620 dsl_dataset_rele(ds
, tag
);
621 dsl_pool_rele(dp
, tag
);
628 dmu_objset_own_impl(dsl_dataset_t
*ds
, dmu_objset_type_t type
,
629 boolean_t readonly
, void *tag
, objset_t
**osp
)
633 err
= dmu_objset_from_ds(ds
, osp
);
635 dsl_dataset_disown(ds
, tag
);
636 } else if (type
!= DMU_OST_ANY
&& type
!= (*osp
)->os_phys
->os_type
) {
637 dsl_dataset_disown(ds
, tag
);
638 return (SET_ERROR(EINVAL
));
639 } else if (!readonly
&& dsl_dataset_is_snapshot(ds
)) {
640 dsl_dataset_disown(ds
, tag
);
641 return (SET_ERROR(EROFS
));
647 * dsl_pool must not be held when this is called.
648 * Upon successful return, there will be a longhold on the dataset,
649 * and the dsl_pool will not be held.
652 dmu_objset_own(const char *name
, dmu_objset_type_t type
,
653 boolean_t readonly
, void *tag
, objset_t
**osp
)
659 err
= dsl_pool_hold(name
, FTAG
, &dp
);
662 err
= dsl_dataset_own(dp
, name
, tag
, &ds
);
664 dsl_pool_rele(dp
, FTAG
);
667 err
= dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
);
668 dsl_pool_rele(dp
, FTAG
);
670 if (err
== 0 && dmu_objset_userobjspace_upgradable(*osp
))
671 dmu_objset_userobjspace_upgrade(*osp
);
677 dmu_objset_own_obj(dsl_pool_t
*dp
, uint64_t obj
, dmu_objset_type_t type
,
678 boolean_t readonly
, void *tag
, objset_t
**osp
)
683 err
= dsl_dataset_own_obj(dp
, obj
, tag
, &ds
);
687 return (dmu_objset_own_impl(ds
, type
, readonly
, tag
, osp
));
691 dmu_objset_rele(objset_t
*os
, void *tag
)
693 dsl_pool_t
*dp
= dmu_objset_pool(os
);
694 dsl_dataset_rele(os
->os_dsl_dataset
, tag
);
695 dsl_pool_rele(dp
, tag
);
699 * When we are called, os MUST refer to an objset associated with a dataset
700 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
701 * == tag. We will then release and reacquire ownership of the dataset while
702 * holding the pool config_rwlock to avoid intervening namespace or ownership
705 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
706 * release the hold on its dataset and acquire a new one on the dataset of the
707 * same name so that it can be partially torn down and reconstructed.
710 dmu_objset_refresh_ownership(objset_t
*os
, void *tag
)
713 dsl_dataset_t
*ds
, *newds
;
714 char name
[ZFS_MAX_DATASET_NAME_LEN
];
716 ds
= os
->os_dsl_dataset
;
717 VERIFY3P(ds
, !=, NULL
);
718 VERIFY3P(ds
->ds_owner
, ==, tag
);
719 VERIFY(dsl_dataset_long_held(ds
));
721 dsl_dataset_name(ds
, name
);
722 dp
= dmu_objset_pool(os
);
723 dsl_pool_config_enter(dp
, FTAG
);
724 dmu_objset_disown(os
, tag
);
725 VERIFY0(dsl_dataset_own(dp
, name
, tag
, &newds
));
726 VERIFY3P(newds
, ==, os
->os_dsl_dataset
);
727 dsl_pool_config_exit(dp
, FTAG
);
731 dmu_objset_disown(objset_t
*os
, void *tag
)
734 * Stop upgrading thread
736 dmu_objset_upgrade_stop(os
);
737 dsl_dataset_disown(os
->os_dsl_dataset
, tag
);
741 dmu_objset_evict_dbufs(objset_t
*os
)
746 dn_marker
= kmem_alloc(sizeof (dnode_t
), KM_SLEEP
);
748 mutex_enter(&os
->os_lock
);
749 dn
= list_head(&os
->os_dnodes
);
752 * Skip dnodes without holds. We have to do this dance
753 * because dnode_add_ref() only works if there is already a
754 * hold. If the dnode has no holds, then it has no dbufs.
756 if (dnode_add_ref(dn
, FTAG
)) {
757 list_insert_after(&os
->os_dnodes
, dn
, dn_marker
);
758 mutex_exit(&os
->os_lock
);
760 dnode_evict_dbufs(dn
);
761 dnode_rele(dn
, FTAG
);
763 mutex_enter(&os
->os_lock
);
764 dn
= list_next(&os
->os_dnodes
, dn_marker
);
765 list_remove(&os
->os_dnodes
, dn_marker
);
767 dn
= list_next(&os
->os_dnodes
, dn
);
770 mutex_exit(&os
->os_lock
);
772 kmem_free(dn_marker
, sizeof (dnode_t
));
774 if (DMU_USERUSED_DNODE(os
) != NULL
) {
775 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os
));
776 dnode_evict_dbufs(DMU_USERUSED_DNODE(os
));
778 dnode_evict_dbufs(DMU_META_DNODE(os
));
782 * Objset eviction processing is split into into two pieces.
783 * The first marks the objset as evicting, evicts any dbufs that
784 * have a refcount of zero, and then queues up the objset for the
785 * second phase of eviction. Once os->os_dnodes has been cleared by
786 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
787 * The second phase closes the special dnodes, dequeues the objset from
788 * the list of those undergoing eviction, and finally frees the objset.
790 * NOTE: Due to asynchronous eviction processing (invocation of
791 * dnode_buf_pageout()), it is possible for the meta dnode for the
792 * objset to have no holds even though os->os_dnodes is not empty.
795 dmu_objset_evict(objset_t
*os
)
799 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
801 for (t
= 0; t
< TXG_SIZE
; t
++)
802 ASSERT(!dmu_objset_is_dirty(os
, t
));
805 dsl_prop_unregister_all(ds
, os
);
810 dmu_objset_evict_dbufs(os
);
812 mutex_enter(&os
->os_lock
);
813 spa_evicting_os_register(os
->os_spa
, os
);
814 if (list_is_empty(&os
->os_dnodes
)) {
815 mutex_exit(&os
->os_lock
);
816 dmu_objset_evict_done(os
);
818 mutex_exit(&os
->os_lock
);
823 dmu_objset_evict_done(objset_t
*os
)
825 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
827 dnode_special_close(&os
->os_meta_dnode
);
828 if (DMU_USERUSED_DNODE(os
)) {
829 dnode_special_close(&os
->os_userused_dnode
);
830 dnode_special_close(&os
->os_groupused_dnode
);
832 zil_free(os
->os_zil
);
834 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
837 * This is a barrier to prevent the objset from going away in
838 * dnode_move() until we can safely ensure that the objset is still in
839 * use. We consider the objset valid before the barrier and invalid
842 rw_enter(&os_lock
, RW_READER
);
845 mutex_destroy(&os
->os_lock
);
846 mutex_destroy(&os
->os_userused_lock
);
847 mutex_destroy(&os
->os_obj_lock
);
848 mutex_destroy(&os
->os_user_ptr_lock
);
849 mutex_destroy(&os
->os_upgrade_lock
);
850 for (int i
= 0; i
< TXG_SIZE
; i
++) {
851 multilist_destroy(os
->os_dirty_dnodes
[i
]);
853 spa_evicting_os_deregister(os
->os_spa
, os
);
854 kmem_free(os
, sizeof (objset_t
));
858 dmu_objset_snap_cmtime(objset_t
*os
)
860 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
863 /* called from dsl for meta-objset */
865 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
866 dmu_objset_type_t type
, dmu_tx_t
*tx
)
871 ASSERT(dmu_tx_is_syncing(tx
));
874 VERIFY0(dmu_objset_from_ds(ds
, &os
));
876 VERIFY0(dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
878 mdn
= DMU_META_DNODE(os
);
880 dnode_allocate(mdn
, DMU_OT_DNODE
, DNODE_BLOCK_SIZE
, DN_MAX_INDBLKSHIFT
,
881 DMU_OT_NONE
, 0, DNODE_MIN_SLOTS
, tx
);
884 * We don't want to have to increase the meta-dnode's nlevels
885 * later, because then we could do it in quescing context while
886 * we are also accessing it in open context.
