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 https://opensource.org/licenses/CDDL-1.0.
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, 2020 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.
31 * Copyright (c) 2017 Open-E, Inc. All Rights Reserved.
32 * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
33 * Copyright (c) 2019, Klara Inc.
34 * Copyright (c) 2019, Allan Jude
35 * Copyright (c) 2022 Hewlett Packard Enterprise Development LP.
38 /* Portions Copyright 2010 Robert Milkowski */
41 #include <sys/zfs_context.h>
42 #include <sys/dmu_objset.h>
43 #include <sys/dsl_dir.h>
44 #include <sys/dsl_dataset.h>
45 #include <sys/dsl_prop.h>
46 #include <sys/dsl_pool.h>
47 #include <sys/dsl_synctask.h>
48 #include <sys/dsl_deleg.h>
49 #include <sys/dnode.h>
52 #include <sys/dmu_tx.h>
55 #include <sys/dmu_impl.h>
56 #include <sys/zfs_ioctl.h>
58 #include <sys/zfs_onexit.h>
59 #include <sys/dsl_destroy.h>
61 #include <sys/zfeature.h>
62 #include <sys/policy.h>
63 #include <sys/spa_impl.h>
64 #include <sys/dmu_recv.h>
65 #include <sys/zfs_project.h>
66 #include "zfs_namecheck.h"
67 #include <sys/vdev_impl.h>
71 * Needed to close a window in dnode_move() that allows the objset to be freed
72 * before it can be safely accessed.
77 * Tunable to overwrite the maximum number of threads for the parallelization
78 * of dmu_objset_find_dp, needed to speed up the import of pools with many
80 * Default is 4 times the number of leaf vdevs.
82 static const int dmu_find_threads
= 0;
85 * Backfill lower metadnode objects after this many have been freed.
86 * Backfilling negatively impacts object creation rates, so only do it
87 * if there are enough holes to fill.
89 static const int dmu_rescan_dnode_threshold
= 1 << DN_MAX_INDBLKSHIFT
;
91 static const char *upgrade_tag
= "upgrade_tag";
93 static void dmu_objset_find_dp_cb(void *arg
);
95 static void dmu_objset_upgrade(objset_t
*os
, dmu_objset_upgrade_cb_t cb
);
96 static void dmu_objset_upgrade_stop(objset_t
*os
);
101 rw_init(&os_lock
, NULL
, RW_DEFAULT
, NULL
);
105 dmu_objset_fini(void)
107 rw_destroy(&os_lock
);
111 dmu_objset_spa(objset_t
*os
)
117 dmu_objset_zil(objset_t
*os
)
123 dmu_objset_pool(objset_t
*os
)
127 if ((ds
= os
->os_dsl_dataset
) != NULL
&& ds
->ds_dir
)
128 return (ds
->ds_dir
->dd_pool
);
130 return (spa_get_dsl(os
->os_spa
));
134 dmu_objset_ds(objset_t
*os
)
136 return (os
->os_dsl_dataset
);
140 dmu_objset_type(objset_t
*os
)
142 return (os
->os_phys
->os_type
);
146 dmu_objset_name(objset_t
*os
, char *buf
)
148 dsl_dataset_name(os
->os_dsl_dataset
, buf
);
152 dmu_objset_id(objset_t
*os
)
154 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
156 return (ds
? ds
->ds_object
: 0);
160 dmu_objset_dnodesize(objset_t
*os
)
162 return (os
->os_dnodesize
);
166 dmu_objset_syncprop(objset_t
*os
)
168 return (os
->os_sync
);
172 dmu_objset_logbias(objset_t
*os
)
174 return (os
->os_logbias
);
178 checksum_changed_cb(void *arg
, uint64_t newval
)
183 * Inheritance should have been done by now.
185 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
187 os
->os_checksum
= zio_checksum_select(newval
, ZIO_CHECKSUM_ON_VALUE
);
191 compression_changed_cb(void *arg
, uint64_t newval
)
196 * Inheritance and range checking should have been done by now.
198 ASSERT(newval
!= ZIO_COMPRESS_INHERIT
);
200 os
->os_compress
= zio_compress_select(os
->os_spa
,
201 ZIO_COMPRESS_ALGO(newval
), ZIO_COMPRESS_ON
);
202 os
->os_complevel
= zio_complevel_select(os
->os_spa
, os
->os_compress
,
203 ZIO_COMPRESS_LEVEL(newval
), ZIO_COMPLEVEL_DEFAULT
);
207 copies_changed_cb(void *arg
, uint64_t newval
)
212 * Inheritance and range checking should have been done by now.
215 ASSERT(newval
<= spa_max_replication(os
->os_spa
));
217 os
->os_copies
= newval
;
221 dedup_changed_cb(void *arg
, uint64_t newval
)
224 spa_t
*spa
= os
->os_spa
;
225 enum zio_checksum checksum
;
228 * Inheritance should have been done by now.
230 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
232 checksum
= zio_checksum_dedup_select(spa
, newval
, ZIO_CHECKSUM_OFF
);
234 os
->os_dedup_checksum
= checksum
& ZIO_CHECKSUM_MASK
;
235 os
->os_dedup_verify
= !!(checksum
& ZIO_CHECKSUM_VERIFY
);
239 primary_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_primary_cache
= newval
;
253 secondary_cache_changed_cb(void *arg
, uint64_t newval
)
258 * Inheritance and range checking should have been done by now.
260 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
261 newval
== ZFS_CACHE_METADATA
);
263 os
->os_secondary_cache
= newval
;
267 sync_changed_cb(void *arg
, uint64_t newval
)
272 * Inheritance and range checking should have been done by now.
274 ASSERT(newval
== ZFS_SYNC_STANDARD
|| newval
== ZFS_SYNC_ALWAYS
||
275 newval
== ZFS_SYNC_DISABLED
);
277 os
->os_sync
= newval
;
279 zil_set_sync(os
->os_zil
, newval
);
283 redundant_metadata_changed_cb(void *arg
, uint64_t newval
)
288 * Inheritance and range checking should have been done by now.
290 ASSERT(newval
== ZFS_REDUNDANT_METADATA_ALL
||
291 newval
== ZFS_REDUNDANT_METADATA_MOST
||
292 newval
== ZFS_REDUNDANT_METADATA_SOME
||
293 newval
== ZFS_REDUNDANT_METADATA_NONE
);
295 os
->os_redundant_metadata
= newval
;
299 dnodesize_changed_cb(void *arg
, uint64_t newval
)
304 case ZFS_DNSIZE_LEGACY
:
305 os
->os_dnodesize
= DNODE_MIN_SIZE
;
307 case ZFS_DNSIZE_AUTO
:
309 * Choose a dnode size that will work well for most
310 * workloads if the user specified "auto". Future code
311 * improvements could dynamically select a dnode size
312 * based on observed workload patterns.
314 os
->os_dnodesize
= DNODE_MIN_SIZE
* 2;
321 os
->os_dnodesize
= newval
;
327 smallblk_changed_cb(void *arg
, uint64_t newval
)
332 * Inheritance and range checking should have been done by now.
334 ASSERT(newval
<= SPA_MAXBLOCKSIZE
);
335 ASSERT(ISP2(newval
));
337 os
->os_zpl_special_smallblock
= newval
;
341 logbias_changed_cb(void *arg
, uint64_t newval
)
345 ASSERT(newval
== ZFS_LOGBIAS_LATENCY
||
346 newval
== ZFS_LOGBIAS_THROUGHPUT
);
347 os
->os_logbias
= newval
;
349 zil_set_logbias(os
->os_zil
, newval
);
353 recordsize_changed_cb(void *arg
, uint64_t newval
)
357 os
->os_recordsize
= newval
;
361 dmu_objset_byteswap(void *buf
, size_t size
)
363 objset_phys_t
*osp
= buf
;
365 ASSERT(size
== OBJSET_PHYS_SIZE_V1
|| size
== OBJSET_PHYS_SIZE_V2
||
366 size
== sizeof (objset_phys_t
));
367 dnode_byteswap(&osp
->os_meta_dnode
);
368 byteswap_uint64_array(&osp
->os_zil_header
, sizeof (zil_header_t
));
369 osp
->os_type
= BSWAP_64(osp
->os_type
);
370 osp
->os_flags
= BSWAP_64(osp
->os_flags
);
371 if (size
>= OBJSET_PHYS_SIZE_V2
) {
372 dnode_byteswap(&osp
->os_userused_dnode
);
373 dnode_byteswap(&osp
->os_groupused_dnode
);
374 if (size
>= sizeof (objset_phys_t
))
375 dnode_byteswap(&osp
->os_projectused_dnode
);
380 * The hash is a CRC-based hash of the objset_t pointer and the object number.
383 dnode_hash(const objset_t
*os
, uint64_t obj
)
385 uintptr_t osv
= (uintptr_t)os
;
386 uint64_t crc
= -1ULL;
388 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
390 * The low 6 bits of the pointer don't have much entropy, because
391 * the objset_t is larger than 2^6 bytes long.
393 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (osv
>> 6)) & 0xFF];
394 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 0)) & 0xFF];
395 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 8)) & 0xFF];
396 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 16)) & 0xFF];
398 crc
^= (osv
>>14) ^ (obj
>>24);
404 dnode_multilist_index_func(multilist_t
*ml
, void *obj
)
409 * The low order bits of the hash value are thought to be
410 * distributed evenly. Otherwise, in the case that the multilist
411 * has a power of two number of sublists, each sublists' usage
412 * would not be evenly distributed. In this context full 64bit
413 * division would be a waste of time, so limit it to 32 bits.
415 return ((unsigned int)dnode_hash(dn
->dn_objset
, dn
->dn_object
) %
416 multilist_get_num_sublists(ml
));
419 static inline boolean_t
420 dmu_os_is_l2cacheable(objset_t
*os
)
423 zfs_cache_type_t cache
= os
->os_secondary_cache
;
424 blkptr_t
*bp
= os
->os_rootbp
;
426 if (bp
!= NULL
&& !BP_IS_HOLE(bp
)) {
427 uint64_t vdev
= DVA_GET_VDEV(bp
->blk_dva
);
428 vdev_t
*rvd
= os
->os_spa
->spa_root_vdev
;
430 if (vdev
< rvd
->vdev_children
)
431 vd
= rvd
->vdev_child
[vdev
];
433 if (cache
== ZFS_CACHE_ALL
|| cache
== ZFS_CACHE_METADATA
) {
437 if ((vd
->vdev_alloc_bias
!= VDEV_BIAS_SPECIAL
&&
438 vd
->vdev_alloc_bias
!= VDEV_BIAS_DEDUP
) ||
439 l2arc_exclude_special
== 0)
448 * Instantiates the objset_t in-memory structure corresponding to the
449 * objset_phys_t that's pointed to by the specified blkptr_t.
452 dmu_objset_open_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
458 ASSERT(ds
== NULL
|| MUTEX_HELD(&ds
->ds_opening_lock
));
459 ASSERT(!BP_IS_REDACTED(bp
));
462 * We need the pool config lock to get properties.
464 ASSERT(ds
== NULL
|| dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
467 * The $ORIGIN dataset (if it exists) doesn't have an associated
468 * objset, so there's no reason to open it. The $ORIGIN dataset
469 * will not exist on pools older than SPA_VERSION_ORIGIN.
