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, 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
37 /* Portions Copyright 2010 Robert Milkowski */
40 #include <sys/zfs_context.h>
41 #include <sys/dmu_objset.h>
42 #include <sys/dsl_dir.h>
43 #include <sys/dsl_dataset.h>
44 #include <sys/dsl_prop.h>
45 #include <sys/dsl_pool.h>
46 #include <sys/dsl_synctask.h>
47 #include <sys/dsl_deleg.h>
48 #include <sys/dnode.h>
51 #include <sys/dmu_tx.h>
54 #include <sys/dmu_impl.h>
55 #include <sys/zfs_ioctl.h>
57 #include <sys/zfs_onexit.h>
58 #include <sys/dsl_destroy.h>
60 #include <sys/zfeature.h>
61 #include <sys/policy.h>
62 #include <sys/spa_impl.h>
63 #include <sys/dmu_recv.h>
64 #include <sys/zfs_project.h>
65 #include "zfs_namecheck.h"
66 #include <sys/vdev_impl.h>
70 * Needed to close a window in dnode_move() that allows the objset to be freed
71 * before it can be safely accessed.
76 * Tunable to overwrite the maximum number of threads for the parallelization
77 * of dmu_objset_find_dp, needed to speed up the import of pools with many
79 * Default is 4 times the number of leaf vdevs.
81 static const int dmu_find_threads
= 0;
84 * Backfill lower metadnode objects after this many have been freed.
85 * Backfilling negatively impacts object creation rates, so only do it
86 * if there are enough holes to fill.
88 static const int dmu_rescan_dnode_threshold
= 1 << DN_MAX_INDBLKSHIFT
;
90 static const char *upgrade_tag
= "upgrade_tag";
92 static void dmu_objset_find_dp_cb(void *arg
);
94 static void dmu_objset_upgrade(objset_t
*os
, dmu_objset_upgrade_cb_t cb
);
95 static void dmu_objset_upgrade_stop(objset_t
*os
);
100 rw_init(&os_lock
, NULL
, RW_DEFAULT
, NULL
);
104 dmu_objset_fini(void)
106 rw_destroy(&os_lock
);
110 dmu_objset_spa(objset_t
*os
)
116 dmu_objset_zil(objset_t
*os
)
122 dmu_objset_pool(objset_t
*os
)
126 if ((ds
= os
->os_dsl_dataset
) != NULL
&& ds
->ds_dir
)
127 return (ds
->ds_dir
->dd_pool
);
129 return (spa_get_dsl(os
->os_spa
));
133 dmu_objset_ds(objset_t
*os
)
135 return (os
->os_dsl_dataset
);
139 dmu_objset_type(objset_t
*os
)
141 return (os
->os_phys
->os_type
);
145 dmu_objset_name(objset_t
*os
, char *buf
)
147 dsl_dataset_name(os
->os_dsl_dataset
, buf
);
151 dmu_objset_id(objset_t
*os
)
153 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
155 return (ds
? ds
->ds_object
: 0);
159 dmu_objset_dnodesize(objset_t
*os
)
161 return (os
->os_dnodesize
);
165 dmu_objset_syncprop(objset_t
*os
)
167 return (os
->os_sync
);
171 dmu_objset_logbias(objset_t
*os
)
173 return (os
->os_logbias
);
177 checksum_changed_cb(void *arg
, uint64_t newval
)
182 * Inheritance should have been done by now.
184 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
186 os
->os_checksum
= zio_checksum_select(newval
, ZIO_CHECKSUM_ON_VALUE
);
190 compression_changed_cb(void *arg
, uint64_t newval
)
195 * Inheritance and range checking should have been done by now.
197 ASSERT(newval
!= ZIO_COMPRESS_INHERIT
);
199 os
->os_compress
= zio_compress_select(os
->os_spa
,
200 ZIO_COMPRESS_ALGO(newval
), ZIO_COMPRESS_ON
);
201 os
->os_complevel
= zio_complevel_select(os
->os_spa
, os
->os_compress
,
202 ZIO_COMPRESS_LEVEL(newval
), ZIO_COMPLEVEL_DEFAULT
);
206 copies_changed_cb(void *arg
, uint64_t newval
)
211 * Inheritance and range checking should have been done by now.
214 ASSERT(newval
<= spa_max_replication(os
->os_spa
));
216 os
->os_copies
= newval
;
220 dedup_changed_cb(void *arg
, uint64_t newval
)
223 spa_t
*spa
= os
->os_spa
;
224 enum zio_checksum checksum
;
227 * Inheritance should have been done by now.
229 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
231 checksum
= zio_checksum_dedup_select(spa
, newval
, ZIO_CHECKSUM_OFF
);
233 os
->os_dedup_checksum
= checksum
& ZIO_CHECKSUM_MASK
;
234 os
->os_dedup_verify
= !!(checksum
& ZIO_CHECKSUM_VERIFY
);
238 primary_cache_changed_cb(void *arg
, uint64_t newval
)
243 * Inheritance and range checking should have been done by now.
245 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
246 newval
== ZFS_CACHE_METADATA
);
248 os
->os_primary_cache
= newval
;
252 secondary_cache_changed_cb(void *arg
, uint64_t newval
)
257 * Inheritance and range checking should have been done by now.
259 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
260 newval
== ZFS_CACHE_METADATA
);
262 os
->os_secondary_cache
= newval
;
266 sync_changed_cb(void *arg
, uint64_t newval
)
271 * Inheritance and range checking should have been done by now.
273 ASSERT(newval
== ZFS_SYNC_STANDARD
|| newval
== ZFS_SYNC_ALWAYS
||
274 newval
== ZFS_SYNC_DISABLED
);
276 os
->os_sync
= newval
;
278 zil_set_sync(os
->os_zil
, newval
);
282 redundant_metadata_changed_cb(void *arg
, uint64_t newval
)
287 * Inheritance and range checking should have been done by now.
289 ASSERT(newval
== ZFS_REDUNDANT_METADATA_ALL
||
290 newval
== ZFS_REDUNDANT_METADATA_MOST
);
292 os
->os_redundant_metadata
= newval
;
296 dnodesize_changed_cb(void *arg
, uint64_t newval
)
301 case ZFS_DNSIZE_LEGACY
:
302 os
->os_dnodesize
= DNODE_MIN_SIZE
;
304 case ZFS_DNSIZE_AUTO
:
306 * Choose a dnode size that will work well for most
307 * workloads if the user specified "auto". Future code
308 * improvements could dynamically select a dnode size
309 * based on observed workload patterns.
311 os
->os_dnodesize
= DNODE_MIN_SIZE
* 2;
318 os
->os_dnodesize
= newval
;
324 smallblk_changed_cb(void *arg
, uint64_t newval
)
329 * Inheritance and range checking should have been done by now.
331 ASSERT(newval
<= SPA_MAXBLOCKSIZE
);
332 ASSERT(ISP2(newval
));
334 os
->os_zpl_special_smallblock
= newval
;
338 logbias_changed_cb(void *arg
, uint64_t newval
)
342 ASSERT(newval
== ZFS_LOGBIAS_LATENCY
||
343 newval
== ZFS_LOGBIAS_THROUGHPUT
);
344 os
->os_logbias
= newval
;
346 zil_set_logbias(os
->os_zil
, newval
);
350 recordsize_changed_cb(void *arg
, uint64_t newval
)
354 os
->os_recordsize
= newval
;
358 dmu_objset_byteswap(void *buf
, size_t size
)
360 objset_phys_t
*osp
= buf
;
362 ASSERT(size
== OBJSET_PHYS_SIZE_V1
|| size
== OBJSET_PHYS_SIZE_V2
||
363 size
== sizeof (objset_phys_t
));
364 dnode_byteswap(&osp
->os_meta_dnode
);
365 byteswap_uint64_array(&osp
->os_zil_header
, sizeof (zil_header_t
));
366 osp
->os_type
= BSWAP_64(osp
->os_type
);
367 osp
->os_flags
= BSWAP_64(osp
->os_flags
);
368 if (size
>= OBJSET_PHYS_SIZE_V2
) {
369 dnode_byteswap(&osp
->os_userused_dnode
);
370 dnode_byteswap(&osp
->os_groupused_dnode
);
371 if (size
>= sizeof (objset_phys_t
))
372 dnode_byteswap(&osp
->os_projectused_dnode
);
377 * The hash is a CRC-based hash of the objset_t pointer and the object number.
380 dnode_hash(const objset_t
*os
, uint64_t obj
)
382 uintptr_t osv
= (uintptr_t)os
;
383 uint64_t crc
= -1ULL;
385 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
387 * The low 6 bits of the pointer don't have much entropy, because
388 * the objset_t is larger than 2^6 bytes long.
390 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (osv
>> 6)) & 0xFF];
391 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 0)) & 0xFF];
392 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 8)) & 0xFF];
393 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 16)) & 0xFF];
395 crc
^= (osv
>>14) ^ (obj
>>24);
401 dnode_multilist_index_func(multilist_t
*ml
, void *obj
)
406 * The low order bits of the hash value are thought to be
407 * distributed evenly. Otherwise, in the case that the multilist
408 * has a power of two number of sublists, each sublists' usage
409 * would not be evenly distributed. In this context full 64bit
410 * division would be a waste of time, so limit it to 32 bits.
412 return ((unsigned int)dnode_hash(dn
->dn_objset
, dn
->dn_object
) %
413 multilist_get_num_sublists(ml
));
416 static inline boolean_t
417 dmu_os_is_l2cacheable(objset_t
*os
)
420 zfs_cache_type_t cache
= os
->os_secondary_cache
;
421 blkptr_t
*bp
= os
->os_rootbp
;
423 if (bp
!= NULL
&& !BP_IS_HOLE(bp
)) {
424 uint64_t vdev
= DVA_GET_VDEV(bp
->blk_dva
);
425 vdev_t
*rvd
= os
->os_spa
->spa_root_vdev
;
427 if (vdev
< rvd
->vdev_children
)
428 vd
= rvd
->vdev_child
[vdev
];
430 if (cache
== ZFS_CACHE_ALL
|| cache
== ZFS_CACHE_METADATA
) {
434 if ((vd
->vdev_alloc_bias
!= VDEV_BIAS_SPECIAL
&&
435 vd
->vdev_alloc_bias
!= VDEV_BIAS_DEDUP
) ||
436 l2arc_exclude_special
== 0)
445 * Instantiates the objset_t in-memory structure corresponding to the
446 * objset_phys_t that's pointed to by the specified blkptr_t.
449 dmu_objset_open_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
455 ASSERT(ds
== NULL
|| MUTEX_HELD(&ds
->ds_opening_lock
));
456 ASSERT(!BP_IS_REDACTED(bp
));
459 * We need the pool config lock to get properties.
461 ASSERT(ds
== NULL
|| dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
464 * The $ORIGIN dataset (if it exists) doesn't have an associated
465 * objset, so there's no reason to open it. The $ORIGIN dataset
466 * will not exist on pools older than SPA_VERSION_ORIGIN.
