4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
26 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
29 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
30 * Copyright 2017 Nexenta Systems, Inc.
33 /* Portions Copyright 2010 Robert Milkowski */
35 #include <sys/zfeature.h>
37 #include <sys/zfs_context.h>
38 #include <sys/dmu_objset.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/dsl_dataset.h>
41 #include <sys/dsl_prop.h>
42 #include <sys/dsl_pool.h>
43 #include <sys/dsl_synctask.h>
44 #include <sys/dsl_deleg.h>
45 #include <sys/dnode.h>
48 #include <sys/dmu_tx.h>
51 #include <sys/dmu_impl.h>
52 #include <sys/zfs_ioctl.h>
54 #include <sys/zfs_onexit.h>
55 #include <sys/dsl_destroy.h>
57 #include <sys/policy.h>
58 #include <sys/spa_impl.h>
59 #include <sys/dmu_send.h>
62 * Needed to close a window in dnode_move() that allows the objset to be freed
63 * before it can be safely accessed.
68 * Tunable to overwrite the maximum number of threads for the parallelization
69 * of dmu_objset_find_dp, needed to speed up the import of pools with many
71 * Default is 4 times the number of leaf vdevs.
73 int dmu_find_threads
= 0;
76 * Backfill lower metadnode objects after this many have been freed.
77 * Backfilling negatively impacts object creation rates, so only do it
78 * if there are enough holes to fill.
80 int dmu_rescan_dnode_threshold
= 1 << DN_MAX_INDBLKSHIFT
;
82 static void dmu_objset_find_dp_cb(void *arg
);
84 static void dmu_objset_upgrade(objset_t
*os
, dmu_objset_upgrade_cb_t cb
);
85 static void dmu_objset_upgrade_stop(objset_t
*os
);
90 rw_init(&os_lock
, NULL
, RW_DEFAULT
, NULL
);
100 dmu_objset_spa(objset_t
*os
)
106 dmu_objset_zil(objset_t
*os
)
112 dmu_objset_pool(objset_t
*os
)
116 if ((ds
= os
->os_dsl_dataset
) != NULL
&& ds
->ds_dir
)
117 return (ds
->ds_dir
->dd_pool
);
119 return (spa_get_dsl(os
->os_spa
));
123 dmu_objset_ds(objset_t
*os
)
125 return (os
->os_dsl_dataset
);
129 dmu_objset_type(objset_t
*os
)
131 return (os
->os_phys
->os_type
);
135 dmu_objset_name(objset_t
*os
, char *buf
)
137 dsl_dataset_name(os
->os_dsl_dataset
, buf
);
141 dmu_objset_id(objset_t
*os
)
143 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
145 return (ds
? ds
->ds_object
: 0);
149 dmu_objset_dnodesize(objset_t
*os
)
151 return (os
->os_dnodesize
);
155 dmu_objset_syncprop(objset_t
*os
)
157 return (os
->os_sync
);
161 dmu_objset_logbias(objset_t
*os
)
163 return (os
->os_logbias
);
167 checksum_changed_cb(void *arg
, uint64_t newval
)
172 * Inheritance should have been done by now.
174 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
176 os
->os_checksum
= zio_checksum_select(newval
, ZIO_CHECKSUM_ON_VALUE
);
180 compression_changed_cb(void *arg
, uint64_t newval
)
185 * Inheritance and range checking should have been done by now.
187 ASSERT(newval
!= ZIO_COMPRESS_INHERIT
);
189 os
->os_compress
= zio_compress_select(os
->os_spa
, newval
,
194 copies_changed_cb(void *arg
, uint64_t newval
)
199 * Inheritance and range checking should have been done by now.
202 ASSERT(newval
<= spa_max_replication(os
->os_spa
));
204 os
->os_copies
= newval
;
208 dedup_changed_cb(void *arg
, uint64_t newval
)
211 spa_t
*spa
= os
->os_spa
;
212 enum zio_checksum checksum
;
215 * Inheritance should have been done by now.
217 ASSERT(newval
!= ZIO_CHECKSUM_INHERIT
);
219 checksum
= zio_checksum_dedup_select(spa
, newval
, ZIO_CHECKSUM_OFF
);
221 os
->os_dedup_checksum
= checksum
& ZIO_CHECKSUM_MASK
;
222 os
->os_dedup_verify
= !!(checksum
& ZIO_CHECKSUM_VERIFY
);
226 primary_cache_changed_cb(void *arg
, uint64_t newval
)
231 * Inheritance and range checking should have been done by now.
233 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
234 newval
== ZFS_CACHE_METADATA
);
236 os
->os_primary_cache
= newval
;
240 secondary_cache_changed_cb(void *arg
, uint64_t newval
)
245 * Inheritance and range checking should have been done by now.
247 ASSERT(newval
== ZFS_CACHE_ALL
|| newval
== ZFS_CACHE_NONE
||
248 newval
== ZFS_CACHE_METADATA
);
250 os
->os_secondary_cache
= newval
;
254 sync_changed_cb(void *arg
, uint64_t newval
)
259 * Inheritance and range checking should have been done by now.
261 ASSERT(newval
== ZFS_SYNC_STANDARD
|| newval
== ZFS_SYNC_ALWAYS
||
262 newval
== ZFS_SYNC_DISABLED
);
264 os
->os_sync
= newval
;
266 zil_set_sync(os
->os_zil
, newval
);
270 redundant_metadata_changed_cb(void *arg
, uint64_t newval
)
275 * Inheritance and range checking should have been done by now.
277 ASSERT(newval
== ZFS_REDUNDANT_METADATA_ALL
||
278 newval
== ZFS_REDUNDANT_METADATA_MOST
);
280 os
->os_redundant_metadata
= newval
;
284 dnodesize_changed_cb(void *arg
, uint64_t newval
)
289 case ZFS_DNSIZE_LEGACY
:
290 os
->os_dnodesize
= DNODE_MIN_SIZE
;
292 case ZFS_DNSIZE_AUTO
:
294 * Choose a dnode size that will work well for most
295 * workloads if the user specified "auto". Future code
296 * improvements could dynamically select a dnode size
297 * based on observed workload patterns.
299 os
->os_dnodesize
= DNODE_MIN_SIZE
* 2;
306 os
->os_dnodesize
= newval
;
312 logbias_changed_cb(void *arg
, uint64_t newval
)
316 ASSERT(newval
== ZFS_LOGBIAS_LATENCY
||
317 newval
== ZFS_LOGBIAS_THROUGHPUT
);
318 os
->os_logbias
= newval
;
320 zil_set_logbias(os
->os_zil
, newval
);
324 recordsize_changed_cb(void *arg
, uint64_t newval
)
328 os
->os_recordsize
= newval
;
332 dmu_objset_byteswap(void *buf
, size_t size
)
334 objset_phys_t
*osp
= buf
;
336 ASSERT(size
== OBJSET_OLD_PHYS_SIZE
|| size
== sizeof (objset_phys_t
));
337 dnode_byteswap(&osp
->os_meta_dnode
);
338 byteswap_uint64_array(&osp
->os_zil_header
, sizeof (zil_header_t
));
339 osp
->os_type
= BSWAP_64(osp
->os_type
);
340 osp
->os_flags
= BSWAP_64(osp
->os_flags
);
341 if (size
== sizeof (objset_phys_t
)) {
342 dnode_byteswap(&osp
->os_userused_dnode
);
343 dnode_byteswap(&osp
->os_groupused_dnode
);
348 * The hash is a CRC-based hash of the objset_t pointer and the object number.
351 dnode_hash(const objset_t
*os
, uint64_t obj
)
353 uintptr_t osv
= (uintptr_t)os
;
354 uint64_t crc
= -1ULL;
356 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
358 * The low 6 bits of the pointer don't have much entropy, because
359 * the objset_t is larger than 2^6 bytes long.
361 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (osv
>> 6)) & 0xFF];
362 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 0)) & 0xFF];
363 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 8)) & 0xFF];
364 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (obj
>> 16)) & 0xFF];
366 crc
^= (osv
>>14) ^ (obj
>>24);
372 dnode_multilist_index_func(multilist_t
*ml
, void *obj
)
375 return (dnode_hash(dn
->dn_objset
, dn
->dn_object
) %
376 multilist_get_num_sublists(ml
));
380 dmu_objset_open_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
386 ASSERT(ds
== NULL
|| MUTEX_HELD(&ds
->ds_opening_lock
));
388 os
= kmem_zalloc(sizeof (objset_t
), KM_SLEEP
);
389 os
->os_dsl_dataset
= ds
;
392 if (!BP_IS_HOLE(os
->os_rootbp
)) {
393 arc_flags_t aflags
= ARC_FLAG_WAIT
;
395 enum zio_flag zio_flags
= ZIO_FLAG_CANFAIL
;
396 SET_BOOKMARK(&zb
, ds
? ds
->ds_object
: DMU_META_OBJSET
,
397 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
399 if (DMU_OS_IS_L2CACHEABLE(os
))
400 aflags
|= ARC_FLAG_L2CACHE
;
402 if (ds
!= NULL
&& ds
->ds_dir
->dd_crypto_obj
!= 0) {
403 ASSERT3U(BP_GET_COMPRESS(bp
), ==, ZIO_COMPRESS_OFF
);
404 ASSERT(BP_IS_AUTHENTICATED(bp
));
405 zio_flags
|= ZIO_FLAG_RAW
;
408 dprintf_bp(os
->os_rootbp
, "reading %s", "");
409 err
= arc_read(NULL
, spa
, os
->os_rootbp
,
410 arc_getbuf_func
, &os
->os_phys_buf
,
411 ZIO_PRIORITY_SYNC_READ
, zio_flags
, &aflags
, &zb
);
413 kmem_free(os
, sizeof (objset_t
));
414 /* convert checksum errors into IO errors */
416 err
= SET_ERROR(EIO
);
420 /* Increase the blocksize if we are permitted. */
421 if (spa_version(spa
) >= SPA_VERSION_USERSPACE
&&
422 arc_buf_size(os
->os_phys_buf
) < sizeof (objset_phys_t
)) {
423 arc_buf_t
*buf
= arc_alloc_buf(spa
, &os
->os_phys_buf
,
424 ARC_BUFC_METADATA
, sizeof (objset_phys_t
));
425 bzero(buf
->b_data
, sizeof (objset_phys_t
));
426 bcopy(os
->os_phys_buf
->b_data
, buf
->b_data
,
427 arc_buf_size(os
->os_phys_buf
));
428 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
429 os
->os_phys_buf
= buf
;
432 os
->os_phys
= os
->os_phys_buf
->b_data
;
433 os
->os_flags
= os
->os_phys
->os_flags
;
435 int size
= spa_version(spa
) >= SPA_VERSION_USERSPACE
?