471 if (ds
!= NULL
&& spa_get_dsl(spa
) != NULL
&&
472 spa_get_dsl(spa
)->dp_origin_snap
!= NULL
) {
473 ASSERT3P(ds
->ds_dir
, !=,
474 spa_get_dsl(spa
)->dp_origin_snap
->ds_dir
);
477 os
= kmem_zalloc(sizeof (objset_t
), KM_SLEEP
);
478 os
->os_dsl_dataset
= ds
;
481 if (!BP_IS_HOLE(os
->os_rootbp
)) {
482 arc_flags_t aflags
= ARC_FLAG_WAIT
;
485 zio_flag_t zio_flags
= ZIO_FLAG_CANFAIL
;
486 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
487 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
489 if (dmu_os_is_l2cacheable(os
))
490 aflags
|= ARC_FLAG_L2CACHE
;
492 if (ds
!= NULL
&& ds
->ds_dir
->dd_crypto_obj
!= 0) {
493 ASSERT3U(BP_GET_COMPRESS(bp
), ==, ZIO_COMPRESS_OFF
);
494 ASSERT(BP_IS_AUTHENTICATED(bp
));
495 zio_flags
|= ZIO_FLAG_RAW
;
498 dprintf_bp(os
->os_rootbp
, "reading %s", "");
499 err
= arc_read(NULL
, spa
, os
->os_rootbp
,
500 arc_getbuf_func
, &os
->os_phys_buf
,
501 ZIO_PRIORITY_SYNC_READ
, zio_flags
, &aflags
, &zb
);
503 kmem_free(os
, sizeof (objset_t
));
504 /* convert checksum errors into IO errors */
506 err
= SET_ERROR(EIO
);
510 if (spa_version(spa
) < SPA_VERSION_USERSPACE
)
511 size
= OBJSET_PHYS_SIZE_V1
;
512 else if (!spa_feature_is_enabled(spa
,
513 SPA_FEATURE_PROJECT_QUOTA
))
514 size
= OBJSET_PHYS_SIZE_V2
;
516 size
= sizeof (objset_phys_t
);
518 /* Increase the blocksize if we are permitted. */
519 if (arc_buf_size(os
->os_phys_buf
) < size
) {
520 arc_buf_t
*buf
= arc_alloc_buf(spa
, &os
->os_phys_buf
,
521 ARC_BUFC_METADATA
, size
);
522 memset(buf
->b_data
, 0, size
);
523 memcpy(buf
->b_data
, os
->os_phys_buf
->b_data
,
524 arc_buf_size(os
->os_phys_buf
));
525 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
526 os
->os_phys_buf
= buf
;
529 os
->os_phys
= os
->os_phys_buf
->b_data
;
530 os
->os_flags
= os
->os_phys
->os_flags
;
532 int size
= spa_version(spa
) >= SPA_VERSION_USERSPACE
?
533 sizeof (objset_phys_t
) : OBJSET_PHYS_SIZE_V1
;
534 os
->os_phys_buf
= arc_alloc_buf(spa
, &os
->os_phys_buf
,
535 ARC_BUFC_METADATA
, size
);
536 os
->os_phys
= os
->os_phys_buf
->b_data
;
537 memset(os
->os_phys
, 0, size
);
540 * These properties will be filled in by the logic in zfs_get_zplprop()
541 * when they are queried for the first time.
543 os
->os_version
= OBJSET_PROP_UNINITIALIZED
;
544 os
->os_normalization
= OBJSET_PROP_UNINITIALIZED
;
545 os
->os_utf8only
= OBJSET_PROP_UNINITIALIZED
;
546 os
->os_casesensitivity
= OBJSET_PROP_UNINITIALIZED
;
549 * Note: the changed_cb will be called once before the register
550 * func returns, thus changing the checksum/compression from the
551 * default (fletcher2/off). Snapshots don't need to know about
552 * checksum/compression/copies.
555 os
->os_encrypted
= (ds
->ds_dir
->dd_crypto_obj
!= 0);
557 err
= dsl_prop_register(ds
,
558 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
559 primary_cache_changed_cb
, os
);
561 err
= dsl_prop_register(ds
,
562 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
563 secondary_cache_changed_cb
, os
);
565 if (!ds
->ds_is_snapshot
) {
567 err
= dsl_prop_register(ds
,
568 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
569 checksum_changed_cb
, os
);
572 err
= dsl_prop_register(ds
,
573 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
574 compression_changed_cb
, os
);
577 err
= dsl_prop_register(ds
,
578 zfs_prop_to_name(ZFS_PROP_COPIES
),
579 copies_changed_cb
, os
);
582 err
= dsl_prop_register(ds
,
583 zfs_prop_to_name(ZFS_PROP_DEDUP
),
584 dedup_changed_cb
, os
);
587 err
= dsl_prop_register(ds
,
588 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
589 logbias_changed_cb
, os
);
592 err
= dsl_prop_register(ds
,
593 zfs_prop_to_name(ZFS_PROP_SYNC
),
594 sync_changed_cb
, os
);
597 err
= dsl_prop_register(ds
,
599 ZFS_PROP_REDUNDANT_METADATA
),
600 redundant_metadata_changed_cb
, os
);
603 err
= dsl_prop_register(ds
,
604 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
605 recordsize_changed_cb
, os
);
608 err
= dsl_prop_register(ds
,
609 zfs_prop_to_name(ZFS_PROP_DNODESIZE
),
610 dnodesize_changed_cb
, os
);
613 err
= dsl_prop_register(ds
,
615 ZFS_PROP_SPECIAL_SMALL_BLOCKS
),
616 smallblk_changed_cb
, os
);
620 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
621 kmem_free(os
, sizeof (objset_t
));
625 /* It's the meta-objset. */
626 os
->os_checksum
= ZIO_CHECKSUM_FLETCHER_4
;
627 os
->os_compress
= ZIO_COMPRESS_ON
;
628 os
->os_complevel
= ZIO_COMPLEVEL_DEFAULT
;
629 os
->os_encrypted
= B_FALSE
;
630 os
->os_copies
= spa_max_replication(spa
);
631 os
->os_dedup_checksum
= ZIO_CHECKSUM_OFF
;
632 os
->os_dedup_verify
= B_FALSE
;
633 os
->os_logbias
= ZFS_LOGBIAS_LATENCY
;
634 os
->os_sync
= ZFS_SYNC_STANDARD
;
635 os
->os_primary_cache
= ZFS_CACHE_ALL
;
636 os
->os_secondary_cache
= ZFS_CACHE_ALL
;
637 os
->os_dnodesize
= DNODE_MIN_SIZE
;
640 if (ds
== NULL
|| !ds
->ds_is_snapshot
)
641 os
->os_zil_header
= os
->os_phys
->os_zil_header
;
642 os
->os_zil
= zil_alloc(os
, &os
->os_zil_header
);
644 for (i
= 0; i
< TXG_SIZE
; i
++) {
645 multilist_create(&os
->os_dirty_dnodes
[i
], sizeof (dnode_t
),
646 offsetof(dnode_t
, dn_dirty_link
[i
]),
647 dnode_multilist_index_func
);
649 list_create(&os
->os_dnodes
, sizeof (dnode_t
),
650 offsetof(dnode_t
, dn_link
));
651 list_create(&os
->os_downgraded_dbufs
, sizeof (dmu_buf_impl_t
),
652 offsetof(dmu_buf_impl_t
, db_link
));
654 list_link_init(&os
->os_evicting_node
);
656 mutex_init(&os
->os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
657 mutex_init(&os
->os_userused_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
658 mutex_init(&os
->os_obj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
659 mutex_init(&os
->os_user_ptr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
660 os
->os_obj_next_percpu_len
= boot_ncpus
;
661 os
->os_obj_next_percpu
= kmem_zalloc(os
->os_obj_next_percpu_len
*
662 sizeof (os
->os_obj_next_percpu
[0]), KM_SLEEP
);
664 dnode_special_open(os
, &os
->os_phys
->os_meta_dnode
,
665 DMU_META_DNODE_OBJECT
, &os
->os_meta_dnode
);
666 if (OBJSET_BUF_HAS_USERUSED(os
->os_phys_buf
)) {
667 dnode_special_open(os
, &os
->os_phys
->os_userused_dnode
,
668 DMU_USERUSED_OBJECT
, &os
->os_userused_dnode
);
669 dnode_special_open(os
, &os
->os_phys
->os_groupused_dnode
,
670 DMU_GROUPUSED_OBJECT
, &os
->os_groupused_dnode
);
671 if (OBJSET_BUF_HAS_PROJECTUSED(os
->os_phys_buf
))
672 dnode_special_open(os
,
673 &os
->os_phys
->os_projectused_dnode
,
674 DMU_PROJECTUSED_OBJECT
, &os
->os_projectused_dnode
);
677 mutex_init(&os
->os_upgrade_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
684 dmu_objset_from_ds(dsl_dataset_t
*ds
, objset_t
**osp
)
689 * We need the pool_config lock to manipulate the dsl_dataset_t.
690 * Even if the dataset is long-held, we need the pool_config lock
691 * to open the objset, as it needs to get properties.
693 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
695 mutex_enter(&ds
->ds_opening_lock
);
696 if (ds
->ds_objset
== NULL
) {
698 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
699 err
= dmu_objset_open_impl(dsl_dataset_get_spa(ds
),
700 ds
, dsl_dataset_get_blkptr(ds
), &os
);
701 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
704 mutex_enter(&ds
->ds_lock
);
705 ASSERT(ds
->ds_objset
== NULL
);
707 mutex_exit(&ds
->ds_lock
);
710 *osp
= ds
->ds_objset
;
711 mutex_exit(&ds
->ds_opening_lock
);
716 * Holds the pool while the objset is held. Therefore only one objset
717 * can be held at a time.
720 dmu_objset_hold_flags(const char *name
, boolean_t decrypt
, const void *tag
,
726 ds_hold_flags_t flags
;
728 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
729 err
= dsl_pool_hold(name
, tag
, &dp
);
732 err
= dsl_dataset_hold_flags(dp
, name
, flags
, tag
, &ds
);
734 dsl_pool_rele(dp
, tag
);
738 err
= dmu_objset_from_ds(ds
, osp
);
740 dsl_dataset_rele(ds
, tag
);
741 dsl_pool_rele(dp
, tag
);
748 dmu_objset_hold(const char *name
, const void *tag
, objset_t
**osp
)
750 return (dmu_objset_hold_flags(name
, B_FALSE
, tag
, osp
));
754 dmu_objset_own_impl(dsl_dataset_t
*ds
, dmu_objset_type_t type
,
755 boolean_t readonly
, boolean_t decrypt
, const void *tag
, objset_t
**osp
)
759 int err
= dmu_objset_from_ds(ds
, osp
);
762 } else if (type
!= DMU_OST_ANY
&& type
!= (*osp
)->os_phys
->os_type
) {
763 return (SET_ERROR(EINVAL
));
764 } else if (!readonly
&& dsl_dataset_is_snapshot(ds
)) {
765 return (SET_ERROR(EROFS
));
766 } else if (!readonly
&& decrypt
&&
767 dsl_dir_incompatible_encryption_version(ds
->ds_dir
)) {
768 return (SET_ERROR(EROFS
));
771 /* if we are decrypting, we can now check MACs in os->os_phys_buf */
772 if (decrypt
&& arc_is_unauthenticated((*osp
)->os_phys_buf
)) {
775 SET_BOOKMARK(&zb
, ds
->ds_object
, ZB_ROOT_OBJECT
,
776 ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
777 err
= arc_untransform((*osp
)->os_phys_buf
, (*osp
)->os_spa
,
782 ASSERT0(arc_is_unauthenticated((*osp
)->os_phys_buf
));
789 * dsl_pool must not be held when this is called.
790 * Upon successful return, there will be a longhold on the dataset,
791 * and the dsl_pool will not be held.
794 dmu_objset_own(const char *name
, dmu_objset_type_t type
,
795 boolean_t readonly
, boolean_t decrypt
, const void *tag
, objset_t
**osp
)
800 ds_hold_flags_t flags
;
802 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
803 err
= dsl_pool_hold(name
, FTAG
, &dp
);
806 err
= dsl_dataset_own(dp
, name
, flags
, tag
, &ds
);
808 dsl_pool_rele(dp
, FTAG
);
811 err
= dmu_objset_own_impl(ds
, type
, readonly
, decrypt
, tag
, osp
);
813 dsl_dataset_disown(ds
, flags
, tag
);
814 dsl_pool_rele(dp
, FTAG
);
819 * User accounting requires the dataset to be decrypted and rw.
820 * We also don't begin user accounting during claiming to help
821 * speed up pool import times and to keep this txg reserved
822 * completely for recovery work.