468 if (ds
!= NULL
&& spa_get_dsl(spa
) != NULL
&&
469 spa_get_dsl(spa
)->dp_origin_snap
!= NULL
) {
470 ASSERT3P(ds
->ds_dir
, !=,
471 spa_get_dsl(spa
)->dp_origin_snap
->ds_dir
);
474 os
= kmem_zalloc(sizeof (objset_t
), KM_SLEEP
);
475 os
->os_dsl_dataset
= ds
;
478 if (!BP_IS_HOLE(os
->os_rootbp
)) {
479 arc_flags_t aflags
= ARC_FLAG_WAIT
;
482 enum zio_flag zio_flags
= ZIO_FLAG_CANFAIL
;
483 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
484 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
486 if (dmu_os_is_l2cacheable(os
))
487 aflags
|= ARC_FLAG_L2CACHE
;
489 if (ds
!= NULL
&& ds
->ds_dir
->dd_crypto_obj
!= 0) {
490 ASSERT3U(BP_GET_COMPRESS(bp
), ==, ZIO_COMPRESS_OFF
);
491 ASSERT(BP_IS_AUTHENTICATED(bp
));
492 zio_flags
|= ZIO_FLAG_RAW
;
495 dprintf_bp(os
->os_rootbp
, "reading %s", "");
496 err
= arc_read(NULL
, spa
, os
->os_rootbp
,
497 arc_getbuf_func
, &os
->os_phys_buf
,
498 ZIO_PRIORITY_SYNC_READ
, zio_flags
, &aflags
, &zb
);
500 kmem_free(os
, sizeof (objset_t
));
501 /* convert checksum errors into IO errors */
503 err
= SET_ERROR(EIO
);
507 if (spa_version(spa
) < SPA_VERSION_USERSPACE
)
508 size
= OBJSET_PHYS_SIZE_V1
;
509 else if (!spa_feature_is_enabled(spa
,
510 SPA_FEATURE_PROJECT_QUOTA
))
511 size
= OBJSET_PHYS_SIZE_V2
;
513 size
= sizeof (objset_phys_t
);
515 /* Increase the blocksize if we are permitted. */
516 if (arc_buf_size(os
->os_phys_buf
) < size
) {
517 arc_buf_t
*buf
= arc_alloc_buf(spa
, &os
->os_phys_buf
,
518 ARC_BUFC_METADATA
, size
);
519 bzero(buf
->b_data
, size
);
520 bcopy(os
->os_phys_buf
->b_data
, buf
->b_data
,
521 arc_buf_size(os
->os_phys_buf
));
522 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
523 os
->os_phys_buf
= buf
;
526 os
->os_phys
= os
->os_phys_buf
->b_data
;
527 os
->os_flags
= os
->os_phys
->os_flags
;
529 int size
= spa_version(spa
) >= SPA_VERSION_USERSPACE
?
530 sizeof (objset_phys_t
) : OBJSET_PHYS_SIZE_V1
;
531 os
->os_phys_buf
= arc_alloc_buf(spa
, &os
->os_phys_buf
,
532 ARC_BUFC_METADATA
, size
);
533 os
->os_phys
= os
->os_phys_buf
->b_data
;
534 bzero(os
->os_phys
, size
);
537 * These properties will be filled in by the logic in zfs_get_zplprop()
538 * when they are queried for the first time.
540 os
->os_version
= OBJSET_PROP_UNINITIALIZED
;
541 os
->os_normalization
= OBJSET_PROP_UNINITIALIZED
;
542 os
->os_utf8only
= OBJSET_PROP_UNINITIALIZED
;
543 os
->os_casesensitivity
= OBJSET_PROP_UNINITIALIZED
;
546 * Note: the changed_cb will be called once before the register
547 * func returns, thus changing the checksum/compression from the
548 * default (fletcher2/off). Snapshots don't need to know about
549 * checksum/compression/copies.
552 os
->os_encrypted
= (ds
->ds_dir
->dd_crypto_obj
!= 0);
554 err
= dsl_prop_register(ds
,
555 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
556 primary_cache_changed_cb
, os
);
558 err
= dsl_prop_register(ds
,
559 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
560 secondary_cache_changed_cb
, os
);
562 if (!ds
->ds_is_snapshot
) {
564 err
= dsl_prop_register(ds
,
565 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
566 checksum_changed_cb
, os
);
569 err
= dsl_prop_register(ds
,
570 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
571 compression_changed_cb
, os
);
574 err
= dsl_prop_register(ds
,
575 zfs_prop_to_name(ZFS_PROP_COPIES
),
576 copies_changed_cb
, os
);
579 err
= dsl_prop_register(ds
,
580 zfs_prop_to_name(ZFS_PROP_DEDUP
),
581 dedup_changed_cb
, os
);
584 err
= dsl_prop_register(ds
,
585 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
586 logbias_changed_cb
, os
);
589 err
= dsl_prop_register(ds
,
590 zfs_prop_to_name(ZFS_PROP_SYNC
),
591 sync_changed_cb
, os
);
594 err
= dsl_prop_register(ds
,
596 ZFS_PROP_REDUNDANT_METADATA
),
597 redundant_metadata_changed_cb
, os
);
600 err
= dsl_prop_register(ds
,
601 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
602 recordsize_changed_cb
, os
);
605 err
= dsl_prop_register(ds
,
606 zfs_prop_to_name(ZFS_PROP_DNODESIZE
),
607 dnodesize_changed_cb
, os
);
610 err
= dsl_prop_register(ds
,
612 ZFS_PROP_SPECIAL_SMALL_BLOCKS
),
613 smallblk_changed_cb
, os
);
617 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
618 kmem_free(os
, sizeof (objset_t
));
622 /* It's the meta-objset. */
623 os
->os_checksum
= ZIO_CHECKSUM_FLETCHER_4
;
624 os
->os_compress
= ZIO_COMPRESS_ON
;
625 os
->os_complevel
= ZIO_COMPLEVEL_DEFAULT
;
626 os
->os_encrypted
= B_FALSE
;
627 os
->os_copies
= spa_max_replication(spa
);
628 os
->os_dedup_checksum
= ZIO_CHECKSUM_OFF
;
629 os
->os_dedup_verify
= B_FALSE
;
630 os
->os_logbias
= ZFS_LOGBIAS_LATENCY
;
631 os
->os_sync
= ZFS_SYNC_STANDARD
;
632 os
->os_primary_cache
= ZFS_CACHE_ALL
;
633 os
->os_secondary_cache
= ZFS_CACHE_ALL
;
634 os
->os_dnodesize
= DNODE_MIN_SIZE
;
637 if (ds
== NULL
|| !ds
->ds_is_snapshot
)
638 os
->os_zil_header
= os
->os_phys
->os_zil_header
;
639 os
->os_zil
= zil_alloc(os
, &os
->os_zil_header
);
641 for (i
= 0; i
< TXG_SIZE
; i
++) {
642 multilist_create(&os
->os_dirty_dnodes
[i
], sizeof (dnode_t
),
643 offsetof(dnode_t
, dn_dirty_link
[i
]),
644 dnode_multilist_index_func
);
646 list_create(&os
->os_dnodes
, sizeof (dnode_t
),
647 offsetof(dnode_t
, dn_link
));
648 list_create(&os
->os_downgraded_dbufs
, sizeof (dmu_buf_impl_t
),
649 offsetof(dmu_buf_impl_t
, db_link
));
651 list_link_init(&os
->os_evicting_node
);
653 mutex_init(&os
->os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
654 mutex_init(&os
->os_userused_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
655 mutex_init(&os
->os_obj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
656 mutex_init(&os
->os_user_ptr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
657 os
->os_obj_next_percpu_len
= boot_ncpus
;
658 os
->os_obj_next_percpu
= kmem_zalloc(os
->os_obj_next_percpu_len
*
659 sizeof (os
->os_obj_next_percpu
[0]), KM_SLEEP
);
661 dnode_special_open(os
, &os
->os_phys
->os_meta_dnode
,
662 DMU_META_DNODE_OBJECT
, &os
->os_meta_dnode
);
663 if (OBJSET_BUF_HAS_USERUSED(os
->os_phys_buf
)) {
664 dnode_special_open(os
, &os
->os_phys
->os_userused_dnode
,
665 DMU_USERUSED_OBJECT
, &os
->os_userused_dnode
);
666 dnode_special_open(os
, &os
->os_phys
->os_groupused_dnode
,
667 DMU_GROUPUSED_OBJECT
, &os
->os_groupused_dnode
);
668 if (OBJSET_BUF_HAS_PROJECTUSED(os
->os_phys_buf
))
669 dnode_special_open(os
,
670 &os
->os_phys
->os_projectused_dnode
,
671 DMU_PROJECTUSED_OBJECT
, &os
->os_projectused_dnode
);
674 mutex_init(&os
->os_upgrade_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
681 dmu_objset_from_ds(dsl_dataset_t
*ds
, objset_t
**osp
)
686 * We need the pool_config lock to manipulate the dsl_dataset_t.
687 * Even if the dataset is long-held, we need the pool_config lock
688 * to open the objset, as it needs to get properties.
690 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
));
692 mutex_enter(&ds
->ds_opening_lock
);
693 if (ds
->ds_objset
== NULL
) {
695 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
696 err
= dmu_objset_open_impl(dsl_dataset_get_spa(ds
),
697 ds
, dsl_dataset_get_blkptr(ds
), &os
);
698 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
701 mutex_enter(&ds
->ds_lock
);
702 ASSERT(ds
->ds_objset
== NULL
);
704 mutex_exit(&ds
->ds_lock
);
707 *osp
= ds
->ds_objset
;
708 mutex_exit(&ds
->ds_opening_lock
);
713 * Holds the pool while the objset is held. Therefore only one objset
714 * can be held at a time.
717 dmu_objset_hold_flags(const char *name
, boolean_t decrypt
, void *tag
,
723 ds_hold_flags_t flags
;
725 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
726 err
= dsl_pool_hold(name
, tag
, &dp
);
729 err
= dsl_dataset_hold_flags(dp
, name
, flags
, tag
, &ds
);
731 dsl_pool_rele(dp
, tag
);
735 err
= dmu_objset_from_ds(ds
, osp
);
737 dsl_dataset_rele(ds
, tag
);
738 dsl_pool_rele(dp
, tag
);
745 dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
)
747 return (dmu_objset_hold_flags(name
, B_FALSE
, tag
, osp
));
751 dmu_objset_own_impl(dsl_dataset_t
*ds
, dmu_objset_type_t type
,
752 boolean_t readonly
, boolean_t decrypt
, void *tag
, objset_t
**osp
)
756 int err
= dmu_objset_from_ds(ds
, osp
);
759 } else if (type
!= DMU_OST_ANY
&& type
!= (*osp
)->os_phys
->os_type
) {
760 return (SET_ERROR(EINVAL
));
761 } else if (!readonly
&& dsl_dataset_is_snapshot(ds
)) {
762 return (SET_ERROR(EROFS
));
763 } else if (!readonly
&& decrypt
&&
764 dsl_dir_incompatible_encryption_version(ds
->ds_dir
)) {
765 return (SET_ERROR(EROFS
));
768 /* if we are decrypting, we can now check MACs in os->os_phys_buf */
769 if (decrypt
&& arc_is_unauthenticated((*osp
)->os_phys_buf
)) {
772 SET_BOOKMARK(&zb
, ds
->ds_object
, ZB_ROOT_OBJECT
,
773 ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
774 err
= arc_untransform((*osp
)->os_phys_buf
, (*osp
)->os_spa
,
779 ASSERT0(arc_is_unauthenticated((*osp
)->os_phys_buf
));
786 * dsl_pool must not be held when this is called.
787 * Upon successful return, there will be a longhold on the dataset,
788 * and the dsl_pool will not be held.
791 dmu_objset_own(const char *name
, dmu_objset_type_t type
,
792 boolean_t readonly
, boolean_t decrypt
, void *tag
, objset_t
**osp
)
797 ds_hold_flags_t flags
;
799 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
800 err
= dsl_pool_hold(name
, FTAG
, &dp
);
803 err
= dsl_dataset_own(dp
, name
, flags
, tag
, &ds
);
805 dsl_pool_rele(dp
, FTAG
);
808 err
= dmu_objset_own_impl(ds
, type
, readonly
, decrypt
, tag
, osp
);
810 dsl_dataset_disown(ds
, flags
, tag
);
811 dsl_pool_rele(dp
, FTAG
);
816 * User accounting requires the dataset to be decrypted and rw.
817 * We also don't begin user accounting during claiming to help
818 * speed up pool import times and to keep this txg reserved
819 * completely for recovery work.