888 * This precaution is not necessary for the MOS (ds == NULL),
889 * because the MOS is only updated in syncing context.
890 * This is most fortunate: the MOS is the only objset that
891 * needs to be synced multiple times as spa_sync() iterates
892 * to convergence, so minimizing its dn_nlevels matters.
898 * Determine the number of levels necessary for the meta-dnode
899 * to contain DN_MAX_OBJECT dnodes. Note that in order to
900 * ensure that we do not overflow 64 bits, there has to be
901 * a nlevels that gives us a number of blocks > DN_MAX_OBJECT
902 * but < 2^64. Therefore,
903 * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be
904 * less than (64 - log2(DN_MAX_OBJECT)) (16).
906 while ((uint64_t)mdn
->dn_nblkptr
<<
907 (mdn
->dn_datablkshift
- DNODE_SHIFT
+
908 (levels
- 1) * (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
912 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
913 mdn
->dn_nlevels
= levels
;
916 ASSERT(type
!= DMU_OST_NONE
);
917 ASSERT(type
!= DMU_OST_ANY
);
918 ASSERT(type
< DMU_OST_NUMTYPES
);
919 os
->os_phys
->os_type
= type
;
920 if (dmu_objset_userused_enabled(os
)) {
921 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
922 if (dmu_objset_userobjused_enabled(os
)) {
923 ds
->ds_feature_activation_needed
[
924 SPA_FEATURE_USEROBJ_ACCOUNTING
] = B_TRUE
;
925 os
->os_phys
->os_flags
|=
926 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
;
928 os
->os_flags
= os
->os_phys
->os_flags
;
931 dsl_dataset_dirty(ds
, tx
);
936 typedef struct dmu_objset_create_arg
{
937 const char *doca_name
;
939 void (*doca_userfunc
)(objset_t
*os
, void *arg
,
940 cred_t
*cr
, dmu_tx_t
*tx
);
942 dmu_objset_type_t doca_type
;
944 } dmu_objset_create_arg_t
;
948 dmu_objset_create_check(void *arg
, dmu_tx_t
*tx
)
950 dmu_objset_create_arg_t
*doca
= arg
;
951 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
956 if (strchr(doca
->doca_name
, '@') != NULL
)
957 return (SET_ERROR(EINVAL
));
959 if (strlen(doca
->doca_name
) >= ZFS_MAX_DATASET_NAME_LEN
)
960 return (SET_ERROR(ENAMETOOLONG
));
962 error
= dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
);
966 dsl_dir_rele(pdd
, FTAG
);
967 return (SET_ERROR(EEXIST
));
969 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
971 dsl_dir_rele(pdd
, FTAG
);
977 dmu_objset_create_sync(void *arg
, dmu_tx_t
*tx
)
979 dmu_objset_create_arg_t
*doca
= arg
;
980 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
988 VERIFY0(dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
));
990 obj
= dsl_dataset_create_sync(pdd
, tail
, NULL
, doca
->doca_flags
,
991 doca
->doca_cred
, tx
);
993 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
994 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
995 bp
= dsl_dataset_get_blkptr(ds
);
996 os
= dmu_objset_create_impl(pdd
->dd_pool
->dp_spa
,
997 ds
, bp
, doca
->doca_type
, tx
);
998 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
1000 if (doca
->doca_userfunc
!= NULL
) {
1001 doca
->doca_userfunc(os
, doca
->doca_userarg
,
1002 doca
->doca_cred
, tx
);
1005 spa_history_log_internal_ds(ds
, "create", tx
, "");
1006 zvol_create_minors(dp
->dp_spa
, doca
->doca_name
, B_TRUE
);
1008 dsl_dataset_rele(ds
, FTAG
);
1009 dsl_dir_rele(pdd
, FTAG
);
1013 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
1014 void (*func
)(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
), void *arg
)
1016 dmu_objset_create_arg_t doca
;
1018 doca
.doca_name
= name
;
1019 doca
.doca_cred
= CRED();
1020 doca
.doca_flags
= flags
;
1021 doca
.doca_userfunc
= func
;
1022 doca
.doca_userarg
= arg
;
1023 doca
.doca_type
= type
;
1025 return (dsl_sync_task(name
,
1026 dmu_objset_create_check
, dmu_objset_create_sync
, &doca
,
1027 5, ZFS_SPACE_CHECK_NORMAL
));
1030 typedef struct dmu_objset_clone_arg
{
1031 const char *doca_clone
;
1032 const char *doca_origin
;
1034 } dmu_objset_clone_arg_t
;
1038 dmu_objset_clone_check(void *arg
, dmu_tx_t
*tx
)
1040 dmu_objset_clone_arg_t
*doca
= arg
;
1044 dsl_dataset_t
*origin
;
1045 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1047 if (strchr(doca
->doca_clone
, '@') != NULL
)
1048 return (SET_ERROR(EINVAL
));
1050 if (strlen(doca
->doca_clone
) >= ZFS_MAX_DATASET_NAME_LEN
)
1051 return (SET_ERROR(ENAMETOOLONG
));
1053 error
= dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
);
1057 dsl_dir_rele(pdd
, FTAG
);
1058 return (SET_ERROR(EEXIST
));
1061 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
1064 dsl_dir_rele(pdd
, FTAG
);
1065 return (SET_ERROR(EDQUOT
));
1067 dsl_dir_rele(pdd
, FTAG
);
1069 error
= dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
);
1073 /* You can only clone snapshots, not the head datasets. */
1074 if (!origin
->ds_is_snapshot
) {
1075 dsl_dataset_rele(origin
, FTAG
);
1076 return (SET_ERROR(EINVAL
));
1078 dsl_dataset_rele(origin
, FTAG
);
1084 dmu_objset_clone_sync(void *arg
, dmu_tx_t
*tx
)
1086 dmu_objset_clone_arg_t
*doca
= arg
;
1087 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1090 dsl_dataset_t
*origin
, *ds
;
1092 char namebuf
[ZFS_MAX_DATASET_NAME_LEN
];
1094 VERIFY0(dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
));
1095 VERIFY0(dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
));
1097 obj
= dsl_dataset_create_sync(pdd
, tail
, origin
, 0,
1098 doca
->doca_cred
, tx
);
1100 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
1101 dsl_dataset_name(origin
, namebuf
);
1102 spa_history_log_internal_ds(ds
, "clone", tx
,
1103 "origin=%s (%llu)", namebuf
, origin
->ds_object
);
1104 zvol_create_minors(dp
->dp_spa
, doca
->doca_clone
, B_TRUE
);
1105 dsl_dataset_rele(ds
, FTAG
);
1106 dsl_dataset_rele(origin
, FTAG
);
1107 dsl_dir_rele(pdd
, FTAG
);
1111 dmu_objset_clone(const char *clone
, const char *origin
)
1113 dmu_objset_clone_arg_t doca
;
1115 doca
.doca_clone
= clone
;
1116 doca
.doca_origin
= origin
;
1117 doca
.doca_cred
= CRED();
1119 return (dsl_sync_task(clone
,
1120 dmu_objset_clone_check
, dmu_objset_clone_sync
, &doca
,
1121 5, ZFS_SPACE_CHECK_NORMAL
));
1125 dmu_objset_snapshot_one(const char *fsname
, const char *snapname
)
1128 char *longsnap