824 if (!readonly
&& !dp
->dp_spa
->spa_claiming
&&
825 (ds
->ds_dir
->dd_crypto_obj
== 0 || decrypt
)) {
826 if (dmu_objset_userobjspace_upgradable(*osp
) ||
827 dmu_objset_projectquota_upgradable(*osp
)) {
828 dmu_objset_id_quota_upgrade(*osp
);
829 } else if (dmu_objset_userused_enabled(*osp
)) {
830 dmu_objset_userspace_upgrade(*osp
);
834 dsl_pool_rele(dp
, FTAG
);
839 dmu_objset_own_obj(dsl_pool_t
*dp
, uint64_t obj
, dmu_objset_type_t type
,
840 boolean_t readonly
, boolean_t decrypt
, const void *tag
, objset_t
**osp
)
844 ds_hold_flags_t flags
;
846 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
847 err
= dsl_dataset_own_obj(dp
, obj
, flags
, tag
, &ds
);
851 err
= dmu_objset_own_impl(ds
, type
, readonly
, decrypt
, tag
, osp
);
853 dsl_dataset_disown(ds
, flags
, tag
);
861 dmu_objset_rele_flags(objset_t
*os
, boolean_t decrypt
, const void *tag
)
863 ds_hold_flags_t flags
;
864 dsl_pool_t
*dp
= dmu_objset_pool(os
);
866 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
867 dsl_dataset_rele_flags(os
->os_dsl_dataset
, flags
, tag
);
868 dsl_pool_rele(dp
, tag
);
872 dmu_objset_rele(objset_t
*os
, const void *tag
)
874 dmu_objset_rele_flags(os
, B_FALSE
, tag
);
878 * When we are called, os MUST refer to an objset associated with a dataset
879 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
880 * == tag. We will then release and reacquire ownership of the dataset while
881 * holding the pool config_rwlock to avoid intervening namespace or ownership
884 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
885 * release the hold on its dataset and acquire a new one on the dataset of the
886 * same name so that it can be partially torn down and reconstructed.
889 dmu_objset_refresh_ownership(dsl_dataset_t
*ds
, dsl_dataset_t
**newds
,
890 boolean_t decrypt
, const void *tag
)
893 char name
[ZFS_MAX_DATASET_NAME_LEN
];
894 ds_hold_flags_t flags
;
896 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
897 VERIFY3P(ds
, !=, NULL
);
898 VERIFY3P(ds
->ds_owner
, ==, tag
);
899 VERIFY(dsl_dataset_long_held(ds
));
901 dsl_dataset_name(ds
, name
);
902 dp
= ds
->ds_dir
->dd_pool
;
903 dsl_pool_config_enter(dp
, FTAG
);
904 dsl_dataset_disown(ds
, flags
, tag
);
905 VERIFY0(dsl_dataset_own(dp
, name
, flags
, tag
, newds
));
906 dsl_pool_config_exit(dp
, FTAG
);
910 dmu_objset_disown(objset_t
*os
, boolean_t decrypt
, const void *tag
)
912 ds_hold_flags_t flags
;
914 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
916 * Stop upgrading thread
918 dmu_objset_upgrade_stop(os
);
919 dsl_dataset_disown(os
->os_dsl_dataset
, flags
, tag
);
923 dmu_objset_evict_dbufs(objset_t
*os
)
928 dn_marker
= kmem_alloc(sizeof (dnode_t
), KM_SLEEP
);
930 mutex_enter(&os
->os_lock
);
931 dn
= list_head(&os
->os_dnodes
);
934 * Skip dnodes without holds. We have to do this dance
935 * because dnode_add_ref() only works if there is already a
936 * hold. If the dnode has no holds, then it has no dbufs.
938 if (dnode_add_ref(dn
, FTAG
)) {
939 list_insert_after(&os
->os_dnodes
, dn
, dn_marker
);
940 mutex_exit(&os
->os_lock
);
942 dnode_evict_dbufs(dn
);
943 dnode_rele(dn
, FTAG
);
945 mutex_enter(&os
->os_lock
);
946 dn
= list_next(&os
->os_dnodes
, dn_marker
);
947 list_remove(&os
->os_dnodes
, dn_marker
);
949 dn
= list_next(&os
->os_dnodes
, dn
);
952 mutex_exit(&os
->os_lock
);
954 kmem_free(dn_marker
, sizeof (dnode_t
));
956 if (DMU_USERUSED_DNODE(os
) != NULL
) {
957 if (DMU_PROJECTUSED_DNODE(os
) != NULL
)
958 dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os
));
959 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os
));
960 dnode_evict_dbufs(DMU_USERUSED_DNODE(os
));
962 dnode_evict_dbufs(DMU_META_DNODE(os
));
966 * Objset eviction processing is split into into two pieces.
967 * The first marks the objset as evicting, evicts any dbufs that
968 * have a refcount of zero, and then queues up the objset for the
969 * second phase of eviction. Once os->os_dnodes has been cleared by
970 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
971 * The second phase closes the special dnodes, dequeues the objset from
972 * the list of those undergoing eviction, and finally frees the objset.
974 * NOTE: Due to asynchronous eviction processing (invocation of
975 * dnode_buf_pageout()), it is possible for the meta dnode for the
976 * objset to have no holds even though os->os_dnodes is not empty.
979 dmu_objset_evict(objset_t
*os
)
981 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
983 for (int t
= 0; t
< TXG_SIZE
; t
++)
984 ASSERT(!dmu_objset_is_dirty(os
, t
));
987 dsl_prop_unregister_all(ds
, os
);
992 dmu_objset_evict_dbufs(os
);
994 mutex_enter(&os
->os_lock
);
995 spa_evicting_os_register(os
->os_spa
, os
);
996 if (list_is_empty(&os
->os_dnodes
)) {
997 mutex_exit(&os
->os_lock
);
998 dmu_objset_evict_done(os
);
1000 mutex_exit(&os
->os_lock
);
1007 dmu_objset_evict_done(objset_t
*os
)
1009 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
1011 dnode_special_close(&os
->os_meta_dnode
);
1012 if (DMU_USERUSED_DNODE(os
)) {
1013 if (DMU_PROJECTUSED_DNODE(os
))
1014 dnode_special_close(&os
->os_projectused_dnode
);
1015 dnode_special_close(&os
->os_userused_dnode
);
1016 dnode_special_close(&os
->os_groupused_dnode
);
1018 zil_free(os
->os_zil
);
1020 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
1023 * This is a barrier to prevent the objset from going away in
1024 * dnode_move() until we can safely ensure that the objset is still in
1025 * use. We consider the objset valid before the barrier and invalid
1026 * after the barrier.
1028 rw_enter(&os_lock
, RW_READER
);
1031 kmem_free(os
->os_obj_next_percpu
,
1032 os
->os_obj_next_percpu_len
* sizeof (os
->os_obj_next_percpu
[0]));
1034 mutex_destroy(&os
->os_lock
);
1035 mutex_destroy(&os
->os_userused_lock
);
1036 mutex_destroy(&os
->os_obj_lock
);
1037 mutex_destroy(&os
->os_user_ptr_lock
);
1038 mutex_destroy(&os
->os_upgrade_lock
);
1039 for (int i
= 0; i
< TXG_SIZE
; i
++)
1040 multilist_destroy(&os
->os_dirty_dnodes
[i
]);
1041 spa_evicting_os_deregister(os
->os_spa
, os
);
1042 kmem_free(os
, sizeof (objset_t
));
1046 dmu_objset_snap_cmtime(objset_t
*os
)
1048 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
1052 dmu_objset_create_impl_dnstats(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
1053 dmu_objset_type_t type
, int levels
, int blksz
, int ibs
, dmu_tx_t
*tx
)
1058 ASSERT(dmu_tx_is_syncing(tx
));
1061 blksz
= DNODE_BLOCK_SIZE
;
1063 ibs
= DN_MAX_INDBLKSHIFT
;
1066 VERIFY0(dmu_objset_from_ds(ds
, &os
));
1068 VERIFY0(dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
1070 mdn
= DMU_META_DNODE(os
);
1072 dnode_allocate(mdn
, DMU_OT_DNODE
, blksz
, ibs
, DMU_OT_NONE
, 0,
1073 DNODE_MIN_SLOTS
, tx
);
1076 * We don't want to have to increase the meta-dnode's nlevels
1077 * later, because then we could do it in quiescing context while
1078 * we are also accessing it in open context.
1080 * This precaution is not necessary for the MOS (ds == NULL),
1081 * because the MOS is only updated in syncing context.
1082 * This is most fortunate: the MOS is the only objset that
1083 * needs to be synced multiple times as spa_sync() iterates
1084 * to convergence, so minimizing its dn_nlevels matters.
1091 * Determine the number of levels necessary for the
1092 * meta-dnode to contain DN_MAX_OBJECT dnodes. Note
1093 * that in order to ensure that we do not overflow
1094 * 64 bits, there has to be a nlevels that gives us a
1095 * number of blocks > DN_MAX_OBJECT but < 2^64.
1096 * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)
1097 * (10) must be less than (64 - log2(DN_MAX_OBJECT))
1100 while ((uint64_t)mdn
->dn_nblkptr
<<
1101 (mdn
->dn_datablkshift
- DNODE_SHIFT
+ (levels
- 1) *
1102 (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
1107 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
1108 mdn
->dn_nlevels
= levels
;
1111 ASSERT(type
!= DMU_OST_NONE
);
1112 ASSERT(type
!= DMU_OST_ANY
);
1113 ASSERT(type
< DMU_OST_NUMTYPES
);
1114 os
->os_phys
->os_type
= type
;
1117 * Enable user accounting if it is enabled and this is not an
1118 * encrypted receive.