821 if (!readonly
&& !dp
->dp_spa
->spa_claiming
&&
822 (ds
->ds_dir
->dd_crypto_obj
== 0 || decrypt
)) {
823 if (dmu_objset_userobjspace_upgradable(*osp
) ||
824 dmu_objset_projectquota_upgradable(*osp
)) {
825 dmu_objset_id_quota_upgrade(*osp
);
826 } else if (dmu_objset_userused_enabled(*osp
)) {
827 dmu_objset_userspace_upgrade(*osp
);
831 dsl_pool_rele(dp
, FTAG
);
836 dmu_objset_own_obj(dsl_pool_t
*dp
, uint64_t obj
, dmu_objset_type_t type
,
837 boolean_t readonly
, boolean_t decrypt
, void *tag
, objset_t
**osp
)
841 ds_hold_flags_t flags
;
843 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
844 err
= dsl_dataset_own_obj(dp
, obj
, flags
, tag
, &ds
);
848 err
= dmu_objset_own_impl(ds
, type
, readonly
, decrypt
, tag
, osp
);
850 dsl_dataset_disown(ds
, flags
, tag
);
858 dmu_objset_rele_flags(objset_t
*os
, boolean_t decrypt
, void *tag
)
860 ds_hold_flags_t flags
;
861 dsl_pool_t
*dp
= dmu_objset_pool(os
);
863 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
864 dsl_dataset_rele_flags(os
->os_dsl_dataset
, flags
, tag
);
865 dsl_pool_rele(dp
, tag
);
869 dmu_objset_rele(objset_t
*os
, void *tag
)
871 dmu_objset_rele_flags(os
, B_FALSE
, tag
);
875 * When we are called, os MUST refer to an objset associated with a dataset
876 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
877 * == tag. We will then release and reacquire ownership of the dataset while
878 * holding the pool config_rwlock to avoid intervening namespace or ownership
881 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
882 * release the hold on its dataset and acquire a new one on the dataset of the
883 * same name so that it can be partially torn down and reconstructed.
886 dmu_objset_refresh_ownership(dsl_dataset_t
*ds
, dsl_dataset_t
**newds
,
887 boolean_t decrypt
, void *tag
)
890 char name
[ZFS_MAX_DATASET_NAME_LEN
];
891 ds_hold_flags_t flags
;
893 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
894 VERIFY3P(ds
, !=, NULL
);
895 VERIFY3P(ds
->ds_owner
, ==, tag
);
896 VERIFY(dsl_dataset_long_held(ds
));
898 dsl_dataset_name(ds
, name
);
899 dp
= ds
->ds_dir
->dd_pool
;
900 dsl_pool_config_enter(dp
, FTAG
);
901 dsl_dataset_disown(ds
, flags
, tag
);
902 VERIFY0(dsl_dataset_own(dp
, name
, flags
, tag
, newds
));
903 dsl_pool_config_exit(dp
, FTAG
);
907 dmu_objset_disown(objset_t
*os
, boolean_t decrypt
, void *tag
)
909 ds_hold_flags_t flags
;
911 flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: DS_HOLD_FLAG_NONE
;
913 * Stop upgrading thread
915 dmu_objset_upgrade_stop(os
);
916 dsl_dataset_disown(os
->os_dsl_dataset
, flags
, tag
);
920 dmu_objset_evict_dbufs(objset_t
*os
)
925 dn_marker
= kmem_alloc(sizeof (dnode_t
), KM_SLEEP
);
927 mutex_enter(&os
->os_lock
);
928 dn
= list_head(&os
->os_dnodes
);
931 * Skip dnodes without holds. We have to do this dance
932 * because dnode_add_ref() only works if there is already a
933 * hold. If the dnode has no holds, then it has no dbufs.
935 if (dnode_add_ref(dn
, FTAG
)) {
936 list_insert_after(&os
->os_dnodes
, dn
, dn_marker
);
937 mutex_exit(&os
->os_lock
);
939 dnode_evict_dbufs(dn
);
940 dnode_rele(dn
, FTAG
);
942 mutex_enter(&os
->os_lock
);
943 dn
= list_next(&os
->os_dnodes
, dn_marker
);
944 list_remove(&os
->os_dnodes
, dn_marker
);
946 dn
= list_next(&os
->os_dnodes
, dn
);
949 mutex_exit(&os
->os_lock
);
951 kmem_free(dn_marker
, sizeof (dnode_t
));
953 if (DMU_USERUSED_DNODE(os
) != NULL
) {
954 if (DMU_PROJECTUSED_DNODE(os
) != NULL
)
955 dnode_evict_dbufs(DMU_PROJECTUSED_DNODE(os
));
956 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os
));
957 dnode_evict_dbufs(DMU_USERUSED_DNODE(os
));
959 dnode_evict_dbufs(DMU_META_DNODE(os
));
963 * Objset eviction processing is split into into two pieces.
964 * The first marks the objset as evicting, evicts any dbufs that
965 * have a refcount of zero, and then queues up the objset for the
966 * second phase of eviction. Once os->os_dnodes has been cleared by
967 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
968 * The second phase closes the special dnodes, dequeues the objset from
969 * the list of those undergoing eviction, and finally frees the objset.
971 * NOTE: Due to asynchronous eviction processing (invocation of
972 * dnode_buf_pageout()), it is possible for the meta dnode for the
973 * objset to have no holds even though os->os_dnodes is not empty.
976 dmu_objset_evict(objset_t
*os
)
978 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
980 for (int t
= 0; t
< TXG_SIZE
; t
++)
981 ASSERT(!dmu_objset_is_dirty(os
, t
));
984 dsl_prop_unregister_all(ds
, os
);
989 dmu_objset_evict_dbufs(os
);
991 mutex_enter(&os
->os_lock
);
992 spa_evicting_os_register(os
->os_spa
, os
);
993 if (list_is_empty(&os
->os_dnodes
)) {
994 mutex_exit(&os
->os_lock
);
995 dmu_objset_evict_done(os
);
997 mutex_exit(&os
->os_lock
);
1004 dmu_objset_evict_done(objset_t
*os
)
1006 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
1008 dnode_special_close(&os
->os_meta_dnode
);
1009 if (DMU_USERUSED_DNODE(os
)) {
1010 if (DMU_PROJECTUSED_DNODE(os
))
1011 dnode_special_close(&os
->os_projectused_dnode
);
1012 dnode_special_close(&os
->os_userused_dnode
);
1013 dnode_special_close(&os
->os_groupused_dnode
);
1015 zil_free(os
->os_zil
);
1017 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
1020 * This is a barrier to prevent the objset from going away in
1021 * dnode_move() until we can safely ensure that the objset is still in
1022 * use. We consider the objset valid before the barrier and invalid
1023 * after the barrier.
1025 rw_enter(&os_lock
, RW_READER
);
1028 kmem_free(os
->os_obj_next_percpu
,
1029 os
->os_obj_next_percpu_len
* sizeof (os
->os_obj_next_percpu
[0]));
1031 mutex_destroy(&os
->os_lock
);
1032 mutex_destroy(&os
->os_userused_lock
);
1033 mutex_destroy(&os
->os_obj_lock
);
1034 mutex_destroy(&os
->os_user_ptr_lock
);
1035 mutex_destroy(&os
->os_upgrade_lock
);
1036 for (int i
= 0; i
< TXG_SIZE
; i
++)
1037 multilist_destroy(&os
->os_dirty_dnodes
[i
]);
1038 spa_evicting_os_deregister(os
->os_spa
, os
);
1039 kmem_free(os
, sizeof (objset_t
));
1043 dmu_objset_snap_cmtime(objset_t
*os
)
1045 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
1049 dmu_objset_create_impl_dnstats(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
1050 dmu_objset_type_t type
, int levels
, int blksz
, int ibs
, dmu_tx_t
*tx
)
1055 ASSERT(dmu_tx_is_syncing(tx
));
1058 blksz
= DNODE_BLOCK_SIZE
;
1060 ibs
= DN_MAX_INDBLKSHIFT
;
1063 VERIFY0(dmu_objset_from_ds(ds
, &os
));
1065 VERIFY0(dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
1067 mdn
= DMU_META_DNODE(os
);
1069 dnode_allocate(mdn
, DMU_OT_DNODE
, blksz
, ibs
, DMU_OT_NONE
, 0,
1070 DNODE_MIN_SLOTS
, tx
);
1073 * We don't want to have to increase the meta-dnode's nlevels
1074 * later, because then we could do it in quiescing context while
1075 * we are also accessing it in open context.
1077 * This precaution is not necessary for the MOS (ds == NULL),
1078 * because the MOS is only updated in syncing context.
1079 * This is most fortunate: the MOS is the only objset that
1080 * needs to be synced multiple times as spa_sync() iterates
1081 * to convergence, so minimizing its dn_nlevels matters.
1088 * Determine the number of levels necessary for the
1089 * meta-dnode to contain DN_MAX_OBJECT dnodes. Note
1090 * that in order to ensure that we do not overflow
1091 * 64 bits, there has to be a nlevels that gives us a
1092 * number of blocks > DN_MAX_OBJECT but < 2^64.
1093 * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)
1094 * (10) must be less than (64 - log2(DN_MAX_OBJECT))
1097 while ((uint64_t)mdn
->dn_nblkptr
<<
1098 (mdn
->dn_datablkshift
- DNODE_SHIFT
+ (levels
- 1) *
1099 (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
1104 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
1105 mdn
->dn_nlevels
= levels
;
1108 ASSERT(type
!= DMU_OST_NONE
);
1109 ASSERT(type
!= DMU_OST_ANY
);
1110 ASSERT(type
< DMU_OST_NUMTYPES
);
1111 os
->os_phys
->os_type
= type
;
1114 * Enable user accounting if it is enabled and this is not an
1115 * encrypted receive.