= kmem_asprintf("%s@%s", fsname
, snapname
);
1129 nvlist_t
*snaps
= fnvlist_alloc();
1131 fnvlist_add_boolean(snaps
, longsnap
);
1133 err
= dsl_dataset_snapshot(snaps
, NULL
, NULL
);
1134 fnvlist_free(snaps
);
1139 dmu_objset_upgrade_task_cb(void *data
)
1141 objset_t
*os
= data
;
1143 mutex_enter(&os
->os_upgrade_lock
);
1144 os
->os_upgrade_status
= EINTR
;
1145 if (!os
->os_upgrade_exit
) {
1146 mutex_exit(&os
->os_upgrade_lock
);
1148 os
->os_upgrade_status
= os
->os_upgrade_cb(os
);
1149 mutex_enter(&os
->os_upgrade_lock
);
1151 os
->os_upgrade_exit
= B_TRUE
;
1152 os
->os_upgrade_id
= 0;
1153 mutex_exit(&os
->os_upgrade_lock
);
1157 dmu_objset_upgrade(objset_t
*os
, dmu_objset_upgrade_cb_t cb
)
1159 if (os
->os_upgrade_id
!= 0)
1162 mutex_enter(&os
->os_upgrade_lock
);
1163 if (os
->os_upgrade_id
== 0 && os
->os_upgrade_status
== 0) {
1164 os
->os_upgrade_exit
= B_FALSE
;
1165 os
->os_upgrade_cb
= cb
;
1166 os
->os_upgrade_id
= taskq_dispatch(
1167 os
->os_spa
->spa_upgrade_taskq
,
1168 dmu_objset_upgrade_task_cb
, os
, TQ_SLEEP
);
1169 if (os
->os_upgrade_id
== TASKQID_INVALID
)
1170 os
->os_upgrade_status
= ENOMEM
;
1172 mutex_exit(&os
->os_upgrade_lock
);
1176 dmu_objset_upgrade_stop(objset_t
*os
)
1178 mutex_enter(&os
->os_upgrade_lock
);
1179 os
->os_upgrade_exit
= B_TRUE
;
1180 if (os
->os_upgrade_id
!= 0) {
1181 taskqid_t id
= os
->os_upgrade_id
;
1183 os
->os_upgrade_id
= 0;
1184 mutex_exit(&os
->os_upgrade_lock
);
1186 taskq_cancel_id(os
->os_spa
->spa_upgrade_taskq
, id
);
1188 mutex_exit(&os
->os_upgrade_lock
);
1193 dmu_objset_sync_dnodes(multilist_sublist_t
*list
, dmu_tx_t
*tx
)
1197 while ((dn
= multilist_sublist_head(list
)) != NULL
) {
1198 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1199 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1201 * Initialize dn_zio outside dnode_sync() because the
1202 * meta-dnode needs to set it ouside dnode_sync().
1204 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1207 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1208 multilist_sublist_remove(list
, dn
);
1210 multilist_t
*newlist
= dn
->dn_objset
->os_synced_dnodes
;
1211 if (newlist
!= NULL
) {
1212 (void) dnode_add_ref(dn
, newlist
);
1213 multilist_insert(newlist
, dn
);
1222 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1226 blkptr_t
*bp
= zio
->io_bp
;
1228 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1230 ASSERT(!BP_IS_EMBEDDED(bp
));
1231 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_OBJSET
);
1232 ASSERT0(BP_GET_LEVEL(bp
));
1235 * Update rootbp fill count: it should be the number of objects
1236 * allocated in the object set (not counting the "special"
1237 * objects that are stored in the objset_phys_t -- the meta
1238 * dnode and user/group accounting objects).
1241 for (i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1242 bp
->blk_fill
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
1243 if (os
->os_dsl_dataset
!= NULL
)
1244 rrw_enter(&os
->os_dsl_dataset
->ds_bp_rwlock
, RW_WRITER
, FTAG
);
1245 *os
->os_rootbp
= *bp
;
1246 if (os
->os_dsl_dataset
!= NULL
)
1247 rrw_exit(&os
->os_dsl_dataset
->ds_bp_rwlock
, FTAG
);
1252 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1254 blkptr_t
*bp
= zio
->io_bp
;
1255 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1258 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1259 ASSERT(BP_EQUAL(bp
, bp_orig
));
1261 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1262 dmu_tx_t
*tx
= os
->os_synctx
;
1264 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1265 dsl_dataset_block_born(ds
, bp
, tx
);
1267 kmem_free(bp
, sizeof (*bp
));
1270 typedef struct sync_dnodes_arg
{
1271 multilist_t
*sda_list
;
1272 int sda_sublist_idx
;
1273 multilist_t
*sda_newlist
;
1275 } sync_dnodes_arg_t
;
1278 sync_dnodes_task(void *arg
)
1280 sync_dnodes_arg_t
*sda
= arg
;
1282 multilist_sublist_t
*ms
=
1283 multilist_sublist_lock(sda
->sda_list
, sda
->sda_sublist_idx
);
1285 dmu_objset_sync_dnodes(ms
, sda
->sda_tx
);
1287 multilist_sublist_unlock(ms
);
1289 kmem_free(sda
, sizeof (*sda
));
1293 /* called from dsl */
1295 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1298 zbookmark_phys_t zb
;
1302 dbuf_dirty_record_t
*dr
;
1303 blkptr_t
*blkptr_copy
= kmem_alloc(sizeof (*os
->os_rootbp
), KM_SLEEP
);
1304 *blkptr_copy
= *os
->os_rootbp
;
1306 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", tx
->tx_txg
);
1308 ASSERT(dmu_tx_is_syncing(tx
));
1309 /* XXX the write_done callback should really give us the tx... */
1312 if (os
->os_dsl_dataset
== NULL
) {
1314 * This is the MOS. If we have upgraded,
1315 * spa_max_replication() could change, so reset
1318 os
->os_copies
= spa_max_replication(os
->os_spa
);
1322 * Create the root block IO
1324 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1325 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1326 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1327 arc_release(os
->os_phys_buf
, &os
->os_phys_buf
);
1329 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1331 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1332 blkptr_copy
, os
->os_phys_buf
, DMU_OS_IS_L2CACHEABLE(os
),
1333 &zp
, dmu_objset_write_ready
, NULL
, NULL
, dmu_objset_write_done
,
1334 os
, ZIO_PRIORITY_ASYNC_WRITE
, ZIO_FLAG_MUSTSUCCEED
, &zb
);
1337 * Sync special dnodes - the parent IO for the sync is the root block
1339 DMU_META_DNODE(os
)->dn_zio
= zio
;
1340 dnode_sync(DMU_META_DNODE(os
), tx
);
1342 os
->os_phys
->os_flags
= os
->os_flags
;
1344 if (DMU_USERUSED_DNODE(os
) &&
1345 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1346 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1347 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1348 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1349 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1352 txgoff
= tx
->tx_txg
& TXG_MASK
;
1354 if (dmu_objset_userused_enabled(os
)) {
1356 * We must create the list here because it uses the
1357 * dn_dirty_link[] of this txg. But it may already
1358 * exist because we call dsl_dataset_sync() twice per txg.