1120 if (dmu_objset_userused_enabled(os
) &&
1121 (!os
->os_encrypted
|| !dmu_objset_is_receiving(os
))) {
1122 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
1123 if (dmu_objset_userobjused_enabled(os
)) {
1124 ds
->ds_feature_activation
[
1125 SPA_FEATURE_USEROBJ_ACCOUNTING
] = (void *)B_TRUE
;
1126 os
->os_phys
->os_flags
|=
1127 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
;
1129 if (dmu_objset_projectquota_enabled(os
)) {
1130 ds
->ds_feature_activation
[
1131 SPA_FEATURE_PROJECT_QUOTA
] = (void *)B_TRUE
;
1132 os
->os_phys
->os_flags
|=
1133 OBJSET_FLAG_PROJECTQUOTA_COMPLETE
;
1135 os
->os_flags
= os
->os_phys
->os_flags
;
1138 dsl_dataset_dirty(ds
, tx
);
1143 /* called from dsl for meta-objset */
1145 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
1146 dmu_objset_type_t type
, dmu_tx_t
*tx
)
1148 return (dmu_objset_create_impl_dnstats(spa
, ds
, bp
, type
, 0, 0, 0, tx
));
1151 typedef struct dmu_objset_create_arg
{
1152 const char *doca_name
;
1155 void (*doca_userfunc
)(objset_t
*os
, void *arg
,
1156 cred_t
*cr
, dmu_tx_t
*tx
);
1158 dmu_objset_type_t doca_type
;
1159 uint64_t doca_flags
;
1160 dsl_crypto_params_t
*doca_dcp
;
1161 } dmu_objset_create_arg_t
;
1164 dmu_objset_create_check(void *arg
, dmu_tx_t
*tx
)
1166 dmu_objset_create_arg_t
*doca
= arg
;
1167 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1169 dsl_dataset_t
*parentds
;
1174 if (strchr(doca
->doca_name
, '@') != NULL
)
1175 return (SET_ERROR(EINVAL
));
1177 if (strlen(doca
->doca_name
) >= ZFS_MAX_DATASET_NAME_LEN
)
1178 return (SET_ERROR(ENAMETOOLONG
));
1180 if (dataset_nestcheck(doca
->doca_name
) != 0)
1181 return (SET_ERROR(ENAMETOOLONG
));
1183 error
= dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
);
1187 dsl_dir_rele(pdd
, FTAG
);
1188 return (SET_ERROR(EEXIST
));
1191 error
= dmu_objset_create_crypt_check(pdd
, doca
->doca_dcp
, NULL
);
1193 dsl_dir_rele(pdd
, FTAG
);
1197 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
1198 doca
->doca_cred
, doca
->doca_proc
);
1200 dsl_dir_rele(pdd
, FTAG
);
1204 /* can't create below anything but filesystems (eg. no ZVOLs) */
1205 error
= dsl_dataset_hold_obj(pdd
->dd_pool
,
1206 dsl_dir_phys(pdd
)->dd_head_dataset_obj
, FTAG
, &parentds
);
1208 dsl_dir_rele(pdd
, FTAG
);
1211 error
= dmu_objset_from_ds(parentds
, &parentos
);
1213 dsl_dataset_rele(parentds
, FTAG
);
1214 dsl_dir_rele(pdd
, FTAG
);
1217 if (dmu_objset_type(parentos
) != DMU_OST_ZFS
) {
1218 dsl_dataset_rele(parentds
, FTAG
);
1219 dsl_dir_rele(pdd
, FTAG
);
1220 return (SET_ERROR(ZFS_ERR_WRONG_PARENT
));
1222 dsl_dataset_rele(parentds
, FTAG
);
1223 dsl_dir_rele(pdd
, FTAG
);
1229 dmu_objset_create_sync(void *arg
, dmu_tx_t
*tx
)
1231 dmu_objset_create_arg_t
*doca
= arg
;
1232 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1233 spa_t
*spa
= dp
->dp_spa
;
1242 VERIFY0(dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
));
1244 obj
= dsl_dataset_create_sync(pdd
, tail
, NULL
, doca
->doca_flags
,
1245 doca
->doca_cred
, doca
->doca_dcp
, tx
);
1247 VERIFY0(dsl_dataset_hold_obj_flags(pdd
->dd_pool
, obj
,
1248 DS_HOLD_FLAG_DECRYPT
, FTAG
, &ds
));
1249 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1250 bp
= dsl_dataset_get_blkptr(ds
);
1251 os
= dmu_objset_create_impl(spa
, ds
, bp
, doca
->doca_type
, tx
);
1252 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
1254 if (doca
->doca_userfunc
!= NULL
) {
1255 doca
->doca_userfunc(os
, doca
->doca_userarg
,
1256 doca
->doca_cred
, tx
);
1260 * The doca_userfunc() may write out some data that needs to be
1261 * encrypted if the dataset is encrypted (specifically the root
1262 * directory). This data must be written out before the encryption
1263 * key mapping is removed by dsl_dataset_rele_flags(). Force the
1264 * I/O to occur immediately by invoking the relevant sections of
1267 if (os
->os_encrypted
) {
1268 dsl_dataset_t
*tmpds
= NULL
;
1269 boolean_t need_sync_done
= B_FALSE
;
1271 mutex_enter(&ds
->ds_lock
);
1272 ds
->ds_owner
= FTAG
;
1273 mutex_exit(&ds
->ds_lock
);
1275 rzio
= zio_root(spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
1276 tmpds
= txg_list_remove_this(&dp
->dp_dirty_datasets
, ds
,
1278 if (tmpds
!= NULL
) {
1279 dsl_dataset_sync(ds
, rzio
, tx
);
1280 need_sync_done
= B_TRUE
;
1282 VERIFY0(zio_wait(rzio
));
1284 dmu_objset_sync_done(os
, tx
);
1285 taskq_wait(dp
->dp_sync_taskq
);
1286 if (txg_list_member(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
1287 ASSERT3P(ds
->ds_key_mapping
, !=, NULL
);
1288 key_mapping_rele(spa
, ds
->ds_key_mapping
, ds
);
1291 rzio
= zio_root(spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
1292 tmpds
= txg_list_remove_this(&dp
->dp_dirty_datasets
, ds
,
1294 if (tmpds
!= NULL
) {
1295 dmu_buf_rele(ds
->ds_dbuf
, ds
);
1296 dsl_dataset_sync(ds
, rzio
, tx
);
1298 VERIFY0(zio_wait(rzio
));
1300 if (need_sync_done
) {
1301 ASSERT3P(ds
->ds_key_mapping
, !=, NULL
);
1302 key_mapping_rele(spa
, ds
->ds_key_mapping
, ds
);
1303 dsl_dataset_sync_done(ds
, tx
);
1306 mutex_enter(&ds
->ds_lock
);
1307 ds
->ds_owner
= NULL
;
1308 mutex_exit(&ds
->ds_lock
);
1311 spa_history_log_internal_ds(ds
, "create", tx
, " ");
1313 dsl_dataset_rele_flags(ds
, DS_HOLD_FLAG_DECRYPT
, FTAG
);
1314 dsl_dir_rele(pdd
, FTAG
);
1318 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
1319 dsl_crypto_params_t
*dcp
, dmu_objset_create_sync_func_t func
, void *arg
)
1321 dmu_objset_create_arg_t doca
;
1322 dsl_crypto_params_t tmp_dcp
= { 0 };
1324 doca
.doca_name
= name
;
1325 doca
.doca_cred
= CRED();
1326 doca
.doca_proc
= curproc
;
1327 doca
.doca_flags
= flags
;
1328 doca
.doca_userfunc
= func
;
1329 doca
.doca_userarg
= arg
;
1330 doca
.doca_type
= type
;
1333 * Some callers (mostly for testing) do not provide a dcp on their
1334 * own but various code inside the sync task will require it to be
1335 * allocated. Rather than adding NULL checks throughout this code
1336 * or adding dummy dcp's to all of the callers we simply create a
1337 * dummy one here and use that. This zero dcp will have the same
1338 * effect as asking for inheritance of all encryption params.
1340 doca
.doca_dcp
= (dcp
!= NULL
) ? dcp
: &tmp_dcp
;
1342 int rv
= dsl_sync_task(name
,
1343 dmu_objset_create_check
, dmu_objset_create_sync
, &doca
,
1344 6, ZFS_SPACE_CHECK_NORMAL
);
1347 zvol_create_minor(name
);
1351 typedef struct dmu_objset_clone_arg
{
1352 const char *doca_clone
;
1353 const char *doca_origin
;
1356 } dmu_objset_clone_arg_t
;
1359 dmu_objset_clone_check(void *arg
, dmu_tx_t
*tx
)
1361 dmu_objset_clone_arg_t
*doca
= arg
;
1365 dsl_dataset_t
*origin
;
1366 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1368 if (strchr(doca
->doca_clone
, '@') != NULL
)
1369 return (SET_ERROR(EINVAL
));
1371 if (strlen(doca
->doca_clone
) >= ZFS_MAX_DATASET_NAME_LEN
)
1372 return (SET_ERROR(ENAMETOOLONG
));
1374 error
= dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
);
1378 dsl_dir_rele(pdd
, FTAG
);
1379 return (SET_ERROR(EEXIST
));
1382 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
1383 doca
->doca_cred
, doca
->doca_proc
);
1385 dsl_dir_rele(pdd
, FTAG
);
1386 return (SET_ERROR(EDQUOT
));
1389 error
= dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
);
1391 dsl_dir_rele(pdd
, FTAG
);
1395 /* You can only clone snapshots, not the head datasets. */
1396 if (!origin
->ds_is_snapshot
) {
1397 dsl_dataset_rele(origin
, FTAG
);
1398 dsl_dir_rele(pdd
, FTAG
);
1399 return (SET_ERROR(EINVAL
));
1402 dsl_dataset_rele(origin
, FTAG
);
1403 dsl_dir_rele(pdd
, FTAG
);
1409 dmu_objset_clone_sync(void *arg
, dmu_tx_t
*tx
)
1411 dmu_objset_clone_arg_t
*doca
= arg
;
1412 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1415 dsl_dataset_t
*origin
, *ds
;
1417 char namebuf
[ZFS_MAX_DATASET_NAME_LEN
];
1419 VERIFY0(dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
));
1420 VERIFY0(dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
));
1422 obj
= dsl_dataset_create_sync(pdd
, tail
, origin
, 0,
1423 doca
->doca_cred
, NULL
, tx
);
1425 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
1426 dsl_dataset_name(origin
, namebuf
);
1427 spa_history_log_internal_ds(ds
, "clone", tx
,
1428 "origin=%s (%llu)", namebuf
, (u_longlong_t
)origin
->ds_object
);
1429 dsl_dataset_rele(ds
, FTAG
);
1430 dsl_dataset_rele(origin
, FTAG
);
1431 dsl_dir_rele(pdd
, FTAG
);
1435 dmu_objset_clone(const char *clone
, const char *origin
)
1437 dmu_objset_clone_arg_t doca
;
1439 doca
.doca_clone
= clone
;
1440 doca
.doca_origin
= origin
;
1441 doca
.doca_cred
= CRED();
1442 doca
.doca_proc
= curproc
;
1444 int rv
= dsl_sync_task(clone
,
1445 dmu_objset_clone_check
, dmu_objset_clone_sync
, &doca
,
1446 6, ZFS_SPACE_CHECK_NORMAL
);
1449 zvol_create_minor(clone
);
1455 dmu_objset_snapshot_one(const char *fsname
, const char *snapname
)
1458 char *longsnap
= kmem_asprintf("%s@%s", fsname
, snapname
);
1459 nvlist_t
*snaps
= fnvlist_alloc();
1461 fnvlist_add_boolean(snaps
, longsnap
);
1462 kmem_strfree(longsnap
);
1463 err
= dsl_dataset_snapshot(snaps
, NULL
, NULL
);
1464 fnvlist_free(snaps
);
1469 dmu_objset_upgrade_task_cb(void *data
)
1471 objset_t
*os
= data
;
1473 mutex_enter(&os
->os_upgrade_lock
);
1474 os
->os_upgrade_status
= EINTR
;
1475 if (!os
->os_upgrade_exit
) {
1478 mutex_exit(&os
->os_upgrade_lock
);
1480 status
= os
->os_upgrade_cb(os
);
1482 mutex_enter(&os
->os_upgrade_lock
);
1484 os
->os_upgrade_status
= status
;
1486 os
->os_upgrade_exit
= B_TRUE
;
1487 os
->os_upgrade_id
= 0;
1488 mutex_exit(&os
->os_upgrade_lock
);
1489 dsl_dataset_long_rele(dmu_objset_ds(os
), upgrade_tag
);
1493 dmu_objset_upgrade(objset_t
*os
, dmu_objset_upgrade_cb_t cb
)
1495 if (os
->os_upgrade_id
!= 0)
1498 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
1499 dsl_dataset_long_hold(dmu_objset_ds(os
), upgrade_tag
);
1501 mutex_enter(&os
->os_upgrade_lock
);
1502 if (os
->os_upgrade_id
== 0 && os
->os_upgrade_status
== 0) {
1503 os
->os_upgrade_exit
= B_FALSE
;
1504 os
->os_upgrade_cb
= cb
;
1505 os
->os_upgrade_id
= taskq_dispatch(
1506 os
->os_spa
->spa_upgrade_taskq
,
1507 dmu_objset_upgrade_task_cb
, os
, TQ_SLEEP
);
1508 if (os
->os_upgrade_id
== TASKQID_INVALID
) {
1509 dsl_dataset_long_rele(dmu_objset_ds(os
), upgrade_tag
);
1510 os
->os_upgrade_status
= ENOMEM
;
1513 dsl_dataset_long_rele(dmu_objset_ds(os
), upgrade_tag
);
1515 mutex_exit(&os
->os_upgrade_lock
);
1519 dmu_objset_upgrade_stop(objset_t
*os
)
1521 mutex_enter(&os
->os_upgrade_lock
);
1522 os
->os_upgrade_exit
= B_TRUE
;
1523 if (os
->os_upgrade_id
!= 0) {
1524 taskqid_t id
= os
->os_upgrade_id
;
1526 os
->os_upgrade_id
= 0;
1527 mutex_exit(&os
->os_upgrade_lock
);
1529 if ((taskq_cancel_id(os
->os_spa
->spa_upgrade_taskq
, id
)) == 0) {
1530 dsl_dataset_long_rele(dmu_objset_ds(os
), upgrade_tag
);
1532 txg_wait_synced(os
->os_spa
->spa_dsl_pool
, 0);
1534 mutex_exit(&os
->os_upgrade_lock
);
1539 dmu_objset_sync_dnodes(multilist_sublist_t
*list
, dmu_tx_t
*tx
)
1543 while ((dn
= multilist_sublist_head(list
)) != NULL
) {
1544 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1545 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1547 * Initialize dn_zio outside dnode_sync() because the
1548 * meta-dnode needs to set it outside dnode_sync().