436 sizeof (objset_phys_t
) : OBJSET_OLD_PHYS_SIZE
;
437 os
->os_phys_buf
= arc_alloc_buf(spa
, &os
->os_phys_buf
,
438 ARC_BUFC_METADATA
, size
);
439 os
->os_phys
= os
->os_phys_buf
->b_data
;
440 bzero(os
->os_phys
, size
);
444 * Note: the changed_cb will be called once before the register
445 * func returns, thus changing the checksum/compression from the
446 * default (fletcher2/off). Snapshots don't need to know about
447 * checksum/compression/copies.
450 boolean_t needlock
= B_FALSE
;
452 os
->os_encrypted
= (ds
->ds_dir
->dd_crypto_obj
!= 0);
455 * Note: it's valid to open the objset if the dataset is
456 * long-held, in which case the pool_config lock will not
459 if (!dsl_pool_config_held(dmu_objset_pool(os
))) {
461 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
464 err
= dsl_prop_register(ds
,
465 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE
),
466 primary_cache_changed_cb
, os
);
468 err
= dsl_prop_register(ds
,
469 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE
),
470 secondary_cache_changed_cb
, os
);
472 if (!ds
->ds_is_snapshot
) {
474 err
= dsl_prop_register(ds
,
475 zfs_prop_to_name(ZFS_PROP_CHECKSUM
),
476 checksum_changed_cb
, os
);
479 err
= dsl_prop_register(ds
,
480 zfs_prop_to_name(ZFS_PROP_COMPRESSION
),
481 compression_changed_cb
, os
);
484 err
= dsl_prop_register(ds
,
485 zfs_prop_to_name(ZFS_PROP_COPIES
),
486 copies_changed_cb
, os
);
489 err
= dsl_prop_register(ds
,
490 zfs_prop_to_name(ZFS_PROP_DEDUP
),
491 dedup_changed_cb
, os
);
494 err
= dsl_prop_register(ds
,
495 zfs_prop_to_name(ZFS_PROP_LOGBIAS
),
496 logbias_changed_cb
, os
);
499 err
= dsl_prop_register(ds
,
500 zfs_prop_to_name(ZFS_PROP_SYNC
),
501 sync_changed_cb
, os
);
504 err
= dsl_prop_register(ds
,
506 ZFS_PROP_REDUNDANT_METADATA
),
507 redundant_metadata_changed_cb
, os
);
510 err
= dsl_prop_register(ds
,
511 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
512 recordsize_changed_cb
, os
);
515 err
= dsl_prop_register(ds
,
516 zfs_prop_to_name(ZFS_PROP_DNODESIZE
),
517 dnodesize_changed_cb
, os
);
521 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
523 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
524 kmem_free(os
, sizeof (objset_t
));
528 /* It's the meta-objset. */
529 os
->os_checksum
= ZIO_CHECKSUM_FLETCHER_4
;
530 os
->os_compress
= ZIO_COMPRESS_ON
;
531 os
->os_encrypted
= B_FALSE
;
532 os
->os_copies
= spa_max_replication(spa
);
533 os
->os_dedup_checksum
= ZIO_CHECKSUM_OFF
;
534 os
->os_dedup_verify
= B_FALSE
;
535 os
->os_logbias
= ZFS_LOGBIAS_LATENCY
;
536 os
->os_sync
= ZFS_SYNC_STANDARD
;
537 os
->os_primary_cache
= ZFS_CACHE_ALL
;
538 os
->os_secondary_cache
= ZFS_CACHE_ALL
;
539 os
->os_dnodesize
= DNODE_MIN_SIZE
;
542 if (ds
== NULL
|| !ds
->ds_is_snapshot
)
543 os
->os_zil_header
= os
->os_phys
->os_zil_header
;
544 os
->os_zil
= zil_alloc(os
, &os
->os_zil_header
);
546 for (i
= 0; i
< TXG_SIZE
; i
++) {
547 os
->os_dirty_dnodes
[i
] = multilist_create(sizeof (dnode_t
),
548 offsetof(dnode_t
, dn_dirty_link
[i
]),
549 dnode_multilist_index_func
);
551 list_create(&os
->os_dnodes
, sizeof (dnode_t
),
552 offsetof(dnode_t
, dn_link
));
553 list_create(&os
->os_downgraded_dbufs
, sizeof (dmu_buf_impl_t
),
554 offsetof(dmu_buf_impl_t
, db_link
));
556 list_link_init(&os
->os_evicting_node
);
558 mutex_init(&os
->os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
559 mutex_init(&os
->os_userused_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
560 mutex_init(&os
->os_obj_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
561 mutex_init(&os
->os_user_ptr_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
562 os
->os_obj_next_percpu_len
= boot_ncpus
;
563 os
->os_obj_next_percpu
= kmem_zalloc(os
->os_obj_next_percpu_len
*
564 sizeof (os
->os_obj_next_percpu
[0]), KM_SLEEP
);
566 dnode_special_open(os
, &os
->os_phys
->os_meta_dnode
,
567 DMU_META_DNODE_OBJECT
, &os
->os_meta_dnode
);
568 if (arc_buf_size(os
->os_phys_buf
) >= sizeof (objset_phys_t
)) {
569 dnode_special_open(os
, &os
->os_phys
->os_userused_dnode
,
570 DMU_USERUSED_OBJECT
, &os
->os_userused_dnode
);
571 dnode_special_open(os
, &os
->os_phys
->os_groupused_dnode
,
572 DMU_GROUPUSED_OBJECT
, &os
->os_groupused_dnode
);
575 mutex_init(&os
->os_upgrade_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
582 dmu_objset_from_ds(dsl_dataset_t
*ds
, objset_t
**osp
)
587 * We shouldn't be doing anything with dsl_dataset_t's unless the
588 * pool_config lock is held, or the dataset is long-held.
590 ASSERT(dsl_pool_config_held(ds
->ds_dir
->dd_pool
) ||
591 dsl_dataset_long_held(ds
));
593 mutex_enter(&ds
->ds_opening_lock
);
594 if (ds
->ds_objset
== NULL
) {
596 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
597 err
= dmu_objset_open_impl(dsl_dataset_get_spa(ds
),
598 ds
, dsl_dataset_get_blkptr(ds
), &os
);
599 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
602 mutex_enter(&ds
->ds_lock
);
603 ASSERT(ds
->ds_objset
== NULL
);
605 mutex_exit(&ds
->ds_lock
);
608 *osp
= ds
->ds_objset
;
609 mutex_exit(&ds
->ds_opening_lock
);
614 * Holds the pool while the objset is held. Therefore only one objset
615 * can be held at a time.
618 dmu_objset_hold_flags(const char *name
, boolean_t decrypt
, void *tag
,
624 ds_hold_flags_t flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: 0;
626 err
= dsl_pool_hold(name
, tag
, &dp
);
629 err
= dsl_dataset_hold_flags(dp
, name
, flags
, tag
, &ds
);
631 dsl_pool_rele(dp
, tag
);
635 err
= dmu_objset_from_ds(ds
, osp
);
637 dsl_dataset_rele(ds
, tag
);
638 dsl_pool_rele(dp
, tag
);
645 dmu_objset_hold(const char *name
, void *tag
, objset_t
**osp
)
647 return (dmu_objset_hold_flags(name
, B_FALSE
, tag
, osp
));
651 dmu_objset_own_impl(dsl_dataset_t
*ds
, dmu_objset_type_t type
,
652 boolean_t readonly
, boolean_t decrypt
, void *tag
, objset_t
**osp
)
656 err
= dmu_objset_from_ds(ds
, osp
);
659 } else if (type
!= DMU_OST_ANY
&& type
!= (*osp
)->os_phys
->os_type
) {
660 return (SET_ERROR(EINVAL
));
661 } else if (!readonly
&& dsl_dataset_is_snapshot(ds
)) {
662 return (SET_ERROR(EROFS
));
665 /* if we are decrypting, we can now check MACs in os->os_phys_buf */
666 if (decrypt
&& arc_is_unauthenticated((*osp
)->os_phys_buf
)) {
667 err
= arc_untransform((*osp
)->os_phys_buf
, (*osp
)->os_spa
,
668 ds
->ds_object
, B_FALSE
);
672 ASSERT0(arc_is_unauthenticated((*osp
)->os_phys_buf
));
679 * dsl_pool must not be held when this is called.
680 * Upon successful return, there will be a longhold on the dataset,
681 * and the dsl_pool will not be held.