1117 if (dmu_objset_userused_enabled(os
) &&
1118 (!os
->os_encrypted
|| !dmu_objset_is_receiving(os
))) {
1119 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
1120 if (dmu_objset_userobjused_enabled(os
)) {
1121 ds
->ds_feature_activation
[
1122 SPA_FEATURE_USEROBJ_ACCOUNTING
] = (void *)B_TRUE
;
1123 os
->os_phys
->os_flags
|=
1124 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
;
1126 if (dmu_objset_projectquota_enabled(os
)) {
1127 ds
->ds_feature_activation
[
1128 SPA_FEATURE_PROJECT_QUOTA
] = (void *)B_TRUE
;
1129 os
->os_phys
->os_flags
|=
1130 OBJSET_FLAG_PROJECTQUOTA_COMPLETE
;
1132 os
->os_flags
= os
->os_phys
->os_flags
;
1135 dsl_dataset_dirty(ds
, tx
);
1140 /* called from dsl for meta-objset */
1142 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
1143 dmu_objset_type_t type
, dmu_tx_t
*tx
)
1145 return (dmu_objset_create_impl_dnstats(spa
, ds
, bp
, type
, 0, 0, 0, tx
));
1148 typedef struct dmu_objset_create_arg
{
1149 const char *doca_name
;
1152 void (*doca_userfunc
)(objset_t
*os
, void *arg
,
1153 cred_t
*cr
, dmu_tx_t
*tx
);
1155 dmu_objset_type_t doca_type
;
1156 uint64_t doca_flags
;
1157 dsl_crypto_params_t
*doca_dcp
;
1158 } dmu_objset_create_arg_t
;
1161 dmu_objset_create_check(void *arg
, dmu_tx_t
*tx
)
1163 dmu_objset_create_arg_t
*doca
= arg
;
1164 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1166 dsl_dataset_t
*parentds
;
1171 if (strchr(doca
->doca_name
, '@') != NULL
)
1172 return (SET_ERROR(EINVAL
));
1174 if (strlen(doca
->doca_name
) >= ZFS_MAX_DATASET_NAME_LEN
)
1175 return (SET_ERROR(ENAMETOOLONG
));
1177 if (dataset_nestcheck(doca
->doca_name
) != 0)
1178 return (SET_ERROR(ENAMETOOLONG
));
1180 error
= dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
);
1184 dsl_dir_rele(pdd
, FTAG
);
1185 return (SET_ERROR(EEXIST
));
1188 error
= dmu_objset_create_crypt_check(pdd
, doca
->doca_dcp
, NULL
);
1190 dsl_dir_rele(pdd
, FTAG
);
1194 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
1195 doca
->doca_cred
, doca
->doca_proc
);
1197 dsl_dir_rele(pdd
, FTAG
);
1201 /* can't create below anything but filesystems (eg. no ZVOLs) */
1202 error
= dsl_dataset_hold_obj(pdd
->dd_pool
,
1203 dsl_dir_phys(pdd
)->dd_head_dataset_obj
, FTAG
, &parentds
);
1205 dsl_dir_rele(pdd
, FTAG
);
1208 error
= dmu_objset_from_ds(parentds
, &parentos
);
1210 dsl_dataset_rele(parentds
, FTAG
);
1211 dsl_dir_rele(pdd
, FTAG
);
1214 if (dmu_objset_type(parentos
) != DMU_OST_ZFS
) {
1215 dsl_dataset_rele(parentds
, FTAG
);
1216 dsl_dir_rele(pdd
, FTAG
);
1217 return (SET_ERROR(ZFS_ERR_WRONG_PARENT
));
1219 dsl_dataset_rele(parentds
, FTAG
);
1220 dsl_dir_rele(pdd
, FTAG
);
1226 dmu_objset_create_sync(void *arg
, dmu_tx_t
*tx
)
1228 dmu_objset_create_arg_t
*doca
= arg
;
1229 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1230 spa_t
*spa
= dp
->dp_spa
;
1239 VERIFY0(dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
));
1241 obj
= dsl_dataset_create_sync(pdd
, tail
, NULL
, doca
->doca_flags
,
1242 doca
->doca_cred
, doca
->doca_dcp
, tx
);
1244 VERIFY0(dsl_dataset_hold_obj_flags(pdd
->dd_pool
, obj
,
1245 DS_HOLD_FLAG_DECRYPT
, FTAG
, &ds
));
1246 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1247 bp
= dsl_dataset_get_blkptr(ds
);
1248 os
= dmu_objset_create_impl(spa
, ds
, bp
, doca
->doca_type
, tx
);
1249 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
1251 if (doca
->doca_userfunc
!= NULL
) {
1252 doca
->doca_userfunc(os
, doca
->doca_userarg
,
1253 doca
->doca_cred
, tx
);
1257 * The doca_userfunc() may write out some data that needs to be
1258 * encrypted if the dataset is encrypted (specifically the root
1259 * directory). This data must be written out before the encryption
1260 * key mapping is removed by dsl_dataset_rele_flags(). Force the
1261 * I/O to occur immediately by invoking the relevant sections of
1264 if (os
->os_encrypted
) {
1265 dsl_dataset_t
*tmpds
= NULL
;
1266 boolean_t need_sync_done
= B_FALSE
;
1268 mutex_enter(&ds
->ds_lock
);
1269 ds
->ds_owner
= FTAG
;
1270 mutex_exit(&ds
->ds_lock
);
1272 rzio
= zio_root(spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
1273 tmpds
= txg_list_remove_this(&dp
->dp_dirty_datasets
, ds
,
1275 if (tmpds
!= NULL
) {
1276 dsl_dataset_sync(ds
, rzio
, tx
);
1277 need_sync_done
= B_TRUE
;
1279 VERIFY0(zio_wait(rzio
));
1281 dmu_objset_sync_done(os
, tx
);
1282 taskq_wait(dp
->dp_sync_taskq
);
1283 if (txg_list_member(&dp
->dp_dirty_datasets
, ds
, tx
->tx_txg
)) {
1284 ASSERT3P(ds
->ds_key_mapping
, !=, NULL
);
1285 key_mapping_rele(spa
, ds
->ds_key_mapping
, ds
);
1288 rzio
= zio_root(spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
1289 tmpds
= txg_list_remove_this(&dp
->dp_dirty_datasets
, ds
,
1291 if (tmpds
!= NULL
) {
1292 dmu_buf_rele(ds
->ds_dbuf
, ds
);
1293 dsl_dataset_sync(ds
, rzio
, tx
);
1295 VERIFY0(zio_wait(rzio
));
1297 if (need_sync_done
) {
1298 ASSERT3P(ds
->ds_key_mapping
, !=, NULL
);
1299 key_mapping_rele(spa
, ds
->ds_key_mapping
, ds
);
1300 dsl_dataset_sync_done(ds
, tx
);
1303 mutex_enter(&ds
->ds_lock
);
1304 ds
->ds_owner
= NULL
;
1305 mutex_exit(&ds
->ds_lock
);
1308 spa_history_log_internal_ds(ds
, "create", tx
, " ");
1310 dsl_dataset_rele_flags(ds
, DS_HOLD_FLAG_DECRYPT
, FTAG
);
1311 dsl_dir_rele(pdd
, FTAG
);
1315 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
1316 dsl_crypto_params_t
*dcp
, dmu_objset_create_sync_func_t func
, void *arg
)
1318 dmu_objset_create_arg_t doca
;
1319 dsl_crypto_params_t tmp_dcp
= { 0 };
1321 doca
.doca_name
= name
;
1322 doca
.doca_cred
= CRED();
1323 doca
.doca_proc
= curproc
;
1324 doca
.doca_flags
= flags
;
1325 doca
.doca_userfunc
= func
;
1326 doca
.doca_userarg
= arg
;
1327 doca
.doca_type
= type
;
1330 * Some callers (mostly for testing) do not provide a dcp on their
1331 * own but various code inside the sync task will require it to be
1332 * allocated. Rather than adding NULL checks throughout this code
1333 * or adding dummy dcp's to all of the callers we simply create a
1334 * dummy one here and use that. This zero dcp will have the same
1335 * effect as asking for inheritance of all encryption params.
1337 doca
.doca_dcp
= (dcp
!= NULL
) ? dcp
: &tmp_dcp
;
1339 int rv
= dsl_sync_task(name
,
1340 dmu_objset_create_check
, dmu_objset_create_sync
, &doca
,
1341 6, ZFS_SPACE_CHECK_NORMAL
);
1344 zvol_create_minor(name
);
1348 typedef struct dmu_objset_clone_arg
{
1349 const char *doca_clone
;
1350 const char *doca_origin
;
1353 } dmu_objset_clone_arg_t
;
1356 dmu_objset_clone_check(void *arg
, dmu_tx_t
*tx
)
1358 dmu_objset_clone_arg_t
*doca
= arg
;
1362 dsl_dataset_t
*origin
;
1363 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1365 if (strchr(doca
->doca_clone
, '@') != NULL
)
1366 return (SET_ERROR(EINVAL
));
1368 if (strlen(doca
->doca_clone
) >= ZFS_MAX_DATASET_NAME_LEN
)
1369 return (SET_ERROR(ENAMETOOLONG
));
1371 error
= dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
);
1375 dsl_dir_rele(pdd
, FTAG
);
1376 return (SET_ERROR(EEXIST
));
1379 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
1380 doca
->doca_cred
, doca
->doca_proc
);
1382 dsl_dir_rele(pdd
, FTAG
);
1383 return (SET_ERROR(EDQUOT
));
1386 error
= dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
);
1388 dsl_dir_rele(pdd
, FTAG
);
1392 /* You can only clone snapshots, not the head datasets. */
1393 if (!origin
->ds_is_snapshot
) {
1394 dsl_dataset_rele(origin
, FTAG
);
1395 dsl_dir_rele(pdd
, FTAG
);
1396 return (SET_ERROR(EINVAL
));
1399 dsl_dataset_rele(origin
, FTAG
);
1400 dsl_dir_rele(pdd
, FTAG
);
1406 dmu_objset_clone_sync(void *arg
, dmu_tx_t
*tx
)
1408 dmu_objset_clone_arg_t
*doca
= arg
;
1409 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1412 dsl_dataset_t
*origin
, *ds
;
1414 char namebuf
[ZFS_MAX_DATASET_NAME_LEN
];
1416 VERIFY0(dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
));
1417 VERIFY0(dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
));
1419 obj
= dsl_dataset_create_sync(pdd
, tail
, origin
, 0,
1420 doca
->doca_cred
, NULL
, tx
);
1422 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
1423 dsl_dataset_name(origin
, namebuf
);
1424 spa_history_log_internal_ds(ds
, "clone", tx
,
1425 "origin=%s (%llu)", namebuf
, (u_longlong_t
)origin
->ds_object
);
1426 dsl_dataset_rele(ds
, FTAG
);
1427 dsl_dataset_rele(origin
, FTAG
);
1428 dsl_dir_rele(pdd
, FTAG
);
1432 dmu_objset_clone(const char *clone
, const char *origin
)
1434 dmu_objset_clone_arg_t doca
;
1436 doca
.doca_clone
= clone
;
1437 doca
.doca_origin
= origin
;
1438 doca
.doca_cred
= CRED();
1439 doca
.doca_proc
= curproc
;
1441 int rv
= dsl_sync_task(clone
,
1442 dmu_objset_clone_check
, dmu_objset_clone_sync
, &doca
,
1443 6, ZFS_SPACE_CHECK_NORMAL
);
1446 zvol_create_minor(clone
);
1452 dmu_objset_snapshot_one(const char *fsname
, const char *snapname
)
1455 char *longsnap
= kmem_asprintf("%s@%s", fsname
, snapname
);
1456 nvlist_t
*snaps
= fnvlist_alloc();
1458 fnvlist_add_boolean(snaps
, longsnap
);
1459 kmem_strfree(longsnap
);
1460 err
= dsl_dataset_snapshot(snaps
, NULL
, NULL
);
1461 fnvlist_free(snaps
);
1466 dmu_objset_upgrade_task_cb(void *data
)
1468 objset_t
*os
= data
;
1470 mutex_enter(&os
->os_upgrade_lock
);
1471 os
->os_upgrade_status
= EINTR
;
1472 if (!os
->os_upgrade_exit
) {
1475 mutex_exit(&os
->os_upgrade_lock
);
1477 status
= os
->os_upgrade_cb(os
);
1479 mutex_enter(&os
->os_upgrade_lock
);
1481 os
->os_upgrade_status
= status
;
1483 os
->os_upgrade_exit
= B_TRUE
;
1484 os
->os_upgrade_id
= 0;
1485 mutex_exit(&os
->os_upgrade_lock
);
1486 dsl_dataset_long_rele(dmu_objset_ds(os
), upgrade_tag
);
1490 dmu_objset_upgrade(objset_t
*os
, dmu_objset_upgrade_cb_t cb
)
1492 if (os
->os_upgrade_id
!= 0)
1495 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
1496 dsl_dataset_long_hold(dmu_objset_ds(os
), upgrade_tag
);
1498 mutex_enter(&os
->os_upgrade_lock
);
1499 if (os
->os_upgrade_id
== 0 && os
->os_upgrade_status
== 0) {
1500 os
->os_upgrade_exit
= B_FALSE
;
1501 os
->os_upgrade_cb
= cb
;
1502 os
->os_upgrade_id
= taskq_dispatch(
1503 os
->os_spa
->spa_upgrade_taskq
,
1504 dmu_objset_upgrade_task_cb
, os
, TQ_SLEEP
);
1505 if (os
->os_upgrade_id
== TASKQID_INVALID
) {
1506 dsl_dataset_long_rele(dmu_objset_ds(os
), upgrade_tag
);
1507 os
->os_upgrade_status
= ENOMEM
;
1510 dsl_dataset_long_rele(dmu_objset_ds(os
), upgrade_tag
);
1512 mutex_exit(&os
->os_upgrade_lock
);
1516 dmu_objset_upgrade_stop(objset_t
*os
)
1518 mutex_enter(&os
->os_upgrade_lock
);
1519 os
->os_upgrade_exit
= B_TRUE
;
1520 if (os
->os_upgrade_id
!= 0) {
1521 taskqid_t id
= os
->os_upgrade_id
;
1523 os
->os_upgrade_id
= 0;
1524 mutex_exit(&os
->os_upgrade_lock
);
1526 if ((taskq_cancel_id(os
->os_spa
->spa_upgrade_taskq
, id
)) == 0) {
1527 dsl_dataset_long_rele(dmu_objset_ds(os
), upgrade_tag
);
1529 txg_wait_synced(os
->os_spa
->spa_dsl_pool
, 0);
1531 mutex_exit(&os
->os_upgrade_lock
);
1536 dmu_objset_sync_dnodes(multilist_sublist_t
*list
, dmu_tx_t
*tx
)
1540 while ((dn
= multilist_sublist_head(list
)) != NULL
) {
1541 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1542 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1544 * Initialize dn_zio outside dnode_sync() because the
1545 * meta-dnode needs to set it outside dnode_sync().