1360 if (os
->os_synced_dnodes
== NULL
) {
1361 os
->os_synced_dnodes
=
1362 multilist_create(sizeof (dnode_t
),
1363 offsetof(dnode_t
, dn_dirty_link
[txgoff
]),
1364 dnode_multilist_index_func
);
1366 ASSERT3U(os
->os_synced_dnodes
->ml_offset
, ==,
1367 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1372 i
< multilist_get_num_sublists(os
->os_dirty_dnodes
[txgoff
]); i
++) {
1373 sync_dnodes_arg_t
*sda
= kmem_alloc(sizeof (*sda
), KM_SLEEP
);
1374 sda
->sda_list
= os
->os_dirty_dnodes
[txgoff
];
1375 sda
->sda_sublist_idx
= i
;
1377 (void) taskq_dispatch(dmu_objset_pool(os
)->dp_sync_taskq
,
1378 sync_dnodes_task
, sda
, 0);
1379 /* callback frees sda */
1381 taskq_wait(dmu_objset_pool(os
)->dp_sync_taskq
);
1383 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1384 while ((dr
= list_head(list
)) != NULL
) {
1385 ASSERT0(dr
->dr_dbuf
->db_level
);
1386 list_remove(list
, dr
);
1388 zio_nowait(dr
->dr_zio
);
1391 /* Enable dnode backfill if enough objects have been freed. */
1392 if (os
->os_freed_dnodes
>= dmu_rescan_dnode_threshold
) {
1393 os
->os_rescan_dnodes
= B_TRUE
;
1394 os
->os_freed_dnodes
= 0;
1398 * Free intent log blocks up to this tx.
1400 zil_sync(os
->os_zil
, tx
);
1401 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1406 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1408 return (!multilist_is_empty(os
->os_dirty_dnodes
[txg
& TXG_MASK
]));
1411 static objset_used_cb_t
*used_cbs
[DMU_OST_NUMTYPES
];
1414 dmu_objset_register_type(dmu_objset_type_t ost
, objset_used_cb_t
*cb
)
1420 dmu_objset_userused_enabled(objset_t
*os
)
1422 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1423 used_cbs
[os
->os_phys
->os_type
] != NULL
&&
1424 DMU_USERUSED_DNODE(os
) != NULL
);
1428 dmu_objset_userobjused_enabled(objset_t
*os
)
1430 return (dmu_objset_userused_enabled(os
) &&
1431 spa_feature_is_enabled(os
->os_spa
, SPA_FEATURE_USEROBJ_ACCOUNTING
));
1434 typedef struct userquota_node
{
1435 /* must be in the first filed, see userquota_update_cache() */
1436 char uqn_id
[20 + DMU_OBJACCT_PREFIX_LEN
];
1438 avl_node_t uqn_node
;
1441 typedef struct userquota_cache
{
1442 avl_tree_t uqc_user_deltas
;
1443 avl_tree_t uqc_group_deltas
;
1444 } userquota_cache_t
;
1447 userquota_compare(const void *l
, const void *r
)
1449 const userquota_node_t
*luqn
= l
;
1450 const userquota_node_t
*ruqn
= r
;
1454 * NB: can only access uqn_id because userquota_update_cache() doesn't
1455 * pass in an entire userquota_node_t.
1457 rv
= strcmp(luqn
->uqn_id
, ruqn
->uqn_id
);
1459 return (AVL_ISIGN(rv
));
1463 do_userquota_cacheflush(objset_t
*os
, userquota_cache_t
*cache
, dmu_tx_t
*tx
)
1466 userquota_node_t
*uqn
;
1468 ASSERT(dmu_tx_is_syncing(tx
));
1471 while ((uqn
= avl_destroy_nodes(&cache
->uqc_user_deltas
,
1472 &cookie
)) != NULL
) {
1474 * os_userused_lock protects against concurrent calls to
1475 * zap_increment_int(). It's needed because zap_increment_int()
1476 * is not thread-safe (i.e. not atomic).
1478 mutex_enter(&os
->os_userused_lock
);
1479 VERIFY0(zap_increment(os
, DMU_USERUSED_OBJECT
,
1480 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1481 mutex_exit(&os
->os_userused_lock
);
1482 kmem_free(uqn
, sizeof (*uqn
));
1484 avl_destroy(&cache
->uqc_user_deltas
);
1487 while ((uqn
= avl_destroy_nodes(&cache
->uqc_group_deltas
,
1488 &cookie
)) != NULL
) {
1489 mutex_enter(&os
->os_userused_lock
);
1490 VERIFY0(zap_increment(os
, DMU_GROUPUSED_OBJECT
,
1491 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1492 mutex_exit(&os
->os_userused_lock
);
1493 kmem_free(uqn
, sizeof (*uqn
));
1495 avl_destroy(&cache
->uqc_group_deltas
);
1499 userquota_update_cache(avl_tree_t
*avl
, const char *id
, int64_t delta
)
1501 userquota_node_t
*uqn
;
1504 ASSERT(strlen(id
) < sizeof (uqn
->uqn_id
));
1506 * Use id directly for searching because uqn_id is the first field of
1507 * userquota_node_t and fields after uqn_id won't be accessed in
1510 uqn
= avl_find(avl
, (const void *)id
, &idx
);
1512 uqn
= kmem_zalloc(sizeof (*uqn
), KM_SLEEP
);
1513 strlcpy(uqn
->uqn_id
, id
, sizeof (uqn
->uqn_id
));
1514 avl_insert(avl
, uqn
, idx
);
1516 uqn
->uqn_delta
+= delta
;
1520 do_userquota_update(userquota_cache_t
*cache
, uint64_t used
, uint64_t flags
,
1521 uint64_t user
, uint64_t group
, boolean_t subtract
)
1523 if ((flags
& DNODE_FLAG_USERUSED_ACCOUNTED
)) {
1524 int64_t delta
= DNODE_MIN_SIZE
+ used
;
1530 (void) sprintf(name
, "%llx", (longlong_t
)user
);
1531 userquota_update_cache(&cache
->uqc_user_deltas
, name
, delta
);
1533 (void) sprintf(name
, "%llx", (longlong_t
)group
);
1534 userquota_update_cache(&cache
->uqc_group_deltas
, name
, delta
);
1539 do_userobjquota_update(userquota_cache_t
*cache
, uint64_t flags
,
1540 uint64_t user
, uint64_t group
, boolean_t subtract
)
1542 if (flags
& DNODE_FLAG_USEROBJUSED_ACCOUNTED
) {
1543 char name
[20 + DMU_OBJACCT_PREFIX_LEN
];
1544 int delta
= subtract
? -1 : 1;
1546 (void) snprintf(name
, sizeof (name
), DMU_OBJACCT_PREFIX
"%llx",
1548 userquota_update_cache(&cache
->uqc_user_deltas
, name
, delta
);
1550 (void) snprintf(name
, sizeof (name
), DMU_OBJACCT_PREFIX
"%llx",
1552 userquota_update_cache(&cache
->uqc_group_deltas
, name
, delta
);
1556 typedef struct userquota_updates_arg
{
1558 int uua_sublist_idx
;
1560 } userquota_updates_arg_t
;
1563 userquota_updates_task(void *arg
)
1565 userquota_updates_arg_t
*uua
= arg
;
1566 objset_t
*os
= uua