1550 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1553 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1554 multilist_sublist_remove(list
, dn
);
1557 * See the comment above dnode_rele_task() for an explanation
1558 * of why this dnode hold is always needed (even when not
1559 * doing user accounting).
1561 multilist_t
*newlist
= &dn
->dn_objset
->os_synced_dnodes
;
1562 (void) dnode_add_ref(dn
, newlist
);
1563 multilist_insert(newlist
, dn
);
1570 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1573 blkptr_t
*bp
= zio
->io_bp
;
1575 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1578 ASSERT(!BP_IS_EMBEDDED(bp
));
1579 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_OBJSET
);
1580 ASSERT0(BP_GET_LEVEL(bp
));
1583 * Update rootbp fill count: it should be the number of objects
1584 * allocated in the object set (not counting the "special"
1585 * objects that are stored in the objset_phys_t -- the meta
1586 * dnode and user/group/project accounting objects).
1588 for (int i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1589 fill
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
1591 BP_SET_FILL(bp
, fill
);
1593 if (os
->os_dsl_dataset
!= NULL
)
1594 rrw_enter(&os
->os_dsl_dataset
->ds_bp_rwlock
, RW_WRITER
, FTAG
);
1595 *os
->os_rootbp
= *bp
;
1596 if (os
->os_dsl_dataset
!= NULL
)
1597 rrw_exit(&os
->os_dsl_dataset
->ds_bp_rwlock
, FTAG
);
1601 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1604 blkptr_t
*bp
= zio
->io_bp
;
1605 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1608 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1609 ASSERT(BP_EQUAL(bp
, bp_orig
));
1611 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1612 dmu_tx_t
*tx
= os
->os_synctx
;
1614 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1615 dsl_dataset_block_born(ds
, bp
, tx
);
1617 kmem_free(bp
, sizeof (*bp
));
1620 typedef struct sync_dnodes_arg
{
1621 multilist_t
*sda_list
;
1622 int sda_sublist_idx
;
1623 multilist_t
*sda_newlist
;
1625 } sync_dnodes_arg_t
;
1628 sync_dnodes_task(void *arg
)
1630 sync_dnodes_arg_t
*sda
= arg
;
1632 multilist_sublist_t
*ms
=
1633 multilist_sublist_lock(sda
->sda_list
, sda
->sda_sublist_idx
);
1635 dmu_objset_sync_dnodes(ms
, sda
->sda_tx
);
1637 multilist_sublist_unlock(ms
);
1639 kmem_free(sda
, sizeof (*sda
));
1643 /* called from dsl */
1645 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1648 zbookmark_phys_t zb
;
1652 dbuf_dirty_record_t
*dr
;
1655 blkptr_t
*blkptr_copy
= kmem_alloc(sizeof (*os
->os_rootbp
), KM_SLEEP
);
1656 *blkptr_copy
= *os
->os_rootbp
;
1658 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", (u_longlong_t
)tx
->tx_txg
);
1660 ASSERT(dmu_tx_is_syncing(tx
));
1661 /* XXX the write_done callback should really give us the tx... */
1664 if (os
->os_dsl_dataset
== NULL
) {
1666 * This is the MOS. If we have upgraded,
1667 * spa_max_replication() could change, so reset
1670 os
->os_copies
= spa_max_replication(os
->os_spa
);
1674 * Create the root block IO
1676 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1677 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1678 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1679 arc_release(os
->os_phys_buf
, &os
->os_phys_buf
);
1681 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1684 * If we are either claiming the ZIL or doing a raw receive, write
1685 * out the os_phys_buf raw. Neither of these actions will effect the
1686 * MAC at this point.
1688 if (os
->os_raw_receive
||
1689 os
->os_next_write_raw
[tx
->tx_txg
& TXG_MASK
]) {
1690 ASSERT(os
->os_encrypted
);
1691 arc_convert_to_raw(os
->os_phys_buf
,
1692 os
->os_dsl_dataset
->ds_object
, ZFS_HOST_BYTEORDER
,
1693 DMU_OT_OBJSET
, NULL
, NULL
, NULL
);
1696 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1697 blkptr_copy
, os
->os_phys_buf
, B_FALSE
, dmu_os_is_l2cacheable(os
),
1698 &zp
, dmu_objset_write_ready
, NULL
, NULL
, dmu_objset_write_done
,
1699 os
, ZIO_PRIORITY_ASYNC_WRITE
, ZIO_FLAG_MUSTSUCCEED
, &zb
);
1702 * Sync special dnodes - the parent IO for the sync is the root block
1704 DMU_META_DNODE(os
)->dn_zio
= zio
;
1705 dnode_sync(DMU_META_DNODE(os
), tx
);
1707 os
->os_phys
->os_flags
= os
->os_flags
;
1709 if (DMU_USERUSED_DNODE(os
) &&
1710 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1711 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1712 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1713 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1714 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1717 if (DMU_PROJECTUSED_DNODE(os
) &&
1718 DMU_PROJECTUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1719 DMU_PROJECTUSED_DNODE(os
)->dn_zio
= zio
;
1720 dnode_sync(DMU_PROJECTUSED_DNODE(os
), tx
);
1723 txgoff
= tx
->tx_txg
& TXG_MASK
;
1726 * We must create the list here because it uses the
1727 * dn_dirty_link[] of this txg. But it may already
1728 * exist because we call dsl_dataset_sync() twice per txg.
1730 if (os
->os_synced_dnodes
.ml_sublists
== NULL
) {
1731 multilist_create(&os
->os_synced_dnodes
, sizeof (dnode_t
),
1732 offsetof(dnode_t
, dn_dirty_link
[txgoff
]),
1733 dnode_multilist_index_func
);
1735 ASSERT3U(os
->os_synced_dnodes
.ml_offset
, ==,
1736 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1739 ml
= &os
->os_dirty_dnodes
[txgoff
];
1740 num_sublists
= multilist_get_num_sublists(ml
);
1741 for (int i
= 0; i
< num_sublists
; i
++) {
1742 if (multilist_sublist_is_empty_idx(ml
, i
))
1744 sync_dnodes_arg_t
*sda
= kmem_alloc(sizeof (*sda
), KM_SLEEP
);
1746 sda
->sda_sublist_idx
= i
;
1748 (void) taskq_dispatch(dmu_objset_pool(os
)->dp_sync_taskq
,
1749 sync_dnodes_task
, sda
, 0);
1750 /* callback frees sda */
1752 taskq_wait(dmu_objset_pool(os
)->dp_sync_taskq
);
1754 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1755 while ((dr
= list_head(list
)) != NULL
) {
1756 ASSERT0(dr
->dr_dbuf
->db_level
);
1757 list_remove(list
, dr
);
1758 zio_nowait(dr
->dr_zio
);
1761 /* Enable dnode backfill if enough objects have been freed. */
1762 if (os
->os_freed_dnodes
>= dmu_rescan_dnode_threshold
) {
1763 os
->os_rescan_dnodes
= B_TRUE
;
1764 os
->os_freed_dnodes
= 0;
1768 * Free intent log blocks up to this tx.
1770 zil_sync(os
->os_zil
, tx
);
1771 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1776 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1778 return (!multilist_is_empty(&os
->os_dirty_dnodes
[txg
& TXG_MASK
]));
1781 static file_info_cb_t
*file_cbs
[DMU_OST_NUMTYPES
];
1784 dmu_objset_register_type(dmu_objset_type_t ost
, file_info_cb_t
*cb
)
1790 dmu_get_file_info(objset_t
*os
, dmu_object_type_t bonustype
, const void *data
,
1791 zfs_file_info_t
*zfi
)
1793 file_info_cb_t
*cb
= file_cbs
[os
->os_phys
->os_type
];
1796 return (cb(bonustype
, data
, zfi
));
1800 dmu_objset_userused_enabled(objset_t
*os
)
1802 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1803 file_cbs
[os
->os_phys
->os_type
] != NULL
&&
1804 DMU_USERUSED_DNODE(os
) != NULL
);
1808 dmu_objset_userobjused_enabled(objset_t
*os
)
1810 return (dmu_objset_userused_enabled(os
) &&
1811 spa_feature_is_enabled(os
->os_spa
, SPA_FEATURE_USEROBJ_ACCOUNTING
));
1815 dmu_objset_projectquota_enabled(objset_t
*os
)
1817 return (file_cbs
[os
->os_phys
->os_type
] != NULL
&&
1818 DMU_PROJECTUSED_DNODE(os
) != NULL
&&
1819 spa_feature_is_enabled(os
->os_spa
, SPA_FEATURE_PROJECT_QUOTA
));
1822 typedef struct userquota_node
{
1823 /* must be in the first filed, see userquota_update_cache() */
1824 char uqn_id
[20 + DMU_OBJACCT_PREFIX_LEN
];
1826 avl_node_t uqn_node
;
1829 typedef struct userquota_cache
{
1830 avl_tree_t uqc_user_deltas
;
1831 avl_tree_t uqc_group_deltas
;
1832 avl_tree_t uqc_project_deltas
;
1833 } userquota_cache_t
;
1836 userquota_compare(const void *l
, const void *r
)
1838 const userquota_node_t
*luqn
= l
;
1839 const userquota_node_t
*ruqn
= r
;
1843 * NB: can only access uqn_id because userquota_update_cache() doesn't
1844 * pass in an entire userquota_node_t.
1846 rv
= strcmp(luqn
->uqn_id
, ruqn
->uqn_id
);
1848 return (TREE_ISIGN(rv
));
1852 do_userquota_cacheflush(objset_t
*os
, userquota_cache_t
*cache
, dmu_tx_t
*tx
)
1855 userquota_node_t
*uqn
;
1857 ASSERT(dmu_tx_is_syncing(tx
));
1860 while ((uqn
= avl_destroy_nodes(&cache
->uqc_user_deltas
,
1861 &cookie
)) != NULL
) {
1863 * os_userused_lock protects against concurrent calls to
1864 * zap_increment_int(). It's needed because zap_increment_int()
1865 * is not thread-safe (i.e. not atomic).