1547 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1550 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1551 multilist_sublist_remove(list
, dn
);
1554 * See the comment above dnode_rele_task() for an explanation
1555 * of why this dnode hold is always needed (even when not
1556 * doing user accounting).
1558 multilist_t
*newlist
= &dn
->dn_objset
->os_synced_dnodes
;
1559 (void) dnode_add_ref(dn
, newlist
);
1560 multilist_insert(newlist
, dn
);
1567 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1570 blkptr_t
*bp
= zio
->io_bp
;
1572 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1575 ASSERT(!BP_IS_EMBEDDED(bp
));
1576 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_OBJSET
);
1577 ASSERT0(BP_GET_LEVEL(bp
));
1580 * Update rootbp fill count: it should be the number of objects
1581 * allocated in the object set (not counting the "special"
1582 * objects that are stored in the objset_phys_t -- the meta
1583 * dnode and user/group/project accounting objects).
1585 for (int i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1586 fill
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
1588 BP_SET_FILL(bp
, fill
);
1590 if (os
->os_dsl_dataset
!= NULL
)
1591 rrw_enter(&os
->os_dsl_dataset
->ds_bp_rwlock
, RW_WRITER
, FTAG
);
1592 *os
->os_rootbp
= *bp
;
1593 if (os
->os_dsl_dataset
!= NULL
)
1594 rrw_exit(&os
->os_dsl_dataset
->ds_bp_rwlock
, FTAG
);
1598 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1601 blkptr_t
*bp
= zio
->io_bp
;
1602 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1605 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1606 ASSERT(BP_EQUAL(bp
, bp_orig
));
1608 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1609 dmu_tx_t
*tx
= os
->os_synctx
;
1611 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1612 dsl_dataset_block_born(ds
, bp
, tx
);
1614 kmem_free(bp
, sizeof (*bp
));
1617 typedef struct sync_dnodes_arg
{
1618 multilist_t
*sda_list
;
1619 int sda_sublist_idx
;
1620 multilist_t
*sda_newlist
;
1622 } sync_dnodes_arg_t
;
1625 sync_dnodes_task(void *arg
)
1627 sync_dnodes_arg_t
*sda
= arg
;
1629 multilist_sublist_t
*ms
=
1630 multilist_sublist_lock(sda
->sda_list
, sda
->sda_sublist_idx
);
1632 dmu_objset_sync_dnodes(ms
, sda
->sda_tx
);
1634 multilist_sublist_unlock(ms
);
1636 kmem_free(sda
, sizeof (*sda
));
1640 /* called from dsl */
1642 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1645 zbookmark_phys_t zb
;
1649 dbuf_dirty_record_t
*dr
;
1652 blkptr_t
*blkptr_copy
= kmem_alloc(sizeof (*os
->os_rootbp
), KM_SLEEP
);
1653 *blkptr_copy
= *os
->os_rootbp
;
1655 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", (u_longlong_t
)tx
->tx_txg
);
1657 ASSERT(dmu_tx_is_syncing(tx
));
1658 /* XXX the write_done callback should really give us the tx... */
1661 if (os
->os_dsl_dataset
== NULL
) {
1663 * This is the MOS. If we have upgraded,
1664 * spa_max_replication() could change, so reset
1667 os
->os_copies
= spa_max_replication(os
->os_spa
);
1671 * Create the root block IO
1673 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1674 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1675 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1676 arc_release(os
->os_phys_buf
, &os
->os_phys_buf
);
1678 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1681 * If we are either claiming the ZIL or doing a raw receive, write
1682 * out the os_phys_buf raw. Neither of these actions will effect the
1683 * MAC at this point.
1685 if (os
->os_raw_receive
||
1686 os
->os_next_write_raw
[tx
->tx_txg
& TXG_MASK
]) {
1687 ASSERT(os
->os_encrypted
);
1688 arc_convert_to_raw(os
->os_phys_buf
,
1689 os
->os_dsl_dataset
->ds_object
, ZFS_HOST_BYTEORDER
,
1690 DMU_OT_OBJSET
, NULL
, NULL
, NULL
);
1693 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1694 blkptr_copy
, os
->os_phys_buf
, dmu_os_is_l2cacheable(os
),
1695 &zp
, dmu_objset_write_ready
, NULL
, NULL
, dmu_objset_write_done
,
1696 os
, ZIO_PRIORITY_ASYNC_WRITE
, ZIO_FLAG_MUSTSUCCEED
, &zb
);
1699 * Sync special dnodes - the parent IO for the sync is the root block
1701 DMU_META_DNODE(os
)->dn_zio
= zio
;
1702 dnode_sync(DMU_META_DNODE(os
), tx
);
1704 os
->os_phys
->os_flags
= os
->os_flags
;
1706 if (DMU_USERUSED_DNODE(os
) &&
1707 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1708 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1709 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1710 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1711 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1714 if (DMU_PROJECTUSED_DNODE(os
) &&
1715 DMU_PROJECTUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1716 DMU_PROJECTUSED_DNODE(os
)->dn_zio
= zio
;
1717 dnode_sync(DMU_PROJECTUSED_DNODE(os
), tx
);
1720 txgoff
= tx
->tx_txg
& TXG_MASK
;
1723 * We must create the list here because it uses the
1724 * dn_dirty_link[] of this txg. But it may already
1725 * exist because we call dsl_dataset_sync() twice per txg.
1727 if (os
->os_synced_dnodes
.ml_sublists
== NULL
) {
1728 multilist_create(&os
->os_synced_dnodes
, sizeof (dnode_t
),
1729 offsetof(dnode_t
, dn_dirty_link
[txgoff
]),
1730 dnode_multilist_index_func
);
1732 ASSERT3U(os
->os_synced_dnodes
.ml_offset
, ==,
1733 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1736 ml
= &os
->os_dirty_dnodes
[txgoff
];
1737 num_sublists
= multilist_get_num_sublists(ml
);
1738 for (int i
= 0; i
< num_sublists
; i
++) {
1739 if (multilist_sublist_is_empty_idx(ml
, i
))
1741 sync_dnodes_arg_t
*sda
= kmem_alloc(sizeof (*sda
), KM_SLEEP
);
1743 sda
->sda_sublist_idx
= i
;
1745 (void) taskq_dispatch(dmu_objset_pool(os
)->dp_sync_taskq
,
1746 sync_dnodes_task
, sda
, 0);
1747 /* callback frees sda */
1749 taskq_wait(dmu_objset_pool(os
)->dp_sync_taskq
);
1751 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1752 while ((dr
= list_head(list
)) != NULL
) {
1753 ASSERT0(dr
->dr_dbuf
->db_level
);
1754 list_remove(list
, dr
);
1755 zio_nowait(dr
->dr_zio
);
1758 /* Enable dnode backfill if enough objects have been freed. */
1759 if (os
->os_freed_dnodes
>= dmu_rescan_dnode_threshold
) {
1760 os
->os_rescan_dnodes
= B_TRUE
;
1761 os
->os_freed_dnodes
= 0;
1765 * Free intent log blocks up to this tx.
1767 zil_sync(os
->os_zil
, tx
);
1768 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1773 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1775 return (!multilist_is_empty(&os
->os_dirty_dnodes
[txg
& TXG_MASK
]));
1778 static file_info_cb_t
*file_cbs
[DMU_OST_NUMTYPES
];
1781 dmu_objset_register_type(dmu_objset_type_t ost
, file_info_cb_t
*cb
)
1787 dmu_get_file_info(objset_t
*os
, dmu_object_type_t bonustype
, const void *data
,
1788 zfs_file_info_t
*zfi
)
1790 file_info_cb_t
*cb
= file_cbs
[os
->os_phys
->os_type
];
1793 return (cb(bonustype
, data
, zfi
));
1797 dmu_objset_userused_enabled(objset_t
*os
)
1799 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1800 file_cbs
[os
->os_phys
->os_type
] != NULL
&&
1801 DMU_USERUSED_DNODE(os
) != NULL
);
1805 dmu_objset_userobjused_enabled(objset_t
*os
)
1807 return (dmu_objset_userused_enabled(os
) &&
1808 spa_feature_is_enabled(os
->os_spa
, SPA_FEATURE_USEROBJ_ACCOUNTING
));
1812 dmu_objset_projectquota_enabled(objset_t
*os
)
1814 return (file_cbs
[os
->os_phys
->os_type
] != NULL
&&
1815 DMU_PROJECTUSED_DNODE(os
) != NULL
&&
1816 spa_feature_is_enabled(os
->os_spa
, SPA_FEATURE_PROJECT_QUOTA
));
1819 typedef struct userquota_node
{
1820 /* must be in the first filed, see userquota_update_cache() */
1821 char uqn_id
[20 + DMU_OBJACCT_PREFIX_LEN
];
1823 avl_node_t uqn_node
;
1826 typedef struct userquota_cache
{
1827 avl_tree_t uqc_user_deltas
;
1828 avl_tree_t uqc_group_deltas
;
1829 avl_tree_t uqc_project_deltas
;
1830 } userquota_cache_t
;
1833 userquota_compare(const void *l
, const void *r
)
1835 const userquota_node_t
*luqn
= l
;
1836 const userquota_node_t
*ruqn
= r
;
1840 * NB: can only access uqn_id because userquota_update_cache() doesn't
1841 * pass in an entire userquota_node_t.
1843 rv
= strcmp(luqn
->uqn_id
, ruqn
->uqn_id
);
1845 return (TREE_ISIGN(rv
));
1849 do_userquota_cacheflush(objset_t
*os
, userquota_cache_t
*cache
, dmu_tx_t
*tx
)
1852 userquota_node_t
*uqn
;
1854 ASSERT(dmu_tx_is_syncing(tx
));
1857 while ((uqn
= avl_destroy_nodes(&cache
->uqc_user_deltas
,
1858 &cookie
)) != NULL
) {
1860 * os_userused_lock protects against concurrent calls to
1861 * zap_increment_int(). It's needed because zap_increment_int()
1862 * is not thread-safe (i.e. not atomic).