->uua_os
;
1567 dmu_tx_t
*tx
= uua
->uua_tx
;
1569 userquota_cache_t cache
= { { 0 } };
1571 multilist_sublist_t
*list
=
1572 multilist_sublist_lock(os
->os_synced_dnodes
, uua
->uua_sublist_idx
);
1574 ASSERT(multilist_sublist_head(list
) == NULL
||
1575 dmu_objset_userused_enabled(os
));
1576 avl_create(&cache
.uqc_user_deltas
, userquota_compare
,
1577 sizeof (userquota_node_t
), offsetof(userquota_node_t
, uqn_node
));
1578 avl_create(&cache
.uqc_group_deltas
, userquota_compare
,
1579 sizeof (userquota_node_t
), offsetof(userquota_node_t
, uqn_node
));
1581 while ((dn
= multilist_sublist_head(list
)) != NULL
) {
1583 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
1584 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
1585 dn
->dn_phys
->dn_flags
&
1586 DNODE_FLAG_USERUSED_ACCOUNTED
);
1588 flags
= dn
->dn_id_flags
;
1590 if (flags
& DN_ID_OLD_EXIST
) {
1591 do_userquota_update(&cache
,
1592 dn
->dn_oldused
, dn
->dn_oldflags
,
1593 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
);
1594 do_userobjquota_update(&cache
, dn
->dn_oldflags
,
1595 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
);
1597 if (flags
& DN_ID_NEW_EXIST
) {
1598 do_userquota_update(&cache
,
1599 DN_USED_BYTES(dn
->dn_phys
), dn
->dn_phys
->dn_flags
,
1600 dn
->dn_newuid
, dn
->dn_newgid
, B_FALSE
);
1601 do_userobjquota_update(&cache
, dn
->dn_phys
->dn_flags
,
1602 dn
->dn_newuid
, dn
->dn_newgid
, B_FALSE
);
1605 mutex_enter(&dn
->dn_mtx
);
1607 dn
->dn_oldflags
= 0;
1608 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
1609 dn
->dn_olduid
= dn
->dn_newuid
;
1610 dn
->dn_oldgid
= dn
->dn_newgid
;
1611 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1612 if (dn
->dn_bonuslen
== 0)
1613 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1615 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1617 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
1618 mutex_exit(&dn
->dn_mtx
);
1620 multilist_sublist_remove(list
, dn
);
1621 dnode_rele(dn
, os
->os_synced_dnodes
);
1623 do_userquota_cacheflush(os
, &cache
, tx
);
1624 multilist_sublist_unlock(list
);
1625 kmem_free(uua
, sizeof (*uua
));
1629 dmu_objset_do_userquota_updates(objset_t
*os
, dmu_tx_t
*tx
)
1631 if (!dmu_objset_userused_enabled(os
))
1634 /* Allocate the user/groupused objects if necessary. */
1635 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
1636 VERIFY0(zap_create_claim(os
,
1637 DMU_USERUSED_OBJECT
,
1638 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1639 VERIFY0(zap_create_claim(os
,
1640 DMU_GROUPUSED_OBJECT
,
1641 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1645 i
< multilist_get_num_sublists(os
->os_synced_dnodes
); i
++) {
1646 userquota_updates_arg_t
*uua
=
1647 kmem_alloc(sizeof (*uua
), KM_SLEEP
);
1649 uua
->uua_sublist_idx
= i
;
1651 /* note: caller does taskq_wait() */
1652 (void) taskq_dispatch(dmu_objset_pool(os
)->dp_sync_taskq
,
1653 userquota_updates_task
, uua
, 0);
1654 /* callback frees uua */
1659 * Returns a pointer to data to find uid/gid from
1661 * If a dirty record for transaction group that is syncing can't
1662 * be found then NULL is returned. In the NULL case it is assumed
1663 * the uid/gid aren't changing.
1666 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
1668 dbuf_dirty_record_t
*dr
, **drp
;
1671 if (db
->db_dirtycnt
== 0)
1672 return (db
->db
.db_data
); /* Nothing is changing */
1674 for (drp
= &db
->db_last_dirty
; (dr
= *drp
) != NULL
; drp
= &dr
->dr_next
)
1675 if (dr
->dr_txg
== tx
->tx_txg
)
1683 DB_DNODE_ENTER(dr
->dr_dbuf
);
1684 dn
= DB_DNODE(dr
->dr_dbuf
);
1686 if (dn
->dn_bonuslen
== 0 &&
1687 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
1688 data
= dr
->dt
.dl
.dr_data
->b_data
;
1690 data
= dr
->dt
.dl
.dr_data
;
1692 DB_DNODE_EXIT(dr
->dr_dbuf
);
1699 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
1701 objset_t
*os
= dn
->dn_objset
;
1703 dmu_buf_impl_t
*db
= NULL
;
1704 uint64_t *user
= NULL
;
1705 uint64_t *group
= NULL
;
1706 int flags
= dn
->dn_id_flags
;
1708 boolean_t have_spill
= B_FALSE
;
1710 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
1713 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
1714 DN_ID_CHKED_SPILL
)))
1717 if (before
&& dn
->dn_bonuslen
!= 0)
1718 data
= DN_BONUS(dn
->dn_phys
);
1719 else if (!before
&& dn
->dn_bonuslen
!= 0) {
1722 mutex_enter(&db
->db_mtx
);
1723 data
= dmu_objset_userquota_find_data(db
, tx
);
1725 data
= DN_BONUS(dn
->dn_phys
);
1727 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
1730 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
1731 rf
|= DB_RF_HAVESTRUCT
;
1732 error
= dmu_spill_hold_by_dnode(dn
,
1733 rf
| DB_RF_MUST_SUCCEED
,
1734 FTAG
, (dmu_buf_t
**)&db
);
1736 mutex_enter(&db
->db_mtx
);
1737 data
= (before
) ? db
->db
.db_data
:
1738 dmu_objset_userquota_find_data(db
, tx
);
1739 have_spill
= B_TRUE
;
1741 mutex_enter(&dn
->dn_mtx
);
1742 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1743 mutex_exit(&dn
->dn_mtx
);
1749 user
= &dn
->dn_olduid
;
1750 group
= &dn
->dn_oldgid
;
1752 user
= &dn
->dn_newuid
;
1753 group
= &dn
->dn_newgid
;
1757 * Must always call the callback in case the object
1758 * type has changed and that type isn't an object type to track
1760 error
= used_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
,
1764 * Preserve existing uid/gid when the callback can't determine
1765 * what the new uid/gid are and the callback returned EEXIST.