1867 mutex_enter(&os
->os_userused_lock
);
1868 VERIFY0(zap_increment(os
, DMU_USERUSED_OBJECT
,
1869 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1870 mutex_exit(&os
->os_userused_lock
);
1871 kmem_free(uqn
, sizeof (*uqn
));
1873 avl_destroy(&cache
->uqc_user_deltas
);
1876 while ((uqn
= avl_destroy_nodes(&cache
->uqc_group_deltas
,
1877 &cookie
)) != NULL
) {
1878 mutex_enter(&os
->os_userused_lock
);
1879 VERIFY0(zap_increment(os
, DMU_GROUPUSED_OBJECT
,
1880 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1881 mutex_exit(&os
->os_userused_lock
);
1882 kmem_free(uqn
, sizeof (*uqn
));
1884 avl_destroy(&cache
->uqc_group_deltas
);
1886 if (dmu_objset_projectquota_enabled(os
)) {
1888 while ((uqn
= avl_destroy_nodes(&cache
->uqc_project_deltas
,
1889 &cookie
)) != NULL
) {
1890 mutex_enter(&os
->os_userused_lock
);
1891 VERIFY0(zap_increment(os
, DMU_PROJECTUSED_OBJECT
,
1892 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1893 mutex_exit(&os
->os_userused_lock
);
1894 kmem_free(uqn
, sizeof (*uqn
));
1896 avl_destroy(&cache
->uqc_project_deltas
);
1901 userquota_update_cache(avl_tree_t
*avl
, const char *id
, int64_t delta
)
1903 userquota_node_t
*uqn
;
1906 ASSERT(strlen(id
) < sizeof (uqn
->uqn_id
));
1908 * Use id directly for searching because uqn_id is the first field of
1909 * userquota_node_t and fields after uqn_id won't be accessed in
1912 uqn
= avl_find(avl
, (const void *)id
, &idx
);
1914 uqn
= kmem_zalloc(sizeof (*uqn
), KM_SLEEP
);
1915 strlcpy(uqn
->uqn_id
, id
, sizeof (uqn
->uqn_id
));
1916 avl_insert(avl
, uqn
, idx
);
1918 uqn
->uqn_delta
+= delta
;
1922 do_userquota_update(objset_t
*os
, userquota_cache_t
*cache
, uint64_t used
,
1923 uint64_t flags
, uint64_t user
, uint64_t group
, uint64_t project
,
1926 if (flags
& DNODE_FLAG_USERUSED_ACCOUNTED
) {
1927 int64_t delta
= DNODE_MIN_SIZE
+ used
;
1933 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)user
);
1934 userquota_update_cache(&cache
->uqc_user_deltas
, name
, delta
);
1936 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)group
);
1937 userquota_update_cache(&cache
->uqc_group_deltas
, name
, delta
);
1939 if (dmu_objset_projectquota_enabled(os
)) {
1940 (void) snprintf(name
, sizeof (name
), "%llx",
1941 (longlong_t
)project
);
1942 userquota_update_cache(&cache
->uqc_project_deltas
,
1949 do_userobjquota_update(objset_t
*os
, userquota_cache_t
*cache
, uint64_t flags
,
1950 uint64_t user
, uint64_t group
, uint64_t project
, boolean_t subtract
)
1952 if (flags
& DNODE_FLAG_USEROBJUSED_ACCOUNTED
) {
1953 char name
[20 + DMU_OBJACCT_PREFIX_LEN
];
1954 int delta
= subtract
? -1 : 1;
1956 (void) snprintf(name
, sizeof (name
), DMU_OBJACCT_PREFIX
"%llx",
1958 userquota_update_cache(&cache
->uqc_user_deltas
, name
, delta
);
1960 (void) snprintf(name
, sizeof (name
), DMU_OBJACCT_PREFIX
"%llx",
1962 userquota_update_cache(&cache
->uqc_group_deltas
, name
, delta
);
1964 if (dmu_objset_projectquota_enabled(os
)) {
1965 (void) snprintf(name
, sizeof (name
),
1966 DMU_OBJACCT_PREFIX
"%llx", (longlong_t
)project
);
1967 userquota_update_cache(&cache
->uqc_project_deltas
,
1973 typedef struct userquota_updates_arg
{
1975 int uua_sublist_idx
;
1977 } userquota_updates_arg_t
;
1980 userquota_updates_task(void *arg
)
1982 userquota_updates_arg_t
*uua
= arg
;
1983 objset_t
*os
= uua
->uua_os
;
1984 dmu_tx_t
*tx
= uua
->uua_tx
;
1986 userquota_cache_t cache
= { { 0 } };
1988 multilist_sublist_t
*list
=
1989 multilist_sublist_lock(&os
->os_synced_dnodes
, uua
->uua_sublist_idx
);
1991 ASSERT(multilist_sublist_head(list
) == NULL
||
1992 dmu_objset_userused_enabled(os
));
1993 avl_create(&cache
.uqc_user_deltas
, userquota_compare
,
1994 sizeof (userquota_node_t
), offsetof(userquota_node_t
, uqn_node
));
1995 avl_create(&cache
.uqc_group_deltas
, userquota_compare
,
1996 sizeof (userquota_node_t
), offsetof(userquota_node_t
, uqn_node
));
1997 if (dmu_objset_projectquota_enabled(os
))
1998 avl_create(&cache
.uqc_project_deltas
, userquota_compare
,
1999 sizeof (userquota_node_t
), offsetof(userquota_node_t
,
2002 while ((dn
= multilist_sublist_head(list
)) != NULL
) {
2004 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
2005 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
2006 dn
->dn_phys
->dn_flags
&
2007 DNODE_FLAG_USERUSED_ACCOUNTED
);
2009 flags
= dn
->dn_id_flags
;
2011 if (flags
& DN_ID_OLD_EXIST
) {
2012 do_userquota_update(os
, &cache
, dn
->dn_oldused
,
2013 dn
->dn_oldflags
, dn
->dn_olduid
, dn
->dn_oldgid
,
2014 dn
->dn_oldprojid
, B_TRUE
);
2015 do_userobjquota_update(os
, &cache
, dn
->dn_oldflags
,
2016 dn
->dn_olduid
, dn
->dn_oldgid
,
2017 dn
->dn_oldprojid
, B_TRUE
);
2019 if (flags
& DN_ID_NEW_EXIST
) {
2020 do_userquota_update(os
, &cache
,
2021 DN_USED_BYTES(dn
->dn_phys
), dn
->dn_phys
->dn_flags
,
2022 dn
->dn_newuid
, dn
->dn_newgid
,
2023 dn
->dn_newprojid
, B_FALSE
);
2024 do_userobjquota_update(os
, &cache
,
2025 dn
->dn_phys
->dn_flags
, dn
->dn_newuid
, dn
->dn_newgid
,
2026 dn
->dn_newprojid
, B_FALSE
);
2029 mutex_enter(&dn
->dn_mtx
);
2031 dn
->dn_oldflags
= 0;
2032 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
2033 dn
->dn_olduid
= dn
->dn_newuid
;
2034 dn
->dn_oldgid
= dn
->dn_newgid
;
2035 dn
->dn_oldprojid
= dn
->dn_newprojid
;
2036 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
2037 if (dn
->dn_bonuslen
== 0)
2038 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
2040 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
2042 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
2043 mutex_exit(&dn
->dn_mtx
);
2045 multilist_sublist_remove(list
, dn
);
2046 dnode_rele(dn
, &os
->os_synced_dnodes
);
2048 do_userquota_cacheflush(os
, &cache
, tx
);
2049 multilist_sublist_unlock(list
);
2050 kmem_free(uua
, sizeof (*uua
));
2054 * Release dnode holds from dmu_objset_sync_dnodes(). When the dnode is being
2055 * synced (i.e. we have issued the zio's for blocks in the dnode), it can't be
2056 * evicted because the block containing the dnode can't be evicted until it is
2057 * written out. However, this hold is necessary to prevent the dnode_t from
2058 * being moved (via dnode_move()) while it's still referenced by
2059 * dbuf_dirty_record_t:dr_dnode. And dr_dnode is needed for
2060 * dirty_lightweight_leaf-type dirty records.
2062 * If we are doing user-object accounting, the dnode_rele() happens from
2063 * userquota_updates_task() instead.
2066 dnode_rele_task(void *arg
)
2068 userquota_updates_arg_t
*uua
= arg
;
2069 objset_t
*os
= uua
->uua_os
;
2071 multilist_sublist_t
*list
=
2072 multilist_sublist_lock(&os
->os_synced_dnodes
, uua
->uua_sublist_idx
);
2075 while ((dn
= multilist_sublist_head(list
)) != NULL
) {
2076 multilist_sublist_remove(list
, dn
);
2077 dnode_rele(dn
, &os
->os_synced_dnodes
);
2079 multilist_sublist_unlock(list
);
2080 kmem_free(uua
, sizeof (*uua
));
2084 * Return TRUE if userquota updates are needed.
2087 dmu_objset_do_userquota_updates_prep(objset_t
*os
, dmu_tx_t
*tx
)
2089 if (!dmu_objset_userused_enabled(os
))
2093 * If this is a raw receive just return and handle accounting
2094 * later when we have the keys loaded. We also don't do user
2095 * accounting during claiming since the datasets are not owned
2096 * for the duration of claiming and this txg should only be
2097 * used for recovery.
2099 if (os
->os_encrypted
&& dmu_objset_is_receiving(os
))
2102 if (tx
->tx_txg
<= os
->os_spa
->spa_claim_max_txg
)
2105 /* Allocate the user/group/project used objects if necessary. */
2106 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
2107 VERIFY0(zap_create_claim(os
,
2108 DMU_USERUSED_OBJECT
,
2109 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
2110 VERIFY0(zap_create_claim(os
,
2111 DMU_GROUPUSED_OBJECT
,
2112 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
2115 if (dmu_objset_projectquota_enabled(os
) &&
2116 DMU_PROJECTUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
2117 VERIFY0(zap_create_claim(os
, DMU_PROJECTUSED_OBJECT
,
2118 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
2124 * Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and
2125 * also release the holds on the dnodes from dmu_objset_sync_dnodes().
2126 * The caller must taskq_wait(dp_sync_taskq).
2129 dmu_objset_sync_done(objset_t
*os
, dmu_tx_t
*tx
)
2131 boolean_t need_userquota
= dmu_objset_do_userquota_updates_prep(os
, tx
);
2133 int num_sublists
= multilist_get_num_sublists(&os
->os_synced_dnodes
);
2134 for (int i
= 0; i
< num_sublists
; i
++) {
2135 userquota_updates_arg_t
*uua
=
2136 kmem_alloc(sizeof (*uua
), KM_SLEEP
);
2138 uua
->uua_sublist_idx
= i
;
2142 * If we don't need to update userquotas, use
2143 * dnode_rele_task() to call dnode_rele()
2145 (void) taskq_dispatch(dmu_objset_pool(os
)->dp_sync_taskq
,
2146 need_userquota
? userquota_updates_task
: dnode_rele_task
,
2148 /* callback frees uua */
2154 * Returns a pointer to data to find uid/gid from
2156 * If a dirty record for transaction group that is syncing can't
2157 * be found then NULL is returned. In the NULL case it is assumed
2158 * the uid/gid aren't changing.
2161 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
2163 dbuf_dirty_record_t
*dr
;
2166 if (db
->db_dirtycnt
== 0)
2167 return (db
->db
.db_data
); /* Nothing is changing */
2169 dr
= dbuf_find_dirty_eq(db
, tx
->tx_txg
);
2174 if (dr
->dr_dnode
->dn_bonuslen
== 0 &&
2175 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
2176 data
= dr
->dt
.dl
.dr_data
->b_data
;
2178 data
= dr
->dt
.dl
.dr_data
;
2185 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
2187 objset_t
*os
= dn
->dn_objset
;
2189 dmu_buf_impl_t
*db
= NULL
;
2190 int flags
= dn
->dn_id_flags
;
2192 boolean_t have_spill
= B_FALSE
;
2194 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
2198 * Raw receives introduce a problem with user accounting. Raw
2199 * receives cannot update the user accounting info because the
2200 * user ids and the sizes are encrypted. To guarantee that we
2201 * never end up with bad user accounting, we simply disable it
2202 * during raw receives. We also disable this for normal receives
2203 * so that an incremental raw receive may be done on top of an
2204 * existing non-raw receive.
2206 if (os
->os_encrypted
&& dmu_objset_is_receiving(os
))
2209 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
2210 DN_ID_CHKED_SPILL
)))
2213 if (before
&& dn
->dn_bonuslen
!= 0)
2214 data
= DN_BONUS(dn
->dn_phys
);
2215 else if (!before
&& dn
->dn_bonuslen
!= 0) {
2218 mutex_enter(&db
->db_mtx
);
2219 data
= dmu_objset_userquota_find_data(db
, tx
);
2221 data
= DN_BONUS(dn
->dn_phys
);
2223 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
2226 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
2227 rf
|= DB_RF_HAVESTRUCT
;
2228 error
= dmu_spill_hold_by_dnode(dn
,
2229 rf
| DB_RF_MUST_SUCCEED
,
2230 FTAG
, (dmu_buf_t
**)&db
);
2232 mutex_enter(&db
->db_mtx
);
2233 data
= (before
) ? db
->db
.db_data
:
2234 dmu_objset_userquota_find_data(db
, tx
);
2235 have_spill
= B_TRUE
;
2237 mutex_enter(&dn
->dn_mtx
);
2238 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
2239 mutex_exit(&dn
->dn_mtx
);
2244 * Must always call the callback in case the object
2245 * type has changed and that type isn't an object type to track
2247 zfs_file_info_t zfi
;
2248 error
= file_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
, &zfi
);
2252 dn
->dn_olduid
= zfi
.zfi_user
;
2253 dn
->dn_oldgid
= zfi
.zfi_group
;
2254 dn
->dn_oldprojid
= zfi
.zfi_project
;
2256 dn
->dn_newuid
= zfi
.zfi_user
;
2257 dn
->dn_newgid
= zfi
.zfi_group
;
2258 dn
->dn_newprojid
= zfi
.zfi_project
;
2262 * Preserve existing uid/gid when the callback can't determine
2263 * what the new uid/gid are and the callback returned EEXIST.