684 dmu_objset_own(const char *name
, dmu_objset_type_t type
,
685 boolean_t readonly
, boolean_t decrypt
, void *tag
, objset_t
**osp
)
690 ds_hold_flags_t flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: 0;
692 err
= dsl_pool_hold(name
, FTAG
, &dp
);
695 err
= dsl_dataset_own(dp
, name
, flags
, tag
, &ds
);
697 dsl_pool_rele(dp
, FTAG
);
700 err
= dmu_objset_own_impl(ds
, type
, readonly
, decrypt
, tag
, osp
);
702 dsl_dataset_disown(ds
, flags
, tag
);
703 dsl_pool_rele(dp
, FTAG
);
707 dsl_pool_rele(dp
, FTAG
);
709 if (dmu_objset_userobjspace_upgradable(*osp
))
710 dmu_objset_userobjspace_upgrade(*osp
);
716 dmu_objset_own_obj(dsl_pool_t
*dp
, uint64_t obj
, dmu_objset_type_t type
,
717 boolean_t readonly
, boolean_t decrypt
, void *tag
, objset_t
**osp
)
721 ds_hold_flags_t flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: 0;
723 err
= dsl_dataset_own_obj(dp
, obj
, flags
, tag
, &ds
);
727 err
= dmu_objset_own_impl(ds
, type
, readonly
, decrypt
, tag
, osp
);
729 dsl_dataset_disown(ds
, flags
, tag
);
737 dmu_objset_rele_flags(objset_t
*os
, boolean_t decrypt
, void *tag
)
739 ds_hold_flags_t flags
= (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: 0;
741 dsl_pool_t
*dp
= dmu_objset_pool(os
);
742 dsl_dataset_rele_flags(os
->os_dsl_dataset
, flags
, tag
);
743 dsl_pool_rele(dp
, tag
);
747 dmu_objset_rele(objset_t
*os
, void *tag
)
749 dmu_objset_rele_flags(os
, B_FALSE
, tag
);
753 * When we are called, os MUST refer to an objset associated with a dataset
754 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
755 * == tag. We will then release and reacquire ownership of the dataset while
756 * holding the pool config_rwlock to avoid intervening namespace or ownership
759 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
760 * release the hold on its dataset and acquire a new one on the dataset of the
761 * same name so that it can be partially torn down and reconstructed.
764 dmu_objset_refresh_ownership(objset_t
*os
, boolean_t decrypt
, void *tag
)
767 dsl_dataset_t
*ds
, *newds
;
768 char name
[ZFS_MAX_DATASET_NAME_LEN
];
770 ds
= os
->os_dsl_dataset
;
771 VERIFY3P(ds
, !=, NULL
);
772 VERIFY3P(ds
->ds_owner
, ==, tag
);
773 VERIFY(dsl_dataset_long_held(ds
));
775 dsl_dataset_name(ds
, name
);
776 dp
= dmu_objset_pool(os
);
777 dsl_pool_config_enter(dp
, FTAG
);
778 dmu_objset_disown(os
, decrypt
, tag
);
779 VERIFY0(dsl_dataset_own(dp
, name
,
780 (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: 0, tag
, &newds
));
781 VERIFY3P(newds
, ==, os
->os_dsl_dataset
);
782 dsl_pool_config_exit(dp
, FTAG
);
786 dmu_objset_disown(objset_t
*os
, boolean_t decrypt
, void *tag
)
789 * Stop upgrading thread
791 dmu_objset_upgrade_stop(os
);
792 dsl_dataset_disown(os
->os_dsl_dataset
,
793 (decrypt
) ? DS_HOLD_FLAG_DECRYPT
: 0, tag
);
797 dmu_objset_evict_dbufs(objset_t
*os
)
802 dn_marker
= kmem_alloc(sizeof (dnode_t
), KM_SLEEP
);
804 mutex_enter(&os
->os_lock
);
805 dn
= list_head(&os
->os_dnodes
);
808 * Skip dnodes without holds. We have to do this dance
809 * because dnode_add_ref() only works if there is already a
810 * hold. If the dnode has no holds, then it has no dbufs.
812 if (dnode_add_ref(dn
, FTAG
)) {
813 list_insert_after(&os
->os_dnodes
, dn
, dn_marker
);
814 mutex_exit(&os
->os_lock
);
816 dnode_evict_dbufs(dn
);
817 dnode_rele(dn
, FTAG
);
819 mutex_enter(&os
->os_lock
);
820 dn
= list_next(&os
->os_dnodes
, dn_marker
);
821 list_remove(&os
->os_dnodes
, dn_marker
);
823 dn
= list_next(&os
->os_dnodes
, dn
);
826 mutex_exit(&os
->os_lock
);
828 kmem_free(dn_marker
, sizeof (dnode_t
));
830 if (DMU_USERUSED_DNODE(os
) != NULL
) {
831 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os
));
832 dnode_evict_dbufs(DMU_USERUSED_DNODE(os
));
834 dnode_evict_dbufs(DMU_META_DNODE(os
));
838 * Objset eviction processing is split into into two pieces.
839 * The first marks the objset as evicting, evicts any dbufs that
840 * have a refcount of zero, and then queues up the objset for the
841 * second phase of eviction. Once os->os_dnodes has been cleared by
842 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
843 * The second phase closes the special dnodes, dequeues the objset from
844 * the list of those undergoing eviction, and finally frees the objset.
846 * NOTE: Due to asynchronous eviction processing (invocation of
847 * dnode_buf_pageout()), it is possible for the meta dnode for the
848 * objset to have no holds even though os->os_dnodes is not empty.
851 dmu_objset_evict(objset_t
*os
)
855 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
857 for (t
= 0; t
< TXG_SIZE
; t
++)
858 ASSERT(!dmu_objset_is_dirty(os
, t
));
861 dsl_prop_unregister_all(ds
, os
);
866 dmu_objset_evict_dbufs(os
);
868 mutex_enter(&os
->os_lock
);
869 spa_evicting_os_register(os
->os_spa
, os
);
870 if (list_is_empty(&os
->os_dnodes
)) {
871 mutex_exit(&os
->os_lock
);
872 dmu_objset_evict_done(os
);
874 mutex_exit(&os
->os_lock
);
881 dmu_objset_evict_done(objset_t
*os
)
883 ASSERT3P(list_head(&os
->os_dnodes
), ==, NULL
);
885 dnode_special_close(&os
->os_meta_dnode
);
886 if (DMU_USERUSED_DNODE(os
)) {
887 dnode_special_close(&os
->os_userused_dnode
);
888 dnode_special_close(&os
->os_groupused_dnode
);
890 zil_free(os
->os_zil
);
892 arc_buf_destroy(os
->os_phys_buf
, &os
->os_phys_buf
);
895 * This is a barrier to prevent the objset from going away in
896 * dnode_move() until we can safely ensure that the objset is still in
897 * use. We consider the objset valid before the barrier and invalid
900 rw_enter(&os_lock
, RW_READER
);
903 kmem_free(os
->os_obj_next_percpu
,
904 os
->os_obj_next_percpu_len
* sizeof (os
->os_obj_next_percpu
[0]));
906 mutex_destroy(&os
->os_lock
);
907 mutex_destroy(&os
->os_userused_lock
);
908 mutex_destroy(&os
->os_obj_lock
);
909 mutex_destroy(&os
->os_user_ptr_lock
);
910 mutex_destroy(&os
->os_upgrade_lock
);
911 for (int i
= 0; i
< TXG_SIZE
; i
++) {
912 multilist_destroy(os
->os_dirty_dnodes
[i
]);
914 spa_evicting_os_deregister(os
->os_spa
, os
);
915 kmem_free(os
, sizeof (objset_t
));
919 dmu_objset_snap_cmtime(objset_t
*os
)
921 return (dsl_dir_snap_cmtime(os
->os_dsl_dataset
->ds_dir
));
925 dmu_objset_create_impl_dnstats(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
926 dmu_objset_type_t type
, int levels
, int blksz
, int ibs
, dmu_tx_t
*tx
)
931 ASSERT(dmu_tx_is_syncing(tx
));
934 blksz
= DNODE_BLOCK_SIZE
;
936 ibs
= DN_MAX_INDBLKSHIFT
;
939 VERIFY0(dmu_objset_from_ds(ds
, &os
));
941 VERIFY0(dmu_objset_open_impl(spa
, NULL
, bp
, &os
));
943 mdn
= DMU_META_DNODE(os
);
945 dnode_allocate(mdn
, DMU_OT_DNODE
, blksz
, ibs
, DMU_OT_NONE
, 0,
946 DNODE_MIN_SLOTS
, tx
);
949 * We don't want to have to increase the meta-dnode's nlevels
950 * later, because then we could do it in quescing context while
951 * we are also accessing it in open context.
953 * This precaution is not necessary for the MOS (ds == NULL),
954 * because the MOS is only updated in syncing context.
955 * This is most fortunate: the MOS is the only objset that
956 * needs to be synced multiple times as spa_sync() iterates
957 * to convergence, so minimizing its dn_nlevels matters.
964 * Determine the number of levels necessary for the
965 * meta-dnode to contain DN_MAX_OBJECT dnodes. Note
966 * that in order to ensure that we do not overflow
967 * 64 bits, there has to be a nlevels that gives us a
968 * number of blocks > DN_MAX_OBJECT but < 2^64.