1864 mutex_enter(&os
->os_userused_lock
);
1865 VERIFY0(zap_increment(os
, DMU_USERUSED_OBJECT
,
1866 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1867 mutex_exit(&os
->os_userused_lock
);
1868 kmem_free(uqn
, sizeof (*uqn
));
1870 avl_destroy(&cache
->uqc_user_deltas
);
1873 while ((uqn
= avl_destroy_nodes(&cache
->uqc_group_deltas
,
1874 &cookie
)) != NULL
) {
1875 mutex_enter(&os
->os_userused_lock
);
1876 VERIFY0(zap_increment(os
, DMU_GROUPUSED_OBJECT
,
1877 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1878 mutex_exit(&os
->os_userused_lock
);
1879 kmem_free(uqn
, sizeof (*uqn
));
1881 avl_destroy(&cache
->uqc_group_deltas
);
1883 if (dmu_objset_projectquota_enabled(os
)) {
1885 while ((uqn
= avl_destroy_nodes(&cache
->uqc_project_deltas
,
1886 &cookie
)) != NULL
) {
1887 mutex_enter(&os
->os_userused_lock
);
1888 VERIFY0(zap_increment(os
, DMU_PROJECTUSED_OBJECT
,
1889 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1890 mutex_exit(&os
->os_userused_lock
);
1891 kmem_free(uqn
, sizeof (*uqn
));
1893 avl_destroy(&cache
->uqc_project_deltas
);
1898 userquota_update_cache(avl_tree_t
*avl
, const char *id
, int64_t delta
)
1900 userquota_node_t
*uqn
;
1903 ASSERT(strlen(id
) < sizeof (uqn
->uqn_id
));
1905 * Use id directly for searching because uqn_id is the first field of
1906 * userquota_node_t and fields after uqn_id won't be accessed in
1909 uqn
= avl_find(avl
, (const void *)id
, &idx
);
1911 uqn
= kmem_zalloc(sizeof (*uqn
), KM_SLEEP
);
1912 strlcpy(uqn
->uqn_id
, id
, sizeof (uqn
->uqn_id
));
1913 avl_insert(avl
, uqn
, idx
);
1915 uqn
->uqn_delta
+= delta
;
1919 do_userquota_update(objset_t
*os
, userquota_cache_t
*cache
, uint64_t used
,
1920 uint64_t flags
, uint64_t user
, uint64_t group
, uint64_t project
,
1923 if (flags
& DNODE_FLAG_USERUSED_ACCOUNTED
) {
1924 int64_t delta
= DNODE_MIN_SIZE
+ used
;
1930 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)user
);
1931 userquota_update_cache(&cache
->uqc_user_deltas
, name
, delta
);
1933 (void) snprintf(name
, sizeof (name
), "%llx", (longlong_t
)group
);
1934 userquota_update_cache(&cache
->uqc_group_deltas
, name
, delta
);
1936 if (dmu_objset_projectquota_enabled(os
)) {
1937 (void) snprintf(name
, sizeof (name
), "%llx",
1938 (longlong_t
)project
);
1939 userquota_update_cache(&cache
->uqc_project_deltas
,
1946 do_userobjquota_update(objset_t
*os
, userquota_cache_t
*cache
, uint64_t flags
,
1947 uint64_t user
, uint64_t group
, uint64_t project
, boolean_t subtract
)
1949 if (flags
& DNODE_FLAG_USEROBJUSED_ACCOUNTED
) {
1950 char name
[20 + DMU_OBJACCT_PREFIX_LEN
];
1951 int delta
= subtract
? -1 : 1;
1953 (void) snprintf(name
, sizeof (name
), DMU_OBJACCT_PREFIX
"%llx",
1955 userquota_update_cache(&cache
->uqc_user_deltas
, name
, delta
);
1957 (void) snprintf(name
, sizeof (name
), DMU_OBJACCT_PREFIX
"%llx",
1959 userquota_update_cache(&cache
->uqc_group_deltas
, name
, delta
);
1961 if (dmu_objset_projectquota_enabled(os
)) {
1962 (void) snprintf(name
, sizeof (name
),
1963 DMU_OBJACCT_PREFIX
"%llx", (longlong_t
)project
);
1964 userquota_update_cache(&cache
->uqc_project_deltas
,
1970 typedef struct userquota_updates_arg
{
1972 int uua_sublist_idx
;
1974 } userquota_updates_arg_t
;
1977 userquota_updates_task(void *arg
)
1979 userquota_updates_arg_t
*uua
= arg
;
1980 objset_t
*os
= uua
->uua_os
;
1981 dmu_tx_t
*tx
= uua
->uua_tx
;
1983 userquota_cache_t cache
= { { 0 } };
1985 multilist_sublist_t
*list
=
1986 multilist_sublist_lock(&os
->os_synced_dnodes
, uua
->uua_sublist_idx
);
1988 ASSERT(multilist_sublist_head(list
) == NULL
||
1989 dmu_objset_userused_enabled(os
));
1990 avl_create(&cache
.uqc_user_deltas
, userquota_compare
,
1991 sizeof (userquota_node_t
), offsetof(userquota_node_t
, uqn_node
));
1992 avl_create(&cache
.uqc_group_deltas
, userquota_compare
,
1993 sizeof (userquota_node_t
), offsetof(userquota_node_t
, uqn_node
));
1994 if (dmu_objset_projectquota_enabled(os
))
1995 avl_create(&cache
.uqc_project_deltas
, userquota_compare
,
1996 sizeof (userquota_node_t
), offsetof(userquota_node_t
,
1999 while ((dn
= multilist_sublist_head(list
)) != NULL
) {
2001 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
2002 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
2003 dn
->dn_phys
->dn_flags
&
2004 DNODE_FLAG_USERUSED_ACCOUNTED
);
2006 flags
= dn
->dn_id_flags
;
2008 if (flags
& DN_ID_OLD_EXIST
) {
2009 do_userquota_update(os
, &cache
, dn
->dn_oldused
,
2010 dn
->dn_oldflags
, dn
->dn_olduid
, dn
->dn_oldgid
,
2011 dn
->dn_oldprojid
, B_TRUE
);
2012 do_userobjquota_update(os
, &cache
, dn
->dn_oldflags
,
2013 dn
->dn_olduid
, dn
->dn_oldgid
,
2014 dn
->dn_oldprojid
, B_TRUE
);
2016 if (flags
& DN_ID_NEW_EXIST
) {
2017 do_userquota_update(os
, &cache
,
2018 DN_USED_BYTES(dn
->dn_phys
), dn
->dn_phys
->dn_flags
,
2019 dn
->dn_newuid
, dn
->dn_newgid
,
2020 dn
->dn_newprojid
, B_FALSE
);
2021 do_userobjquota_update(os
, &cache
,
2022 dn
->dn_phys
->dn_flags
, dn
->dn_newuid
, dn
->dn_newgid
,
2023 dn
->dn_newprojid
, B_FALSE
);
2026 mutex_enter(&dn
->dn_mtx
);
2028 dn
->dn_oldflags
= 0;
2029 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
2030 dn
->dn_olduid
= dn
->dn_newuid
;
2031 dn
->dn_oldgid
= dn
->dn_newgid
;
2032 dn
->dn_oldprojid
= dn
->dn_newprojid
;
2033 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
2034 if (dn
->dn_bonuslen
== 0)
2035 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
2037 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
2039 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
2040 mutex_exit(&dn
->dn_mtx
);
2042 multilist_sublist_remove(list
, dn
);
2043 dnode_rele(dn
, &os
->os_synced_dnodes
);
2045 do_userquota_cacheflush(os
, &cache
, tx
);
2046 multilist_sublist_unlock(list
);
2047 kmem_free(uua
, sizeof (*uua
));
2051 * Release dnode holds from dmu_objset_sync_dnodes(). When the dnode is being
2052 * synced (i.e. we have issued the zio's for blocks in the dnode), it can't be
2053 * evicted because the block containing the dnode can't be evicted until it is
2054 * written out. However, this hold is necessary to prevent the dnode_t from
2055 * being moved (via dnode_move()) while it's still referenced by
2056 * dbuf_dirty_record_t:dr_dnode. And dr_dnode is needed for
2057 * dirty_lightweight_leaf-type dirty records.
2059 * If we are doing user-object accounting, the dnode_rele() happens from
2060 * userquota_updates_task() instead.
2063 dnode_rele_task(void *arg
)
2065 userquota_updates_arg_t
*uua
= arg
;
2066 objset_t
*os
= uua
->uua_os
;
2068 multilist_sublist_t
*list
=
2069 multilist_sublist_lock(&os
->os_synced_dnodes
, uua
->uua_sublist_idx
);
2072 while ((dn
= multilist_sublist_head(list
)) != NULL
) {
2073 multilist_sublist_remove(list
, dn
);
2074 dnode_rele(dn
, &os
->os_synced_dnodes
);
2076 multilist_sublist_unlock(list
);
2077 kmem_free(uua
, sizeof (*uua
));
2081 * Return TRUE if userquota updates are needed.
2084 dmu_objset_do_userquota_updates_prep(objset_t
*os
, dmu_tx_t
*tx
)
2086 if (!dmu_objset_userused_enabled(os
))
2090 * If this is a raw receive just return and handle accounting
2091 * later when we have the keys loaded. We also don't do user
2092 * accounting during claiming since the datasets are not owned
2093 * for the duration of claiming and this txg should only be
2094 * used for recovery.
2096 if (os
->os_encrypted
&& dmu_objset_is_receiving(os
))
2099 if (tx
->tx_txg
<= os
->os_spa
->spa_claim_max_txg
)
2102 /* Allocate the user/group/project used objects if necessary. */
2103 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
2104 VERIFY0(zap_create_claim(os
,
2105 DMU_USERUSED_OBJECT
,
2106 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
2107 VERIFY0(zap_create_claim(os
,
2108 DMU_GROUPUSED_OBJECT
,
2109 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
2112 if (dmu_objset_projectquota_enabled(os
) &&
2113 DMU_PROJECTUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
2114 VERIFY0(zap_create_claim(os
, DMU_PROJECTUSED_OBJECT
,
2115 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
2121 * Dispatch taskq tasks to dp_sync_taskq to update the user accounting, and
2122 * also release the holds on the dnodes from dmu_objset_sync_dnodes().
2123 * The caller must taskq_wait(dp_sync_taskq).
2126 dmu_objset_sync_done(objset_t
*os
, dmu_tx_t
*tx
)
2128 boolean_t need_userquota
= dmu_objset_do_userquota_updates_prep(os
, tx
);
2130 int num_sublists
= multilist_get_num_sublists(&os
->os_synced_dnodes
);
2131 for (int i
= 0; i
< num_sublists
; i
++) {
2132 userquota_updates_arg_t
*uua
=
2133 kmem_alloc(sizeof (*uua
), KM_SLEEP
);
2135 uua
->uua_sublist_idx
= i
;
2139 * If we don't need to update userquotas, use
2140 * dnode_rele_task() to call dnode_rele()
2142 (void) taskq_dispatch(dmu_objset_pool(os
)->dp_sync_taskq
,
2143 need_userquota
? userquota_updates_task
: dnode_rele_task
,
2145 /* callback frees uua */
2151 * Returns a pointer to data to find uid/gid from
2153 * If a dirty record for transaction group that is syncing can't
2154 * be found then NULL is returned. In the NULL case it is assumed
2155 * the uid/gid aren't changing.
2158 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
2160 dbuf_dirty_record_t
*dr
;
2163 if (db
->db_dirtycnt
== 0)
2164 return (db
->db
.db_data
); /* Nothing is changing */
2166 dr
= dbuf_find_dirty_eq(db
, tx
->tx_txg
);
2171 if (dr
->dr_dnode
->dn_bonuslen
== 0 &&
2172 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
2173 data
= dr
->dt
.dl
.dr_data
->b_data
;
2175 data
= dr
->dt
.dl
.dr_data
;
2182 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
2184 objset_t
*os
= dn
->dn_objset
;
2186 dmu_buf_impl_t
*db
= NULL
;
2187 int flags
= dn
->dn_id_flags
;
2189 boolean_t have_spill
= B_FALSE
;
2191 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
2195 * Raw receives introduce a problem with user accounting. Raw
2196 * receives cannot update the user accounting info because the
2197 * user ids and the sizes are encrypted. To guarantee that we
2198 * never end up with bad user accounting, we simply disable it
2199 * during raw receives. We also disable this for normal receives
2200 * so that an incremental raw receive may be done on top of an
2201 * existing non-raw receive.
2203 if (os
->os_encrypted
&& dmu_objset_is_receiving(os
))
2206 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
2207 DN_ID_CHKED_SPILL
)))
2210 if (before
&& dn
->dn_bonuslen
!= 0)
2211 data
= DN_BONUS(dn
->dn_phys
);
2212 else if (!before
&& dn
->dn_bonuslen
!= 0) {
2215 mutex_enter(&db
->db_mtx
);
2216 data
= dmu_objset_userquota_find_data(db
, tx
);
2218 data
= DN_BONUS(dn
->dn_phys
);
2220 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
2223 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
2224 rf
|= DB_RF_HAVESTRUCT
;
2225 error
= dmu_spill_hold_by_dnode(dn
,
2226 rf
| DB_RF_MUST_SUCCEED
,
2227 FTAG
, (dmu_buf_t
**)&db
);
2229 mutex_enter(&db
->db_mtx
);
2230 data
= (before
) ? db
->db
.db_data
:
2231 dmu_objset_userquota_find_data(db
, tx
);
2232 have_spill
= B_TRUE
;
2234 mutex_enter(&dn
->dn_mtx
);
2235 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
2236 mutex_exit(&dn
->dn_mtx
);
2241 * Must always call the callback in case the object
2242 * type has changed and that type isn't an object type to track
2244 zfs_file_info_t zfi
;
2245 error
= file_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
, &zfi
);
2249 dn
->dn_olduid
= zfi
.zfi_user
;
2250 dn
->dn_oldgid
= zfi
.zfi_group
;
2251 dn
->dn_oldprojid
= zfi
.zfi_project
;
2253 dn
->dn_newuid
= zfi
.zfi_user
;
2254 dn
->dn_newgid
= zfi
.zfi_group
;
2255 dn
->dn_newprojid
= zfi
.zfi_project
;
2259 * Preserve existing uid/gid when the callback can't determine
2260 * what the new uid/gid are and the callback returned EEXIST.