2264 * The EEXIST error tells us to just use the existing uid/gid.
2265 * If we don't know what the old values are then just assign
2266 * them to 0, since that is a new file being created.
2268 if (!before
&& data
== NULL
&& error
== EEXIST
) {
2269 if (flags
& DN_ID_OLD_EXIST
) {
2270 dn
->dn_newuid
= dn
->dn_olduid
;
2271 dn
->dn_newgid
= dn
->dn_oldgid
;
2272 dn
->dn_newprojid
= dn
->dn_oldprojid
;
2276 dn
->dn_newprojid
= ZFS_DEFAULT_PROJID
;
2282 mutex_exit(&db
->db_mtx
);
2284 mutex_enter(&dn
->dn_mtx
);
2285 if (error
== 0 && before
)
2286 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
2287 if (error
== 0 && !before
)
2288 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
2291 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
2293 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
2295 mutex_exit(&dn
->dn_mtx
);
2297 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
2301 dmu_objset_userspace_present(objset_t
*os
)
2303 return (os
->os_phys
->os_flags
&
2304 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
2308 dmu_objset_userobjspace_present(objset_t
*os
)
2310 return (os
->os_phys
->os_flags
&
2311 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
);
2315 dmu_objset_projectquota_present(objset_t
*os
)
2317 return (os
->os_phys
->os_flags
&
2318 OBJSET_FLAG_PROJECTQUOTA_COMPLETE
);
2322 dmu_objset_space_upgrade(objset_t
*os
)
2328 * We simply need to mark every object dirty, so that it will be
2329 * synced out and now accounted. If this is called
2330 * concurrently, or if we already did some work before crashing,
2331 * that's fine, since we track each object's accounted state
2335 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
2340 mutex_enter(&os
->os_upgrade_lock
);
2341 if (os
->os_upgrade_exit
)
2342 err
= SET_ERROR(EINTR
);
2343 mutex_exit(&os
->os_upgrade_lock
);
2347 if (issig(JUSTLOOKING
) && issig(FORREAL
))
2348 return (SET_ERROR(EINTR
));
2350 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
2353 tx
= dmu_tx_create(os
);
2354 dmu_tx_hold_bonus(tx
, obj
);
2355 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
2357 dmu_buf_rele(db
, FTAG
);
2361 dmu_buf_will_dirty(db
, tx
);
2362 dmu_buf_rele(db
, FTAG
);
2369 dmu_objset_userspace_upgrade_cb(objset_t
*os
)
2373 if (dmu_objset_userspace_present(os
))
2375 if (dmu_objset_is_snapshot(os
))
2376 return (SET_ERROR(EINVAL
));
2377 if (!dmu_objset_userused_enabled(os
))
2378 return (SET_ERROR(ENOTSUP
));
2380 err
= dmu_objset_space_upgrade(os
);
2384 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
2385 txg_wait_synced(dmu_objset_pool(os
), 0);
2390 dmu_objset_userspace_upgrade(objset_t
*os
)
2392 dmu_objset_upgrade(os
, dmu_objset_userspace_upgrade_cb
);
2396 dmu_objset_id_quota_upgrade_cb(objset_t
*os
)
2400 if (dmu_objset_userobjspace_present(os
) &&
2401 dmu_objset_projectquota_present(os
))
2403 if (dmu_objset_is_snapshot(os
))
2404 return (SET_ERROR(EINVAL
));
2405 if (!dmu_objset_userused_enabled(os
))
2406 return (SET_ERROR(ENOTSUP
));
2407 if (!dmu_objset_projectquota_enabled(os
) &&
2408 dmu_objset_userobjspace_present(os
))
2409 return (SET_ERROR(ENOTSUP
));
2411 err
= dmu_objset_space_upgrade(os
);
2415 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
2416 if (dmu_objset_userobjused_enabled(os
))
2417 os
->os_flags
|= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
;
2418 if (dmu_objset_projectquota_enabled(os
))
2419 os
->os_flags
|= OBJSET_FLAG_PROJECTQUOTA_COMPLETE
;
2421 txg_wait_synced(dmu_objset_pool(os
), 0);
2426 dmu_objset_id_quota_upgrade(objset_t
*os
)
2428 dmu_objset_upgrade(os
, dmu_objset_id_quota_upgrade_cb
);
2432 dmu_objset_userobjspace_upgradable(objset_t
*os
)
2434 return (dmu_objset_type(os
) == DMU_OST_ZFS
&&
2435 !dmu_objset_is_snapshot(os
) &&
2436 dmu_objset_userobjused_enabled(os
) &&
2437 !dmu_objset_userobjspace_present(os
) &&
2438 spa_writeable(dmu_objset_spa(os
)));
2442 dmu_objset_projectquota_upgradable(objset_t
*os
)
2444 return (dmu_objset_type(os
) == DMU_OST_ZFS
&&
2445 !dmu_objset_is_snapshot(os
) &&
2446 dmu_objset_projectquota_enabled(os
) &&
2447 !dmu_objset_projectquota_present(os
) &&
2448 spa_writeable(dmu_objset_spa(os
)));
2452 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
2453 uint64_t *usedobjsp
, uint64_t *availobjsp
)
2455 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
2456 usedobjsp
, availobjsp
);
2460 dmu_objset_fsid_guid(objset_t
*os
)
2462 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
2466 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
2468 stat
->dds_type
= os
->os_phys
->os_type
;
2469 if (os
->os_dsl_dataset
)
2470 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
2474 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
2476 ASSERT(os
->os_dsl_dataset
||
2477 os
->os_phys
->os_type
== DMU_OST_META
);
2479 if (os
->os_dsl_dataset
!= NULL
)
2480 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
2482 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
2483 os
->os_phys
->os_type
);
2484 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
2485 dmu_objset_userspace_present(os
));
2489 dmu_objset_is_snapshot(objset_t
*os
)
2491 if (os
->os_dsl_dataset
!= NULL
)
2492 return (os
->os_dsl_dataset
->ds_is_snapshot
);
2498 dmu_snapshot_realname(objset_t
*os
, const char *name
, char *real
, int maxlen
,
2499 boolean_t
*conflict
)
2501 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
2504 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
2505 return (SET_ERROR(ENOENT
));
2507 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
2508 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, name
, 8, 1, &ignored
,
2509 MT_NORMALIZE
, real
, maxlen
, conflict
));
2513 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
2514 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
2516 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
2517 zap_cursor_t cursor
;
2518 zap_attribute_t attr
;
2520 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
2522 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
2523 return (SET_ERROR(ENOENT
));
2525 zap_cursor_init_serialized(&cursor
,
2526 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
2527 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, *offp
);
2529 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
2530 zap_cursor_fini(&cursor
);
2531 return (SET_ERROR(ENOENT
));
2534 if (strlen(attr
.za_name
) + 1 > namelen
) {
2535 zap_cursor_fini(&cursor
);
2536 return (SET_ERROR(ENAMETOOLONG
));
2539 (void) strlcpy(name
, attr
.za_name
, namelen
);
2541 *idp
= attr
.za_first_integer
;
2543 *case_conflict
= attr
.za_normalization_conflict
;
2544 zap_cursor_advance(&cursor
);
2545 *offp
= zap_cursor_serialize(&cursor
);
2546 zap_cursor_fini(&cursor
);
2552 dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *value
)
2554 return (dsl_dataset_snap_lookup(os
->os_dsl_dataset
, name
, value
));
2558 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
2559 uint64_t *idp
, uint64_t *offp
)
2561 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
2562 zap_cursor_t cursor
;
2563 zap_attribute_t attr
;
2565 /* there is no next dir on a snapshot! */
2566 if (os
->os_dsl_dataset
->ds_object
!=
2567 dsl_dir_phys(dd
)->dd_head_dataset_obj
)
2568 return (SET_ERROR(ENOENT
));
2570 zap_cursor_init_serialized(&cursor
,
2571 dd
->dd_pool
->dp_meta_objset
,
2572 dsl_dir_phys(dd
)->dd_child_dir_zapobj
, *offp
);
2574 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
2575 zap_cursor_fini(&cursor
);
2576 return (SET_ERROR(ENOENT
));
2579 if (strlen(attr
.za_name
) + 1 > namelen
) {
2580 zap_cursor_fini(&cursor
);
2581 return (SET_ERROR(ENAMETOOLONG
));
2584 (void) strlcpy(name
, attr
.za_name
, namelen
);
2586 *idp
= attr
.za_first_integer
;
2587 zap_cursor_advance(&cursor
);
2588 *offp
= zap_cursor_serialize(&cursor
);
2589 zap_cursor_fini(&cursor
);
2594 typedef struct dmu_objset_find_ctx
{
2598 char *dc_ddname
; /* last component of ddobj's name */
2599 int (*dc_func
)(dsl_pool_t
*, dsl_dataset_t
*, void *);
2602 kmutex_t
*dc_error_lock
;
2604 } dmu_objset_find_ctx_t
;
2607 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t
*dcp
)
2609 dsl_pool_t
*dp
= dcp
->dc_dp
;
2613 zap_attribute_t
*attr
;
2617 /* don't process if there already was an error */
2618 if (*dcp
->dc_error
!= 0)
2622 * Note: passing the name (dc_ddname) here is optional, but it
2623 * improves performance because we don't need to call
2624 * zap_value_search() to determine the name.
2626 err
= dsl_dir_hold_obj(dp
, dcp
->dc_ddobj
, dcp
->dc_ddname
, FTAG
, &dd
);
2630 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2631 if (dd
->dd_myname
[0] == '$') {
2632 dsl_dir_rele(dd
, FTAG
);
2636 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
2637 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
2640 * Iterate over all children.
2642 if (dcp
->dc_flags
& DS_FIND_CHILDREN
) {
2643 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2644 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
2645 zap_cursor_retrieve(&zc
, attr
) == 0;
2646 (void) zap_cursor_advance(&zc
)) {
2647 ASSERT3U(attr
->za_integer_length
, ==,
2649 ASSERT3U(attr
->za_num_integers
, ==, 1);
2651 dmu_objset_find_ctx_t
*child_dcp
=
2652 kmem_alloc(sizeof (*child_dcp
), KM_SLEEP
);
2654 child_dcp
->dc_ddobj
= attr
->za_first_integer
;
2655 child_dcp
->dc_ddname
= spa_strdup(attr
->za_name
);
2656 if (dcp
->dc_tq
!= NULL
)
2657 (void) taskq_dispatch(dcp
->dc_tq
,
2658 dmu_objset_find_dp_cb
, child_dcp
, TQ_SLEEP
);
2660 dmu_objset_find_dp_impl(child_dcp
);
2662 zap_cursor_fini(&zc
);
2666 * Iterate over all snapshots.
2668 if (dcp
->dc_flags
& DS_FIND_SNAPSHOTS
) {
2670 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2675 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
2676 dsl_dataset_rele(ds
, FTAG
);
2678 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
2679 zap_cursor_retrieve(&zc
, attr
) == 0;
2680 (void) zap_cursor_advance(&zc
)) {
2681 ASSERT3U(attr
->za_integer_length
, ==,
2683 ASSERT3U(attr
->za_num_integers
, ==, 1);
2685 err
= dsl_dataset_hold_obj(dp
,
2686 attr
->za_first_integer
, FTAG
, &ds
);
2689 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
2690 dsl_dataset_rele(ds
, FTAG
);
2694 zap_cursor_fini(&zc
);
2698 kmem_free(attr
, sizeof (zap_attribute_t
));
2701 dsl_dir_rele(dd
, FTAG
);
2708 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2711 * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2712 * that the dir will remain cached, and we won't have to re-instantiate
2713 * it (which could be expensive due to finding its name via
2714 * zap_value_search()).