969 * Therefore, (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)
970 * (10) must be less than (64 - log2(DN_MAX_OBJECT))
973 while ((uint64_t)mdn
->dn_nblkptr
<<
974 (mdn
->dn_datablkshift
- DNODE_SHIFT
+ (levels
- 1) *
975 (mdn
->dn_indblkshift
- SPA_BLKPTRSHIFT
)) <
980 mdn
->dn_next_nlevels
[tx
->tx_txg
& TXG_MASK
] =
981 mdn
->dn_nlevels
= levels
;
984 ASSERT(type
!= DMU_OST_NONE
);
985 ASSERT(type
!= DMU_OST_ANY
);
986 ASSERT(type
< DMU_OST_NUMTYPES
);
987 os
->os_phys
->os_type
= type
;
990 * Enable user accounting if it is enabled and this is not an
993 if (dmu_objset_userused_enabled(os
) &&
994 (!os
->os_encrypted
|| !dmu_objset_is_receiving(os
))) {
995 os
->os_phys
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
996 if (dmu_objset_userobjused_enabled(os
)) {
997 ds
->ds_feature_activation_needed
[
998 SPA_FEATURE_USEROBJ_ACCOUNTING
] = B_TRUE
;
999 os
->os_phys
->os_flags
|=
1000 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
;
1002 os
->os_flags
= os
->os_phys
->os_flags
;
1005 dsl_dataset_dirty(ds
, tx
);
1010 /* called from dsl for meta-objset */
1012 dmu_objset_create_impl(spa_t
*spa
, dsl_dataset_t
*ds
, blkptr_t
*bp
,
1013 dmu_objset_type_t type
, dmu_tx_t
*tx
)
1015 return (dmu_objset_create_impl_dnstats(spa
, ds
, bp
, type
, 0, 0, 0, tx
));
1018 typedef struct dmu_objset_create_arg
{
1019 const char *doca_name
;
1021 void (*doca_userfunc
)(objset_t
*os
, void *arg
,
1022 cred_t
*cr
, dmu_tx_t
*tx
);
1024 dmu_objset_type_t doca_type
;
1025 uint64_t doca_flags
;
1026 dsl_crypto_params_t
*doca_dcp
;
1027 } dmu_objset_create_arg_t
;
1031 dmu_objset_create_check(void *arg
, dmu_tx_t
*tx
)
1033 dmu_objset_create_arg_t
*doca
= arg
;
1034 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1039 if (strchr(doca
->doca_name
, '@') != NULL
)
1040 return (SET_ERROR(EINVAL
));
1042 if (strlen(doca
->doca_name
) >= ZFS_MAX_DATASET_NAME_LEN
)
1043 return (SET_ERROR(ENAMETOOLONG
));
1045 error
= dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
);
1049 dsl_dir_rele(pdd
, FTAG
);
1050 return (SET_ERROR(EEXIST
));
1053 error
= dmu_objset_create_crypt_check(pdd
, doca
->doca_dcp
);
1055 dsl_dir_rele(pdd
, FTAG
);
1059 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
1062 dsl_dir_rele(pdd
, FTAG
);
1068 dmu_objset_create_sync(void *arg
, dmu_tx_t
*tx
)
1070 dmu_objset_create_arg_t
*doca
= arg
;
1071 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1080 VERIFY0(dsl_dir_hold(dp
, doca
->doca_name
, FTAG
, &pdd
, &tail
));
1082 obj
= dsl_dataset_create_sync(pdd
, tail
, NULL
, doca
->doca_flags
,
1083 doca
->doca_cred
, doca
->doca_dcp
, tx
);
1085 VERIFY0(dsl_dataset_hold_obj_flags(pdd
->dd_pool
, obj
,
1086 DS_HOLD_FLAG_DECRYPT
, FTAG
, &ds
));
1087 rrw_enter(&ds
->ds_bp_rwlock
, RW_READER
, FTAG
);
1088 bp
= dsl_dataset_get_blkptr(ds
);
1089 os
= dmu_objset_create_impl(pdd
->dd_pool
->dp_spa
,
1090 ds
, bp
, doca
->doca_type
, tx
);
1091 rrw_exit(&ds
->ds_bp_rwlock
, FTAG
);
1093 if (doca
->doca_userfunc
!= NULL
) {
1094 doca
->doca_userfunc(os
, doca
->doca_userarg
,
1095 doca
->doca_cred
, tx
);
1099 * The doca_userfunc() will write out some data that needs to be
1100 * encrypted if the dataset is encrypted (specifically the root
1101 * directory). This data must be written out before the encryption
1102 * key mapping is removed by dsl_dataset_rele_flags(). Force the
1103 * I/O to occur immediately by invoking the relevant sections of
1106 if (os
->os_encrypted
) {
1107 dsl_dataset_t
*tmpds
= NULL
;
1108 boolean_t need_sync_done
= B_FALSE
;
1110 rzio
= zio_root(dp
->dp_spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
1111 tmpds
= txg_list_remove(&dp
->dp_dirty_datasets
, tx
->tx_txg
);
1112 if (tmpds
!= NULL
) {
1113 ASSERT3P(ds
, ==, tmpds
);
1114 dsl_dataset_sync(ds
, rzio
, tx
);
1115 need_sync_done
= B_TRUE
;
1117 VERIFY0(zio_wait(rzio
));
1119 dmu_objset_do_userquota_updates(os
, tx
);
1120 taskq_wait(dp
->dp_sync_taskq
);
1122 rzio
= zio_root(dp
->dp_spa
, NULL
, NULL
, ZIO_FLAG_MUSTSUCCEED
);
1123 tmpds
= txg_list_remove(&dp
->dp_dirty_datasets
, tx
->tx_txg
);
1124 if (tmpds
!= NULL
) {
1125 ASSERT3P(ds
, ==, tmpds
);
1126 dmu_buf_rele(ds
->ds_dbuf
, ds
);
1127 dsl_dataset_sync(ds
, rzio
, tx
);
1129 VERIFY0(zio_wait(rzio
));
1132 dsl_dataset_sync_done(ds
, tx
);
1135 spa_history_log_internal_ds(ds
, "create", tx
, "");
1136 zvol_create_minors(dp
->dp_spa
, doca
->doca_name
, B_TRUE
);
1138 dsl_dataset_rele_flags(ds
, DS_HOLD_FLAG_DECRYPT
, FTAG
);
1139 dsl_dir_rele(pdd
, FTAG
);
1143 dmu_objset_create(const char *name
, dmu_objset_type_t type
, uint64_t flags
,
1144 dsl_crypto_params_t
*dcp
, dmu_objset_create_sync_func_t func
, void *arg
)
1146 dmu_objset_create_arg_t doca
;
1147 dsl_crypto_params_t tmp_dcp
= { 0 };
1149 doca
.doca_name
= name
;
1150 doca
.doca_cred
= CRED();
1151 doca
.doca_flags
= flags
;
1152 doca
.doca_userfunc
= func
;
1153 doca
.doca_userarg
= arg
;
1154 doca
.doca_type
= type
;
1157 * Some callers (mostly for testing) do not provide a dcp on their
1158 * own but various code inside the sync task will require it to be
1159 * allocated. Rather than adding NULL checks throughout this code
1160 * or adding dummy dcp's to all of the callers we simply create a
1161 * dummy one here and use that. This zero dcp will have the same
1162 * effect as asking for inheritence of all encryption params.
1164 doca
.doca_dcp
= (dcp
!= NULL
) ? dcp
: &tmp_dcp
;
1166 return (dsl_sync_task(name
,
1167 dmu_objset_create_check
, dmu_objset_create_sync
, &doca
,
1168 6, ZFS_SPACE_CHECK_NORMAL
));
1171 typedef struct dmu_objset_clone_arg
{
1172 const char *doca_clone
;
1173 const char *doca_origin
;
1175 } dmu_objset_clone_arg_t
;
1179 dmu_objset_clone_check(void *arg
, dmu_tx_t
*tx
)
1181 dmu_objset_clone_arg_t
*doca
= arg
;
1185 dsl_dataset_t
*origin
;
1186 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1188 if (strchr(doca
->doca_clone
, '@') != NULL
)
1189 return (SET_ERROR(EINVAL
));
1191 if (strlen(doca
->doca_clone
) >= ZFS_MAX_DATASET_NAME_LEN
)
1192 return (SET_ERROR(ENAMETOOLONG
));
1194 error
= dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
);
1198 dsl_dir_rele(pdd
, FTAG
);
1199 return (SET_ERROR(EEXIST
));
1202 error
= dsl_fs_ss_limit_check(pdd
, 1, ZFS_PROP_FILESYSTEM_LIMIT
, NULL
,
1205 dsl_dir_rele(pdd
, FTAG
);
1206 return (SET_ERROR(EDQUOT
));
1209 error
= dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
);
1211 dsl_dir_rele(pdd
, FTAG
);
1215 /* You can only clone snapshots, not the head datasets. */
1216 if (!origin
->ds_is_snapshot
) {
1217 dsl_dataset_rele(origin
, FTAG
);
1218 dsl_dir_rele(pdd
, FTAG
);
1219 return (SET_ERROR(EINVAL
));
1222 error
= dmu_objset_clone_crypt_check(pdd
, origin
->ds_dir
);
1224 dsl_dataset_rele(origin
, FTAG
);
1225 dsl_dir_rele(pdd
, FTAG
);
1229 dsl_dataset_rele(origin
, FTAG
);
1230 dsl_dir_rele(pdd
, FTAG
);
1236 dmu_objset_clone_sync(void *arg
, dmu_tx_t
*tx
)
1238 dmu_objset_clone_arg_t
*doca
= arg
;
1239 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1242 dsl_dataset_t
*origin
, *ds
;
1244 char namebuf
[ZFS_MAX_DATASET_NAME_LEN
];
1246 VERIFY0(dsl_dir_hold(dp
, doca
->doca_clone
, FTAG
, &pdd
, &tail
));
1247 VERIFY0(dsl_dataset_hold(dp
, doca
->doca_origin
, FTAG
, &origin
));
1249 obj
= dsl_dataset_create_sync(pdd
, tail
, origin
, 0,
1250 doca
->doca_cred