1766 * The EEXIST error tells us to just use the existing uid/gid.
1767 * If we don't know what the old values are then just assign
1768 * them to 0, since that is a new file being created.
1770 if (!before
&& data
== NULL
&& error
== EEXIST
) {
1771 if (flags
& DN_ID_OLD_EXIST
) {
1772 dn
->dn_newuid
= dn
->dn_olduid
;
1773 dn
->dn_newgid
= dn
->dn_oldgid
;
1782 mutex_exit(&db
->db_mtx
);
1784 mutex_enter(&dn
->dn_mtx
);
1785 if (error
== 0 && before
)
1786 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1787 if (error
== 0 && !before
)
1788 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
1791 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1793 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1795 mutex_exit(&dn
->dn_mtx
);
1797 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
1801 dmu_objset_userspace_present(objset_t
*os
)
1803 return (os
->os_phys
->os_flags
&
1804 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
1808 dmu_objset_userobjspace_present(objset_t
*os
)
1810 return (os
->os_phys
->os_flags
&
1811 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
);
1815 dmu_objset_space_upgrade(objset_t
*os
)
1821 * We simply need to mark every object dirty, so that it will be
1822 * synced out and now accounted. If this is called
1823 * concurrently, or if we already did some work before crashing,
1824 * that's fine, since we track each object's accounted state
1828 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
1833 mutex_enter(&os
->os_upgrade_lock
);
1834 if (os
->os_upgrade_exit
)
1835 err
= SET_ERROR(EINTR
);
1836 mutex_exit(&os
->os_upgrade_lock
);
1840 if (issig(JUSTLOOKING
) && issig(FORREAL
))
1841 return (SET_ERROR(EINTR
));
1843 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
1846 tx
= dmu_tx_create(os
);
1847 dmu_tx_hold_bonus(tx
, obj
);
1848 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
1853 dmu_buf_will_dirty(db
, tx
);
1854 dmu_buf_rele(db
, FTAG
);
1861 dmu_objset_userspace_upgrade(objset_t
*os
)
1865 if (dmu_objset_userspace_present(os
))
1867 if (dmu_objset_is_snapshot(os
))
1868 return (SET_ERROR(EINVAL
));
1869 if (!dmu_objset_userused_enabled(os
))
1870 return (SET_ERROR(ENOTSUP
));
1872 err
= dmu_objset_space_upgrade(os
);
1876 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
1877 txg_wait_synced(dmu_objset_pool(os
), 0);
1882 dmu_objset_userobjspace_upgrade_cb(objset_t
*os
)
1886 if (dmu_objset_userobjspace_present(os
))
1888 if (dmu_objset_is_snapshot(os
))
1889 return (SET_ERROR(EINVAL
));
1890 if (!dmu_objset_userobjused_enabled(os
))
1891 return (SET_ERROR(ENOTSUP
));
1893 dmu_objset_ds(os
)->ds_feature_activation_needed
[
1894 SPA_FEATURE_USEROBJ_ACCOUNTING
] = B_TRUE
;
1896 err
= dmu_objset_space_upgrade(os
);
1900 os
->os_flags
|= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
;
1901 txg_wait_synced(dmu_objset_pool(os
), 0);
1906 dmu_objset_userobjspace_upgrade(objset_t
*os
)
1908 dmu_objset_upgrade(os
, dmu_objset_userobjspace_upgrade_cb
);
1912 dmu_objset_userobjspace_upgradable(objset_t
*os
)
1914 return (dmu_objset_type(os
) == DMU_OST_ZFS
&&
1915 !dmu_objset_is_snapshot(os
) &&
1916 dmu_objset_userobjused_enabled(os
) &&
1917 !dmu_objset_userobjspace_present(os
));
1921 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
1922 uint64_t *usedobjsp
, uint64_t *availobjsp
)
1924 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
1925 usedobjsp
, availobjsp
);
1929 dmu_objset_fsid_guid(objset_t
*os
)
1931 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
1935 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
1937 stat
->dds_type
= os
->os_phys
->os_type
;
1938 if (os
->os_dsl_dataset
)
1939 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
1943 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
1945 ASSERT(os
->os_dsl_dataset
||
1946 os
->os_phys
->os_type
== DMU_OST_META
);
1948 if (os
->os_dsl_dataset
!= NULL
)
1949 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
1951 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
1952 os
->os_phys
->os_type
);
1953 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
1954 dmu_objset_userspace_present(os
));
1958 dmu_objset_is_snapshot(objset_t
*os
)
1960 if (os
->os_dsl_dataset
!= NULL
)
1961 return (os
->os_dsl_dataset
->ds_is_snapshot
);
1967 dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
, int maxlen
,
1968 boolean_t
*conflict
)
1970 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1973 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1974 return (SET_ERROR(ENOENT
));
1976 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1977 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, name
, 8, 1, &ignored
,
1978 MT_NORMALIZE
, real
, maxlen
, conflict
));
1982 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
1983 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
1985 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1986 zap_cursor_t cursor
;
1987 zap_attribute_t attr
;
1989 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
1991 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
1992 return (SET_ERROR(ENOENT
));
1994 zap_cursor_init_serialized(&cursor
,
1995 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
1996 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, *offp
);
1998 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
1999 zap_cursor_fini(&cursor
);
2000 return (SET_ERROR(ENOENT
));
2003 if (strlen(attr
.za_name
) + 1 > namelen
) {
2004 zap_cursor_fini(&cursor
);
2005 return (SET_ERROR(ENAMETOOLONG
));
2008 (void) strcpy(name
, attr
.za_name
);
2010 *idp
= attr
.za_first_integer
;
2012 *case_conflict
= attr
.za_normalization_conflict
;
2013 zap_cursor_advance(&cursor
);
2014 *offp
= zap_cursor_serialize(&cursor
);
2015 zap_cursor_fini(&cursor
);
2021 dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *value
)
2023 return (dsl_dataset_snap_lookup(os
->os_dsl_dataset
, name
, value
));
2027 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
2028 uint64_t *idp
, uint64_t *offp
)
2030 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
2031 zap_cursor_t cursor
;
2032 zap_attribute_t attr
;
2034 /* there is no next dir on a snapshot! */
2035 if (os
->os_dsl_dataset
->ds_object
!=
2036 dsl_dir_phys(dd
)->dd_head_dataset_obj
)
2037 return (SET_ERROR(ENOENT
));
2039 zap_cursor_init_serialized(&cursor
,
2040 dd
->dd_pool
->dp_meta_objset
,
2041 dsl_dir_phys(dd
)->dd_child_dir_zapobj
, *offp
);
2043 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
2044 zap_cursor_fini(&cursor
);
2045 return (SET_ERROR(ENOENT
));
2048 if (strlen(attr
.za_name
) + 1 > namelen
) {
2049 zap_cursor_fini(&cursor
);
2050 return (SET_ERROR(ENAMETOOLONG
));
2053 (void) strcpy(name
, attr
.za_name
);
2055 *idp
= attr
.za_first_integer
;
2056 zap_cursor_advance(&cursor
);
2057 *offp
= zap_cursor_serialize(&cursor
);
2058 zap_cursor_fini(&cursor
);
2063 typedef struct dmu_objset_find_ctx
{
2067 char *dc_ddname
; /* last component of ddobj's name */
2068 int (*dc_func
)(dsl_pool_t
*, dsl_dataset_t
*, void *);
2071 kmutex_t
*dc_error_lock
;
2073 } dmu_objset_find_ctx_t
;
2076 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t
*dcp
)
2078 dsl_pool_t
*dp
= dcp
->dc_dp
;
2079 dmu_objset_find_ctx_t
*child_dcp
;
2083 zap_attribute_t
*attr
;
2087 /* don't process if there already was an error */
2088 if (*dcp
->dc_error
!= 0)
2092 * Note: passing the name (dc_ddname) here is optional, but it
2093 * improves performance because we don't need to call
2094 * zap_value_search() to determine the name.