2716 dsl_dir_rele(dd
, FTAG
);
2719 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
2720 dsl_dataset_rele(ds
, FTAG
);
2724 mutex_enter(dcp
->dc_error_lock
);
2725 /* only keep first error */
2726 if (*dcp
->dc_error
== 0)
2727 *dcp
->dc_error
= err
;
2728 mutex_exit(dcp
->dc_error_lock
);
2731 if (dcp
->dc_ddname
!= NULL
)
2732 spa_strfree(dcp
->dc_ddname
);
2733 kmem_free(dcp
, sizeof (*dcp
));
2737 dmu_objset_find_dp_cb(void *arg
)
2739 dmu_objset_find_ctx_t
*dcp
= arg
;
2740 dsl_pool_t
*dp
= dcp
->dc_dp
;
2743 * We need to get a pool_config_lock here, as there are several
2744 * assert(pool_config_held) down the stack. Getting a lock via
2745 * dsl_pool_config_enter is risky, as it might be stalled by a
2746 * pending writer. This would deadlock, as the write lock can
2747 * only be granted when our parent thread gives up the lock.
2748 * The _prio interface gives us priority over a pending writer.
2750 dsl_pool_config_enter_prio(dp
, FTAG
);
2752 dmu_objset_find_dp_impl(dcp
);
2754 dsl_pool_config_exit(dp
, FTAG
);
2758 * Find objsets under and including ddobj, call func(ds) on each.
2759 * The order for the enumeration is completely undefined.
2760 * func is called with dsl_pool_config held.
2763 dmu_objset_find_dp(dsl_pool_t
*dp
, uint64_t ddobj
,
2764 int func(dsl_pool_t
*, dsl_dataset_t
*, void *), void *arg
, int flags
)
2769 dmu_objset_find_ctx_t
*dcp
;
2772 mutex_init(&err_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2773 dcp
= kmem_alloc(sizeof (*dcp
), KM_SLEEP
);
2776 dcp
->dc_ddobj
= ddobj
;
2777 dcp
->dc_ddname
= NULL
;
2778 dcp
->dc_func
= func
;
2780 dcp
->dc_flags
= flags
;
2781 dcp
->dc_error_lock
= &err_lock
;
2782 dcp
->dc_error
= &error
;
2784 if ((flags
& DS_FIND_SERIALIZE
) || dsl_pool_config_held_writer(dp
)) {
2786 * In case a write lock is held we can't make use of
2787 * parallelism, as down the stack of the worker threads
2788 * the lock is asserted via dsl_pool_config_held.
2789 * In case of a read lock this is solved by getting a read
2790 * lock in each worker thread, which isn't possible in case
2791 * of a writer lock. So we fall back to the synchronous path
2793 * In the future it might be possible to get some magic into
2794 * dsl_pool_config_held in a way that it returns true for
2795 * the worker threads so that a single lock held from this
2796 * thread suffices. For now, stay single threaded.
2798 dmu_objset_find_dp_impl(dcp
);
2799 mutex_destroy(&err_lock
);
2804 ntasks
= dmu_find_threads
;
2806 ntasks
= vdev_count_leaves(dp
->dp_spa
) * 4;
2807 tq
= taskq_create("dmu_objset_find", ntasks
, maxclsyspri
, ntasks
,
2810 kmem_free(dcp
, sizeof (*dcp
));
2811 mutex_destroy(&err_lock
);
2813 return (SET_ERROR(ENOMEM
));
2817 /* dcp will be freed by task */
2818 (void) taskq_dispatch(tq
, dmu_objset_find_dp_cb
, dcp
, TQ_SLEEP
);
2821 * PORTING: this code relies on the property of taskq_wait to wait
2822 * until no more tasks are queued and no more tasks are active. As
2823 * we always queue new tasks from within other tasks, task_wait
2824 * reliably waits for the full recursion to finish, even though we
2825 * enqueue new tasks after taskq_wait has been called.
2826 * On platforms other than illumos, taskq_wait may not have this
2831 mutex_destroy(&err_lock
);
2837 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2838 * The dp_config_rwlock must not be held when this is called, and it
2839 * will not be held when the callback is called.
2840 * Therefore this function should only be used when the pool is not changing
2841 * (e.g. in syncing context), or the callback can deal with the possible races.
2844 dmu_objset_find_impl(spa_t
*spa
, const char *name
,
2845 int func(const char *, void *), void *arg
, int flags
)
2848 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2851 zap_attribute_t
*attr
;
2856 dsl_pool_config_enter(dp
, FTAG
);
2858 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, NULL
);
2860 dsl_pool_config_exit(dp
, FTAG
);
2864 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2865 if (dd
->dd_myname
[0] == '$') {
2866 dsl_dir_rele(dd
, FTAG
);
2867 dsl_pool_config_exit(dp
, FTAG
);
2871 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
2872 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
2875 * Iterate over all children.
2877 if (flags
& DS_FIND_CHILDREN
) {
2878 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2879 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
2880 zap_cursor_retrieve(&zc
, attr
) == 0;
2881 (void) zap_cursor_advance(&zc
)) {
2882 ASSERT3U(attr
->za_integer_length
, ==,
2884 ASSERT3U(attr
->za_num_integers
, ==, 1);
2886 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
2887 dsl_pool_config_exit(dp
, FTAG
);
2888 err
= dmu_objset_find_impl(spa
, child
,
2890 dsl_pool_config_enter(dp
, FTAG
);
2891 kmem_strfree(child
);
2895 zap_cursor_fini(&zc
);
2898 dsl_dir_rele(dd
, FTAG
);
2899 dsl_pool_config_exit(dp
, FTAG
);
2900 kmem_free(attr
, sizeof (zap_attribute_t
));
2906 * Iterate over all snapshots.
2908 if (flags
& DS_FIND_SNAPSHOTS
) {
2909 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2914 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
2915 dsl_dataset_rele(ds
, FTAG
);
2917 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
2918 zap_cursor_retrieve(&zc
, attr
) == 0;
2919 (void) zap_cursor_advance(&zc
)) {
2920 ASSERT3U(attr
->za_integer_length
, ==,
2922 ASSERT3U(attr
->za_num_integers
, ==, 1);
2924 child
= kmem_asprintf("%s@%s",
2925 name
, attr
->za_name
);
2926 dsl_pool_config_exit(dp
, FTAG
);
2927 err
= func(child
, arg
);
2928 dsl_pool_config_enter(dp
, FTAG
);
2929 kmem_strfree(child
);
2933 zap_cursor_fini(&zc
);
2937 dsl_dir_rele(dd
, FTAG
);
2938 kmem_free(attr
, sizeof (zap_attribute_t
));
2939 dsl_pool_config_exit(dp
, FTAG
);
2944 /* Apply to self. */
2945 return (func(name
, arg
));
2949 * See comment above dmu_objset_find_impl().
2952 dmu_objset_find(const char *name
, int func(const char *, void *), void *arg
,
2958 error
= spa_open(name
, &spa
, FTAG
);
2961 error
= dmu_objset_find_impl(spa
, name
, func
, arg
, flags
);
2962 spa_close(spa
, FTAG
);
2967 dmu_objset_incompatible_encryption_version(objset_t
*os
)
2969 return (dsl_dir_incompatible_encryption_version(
2970 os
->os_dsl_dataset
->ds_dir
));
2974 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
2976 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2977 os
->os_user_ptr
= user_ptr
;
2981 dmu_objset_get_user(objset_t
*os
)
2983 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2984 return (os
->os_user_ptr
);
2988 * Determine name of filesystem, given name of snapshot.
2989 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2992 dmu_fsname(const char *snapname
, char *buf
)
2994 char *atp
= strchr(snapname
, '@');
2996 return (SET_ERROR(EINVAL
));
2997 if (atp
- snapname
>= ZFS_MAX_DATASET_NAME_LEN
)
2998 return (SET_ERROR(ENAMETOOLONG
));
2999 (void) strlcpy(buf
, snapname
, atp
- snapname
+ 1);
3004 * Call when we think we're going to write/free space in open context
3005 * to track the amount of dirty data in the open txg, which is also the
3006 * amount of memory that can not be evicted until this txg syncs.
3008 * Note that there are two conditions where this can be called from
3011 * [1] When we just created the dataset, in which case we go on with
3012 * updating any accounting of dirty data as usual.
3013 * [2] When we are dirtying MOS data, in which case we only update the
3014 * pool's accounting of dirty data.
3017 dmu_objset_willuse_space(objset_t
*os
, int64_t space
, dmu_tx_t
*tx
)
3019 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
3020 int64_t aspace
= spa_get_worst_case_asize(os
->os_spa
, space
);
3023 dsl_dir_willuse_space(ds
->ds_dir
, aspace
, tx
);
3026 dsl_pool_dirty_space(dmu_tx_pool(tx
), space
, tx
);
3029 #if defined(_KERNEL)
3030 EXPORT_SYMBOL(dmu_objset_zil
);
3031 EXPORT_SYMBOL(dmu_objset_pool
);
3032 EXPORT_SYMBOL(dmu_objset_ds
);
3033 EXPORT_SYMBOL(dmu_objset_type
);
3034 EXPORT_SYMBOL(dmu_objset_name
);
3035 EXPORT_SYMBOL(dmu_objset_hold
);
3036 EXPORT_SYMBOL(dmu_objset_hold_flags
);
3037 EXPORT_SYMBOL(dmu_objset_own
);
3038 EXPORT_SYMBOL(dmu_objset_rele
);
3039 EXPORT_SYMBOL(dmu_objset_rele_flags
);
3040 EXPORT_SYMBOL(dmu_objset_disown
);
3041 EXPORT_SYMBOL(dmu_objset_from_ds
);
3042 EXPORT_SYMBOL(dmu_objset_create
);
3043 EXPORT_SYMBOL(dmu_objset_clone
);
3044 EXPORT_SYMBOL(dmu_objset_stats
);
3045 EXPORT_SYMBOL(dmu_objset_fast_stat
);
3046 EXPORT_SYMBOL(dmu_objset_spa
);
3047 EXPORT_SYMBOL(dmu_objset_space
);
3048 EXPORT_SYMBOL(dmu_objset_fsid_guid
);
3049 EXPORT_SYMBOL(dmu_objset_find
);
3050 EXPORT_SYMBOL(dmu_objset_byteswap
);
3051 EXPORT_SYMBOL(dmu_objset_evict_dbufs
);
3052 EXPORT_SYMBOL(dmu_objset_snap_cmtime
);
3053 EXPORT_SYMBOL(dmu_objset_dnodesize
);
3055 EXPORT_SYMBOL(dmu_objset_sync
);
3056 EXPORT_SYMBOL(dmu_objset_is_dirty
);
3057 EXPORT_SYMBOL(dmu_objset_create_impl_dnstats
);
3058 EXPORT_SYMBOL(dmu_objset_create_impl
);
3059 EXPORT_SYMBOL(dmu_objset_open_impl
);
3060 EXPORT_SYMBOL(dmu_objset_evict
);
3061 EXPORT_SYMBOL(dmu_objset_register_type
);
3062 EXPORT_SYMBOL(dmu_objset_sync_done
);
3063 EXPORT_SYMBOL(dmu_objset_userquota_get_ids
);
3064 EXPORT_SYMBOL(dmu_objset_userused_enabled
);
3065 EXPORT_SYMBOL(dmu_objset_userspace_upgrade
);
3066 EXPORT_SYMBOL(dmu_objset_userspace_present
);
3067 EXPORT_SYMBOL(dmu_objset_userobjused_enabled
);
3068 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable
);
3069 EXPORT_SYMBOL(dmu_objset_userobjspace_present
);
3070 EXPORT_SYMBOL(dmu_objset_projectquota_enabled
);
3071 EXPORT_SYMBOL(dmu_objset_projectquota_present
);
3072 EXPORT_SYMBOL(dmu_objset_projectquota_upgradable
);
3073 EXPORT_SYMBOL(dmu_objset_id_quota_upgrade
);