, NULL
, tx
);
1252 VERIFY0(dsl_dataset_hold_obj(pdd
->dd_pool
, obj
, FTAG
, &ds
));
1253 dsl_dataset_name(origin
, namebuf
);
1254 spa_history_log_internal_ds(ds
, "clone", tx
,
1255 "origin=%s (%llu)", namebuf
, origin
->ds_object
);
1256 zvol_create_minors(dp
->dp_spa
, doca
->doca_clone
, B_TRUE
);
1257 dsl_dataset_rele(ds
, FTAG
);
1258 dsl_dataset_rele(origin
, FTAG
);
1259 dsl_dir_rele(pdd
, FTAG
);
1263 dmu_objset_clone(const char *clone
, const char *origin
)
1265 dmu_objset_clone_arg_t doca
;
1267 doca
.doca_clone
= clone
;
1268 doca
.doca_origin
= origin
;
1269 doca
.doca_cred
= CRED();
1271 return (dsl_sync_task(clone
,
1272 dmu_objset_clone_check
, dmu_objset_clone_sync
, &doca
,
1273 6, ZFS_SPACE_CHECK_NORMAL
));
1277 dmu_objset_snapshot_one(const char *fsname
, const char *snapname
)
1280 char *longsnap
= kmem_asprintf("%s@%s", fsname
, snapname
);
1281 nvlist_t
*snaps
= fnvlist_alloc();
1283 fnvlist_add_boolean(snaps
, longsnap
);
1285 err
= dsl_dataset_snapshot(snaps
, NULL
, NULL
);
1286 fnvlist_free(snaps
);
1291 dmu_objset_upgrade_task_cb(void *data
)
1293 objset_t
*os
= data
;
1295 mutex_enter(&os
->os_upgrade_lock
);
1296 os
->os_upgrade_status
= EINTR
;
1297 if (!os
->os_upgrade_exit
) {
1298 mutex_exit(&os
->os_upgrade_lock
);
1300 os
->os_upgrade_status
= os
->os_upgrade_cb(os
);
1301 mutex_enter(&os
->os_upgrade_lock
);
1303 os
->os_upgrade_exit
= B_TRUE
;
1304 os
->os_upgrade_id
= 0;
1305 mutex_exit(&os
->os_upgrade_lock
);
1309 dmu_objset_upgrade(objset_t
*os
, dmu_objset_upgrade_cb_t cb
)
1311 if (os
->os_upgrade_id
!= 0)
1314 mutex_enter(&os
->os_upgrade_lock
);
1315 if (os
->os_upgrade_id
== 0 && os
->os_upgrade_status
== 0) {
1316 os
->os_upgrade_exit
= B_FALSE
;
1317 os
->os_upgrade_cb
= cb
;
1318 os
->os_upgrade_id
= taskq_dispatch(
1319 os
->os_spa
->spa_upgrade_taskq
,
1320 dmu_objset_upgrade_task_cb
, os
, TQ_SLEEP
);
1321 if (os
->os_upgrade_id
== TASKQID_INVALID
)
1322 os
->os_upgrade_status
= ENOMEM
;
1324 mutex_exit(&os
->os_upgrade_lock
);
1328 dmu_objset_upgrade_stop(objset_t
*os
)
1330 mutex_enter(&os
->os_upgrade_lock
);
1331 os
->os_upgrade_exit
= B_TRUE
;
1332 if (os
->os_upgrade_id
!= 0) {
1333 taskqid_t id
= os
->os_upgrade_id
;
1335 os
->os_upgrade_id
= 0;
1336 mutex_exit(&os
->os_upgrade_lock
);
1338 taskq_cancel_id(os
->os_spa
->spa_upgrade_taskq
, id
);
1340 mutex_exit(&os
->os_upgrade_lock
);
1345 dmu_objset_sync_dnodes(multilist_sublist_t
*list
, dmu_tx_t
*tx
)
1349 while ((dn
= multilist_sublist_head(list
)) != NULL
) {
1350 ASSERT(dn
->dn_object
!= DMU_META_DNODE_OBJECT
);
1351 ASSERT(dn
->dn_dbuf
->db_data_pending
);
1353 * Initialize dn_zio outside dnode_sync() because the
1354 * meta-dnode needs to set it ouside dnode_sync().
1356 dn
->dn_zio
= dn
->dn_dbuf
->db_data_pending
->dr_zio
;
1359 ASSERT3U(dn
->dn_nlevels
, <=, DN_MAX_LEVELS
);
1360 multilist_sublist_remove(list
, dn
);
1362 multilist_t
*newlist
= dn
->dn_objset
->os_synced_dnodes
;
1363 if (newlist
!= NULL
) {
1364 (void) dnode_add_ref(dn
, newlist
);
1365 multilist_insert(newlist
, dn
);
1374 dmu_objset_write_ready(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1378 blkptr_t
*bp
= zio
->io_bp
;
1380 dnode_phys_t
*dnp
= &os
->os_phys
->os_meta_dnode
;
1383 ASSERT(!BP_IS_EMBEDDED(bp
));
1384 ASSERT3U(BP_GET_TYPE(bp
), ==, DMU_OT_OBJSET
);
1385 ASSERT0(BP_GET_LEVEL(bp
));
1388 * Update rootbp fill count: it should be the number of objects
1389 * allocated in the object set (not counting the "special"
1390 * objects that are stored in the objset_phys_t -- the meta
1391 * dnode and user/group accounting objects).
1393 for (i
= 0; i
< dnp
->dn_nblkptr
; i
++)
1394 fill
+= BP_GET_FILL(&dnp
->dn_blkptr
[i
]);
1396 BP_SET_FILL(bp
, fill
);
1398 if (os
->os_dsl_dataset
!= NULL
)
1399 rrw_enter(&os
->os_dsl_dataset
->ds_bp_rwlock
, RW_WRITER
, FTAG
);
1400 *os
->os_rootbp
= *bp
;
1401 if (os
->os_dsl_dataset
!= NULL
)
1402 rrw_exit(&os
->os_dsl_dataset
->ds_bp_rwlock
, FTAG
);
1407 dmu_objset_write_done(zio_t
*zio
, arc_buf_t
*abuf
, void *arg
)
1409 blkptr_t
*bp
= zio
->io_bp
;
1410 blkptr_t
*bp_orig
= &zio
->io_bp_orig
;
1413 if (zio
->io_flags
& ZIO_FLAG_IO_REWRITE
) {
1414 ASSERT(BP_EQUAL(bp
, bp_orig
));
1416 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
1417 dmu_tx_t
*tx
= os
->os_synctx
;
1419 (void) dsl_dataset_block_kill(ds
, bp_orig
, tx
, B_TRUE
);
1420 dsl_dataset_block_born(ds
, bp
, tx
);
1422 kmem_free(bp
, sizeof (*bp
));
1425 typedef struct sync_dnodes_arg
{
1426 multilist_t
*sda_list
;
1427 int sda_sublist_idx
;
1428 multilist_t
*sda_newlist
;
1430 } sync_dnodes_arg_t
;
1433 sync_dnodes_task(void *arg
)
1435 sync_dnodes_arg_t
*sda
= arg
;
1437 multilist_sublist_t
*ms
=
1438 multilist_sublist_lock(sda
->sda_list
, sda
->sda_sublist_idx
);
1440 dmu_objset_sync_dnodes(ms
, sda
->sda_tx
);
1442 multilist_sublist_unlock(ms
);
1444 kmem_free(sda
, sizeof (*sda
));
1448 /* called from dsl */
1450 dmu_objset_sync(objset_t
*os
, zio_t
*pio
, dmu_tx_t
*tx
)
1453 zbookmark_phys_t zb
;
1457 dbuf_dirty_record_t
*dr
;
1458 blkptr_t
*blkptr_copy
= kmem_alloc(sizeof (*os
->os_rootbp
), KM_SLEEP
);
1459 *blkptr_copy
= *os
->os_rootbp
;
1461 dprintf_ds(os
->os_dsl_dataset
, "txg=%llu\n", tx
->tx_txg
);
1463 ASSERT(dmu_tx_is_syncing(tx
));
1464 /* XXX the write_done callback should really give us the tx... */
1467 if (os
->os_dsl_dataset
== NULL
) {
1469 * This is the MOS. If we have upgraded,
1470 * spa_max_replication() could change, so reset
1473 os
->os_copies
= spa_max_replication(os
->os_spa
);
1477 * Create the root block IO
1479 SET_BOOKMARK(&zb
, os
->os_dsl_dataset
?
1480 os
->os_dsl_dataset
->ds_object
: DMU_META_OBJSET
,
1481 ZB_ROOT_OBJECT
, ZB_ROOT_LEVEL
, ZB_ROOT_BLKID
);
1482 arc_release(os
->os_phys_buf
, &os
->os_phys_buf
);
1484 dmu_write_policy(os
, NULL
, 0, 0, &zp
);
1487 * If we are either claiming the ZIL or doing a raw receive write out
1488 * the os_phys_buf raw. Neither of these actions will effect the MAC
1491 if (arc_is_unauthenticated(os
->os_phys_buf
) || os
->os_next_write_raw
) {
1492 ASSERT(os
->os_encrypted
);
1493 os
->os_next_write_raw
= B_FALSE
;
1494 arc_convert_to_raw(os
->os_phys_buf
,
1495 os
->os_dsl_dataset
->ds_object
, ZFS_HOST_BYTEORDER
,
1496 DMU_OT_OBJSET
, NULL
, NULL
, NULL
);
1499 zio
= arc_write(pio
, os
->os_spa
, tx
->tx_txg
,
1500 blkptr_copy
, os
->os_phys_buf
, DMU_OS_IS_L2CACHEABLE(os
),
1501 &zp
, dmu_objset_write_ready
, NULL
, NULL
, dmu_objset_write_done
,
1502 os
, ZIO_PRIORITY_ASYNC_WRITE
, ZIO_FLAG_MUSTSUCCEED
, &zb
);
1505 * Sync special dnodes - the parent IO for the sync is the root block
1507 DMU_META_DNODE(os
)->dn_zio
= zio
;
1508 dnode_sync(DMU_META_DNODE(os
), tx
);
1510 os
->os_phys
->os_flags
= os
->os_flags
;
1512 if (DMU_USERUSED_DNODE(os
) &&
1513 DMU_USERUSED_DNODE(os
)->dn_type
!= DMU_OT_NONE
) {
1514 DMU_USERUSED_DNODE(os
)->dn_zio
= zio
;
1515 dnode_sync(DMU_USERUSED_DNODE(os
), tx
);
1516 DMU_GROUPUSED_DNODE(os
)->dn_zio
= zio
;
1517 dnode_sync(DMU_GROUPUSED_DNODE(os
), tx
);
1520 txgoff
= tx
->tx_txg
& TXG_MASK
;
1522 if (dmu_objset_userused_enabled(os
) &&
1523 (!os
->os_encrypted
|| !dmu_objset_is_receiving(os
))) {
1525 * We must create the list here because it uses the
1526 * dn_dirty_link[] of this txg. But it may already
1527 * exist because we call dsl_dataset_sync() twice per txg.