2096 err
= dsl_dir_hold_obj(dp
, dcp
->dc_ddobj
, dcp
->dc_ddname
, FTAG
, &dd
);
2100 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2101 if (dd
->dd_myname
[0] == '$') {
2102 dsl_dir_rele(dd
, FTAG
);
2106 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
2107 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
2110 * Iterate over all children.
2112 if (dcp
->dc_flags
& DS_FIND_CHILDREN
) {
2113 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2114 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
2115 zap_cursor_retrieve(&zc
, attr
) == 0;
2116 (void) zap_cursor_advance(&zc
)) {
2117 ASSERT3U(attr
->za_integer_length
, ==,
2119 ASSERT3U(attr
->za_num_integers
, ==, 1);
2122 kmem_alloc(sizeof (*child_dcp
), KM_SLEEP
);
2124 child_dcp
->dc_ddobj
= attr
->za_first_integer
;
2125 child_dcp
->dc_ddname
= spa_strdup(attr
->za_name
);
2126 if (dcp
->dc_tq
!= NULL
)
2127 (void) taskq_dispatch(dcp
->dc_tq
,
2128 dmu_objset_find_dp_cb
, child_dcp
, TQ_SLEEP
);
2130 dmu_objset_find_dp_impl(child_dcp
);
2132 zap_cursor_fini(&zc
);
2136 * Iterate over all snapshots.
2138 if (dcp
->dc_flags
& DS_FIND_SNAPSHOTS
) {
2140 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2145 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
2146 dsl_dataset_rele(ds
, FTAG
);
2148 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
2149 zap_cursor_retrieve(&zc
, attr
) == 0;
2150 (void) zap_cursor_advance(&zc
)) {
2151 ASSERT3U(attr
->za_integer_length
, ==,
2153 ASSERT3U(attr
->za_num_integers
, ==, 1);
2155 err
= dsl_dataset_hold_obj(dp
,
2156 attr
->za_first_integer
, FTAG
, &ds
);
2159 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
2160 dsl_dataset_rele(ds
, FTAG
);
2164 zap_cursor_fini(&zc
);
2168 kmem_free(attr
, sizeof (zap_attribute_t
));
2171 dsl_dir_rele(dd
, FTAG
);
2178 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2181 * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2182 * that the dir will remain cached, and we won't have to re-instantiate
2183 * it (which could be expensive due to finding its name via
2184 * zap_value_search()).
2186 dsl_dir_rele(dd
, FTAG
);
2189 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
2190 dsl_dataset_rele(ds
, FTAG
);
2194 mutex_enter(dcp
->dc_error_lock
);
2195 /* only keep first error */
2196 if (*dcp
->dc_error
== 0)
2197 *dcp
->dc_error
= err
;
2198 mutex_exit(dcp
->dc_error_lock
);
2201 if (dcp
->dc_ddname
!= NULL
)
2202 spa_strfree(dcp
->dc_ddname
);
2203 kmem_free(dcp
, sizeof (*dcp
));
2207 dmu_objset_find_dp_cb(void *arg
)
2209 dmu_objset_find_ctx_t
*dcp
= arg
;
2210 dsl_pool_t
*dp
= dcp
->dc_dp
;
2213 * We need to get a pool_config_lock here, as there are several
2214 * asssert(pool_config_held) down the stack. Getting a lock via
2215 * dsl_pool_config_enter is risky, as it might be stalled by a
2216 * pending writer. This would deadlock, as the write lock can
2217 * only be granted when our parent thread gives up the lock.
2218 * The _prio interface gives us priority over a pending writer.
2220 dsl_pool_config_enter_prio(dp
, FTAG
);
2222 dmu_objset_find_dp_impl(dcp
);
2224 dsl_pool_config_exit(dp
, FTAG
);
2228 * Find objsets under and including ddobj, call func(ds) on each.
2229 * The order for the enumeration is completely undefined.
2230 * func is called with dsl_pool_config held.
2233 dmu_objset_find_dp(dsl_pool_t
*dp
, uint64_t ddobj
,
2234 int func(dsl_pool_t
*, dsl_dataset_t
*, void *), void *arg
, int flags
)
2239 dmu_objset_find_ctx_t
*dcp
;
2242 mutex_init(&err_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2243 dcp
= kmem_alloc(sizeof (*dcp
), KM_SLEEP
);
2246 dcp
->dc_ddobj
= ddobj
;
2247 dcp
->dc_ddname
= NULL
;
2248 dcp
->dc_func
= func
;
2250 dcp
->dc_flags
= flags
;
2251 dcp
->dc_error_lock
= &err_lock
;
2252 dcp
->dc_error
= &error
;
2254 if ((flags
& DS_FIND_SERIALIZE
) || dsl_pool_config_held_writer(dp
)) {
2256 * In case a write lock is held we can't make use of
2257 * parallelism, as down the stack of the worker threads
2258 * the lock is asserted via dsl_pool_config_held.
2259 * In case of a read lock this is solved by getting a read
2260 * lock in each worker thread, which isn't possible in case
2261 * of a writer lock. So we fall back to the synchronous path
2263 * In the future it might be possible to get some magic into
2264 * dsl_pool_config_held in a way that it returns true for
2265 * the worker threads so that a single lock held from this
2266 * thread suffices. For now, stay single threaded.
2268 dmu_objset_find_dp_impl(dcp
);
2269 mutex_destroy(&err_lock
);
2274 ntasks
= dmu_find_threads
;
2276 ntasks
= vdev_count_leaves(dp
->dp_spa
) * 4;
2277 tq
= taskq_create("dmu_objset_find", ntasks
, maxclsyspri
, ntasks
,
2280 kmem_free(dcp
, sizeof (*dcp
));
2281 mutex_destroy(&err_lock
);
2283 return (SET_ERROR(ENOMEM
));
2287 /* dcp will be freed by task */
2288 (void) taskq_dispatch(tq
, dmu_objset_find_dp_cb
, dcp
, TQ_SLEEP
);
2291 * PORTING: this code relies on the property of taskq_wait to wait
2292 * until no more tasks are queued and no more tasks are active. As
2293 * we always queue new tasks from within other tasks, task_wait
2294 * reliably waits for the full recursion to finish, even though we
2295 * enqueue new tasks after taskq_wait has been called.