2261 * The EEXIST error tells us to just use the existing uid/gid.
2262 * If we don't know what the old values are then just assign
2263 * them to 0, since that is a new file being created.
2265 if (!before
&& data
== NULL
&& error
== EEXIST
) {
2266 if (flags
& DN_ID_OLD_EXIST
) {
2267 dn
->dn_newuid
= dn
->dn_olduid
;
2268 dn
->dn_newgid
= dn
->dn_oldgid
;
2269 dn
->dn_newprojid
= dn
->dn_oldprojid
;
2273 dn
->dn_newprojid
= ZFS_DEFAULT_PROJID
;
2279 mutex_exit(&db
->db_mtx
);
2281 mutex_enter(&dn
->dn_mtx
);
2282 if (error
== 0 && before
)
2283 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
2284 if (error
== 0 && !before
)
2285 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
2288 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
2290 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
2292 mutex_exit(&dn
->dn_mtx
);
2294 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
2298 dmu_objset_userspace_present(objset_t
*os
)
2300 return (os
->os_phys
->os_flags
&
2301 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
2305 dmu_objset_userobjspace_present(objset_t
*os
)
2307 return (os
->os_phys
->os_flags
&
2308 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
);
2312 dmu_objset_projectquota_present(objset_t
*os
)
2314 return (os
->os_phys
->os_flags
&
2315 OBJSET_FLAG_PROJECTQUOTA_COMPLETE
);
2319 dmu_objset_space_upgrade(objset_t
*os
)
2325 * We simply need to mark every object dirty, so that it will be
2326 * synced out and now accounted. If this is called
2327 * concurrently, or if we already did some work before crashing,
2328 * that's fine, since we track each object's accounted state
2332 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
2337 mutex_enter(&os
->os_upgrade_lock
);
2338 if (os
->os_upgrade_exit
)
2339 err
= SET_ERROR(EINTR
);
2340 mutex_exit(&os
->os_upgrade_lock
);
2344 if (issig(JUSTLOOKING
) && issig(FORREAL
))
2345 return (SET_ERROR(EINTR
));
2347 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
2350 tx
= dmu_tx_create(os
);
2351 dmu_tx_hold_bonus(tx
, obj
);
2352 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
2354 dmu_buf_rele(db
, FTAG
);
2358 dmu_buf_will_dirty(db
, tx
);
2359 dmu_buf_rele(db
, FTAG
);
2366 dmu_objset_userspace_upgrade_cb(objset_t
*os
)
2370 if (dmu_objset_userspace_present(os
))
2372 if (dmu_objset_is_snapshot(os
))
2373 return (SET_ERROR(EINVAL
));
2374 if (!dmu_objset_userused_enabled(os
))
2375 return (SET_ERROR(ENOTSUP
));
2377 err
= dmu_objset_space_upgrade(os
);
2381 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
2382 txg_wait_synced(dmu_objset_pool(os
), 0);
2387 dmu_objset_userspace_upgrade(objset_t
*os
)
2389 dmu_objset_upgrade(os
, dmu_objset_userspace_upgrade_cb
);
2393 dmu_objset_id_quota_upgrade_cb(objset_t
*os
)
2397 if (dmu_objset_userobjspace_present(os
) &&
2398 dmu_objset_projectquota_present(os
))
2400 if (dmu_objset_is_snapshot(os
))
2401 return (SET_ERROR(EINVAL
));
2402 if (!dmu_objset_userused_enabled(os
))
2403 return (SET_ERROR(ENOTSUP
));
2404 if (!dmu_objset_projectquota_enabled(os
) &&
2405 dmu_objset_userobjspace_present(os
))
2406 return (SET_ERROR(ENOTSUP
));
2408 if (dmu_objset_userobjused_enabled(os
))
2409 dmu_objset_ds(os
)->ds_feature_activation
[
2410 SPA_FEATURE_USEROBJ_ACCOUNTING
] = (void *)B_TRUE
;
2411 if (dmu_objset_projectquota_enabled(os
))
2412 dmu_objset_ds(os
)->ds_feature_activation
[
2413 SPA_FEATURE_PROJECT_QUOTA
] = (void *)B_TRUE
;
2415 err
= dmu_objset_space_upgrade(os
);
2419 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
2420 if (dmu_objset_userobjused_enabled(os
))
2421 os
->os_flags
|= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
;
2422 if (dmu_objset_projectquota_enabled(os
))
2423 os
->os_flags
|= OBJSET_FLAG_PROJECTQUOTA_COMPLETE
;
2425 txg_wait_synced(dmu_objset_pool(os
), 0);
2430 dmu_objset_id_quota_upgrade(objset_t
*os
)
2432 dmu_objset_upgrade(os
, dmu_objset_id_quota_upgrade_cb
);
2436 dmu_objset_userobjspace_upgradable(objset_t
*os
)
2438 return (dmu_objset_type(os
) == DMU_OST_ZFS
&&
2439 !dmu_objset_is_snapshot(os
) &&
2440 dmu_objset_userobjused_enabled(os
) &&
2441 !dmu_objset_userobjspace_present(os
) &&
2442 spa_writeable(dmu_objset_spa(os
)));
2446 dmu_objset_projectquota_upgradable(objset_t
*os
)
2448 return (dmu_objset_type(os
) == DMU_OST_ZFS
&&
2449 !dmu_objset_is_snapshot(os
) &&
2450 dmu_objset_projectquota_enabled(os
) &&
2451 !dmu_objset_projectquota_present(os
) &&
2452 spa_writeable(dmu_objset_spa(os
)));
2456 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
2457 uint64_t *usedobjsp
, uint64_t *availobjsp
)
2459 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
2460 usedobjsp
, availobjsp
);
2464 dmu_objset_fsid_guid(objset_t
*os
)
2466 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
2470 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
2472 stat
->dds_type
= os
->os_phys
->os_type
;
2473 if (os
->os_dsl_dataset
)
2474 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
2478 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
2480 ASSERT(os
->os_dsl_dataset
||
2481 os
->os_phys
->os_type
== DMU_OST_META
);
2483 if (os
->os_dsl_dataset
!= NULL
)
2484 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
2486 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
2487 os
->os_phys
->os_type
);
2488 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
2489 dmu_objset_userspace_present(os
));
2493 dmu_objset_is_snapshot(objset_t
*os
)
2495 if (os
->os_dsl_dataset
!= NULL
)
2496 return (os
->os_dsl_dataset
->ds_is_snapshot
);
2502 dmu_snapshot_realname(objset_t
*os
, const char *name
, char *real
, int maxlen
,
2503 boolean_t
*conflict
)
2505 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
2508 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
2509 return (SET_ERROR(ENOENT
));
2511 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
2512 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, name
, 8, 1, &ignored
,
2513 MT_NORMALIZE
, real
, maxlen
, conflict
));
2517 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
2518 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
2520 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
2521 zap_cursor_t cursor
;
2522 zap_attribute_t attr
;
2524 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
2526 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
2527 return (SET_ERROR(ENOENT
));
2529 zap_cursor_init_serialized(&cursor
,
2530 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
2531 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, *offp
);
2533 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
2534 zap_cursor_fini(&cursor
);
2535 return (SET_ERROR(ENOENT
));
2538 if (strlen(attr
.za_name
) + 1 > namelen
) {
2539 zap_cursor_fini(&cursor
);
2540 return (SET_ERROR(ENAMETOOLONG
));
2543 (void) strlcpy(name
, attr
.za_name
, namelen
);
2545 *idp
= attr
.za_first_integer
;
2547 *case_conflict
= attr
.za_normalization_conflict
;
2548 zap_cursor_advance(&cursor
);
2549 *offp
= zap_cursor_serialize(&cursor
);
2550 zap_cursor_fini(&cursor
);
2556 dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *value
)
2558 return (dsl_dataset_snap_lookup(os
->os_dsl_dataset
, name
, value
));
2562 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
2563 uint64_t *idp
, uint64_t *offp
)
2565 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
2566 zap_cursor_t cursor
;
2567 zap_attribute_t attr
;
2569 /* there is no next dir on a snapshot! */
2570 if (os
->os_dsl_dataset
->ds_object
!=
2571 dsl_dir_phys(dd
)->dd_head_dataset_obj
)
2572 return (SET_ERROR(ENOENT
));
2574 zap_cursor_init_serialized(&cursor
,
2575 dd
->dd_pool
->dp_meta_objset
,
2576 dsl_dir_phys(dd
)->dd_child_dir_zapobj
, *offp
);
2578 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
2579 zap_cursor_fini(&cursor
);
2580 return (SET_ERROR(ENOENT
));
2583 if (strlen(attr
.za_name
) + 1 > namelen
) {
2584 zap_cursor_fini(&cursor
);
2585 return (SET_ERROR(ENAMETOOLONG
));
2588 (void) strlcpy(name
, attr
.za_name
, namelen
);
2590 *idp
= attr
.za_first_integer
;
2591 zap_cursor_advance(&cursor
);
2592 *offp
= zap_cursor_serialize(&cursor
);
2593 zap_cursor_fini(&cursor
);
2598 typedef struct dmu_objset_find_ctx
{
2602 char *dc_ddname
; /* last component of ddobj's name */
2603 int (*dc_func
)(dsl_pool_t
*, dsl_dataset_t
*, void *);
2606 kmutex_t
*dc_error_lock
;
2608 } dmu_objset_find_ctx_t
;
2611 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t
*dcp
)
2613 dsl_pool_t
*dp
= dcp
->dc_dp
;
2617 zap_attribute_t
*attr
;
2621 /* don't process if there already was an error */
2622 if (*dcp
->dc_error
!= 0)
2626 * Note: passing the name (dc_ddname) here is optional, but it
2627 * improves performance because we don't need to call
2628 * zap_value_search() to determine the name.
2630 err
= dsl_dir_hold_obj(dp
, dcp
->dc_ddobj
, dcp
->dc_ddname
, FTAG
, &dd
);
2634 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2635 if (dd
->dd_myname
[0] == '$') {
2636 dsl_dir_rele(dd
, FTAG
);
2640 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
2641 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
2644 * Iterate over all children.
2646 if (dcp
->dc_flags
& DS_FIND_CHILDREN
) {
2647 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2648 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
2649 zap_cursor_retrieve(&zc
, attr
) == 0;
2650 (void) zap_cursor_advance(&zc
)) {
2651 ASSERT3U(attr
->za_integer_length
, ==,
2653 ASSERT3U(attr
->za_num_integers
, ==, 1);
2655 dmu_objset_find_ctx_t
*child_dcp
=
2656 kmem_alloc(sizeof (*child_dcp
), KM_SLEEP
);
2658 child_dcp
->dc_ddobj
= attr
->za_first_integer
;
2659 child_dcp
->dc_ddname
= spa_strdup(attr
->za_name
);
2660 if (dcp
->dc_tq
!= NULL
)
2661 (void) taskq_dispatch(dcp
->dc_tq
,
2662 dmu_objset_find_dp_cb
, child_dcp
, TQ_SLEEP
);
2664 dmu_objset_find_dp_impl(child_dcp
);
2666 zap_cursor_fini(&zc
);
2670 * Iterate over all snapshots.
2672 if (dcp
->dc_flags
& DS_FIND_SNAPSHOTS
) {
2674 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2679 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
2680 dsl_dataset_rele(ds
, FTAG
);
2682 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
2683 zap_cursor_retrieve(&zc
, attr
) == 0;
2684 (void) zap_cursor_advance(&zc
)) {
2685 ASSERT3U(attr
->za_integer_length
, ==,
2687 ASSERT3U(attr
->za_num_integers
, ==, 1);
2689 err
= dsl_dataset_hold_obj(dp
,
2690 attr
->za_first_integer
, FTAG
, &ds
);
2693 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
2694 dsl_dataset_rele(ds
, FTAG
);
2698 zap_cursor_fini(&zc
);
2702 kmem_free(attr
, sizeof (zap_attribute_t
));
2705 dsl_dir_rele(dd
, FTAG
);
2712 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2715 * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2716 * that the dir will remain cached, and we won't have to re-instantiate
2717 * it (which could be expensive due to finding its name via
2718 * zap_value_search()).