1529 if (os
->os_synced_dnodes
== NULL
) {
1530 os
->os_synced_dnodes
=
1531 multilist_create(sizeof (dnode_t
),
1532 offsetof(dnode_t
, dn_dirty_link
[txgoff
]),
1533 dnode_multilist_index_func
);
1535 ASSERT3U(os
->os_synced_dnodes
->ml_offset
, ==,
1536 offsetof(dnode_t
, dn_dirty_link
[txgoff
]));
1541 i
< multilist_get_num_sublists(os
->os_dirty_dnodes
[txgoff
]); i
++) {
1542 sync_dnodes_arg_t
*sda
= kmem_alloc(sizeof (*sda
), KM_SLEEP
);
1543 sda
->sda_list
= os
->os_dirty_dnodes
[txgoff
];
1544 sda
->sda_sublist_idx
= i
;
1546 (void) taskq_dispatch(dmu_objset_pool(os
)->dp_sync_taskq
,
1547 sync_dnodes_task
, sda
, 0);
1548 /* callback frees sda */
1550 taskq_wait(dmu_objset_pool(os
)->dp_sync_taskq
);
1552 list
= &DMU_META_DNODE(os
)->dn_dirty_records
[txgoff
];
1553 while ((dr
= list_head(list
)) != NULL
) {
1554 ASSERT0(dr
->dr_dbuf
->db_level
);
1555 list_remove(list
, dr
);
1557 zio_nowait(dr
->dr_zio
);
1560 /* Enable dnode backfill if enough objects have been freed. */
1561 if (os
->os_freed_dnodes
>= dmu_rescan_dnode_threshold
) {
1562 os
->os_rescan_dnodes
= B_TRUE
;
1563 os
->os_freed_dnodes
= 0;
1567 * Free intent log blocks up to this tx.
1569 zil_sync(os
->os_zil
, tx
);
1570 os
->os_phys
->os_zil_header
= os
->os_zil_header
;
1575 dmu_objset_is_dirty(objset_t
*os
, uint64_t txg
)
1577 return (!multilist_is_empty(os
->os_dirty_dnodes
[txg
& TXG_MASK
]));
1580 static objset_used_cb_t
*used_cbs
[DMU_OST_NUMTYPES
];
1583 dmu_objset_register_type(dmu_objset_type_t ost
, objset_used_cb_t
*cb
)
1589 dmu_objset_userused_enabled(objset_t
*os
)
1591 return (spa_version(os
->os_spa
) >= SPA_VERSION_USERSPACE
&&
1592 used_cbs
[os
->os_phys
->os_type
] != NULL
&&
1593 DMU_USERUSED_DNODE(os
) != NULL
);
1597 dmu_objset_userobjused_enabled(objset_t
*os
)
1599 return (dmu_objset_userused_enabled(os
) &&
1600 spa_feature_is_enabled(os
->os_spa
, SPA_FEATURE_USEROBJ_ACCOUNTING
));
1603 typedef struct userquota_node
{
1604 /* must be in the first filed, see userquota_update_cache() */
1605 char uqn_id
[20 + DMU_OBJACCT_PREFIX_LEN
];
1607 avl_node_t uqn_node
;
1610 typedef struct userquota_cache
{
1611 avl_tree_t uqc_user_deltas
;
1612 avl_tree_t uqc_group_deltas
;
1613 } userquota_cache_t
;
1616 userquota_compare(const void *l
, const void *r
)
1618 const userquota_node_t
*luqn
= l
;
1619 const userquota_node_t
*ruqn
= r
;
1623 * NB: can only access uqn_id because userquota_update_cache() doesn't
1624 * pass in an entire userquota_node_t.
1626 rv
= strcmp(luqn
->uqn_id
, ruqn
->uqn_id
);
1628 return (AVL_ISIGN(rv
));
1632 do_userquota_cacheflush(objset_t
*os
, userquota_cache_t
*cache
, dmu_tx_t
*tx
)
1635 userquota_node_t
*uqn
;
1637 ASSERT(dmu_tx_is_syncing(tx
));
1640 while ((uqn
= avl_destroy_nodes(&cache
->uqc_user_deltas
,
1641 &cookie
)) != NULL
) {
1643 * os_userused_lock protects against concurrent calls to
1644 * zap_increment_int(). It's needed because zap_increment_int()
1645 * is not thread-safe (i.e. not atomic).
1647 mutex_enter(&os
->os_userused_lock
);
1648 VERIFY0(zap_increment(os
, DMU_USERUSED_OBJECT
,
1649 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1650 mutex_exit(&os
->os_userused_lock
);
1651 kmem_free(uqn
, sizeof (*uqn
));
1653 avl_destroy(&cache
->uqc_user_deltas
);
1656 while ((uqn
= avl_destroy_nodes(&cache
->uqc_group_deltas
,
1657 &cookie
)) != NULL
) {
1658 mutex_enter(&os
->os_userused_lock
);
1659 VERIFY0(zap_increment(os
, DMU_GROUPUSED_OBJECT
,
1660 uqn
->uqn_id
, uqn
->uqn_delta
, tx
));
1661 mutex_exit(&os
->os_userused_lock
);
1662 kmem_free(uqn
, sizeof (*uqn
));
1664 avl_destroy(&cache
->uqc_group_deltas
);
1668 userquota_update_cache(avl_tree_t
*avl
, const char *id
, int64_t delta
)
1670 userquota_node_t
*uqn
;
1673 ASSERT(strlen(id
) < sizeof (uqn
->uqn_id
));
1675 * Use id directly for searching because uqn_id is the first field of
1676 * userquota_node_t and fields after uqn_id won't be accessed in
1679 uqn
= avl_find(avl
, (const void *)id
, &idx
);
1681 uqn
= kmem_zalloc(sizeof (*uqn
), KM_SLEEP
);
1682 strlcpy(uqn
->uqn_id
, id
, sizeof (uqn
->uqn_id
));
1683 avl_insert(avl
, uqn
, idx
);
1685 uqn
->uqn_delta
+= delta
;
1689 do_userquota_update(userquota_cache_t
*cache
, uint64_t used
, uint64_t flags
,
1690 uint64_t user
, uint64_t group
, boolean_t subtract
)
1692 if ((flags
& DNODE_FLAG_USERUSED_ACCOUNTED
)) {
1693 int64_t delta
= DNODE_MIN_SIZE
+ used
;
1699 (void) sprintf(name
, "%llx", (longlong_t
)user
);
1700 userquota_update_cache(&cache
->uqc_user_deltas
, name
, delta
);
1702 (void) sprintf(name
, "%llx", (longlong_t
)group
);
1703 userquota_update_cache(&cache
->uqc_group_deltas
, name
, delta
);
1708 do_userobjquota_update(userquota_cache_t
*cache
, uint64_t flags
,
1709 uint64_t user
, uint64_t group
, boolean_t subtract
)
1711 if (flags
& DNODE_FLAG_USEROBJUSED_ACCOUNTED
) {
1712 char name
[20 + DMU_OBJACCT_PREFIX_LEN
];
1713 int delta
= subtract
? -1 : 1;
1715 (void) snprintf(name
, sizeof (name
), DMU_OBJACCT_PREFIX
"%llx",
1717 userquota_update_cache(&cache
->uqc_user_deltas
, name
, delta
);
1719 (void) snprintf(name
, sizeof (name
), DMU_OBJACCT_PREFIX
"%llx",
1721 userquota_update_cache(&cache
->uqc_group_deltas
, name
, delta
);
1725 typedef struct userquota_updates_arg
{
1727 int uua_sublist_idx
;
1729 } userquota_updates_arg_t
;
1732 userquota_updates_task(void *arg
)
1734 userquota_updates_arg_t
*uua
= arg
;
1735 objset_t
*os
= uua
->uua_os
;
1736 dmu_tx_t
*tx
= uua
->uua_tx
;
1738 userquota_cache_t cache
= { { 0 } };
1740 multilist_sublist_t
*list
=
1741 multilist_sublist_lock(os
->os_synced_dnodes
, uua
->uua_sublist_idx
);
1743 ASSERT(multilist_sublist_head(list
) == NULL
||
1744 dmu_objset_userused_enabled(os
));
1745 avl_create(&cache
.uqc_user_deltas
, userquota_compare
,
1746 sizeof (userquota_node_t
), offsetof(userquota_node_t
, uqn_node
));
1747 avl_create(&cache
.uqc_group_deltas
, userquota_compare
,
1748 sizeof (userquota_node_t
), offsetof(userquota_node_t
, uqn_node
));
1750 while ((dn
= multilist_sublist_head(list
)) != NULL
) {
1752 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn
->dn_object
));
1753 ASSERT(dn
->dn_phys
->dn_type
== DMU_OT_NONE
||
1754 dn
->dn_phys
->dn_flags
&
1755 DNODE_FLAG_USERUSED_ACCOUNTED
);
1757 flags
= dn
->dn_id_flags
;
1759 if (flags
& DN_ID_OLD_EXIST
) {
1760 do_userquota_update(&cache
,
1761 dn
->dn_oldused
, dn
->dn_oldflags
,
1762 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
);
1763 do_userobjquota_update(&cache
, dn
->dn_oldflags
,
1764 dn
->dn_olduid
, dn
->dn_oldgid
, B_TRUE
);
1766 if (flags
& DN_ID_NEW_EXIST
) {
1767 do_userquota_update(&cache
,
1768 DN_USED_BYTES(dn
->dn_phys
), dn
->dn_phys
->dn_flags
,
1769 dn
->dn_newuid
, dn
->dn_newgid
, B_FALSE
);
1770 do_userobjquota_update(&cache
, dn
->dn_phys
->dn_flags
,
1771 dn
->dn_newuid
, dn
->dn_newgid
, B_FALSE
);
1774 mutex_enter(&dn
->dn_mtx
);
1776 dn
->dn_oldflags
= 0;
1777 if (dn
->dn_id_flags
& DN_ID_NEW_EXIST
) {
1778 dn
->dn_olduid
= dn
->dn_newuid
;
1779 dn
->dn_oldgid
= dn
->dn_newgid
;
1780 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1781 if (dn
->dn_bonuslen
== 0)
1782 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1784 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1786 dn
->dn_id_flags
&= ~(DN_ID_NEW_EXIST
);
1787 mutex_exit(&dn
->dn_mtx
);
1789 multilist_sublist_remove(list
, dn
);
1790 dnode_rele(dn
, os
->os_synced_dnodes
);
1792 do_userquota_cacheflush(os
, &cache
, tx
);
1793 multilist_sublist_unlock(list
);
1794 kmem_free(uua
, sizeof (*uua
));
1798 dmu_objset_do_userquota_updates(objset_t
*os
, dmu_tx_t
*tx
)
1800 if (!dmu_objset_userused_enabled(os
))
1803 /* if this is a raw receive just return and handle accounting later */
1804 if (os
->os_encrypted
&& dmu_objset_is_receiving(os
))
1807 /* Allocate the user/groupused objects if necessary. */
1808 if (DMU_USERUSED_DNODE(os
)->dn_type
== DMU_OT_NONE
) {
1809 VERIFY0(zap_create_claim(os
,
1810 DMU_USERUSED_OBJECT
,
1811 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1812 VERIFY0(zap_create_claim(os
,
1813 DMU_GROUPUSED_OBJECT
,
1814 DMU_OT_USERGROUP_USED
, DMU_OT_NONE
, 0, tx
));
1818 i
< multilist_get_num_sublists(os
->os_synced_dnodes
); i
++) {
1819 userquota_updates_arg_t
*uua
=
1820 kmem_alloc(sizeof (*uua
), KM_SLEEP
);
1822 uua
->uua_sublist_idx
= i
;
1824 /* note: caller does taskq_wait() */
1825 (void) taskq_dispatch(dmu_objset_pool(os
)->dp_sync_taskq
,
1826 userquota_updates_task
, uua
, 0);
1827 /* callback frees uua */
1832 * Returns a pointer to data to find uid/gid from
1834 * If a dirty record for transaction group that is syncing can't
1835 * be found then NULL is returned. In the NULL case it is assumed
1836 * the uid/gid aren't changing.