2296 * On platforms other than illumos, taskq_wait may not have this
2301 mutex_destroy(&err_lock
);
2307 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2308 * The dp_config_rwlock must not be held when this is called, and it
2309 * will not be held when the callback is called.
2310 * Therefore this function should only be used when the pool is not changing
2311 * (e.g. in syncing context), or the callback can deal with the possible races.
2314 dmu_objset_find_impl(spa_t
*spa
, const char *name
,
2315 int func(const char *, void *), void *arg
, int flags
)
2318 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2321 zap_attribute_t
*attr
;
2326 dsl_pool_config_enter(dp
, FTAG
);
2328 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, NULL
);
2330 dsl_pool_config_exit(dp
, FTAG
);
2334 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2335 if (dd
->dd_myname
[0] == '$') {
2336 dsl_dir_rele(dd
, FTAG
);
2337 dsl_pool_config_exit(dp
, FTAG
);
2341 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
2342 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
2345 * Iterate over all children.
2347 if (flags
& DS_FIND_CHILDREN
) {
2348 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2349 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
2350 zap_cursor_retrieve(&zc
, attr
) == 0;
2351 (void) zap_cursor_advance(&zc
)) {
2352 ASSERT3U(attr
->za_integer_length
, ==,
2354 ASSERT3U(attr
->za_num_integers
, ==, 1);
2356 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
2357 dsl_pool_config_exit(dp
, FTAG
);
2358 err
= dmu_objset_find_impl(spa
, child
,
2360 dsl_pool_config_enter(dp
, FTAG
);
2365 zap_cursor_fini(&zc
);
2368 dsl_dir_rele(dd
, FTAG
);
2369 dsl_pool_config_exit(dp
, FTAG
);
2370 kmem_free(attr
, sizeof (zap_attribute_t
));
2376 * Iterate over all snapshots.
2378 if (flags
& DS_FIND_SNAPSHOTS
) {
2379 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2384 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
2385 dsl_dataset_rele(ds
, FTAG
);
2387 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
2388 zap_cursor_retrieve(&zc
, attr
) == 0;
2389 (void) zap_cursor_advance(&zc
)) {
2390 ASSERT3U(attr
->za_integer_length
, ==,
2392 ASSERT3U(attr
->za_num_integers
, ==, 1);
2394 child
= kmem_asprintf("%s@%s",
2395 name
, attr
->za_name
);
2396 dsl_pool_config_exit(dp
, FTAG
);
2397 err
= func(child
, arg
);
2398 dsl_pool_config_enter(dp
, FTAG
);
2403 zap_cursor_fini(&zc
);
2407 dsl_dir_rele(dd
, FTAG
);
2408 kmem_free(attr
, sizeof (zap_attribute_t
));
2409 dsl_pool_config_exit(dp
, FTAG
);
2414 /* Apply to self. */
2415 return (func(name
, arg
));
2419 * See comment above dmu_objset_find_impl().
2422 dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
2428 error
= spa_open(name
, &spa
, FTAG
);
2431 error
= dmu_objset_find_impl(spa
, name
, func
, arg
, flags
);
2432 spa_close(spa
, FTAG
);
2437 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
2439 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2440 os
->os_user_ptr
= user_ptr
;
2444 dmu_objset_get_user(objset_t
*os
)
2446 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2447 return (os
->os_user_ptr
);
2451 * Determine name of filesystem, given name of snapshot.
2452 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2455 dmu_fsname(const char *snapname
, char *buf
)
2457 char *atp
= strchr(snapname
, '@');
2459 return (SET_ERROR(EINVAL
));
2460 if (atp
- snapname
>= ZFS_MAX_DATASET_NAME_LEN
)
2461 return (SET_ERROR(ENAMETOOLONG
));
2462 (void) strlcpy(buf
, snapname
, atp
- snapname
+ 1);
2467 * Call when we think we're going to write/free space in open context to track
2468 * the amount of dirty data in the open txg, which is also the amount
2469 * of memory that can not be evicted until this txg syncs.
2472 dmu_objset_willuse_space(objset_t
*os
, int64_t space
, dmu_tx_t
*tx
)
2474 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
2475 int64_t aspace
= spa_get_worst_case_asize(os
->os_spa
, space
);
2478 dsl_dir_willuse_space(ds
->ds_dir
, aspace
, tx
);
2479 dsl_pool_dirty_space(dmu_tx_pool(tx
), space
, tx
);
2483 #if defined(_KERNEL) && defined(HAVE_SPL)
2484 EXPORT_SYMBOL(dmu_objset_zil
);
2485 EXPORT_SYMBOL(dmu_objset_pool
);
2486 EXPORT_SYMBOL(dmu_objset_ds
);
2487 EXPORT_SYMBOL(dmu_objset_type
);
2488 EXPORT_SYMBOL(dmu_objset_name
);
2489 EXPORT_SYMBOL(dmu_objset_hold
);
2490 EXPORT_SYMBOL(dmu_objset_own
);
2491 EXPORT_SYMBOL(dmu_objset_rele
);
2492 EXPORT_SYMBOL(dmu_objset_disown
);
2493 EXPORT_SYMBOL(dmu_objset_from_ds
);
2494 EXPORT_SYMBOL(dmu_objset_create
);
2495 EXPORT_SYMBOL(dmu_objset_clone
);
2496 EXPORT_SYMBOL(dmu_objset_stats
);
2497 EXPORT_SYMBOL(dmu_objset_fast_stat
);
2498 EXPORT_SYMBOL(dmu_objset_spa
);
2499 EXPORT_SYMBOL(dmu_objset_space
);
2500 EXPORT_SYMBOL(dmu_objset_fsid_guid
);
2501 EXPORT_SYMBOL(dmu_objset_find
);
2502 EXPORT_SYMBOL(dmu_objset_byteswap
);
2503 EXPORT_SYMBOL(dmu_objset_evict_dbufs
);
2504 EXPORT_SYMBOL(dmu_objset_snap_cmtime
);
2505 EXPORT_SYMBOL(dmu_objset_dnodesize
);
2507 EXPORT_SYMBOL(dmu_objset_sync
);
2508 EXPORT_SYMBOL(dmu_objset_is_dirty
);
2509 EXPORT_SYMBOL(dmu_objset_create_impl
);
2510 EXPORT_SYMBOL(dmu_objset_open_impl
);
2511 EXPORT_SYMBOL(dmu_objset_evict
);
2512 EXPORT_SYMBOL(dmu_objset_register_type
);
2513 EXPORT_SYMBOL(dmu_objset_do_userquota_updates
);
2514 EXPORT_SYMBOL(dmu_objset_userquota_get_ids
);
2515 EXPORT_SYMBOL(dmu_objset_userused_enabled
);
2516 EXPORT_SYMBOL(dmu_objset_userspace_upgrade
);
2517 EXPORT_SYMBOL(dmu_objset_userspace_present
);
2518 EXPORT_SYMBOL(dmu_objset_userobjused_enabled
);
2519 EXPORT_SYMBOL(dmu_objset_userobjspace_upgrade
);
2520 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable
);
2521 EXPORT_SYMBOL(dmu_objset_userobjspace_present
);