2720 dsl_dir_rele(dd
, FTAG
);
2723 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
2724 dsl_dataset_rele(ds
, FTAG
);
2728 mutex_enter(dcp
->dc_error_lock
);
2729 /* only keep first error */
2730 if (*dcp
->dc_error
== 0)
2731 *dcp
->dc_error
= err
;
2732 mutex_exit(dcp
->dc_error_lock
);
2735 if (dcp
->dc_ddname
!= NULL
)
2736 spa_strfree(dcp
->dc_ddname
);
2737 kmem_free(dcp
, sizeof (*dcp
));
2741 dmu_objset_find_dp_cb(void *arg
)
2743 dmu_objset_find_ctx_t
*dcp
= arg
;
2744 dsl_pool_t
*dp
= dcp
->dc_dp
;
2747 * We need to get a pool_config_lock here, as there are several
2748 * assert(pool_config_held) down the stack. Getting a lock via
2749 * dsl_pool_config_enter is risky, as it might be stalled by a
2750 * pending writer. This would deadlock, as the write lock can
2751 * only be granted when our parent thread gives up the lock.
2752 * The _prio interface gives us priority over a pending writer.
2754 dsl_pool_config_enter_prio(dp
, FTAG
);
2756 dmu_objset_find_dp_impl(dcp
);
2758 dsl_pool_config_exit(dp
, FTAG
);
2762 * Find objsets under and including ddobj, call func(ds) on each.
2763 * The order for the enumeration is completely undefined.
2764 * func is called with dsl_pool_config held.
2767 dmu_objset_find_dp(dsl_pool_t
*dp
, uint64_t ddobj
,
2768 int func(dsl_pool_t
*, dsl_dataset_t
*, void *), void *arg
, int flags
)
2773 dmu_objset_find_ctx_t
*dcp
;
2776 mutex_init(&err_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2777 dcp
= kmem_alloc(sizeof (*dcp
), KM_SLEEP
);
2780 dcp
->dc_ddobj
= ddobj
;
2781 dcp
->dc_ddname
= NULL
;
2782 dcp
->dc_func
= func
;
2784 dcp
->dc_flags
= flags
;
2785 dcp
->dc_error_lock
= &err_lock
;
2786 dcp
->dc_error
= &error
;
2788 if ((flags
& DS_FIND_SERIALIZE
) || dsl_pool_config_held_writer(dp
)) {
2790 * In case a write lock is held we can't make use of
2791 * parallelism, as down the stack of the worker threads
2792 * the lock is asserted via dsl_pool_config_held.
2793 * In case of a read lock this is solved by getting a read
2794 * lock in each worker thread, which isn't possible in case
2795 * of a writer lock. So we fall back to the synchronous path
2797 * In the future it might be possible to get some magic into
2798 * dsl_pool_config_held in a way that it returns true for
2799 * the worker threads so that a single lock held from this
2800 * thread suffices. For now, stay single threaded.
2802 dmu_objset_find_dp_impl(dcp
);
2803 mutex_destroy(&err_lock
);
2808 ntasks
= dmu_find_threads
;
2810 ntasks
= vdev_count_leaves(dp
->dp_spa
) * 4;
2811 tq
= taskq_create("dmu_objset_find", ntasks
, maxclsyspri
, ntasks
,
2814 kmem_free(dcp
, sizeof (*dcp
));
2815 mutex_destroy(&err_lock
);
2817 return (SET_ERROR(ENOMEM
));
2821 /* dcp will be freed by task */
2822 (void) taskq_dispatch(tq
, dmu_objset_find_dp_cb
, dcp
, TQ_SLEEP
);
2825 * PORTING: this code relies on the property of taskq_wait to wait
2826 * until no more tasks are queued and no more tasks are active. As
2827 * we always queue new tasks from within other tasks, task_wait
2828 * reliably waits for the full recursion to finish, even though we
2829 * enqueue new tasks after taskq_wait has been called.
2830 * On platforms other than illumos, taskq_wait may not have this
2835 mutex_destroy(&err_lock
);
2841 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2842 * The dp_config_rwlock must not be held when this is called, and it
2843 * will not be held when the callback is called.
2844 * Therefore this function should only be used when the pool is not changing
2845 * (e.g. in syncing context), or the callback can deal with the possible races.
2848 dmu_objset_find_impl(spa_t
*spa
, const char *name
,
2849 int func(const char *, void *), void *arg
, int flags
)
2852 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2855 zap_attribute_t
*attr
;
2860 dsl_pool_config_enter(dp
, FTAG
);
2862 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, NULL
);
2864 dsl_pool_config_exit(dp
, FTAG
);
2868 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2869 if (dd
->dd_myname
[0] == '$') {
2870 dsl_dir_rele(dd
, FTAG
);
2871 dsl_pool_config_exit(dp
, FTAG
);
2875 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
2876 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
2879 * Iterate over all children.
2881 if (flags
& DS_FIND_CHILDREN
) {
2882 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2883 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
2884 zap_cursor_retrieve(&zc
, attr
) == 0;
2885 (void) zap_cursor_advance(&zc
)) {
2886 ASSERT3U(attr
->za_integer_length
, ==,
2888 ASSERT3U(attr
->za_num_integers
, ==, 1);
2890 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
2891 dsl_pool_config_exit(dp
, FTAG
);
2892 err
= dmu_objset_find_impl(spa
, child
,
2894 dsl_pool_config_enter(dp
, FTAG
);
2895 kmem_strfree(child
);
2899 zap_cursor_fini(&zc
);
2902 dsl_dir_rele(dd
, FTAG
);
2903 dsl_pool_config_exit(dp
, FTAG
);
2904 kmem_free(attr
, sizeof (zap_attribute_t
));
2910 * Iterate over all snapshots.
2912 if (flags
& DS_FIND_SNAPSHOTS
) {
2913 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2918 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
2919 dsl_dataset_rele(ds
, FTAG
);
2921 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
2922 zap_cursor_retrieve(&zc
, attr
) == 0;
2923 (void) zap_cursor_advance(&zc
)) {
2924 ASSERT3U(attr
->za_integer_length
, ==,
2926 ASSERT3U(attr
->za_num_integers
, ==, 1);
2928 child
= kmem_asprintf("%s@%s",
2929 name
, attr
->za_name
);
2930 dsl_pool_config_exit(dp
, FTAG
);
2931 err
= func(child
, arg
);
2932 dsl_pool_config_enter(dp
, FTAG
);
2933 kmem_strfree(child
);
2937 zap_cursor_fini(&zc
);
2941 dsl_dir_rele(dd
, FTAG
);
2942 kmem_free(attr
, sizeof (zap_attribute_t
));
2943 dsl_pool_config_exit(dp
, FTAG
);
2948 /* Apply to self. */
2949 return (func(name
, arg
));
2953 * See comment above dmu_objset_find_impl().
2956 dmu_objset_find(const char *name
, int func(const char *, void *), void *arg
,
2962 error
= spa_open(name
, &spa
, FTAG
);
2965 error
= dmu_objset_find_impl(spa
, name
, func
, arg
, flags
);
2966 spa_close(spa
, FTAG
);
2971 dmu_objset_incompatible_encryption_version(objset_t
*os
)
2973 return (dsl_dir_incompatible_encryption_version(
2974 os
->os_dsl_dataset
->ds_dir
));
2978 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
2980 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2981 os
->os_user_ptr
= user_ptr
;
2985 dmu_objset_get_user(objset_t
*os
)
2987 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2988 return (os
->os_user_ptr
);
2992 * Determine name of filesystem, given name of snapshot.
2993 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2996 dmu_fsname(const char *snapname
, char *buf
)
2998 char *atp
= strchr(snapname
, '@');
3000 return (SET_ERROR(EINVAL
));
3001 if (atp
- snapname
>= ZFS_MAX_DATASET_NAME_LEN
)
3002 return (SET_ERROR(ENAMETOOLONG
));
3003 (void) strlcpy(buf
, snapname
, atp
- snapname
+ 1);
3008 * Call when we think we're going to write/free space in open context
3009 * to track the amount of dirty data in the open txg, which is also the
3010 * amount of memory that can not be evicted until this txg syncs.
3012 * Note that there are two conditions where this can be called from
3015 * [1] When we just created the dataset, in which case we go on with
3016 * updating any accounting of dirty data as usual.
3017 * [2] When we are dirtying MOS data, in which case we only update the
3018 * pool's accounting of dirty data.
3021 dmu_objset_willuse_space(objset_t
*os
, int64_t space
, dmu_tx_t
*tx
)
3023 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
3024 int64_t aspace
= spa_get_worst_case_asize(os
->os_spa
, space
);
3027 dsl_dir_willuse_space(ds
->ds_dir
, aspace
, tx
);
3030 dsl_pool_dirty_space(dmu_tx_pool(tx
), space
, tx
);
3033 #if defined(_KERNEL)
3034 EXPORT_SYMBOL(dmu_objset_zil
);
3035 EXPORT_SYMBOL(dmu_objset_pool
);
3036 EXPORT_SYMBOL(dmu_objset_ds
);
3037 EXPORT_SYMBOL(dmu_objset_type
);
3038 EXPORT_SYMBOL(dmu_objset_name
);
3039 EXPORT_SYMBOL(dmu_objset_hold
);
3040 EXPORT_SYMBOL(dmu_objset_hold_flags
);
3041 EXPORT_SYMBOL(dmu_objset_own
);
3042 EXPORT_SYMBOL(dmu_objset_rele
);
3043 EXPORT_SYMBOL(dmu_objset_rele_flags
);
3044 EXPORT_SYMBOL(dmu_objset_disown
);
3045 EXPORT_SYMBOL(dmu_objset_from_ds
);
3046 EXPORT_SYMBOL(dmu_objset_create
);
3047 EXPORT_SYMBOL(dmu_objset_clone
);
3048 EXPORT_SYMBOL(dmu_objset_stats
);
3049 EXPORT_SYMBOL(dmu_objset_fast_stat
);
3050 EXPORT_SYMBOL(dmu_objset_spa
);
3051 EXPORT_SYMBOL(dmu_objset_space
);
3052 EXPORT_SYMBOL(dmu_objset_fsid_guid
);
3053 EXPORT_SYMBOL(dmu_objset_find
);
3054 EXPORT_SYMBOL(dmu_objset_byteswap
);
3055 EXPORT_SYMBOL(dmu_objset_evict_dbufs
);
3056 EXPORT_SYMBOL(dmu_objset_snap_cmtime
);
3057 EXPORT_SYMBOL(dmu_objset_dnodesize
);
3059 EXPORT_SYMBOL(dmu_objset_sync
);
3060 EXPORT_SYMBOL(dmu_objset_is_dirty
);
3061 EXPORT_SYMBOL(dmu_objset_create_impl_dnstats
);
3062 EXPORT_SYMBOL(dmu_objset_create_impl
);
3063 EXPORT_SYMBOL(dmu_objset_open_impl
);
3064 EXPORT_SYMBOL(dmu_objset_evict
);
3065 EXPORT_SYMBOL(dmu_objset_register_type
);
3066 EXPORT_SYMBOL(dmu_objset_sync_done
);
3067 EXPORT_SYMBOL(dmu_objset_userquota_get_ids
);
3068 EXPORT_SYMBOL(dmu_objset_userused_enabled
);
3069 EXPORT_SYMBOL(dmu_objset_userspace_upgrade
);
3070 EXPORT_SYMBOL(dmu_objset_userspace_present
);
3071 EXPORT_SYMBOL(dmu_objset_userobjused_enabled
);
3072 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable
);
3073 EXPORT_SYMBOL(dmu_objset_userobjspace_present
);
3074 EXPORT_SYMBOL(dmu_objset_projectquota_enabled
);
3075 EXPORT_SYMBOL(dmu_objset_projectquota_present
);
3076 EXPORT_SYMBOL(dmu_objset_projectquota_upgradable
);
3077 EXPORT_SYMBOL(dmu_objset_id_quota_upgrade
);