1839 dmu_objset_userquota_find_data(dmu_buf_impl_t
*db
, dmu_tx_t
*tx
)
1841 dbuf_dirty_record_t
*dr
, **drp
;
1844 if (db
->db_dirtycnt
== 0)
1845 return (db
->db
.db_data
); /* Nothing is changing */
1847 for (drp
= &db
->db_last_dirty
; (dr
= *drp
) != NULL
; drp
= &dr
->dr_next
)
1848 if (dr
->dr_txg
== tx
->tx_txg
)
1856 DB_DNODE_ENTER(dr
->dr_dbuf
);
1857 dn
= DB_DNODE(dr
->dr_dbuf
);
1859 if (dn
->dn_bonuslen
== 0 &&
1860 dr
->dr_dbuf
->db_blkid
== DMU_SPILL_BLKID
)
1861 data
= dr
->dt
.dl
.dr_data
->b_data
;
1863 data
= dr
->dt
.dl
.dr_data
;
1865 DB_DNODE_EXIT(dr
->dr_dbuf
);
1872 dmu_objset_userquota_get_ids(dnode_t
*dn
, boolean_t before
, dmu_tx_t
*tx
)
1874 objset_t
*os
= dn
->dn_objset
;
1876 dmu_buf_impl_t
*db
= NULL
;
1877 uint64_t *user
= NULL
;
1878 uint64_t *group
= NULL
;
1879 int flags
= dn
->dn_id_flags
;
1881 boolean_t have_spill
= B_FALSE
;
1883 if (!dmu_objset_userused_enabled(dn
->dn_objset
))
1887 * Raw receives introduce a problem with user accounting. Raw
1888 * receives cannot update the user accounting info because the
1889 * user ids and the sizes are encrypted. To guarantee that we
1890 * never end up with bad user accounting, we simply disable it
1891 * during raw receives. We also disable this for normal receives
1892 * so that an incremental raw receive may be done on top of an
1893 * existing non-raw receive.
1895 if (os
->os_encrypted
&& dmu_objset_is_receiving(os
))
1898 if (before
&& (flags
& (DN_ID_CHKED_BONUS
|DN_ID_OLD_EXIST
|
1899 DN_ID_CHKED_SPILL
)))
1902 if (before
&& dn
->dn_bonuslen
!= 0)
1903 data
= DN_BONUS(dn
->dn_phys
);
1904 else if (!before
&& dn
->dn_bonuslen
!= 0) {
1907 mutex_enter(&db
->db_mtx
);
1908 data
= dmu_objset_userquota_find_data(db
, tx
);
1910 data
= DN_BONUS(dn
->dn_phys
);
1912 } else if (dn
->dn_bonuslen
== 0 && dn
->dn_bonustype
== DMU_OT_SA
) {
1915 if (RW_WRITE_HELD(&dn
->dn_struct_rwlock
))
1916 rf
|= DB_RF_HAVESTRUCT
;
1917 error
= dmu_spill_hold_by_dnode(dn
,
1918 rf
| DB_RF_MUST_SUCCEED
,
1919 FTAG
, (dmu_buf_t
**)&db
);
1921 mutex_enter(&db
->db_mtx
);
1922 data
= (before
) ? db
->db
.db_data
:
1923 dmu_objset_userquota_find_data(db
, tx
);
1924 have_spill
= B_TRUE
;
1926 mutex_enter(&dn
->dn_mtx
);
1927 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1928 mutex_exit(&dn
->dn_mtx
);
1934 user
= &dn
->dn_olduid
;
1935 group
= &dn
->dn_oldgid
;
1937 user
= &dn
->dn_newuid
;
1938 group
= &dn
->dn_newgid
;
1942 * Must always call the callback in case the object
1943 * type has changed and that type isn't an object type to track
1945 error
= used_cbs
[os
->os_phys
->os_type
](dn
->dn_bonustype
, data
,
1949 * Preserve existing uid/gid when the callback can't determine
1950 * what the new uid/gid are and the callback returned EEXIST.
1951 * The EEXIST error tells us to just use the existing uid/gid.
1952 * If we don't know what the old values are then just assign
1953 * them to 0, since that is a new file being created.
1955 if (!before
&& data
== NULL
&& error
== EEXIST
) {
1956 if (flags
& DN_ID_OLD_EXIST
) {
1957 dn
->dn_newuid
= dn
->dn_olduid
;
1958 dn
->dn_newgid
= dn
->dn_oldgid
;
1967 mutex_exit(&db
->db_mtx
);
1969 mutex_enter(&dn
->dn_mtx
);
1970 if (error
== 0 && before
)
1971 dn
->dn_id_flags
|= DN_ID_OLD_EXIST
;
1972 if (error
== 0 && !before
)
1973 dn
->dn_id_flags
|= DN_ID_NEW_EXIST
;
1976 dn
->dn_id_flags
|= DN_ID_CHKED_SPILL
;
1978 dn
->dn_id_flags
|= DN_ID_CHKED_BONUS
;
1980 mutex_exit(&dn
->dn_mtx
);
1982 dmu_buf_rele((dmu_buf_t
*)db
, FTAG
);
1986 dmu_objset_userspace_present(objset_t
*os
)
1988 return (os
->os_phys
->os_flags
&
1989 OBJSET_FLAG_USERACCOUNTING_COMPLETE
);
1993 dmu_objset_userobjspace_present(objset_t
*os
)
1995 return (os
->os_phys
->os_flags
&
1996 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
);
2000 dmu_objset_space_upgrade(objset_t
*os
)
2006 * We simply need to mark every object dirty, so that it will be
2007 * synced out and now accounted. If this is called
2008 * concurrently, or if we already did some work before crashing,
2009 * that's fine, since we track each object's accounted state
2013 for (obj
= 0; err
== 0; err
= dmu_object_next(os
, &obj
, FALSE
, 0)) {
2018 mutex_enter(&os
->os_upgrade_lock
);
2019 if (os
->os_upgrade_exit
)
2020 err
= SET_ERROR(EINTR
);
2021 mutex_exit(&os
->os_upgrade_lock
);
2025 if (issig(JUSTLOOKING
) && issig(FORREAL
))
2026 return (SET_ERROR(EINTR
));
2028 objerr
= dmu_bonus_hold(os
, obj
, FTAG
, &db
);
2031 tx
= dmu_tx_create(os
);
2032 dmu_tx_hold_bonus(tx
, obj
);
2033 objerr
= dmu_tx_assign(tx
, TXG_WAIT
);
2038 dmu_buf_will_dirty(db
, tx
);
2039 dmu_buf_rele(db
, FTAG
);
2046 dmu_objset_userspace_upgrade(objset_t
*os
)
2050 if (dmu_objset_userspace_present(os
))
2052 if (dmu_objset_is_snapshot(os
))
2053 return (SET_ERROR(EINVAL
));
2054 if (!dmu_objset_userused_enabled(os
))
2055 return (SET_ERROR(ENOTSUP
));
2057 err
= dmu_objset_space_upgrade(os
);
2061 os
->os_flags
|= OBJSET_FLAG_USERACCOUNTING_COMPLETE
;
2062 txg_wait_synced(dmu_objset_pool(os
), 0);
2067 dmu_objset_userobjspace_upgrade_cb(objset_t
*os
)
2071 if (dmu_objset_userobjspace_present(os
))
2073 if (dmu_objset_is_snapshot(os
))
2074 return (SET_ERROR(EINVAL
));
2075 if (!dmu_objset_userobjused_enabled(os
))
2076 return (SET_ERROR(ENOTSUP
));
2078 dmu_objset_ds(os
)->ds_feature_activation_needed
[
2079 SPA_FEATURE_USEROBJ_ACCOUNTING
] = B_TRUE
;
2081 err
= dmu_objset_space_upgrade(os
);
2085 os
->os_flags
|= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE
;
2086 txg_wait_synced(dmu_objset_pool(os
), 0);
2091 dmu_objset_userobjspace_upgrade(objset_t
*os
)
2093 dmu_objset_upgrade(os
, dmu_objset_userobjspace_upgrade_cb
);
2097 dmu_objset_userobjspace_upgradable(objset_t
*os
)
2099 return (dmu_objset_type(os
) == DMU_OST_ZFS
&&
2100 !dmu_objset_is_snapshot(os
) &&
2101 dmu_objset_userobjused_enabled(os
) &&
2102 !dmu_objset_userobjspace_present(os
));
2106 dmu_objset_space(objset_t
*os
, uint64_t *refdbytesp
, uint64_t *availbytesp
,
2107 uint64_t *usedobjsp
, uint64_t *availobjsp
)
2109 dsl_dataset_space(os
->os_dsl_dataset
, refdbytesp
, availbytesp
,
2110 usedobjsp
, availobjsp
);
2114 dmu_objset_fsid_guid(objset_t
*os
)
2116 return (dsl_dataset_fsid_guid(os
->os_dsl_dataset
));
2120 dmu_objset_fast_stat(objset_t
*os
, dmu_objset_stats_t
*stat
)
2122 stat
->dds_type
= os
->os_phys
->os_type
;
2123 if (os
->os_dsl_dataset
)
2124 dsl_dataset_fast_stat(os
->os_dsl_dataset
, stat
);
2128 dmu_objset_stats(objset_t
*os
, nvlist_t
*nv
)
2130 ASSERT(os
->os_dsl_dataset
||
2131 os
->os_phys
->os_type
== DMU_OST_META
);
2133 if (os
->os_dsl_dataset
!= NULL
)
2134 dsl_dataset_stats(os
->os_dsl_dataset
, nv
);
2136 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_TYPE
,
2137 os
->os_phys
->os_type
);
2138 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_USERACCOUNTING
,
2139 dmu_objset_userspace_present(os
));
2143 dmu_objset_is_snapshot(objset_t
*os
)
2145 if (os
->os_dsl_dataset
!= NULL
)
2146 return (os
->os_dsl_dataset
->ds_is_snapshot
);
2152 dmu_snapshot_realname(objset_t
*os
, char *name
, char *real
, int maxlen
,
2153 boolean_t
*conflict
)
2155 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
2158 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
2159 return (SET_ERROR(ENOENT
));
2161 return (zap_lookup_norm(ds
->ds_dir
->dd_pool
->dp_meta_objset
,
2162 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, name
, 8, 1, &ignored
,
2163 MT_NORMALIZE
, real
, maxlen
, conflict
));
2167 dmu_snapshot_list_next(objset_t
*os
, int namelen
, char *name
,
2168 uint64_t *idp
, uint64_t *offp
, boolean_t
*case_conflict
)
2170 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
2171 zap_cursor_t cursor
;
2172 zap_attribute_t attr
;
2174 ASSERT(dsl_pool_config_held(dmu_objset_pool(os
)));
2176 if (dsl_dataset_phys(ds
)->ds_snapnames_zapobj
== 0)
2177 return (SET_ERROR(ENOENT
));
2179 zap_cursor_init_serialized(&cursor
,
2180 ds
->ds_dir
->dd_pool
->dp_meta_objset
,
2181 dsl_dataset_phys(ds
)->ds_snapnames_zapobj
, *offp
);
2183 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
2184 zap_cursor_fini(&cursor
);
2185 return (SET_ERROR(ENOENT
));
2188 if (strlen(attr
.za_name
) + 1 > namelen
) {
2189 zap_cursor_fini(&cursor
);
2190 return (SET_ERROR(ENAMETOOLONG
));
2193 (void) strcpy(name
, attr
.za_name
);
2195 *idp
= attr
.za_first_integer
;
2197 *case_conflict
= attr
.za_normalization_conflict
;
2198 zap_cursor_advance(&cursor
);
2199 *offp
= zap_cursor_serialize(&cursor
);
2200 zap_cursor_fini(&cursor
);
2206 dmu_snapshot_lookup(objset_t
*os
, const char *name
, uint64_t *value
)
2208 return (dsl_dataset_snap_lookup(os
->os_dsl_dataset
, name
, value
));
2212 dmu_dir_list_next(objset_t
*os
, int namelen
, char *name
,
2213 uint64_t *idp
, uint64_t *offp
)
2215 dsl_dir_t
*dd
= os
->os_dsl_dataset
->ds_dir
;
2216 zap_cursor_t cursor
;
2217 zap_attribute_t attr
;
2219 /* there is no next dir on a snapshot! */
2220 if (os
->os_dsl_dataset
->ds_object
!=
2221 dsl_dir_phys(dd
)->dd_head_dataset_obj
)
2222 return (SET_ERROR(ENOENT
));
2224 zap_cursor_init_serialized(&cursor
,
2225 dd
->dd_pool
->dp_meta_objset
,
2226 dsl_dir_phys(dd
)->dd_child_dir_zapobj
, *offp
);
2228 if (zap_cursor_retrieve(&cursor
, &attr
) != 0) {
2229 zap_cursor_fini(&cursor
);
2230 return (SET_ERROR(ENOENT
));
2233 if (strlen(attr
.za_name
) + 1 > namelen
) {
2234 zap_cursor_fini(&cursor
);
2235 return (SET_ERROR(ENAMETOOLONG
));
2238 (void) strcpy(name
, attr
.za_name
);
2240 *idp
= attr
.za_first_integer
;
2241 zap_cursor_advance(&cursor
);
2242 *offp
= zap_cursor_serialize(&cursor
);
2243 zap_cursor_fini(&cursor
);
2248 typedef struct dmu_objset_find_ctx
{
2252 char *dc_ddname
; /* last component of ddobj's name */
2253 int (*dc_func
)(dsl_pool_t
*, dsl_dataset_t
*, void *);
2256 kmutex_t
*dc_error_lock
;
2258 } dmu_objset_find_ctx_t
;
2261 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t
*dcp
)
2263 dsl_pool_t
*dp
= dcp
->dc_dp
;
2264 dmu_objset_find_ctx_t
*child_dcp
;
2268 zap_attribute_t
*attr
;
2272 /* don't process if there already was an error */
2273 if (*dcp
->dc_error
!= 0)
2277 * Note: passing the name (dc_ddname) here is optional, but it
2278 * improves performance because we don't need to call
2279 * zap_value_search() to determine the name.
2281 err
= dsl_dir_hold_obj(dp
, dcp
->dc_ddobj
, dcp
->dc_ddname
, FTAG
, &dd
);
2285 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2286 if (dd
->dd_myname
[0] == '$') {
2287 dsl_dir_rele(dd
, FTAG
);
2291 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
2292 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
2295 * Iterate over all children.
2297 if (dcp
->dc_flags
& DS_FIND_CHILDREN
) {
2298 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2299 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
2300 zap_cursor_retrieve(&zc
, attr
) == 0;
2301 (void) zap_cursor_advance(&zc
)) {
2302 ASSERT3U(attr
->za_integer_length
, ==,
2304 ASSERT3U(attr
->za_num_integers
, ==, 1);
2307 kmem_alloc(sizeof (*child_dcp
), KM_SLEEP
);
2309 child_dcp
->dc_ddobj
= attr
->za_first_integer
;
2310 child_dcp
->dc_ddname
= spa_strdup(attr
->za_name
);
2311 if (dcp
->dc_tq
!= NULL
)
2312 (void) taskq_dispatch(dcp
->dc_tq
,
2313 dmu_objset_find_dp_cb
, child_dcp
, TQ_SLEEP
);
2315 dmu_objset_find_dp_impl(child_dcp
);
2317 zap_cursor_fini(&zc
);
2321 * Iterate over all snapshots.
2323 if (dcp
->dc_flags
& DS_FIND_SNAPSHOTS
) {
2325 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2330 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
2331 dsl_dataset_rele(ds
, FTAG
);
2333 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
2334 zap_cursor_retrieve(&zc
, attr
) == 0;
2335 (void) zap_cursor_advance(&zc
)) {
2336 ASSERT3U(attr
->za_integer_length
, ==,
2338 ASSERT3U(attr
->za_num_integers
, ==, 1);
2340 err
= dsl_dataset_hold_obj(dp
,
2341 attr
->za_first_integer
, FTAG
, &ds
);
2344 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
2345 dsl_dataset_rele(ds
, FTAG
);
2349 zap_cursor_fini(&zc
);
2353 kmem_free(attr
, sizeof (zap_attribute_t
));
2356 dsl_dir_rele(dd
, FTAG
);
2363 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2366 * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2367 * that the dir will remain cached, and we won't have to re-instantiate
2368 * it (which could be expensive due to finding its name via
2369 * zap_value_search()).
2371 dsl_dir_rele(dd
, FTAG
);
2374 err
= dcp
->dc_func(dp
, ds
, dcp
->dc_arg
);
2375 dsl_dataset_rele(ds
, FTAG
);
2379 mutex_enter(dcp
->dc_error_lock
);
2380 /* only keep first error */
2381 if (*dcp
->dc_error
== 0)
2382 *dcp
->dc_error
= err
;
2383 mutex_exit(dcp
->dc_error_lock
);
2386 if (dcp
->dc_ddname
!= NULL
)
2387 spa_strfree(dcp
->dc_ddname
);
2388 kmem_free(dcp
, sizeof (*dcp
));
2392 dmu_objset_find_dp_cb(void *arg
)
2394 dmu_objset_find_ctx_t
*dcp
= arg
;
2395 dsl_pool_t
*dp
= dcp
->dc_dp
;
2398 * We need to get a pool_config_lock here, as there are several
2399 * asssert(pool_config_held) down the stack. Getting a lock via
2400 * dsl_pool_config_enter is risky, as it might be stalled by a
2401 * pending writer. This would deadlock, as the write lock can
2402 * only be granted when our parent thread gives up the lock.
2403 * The _prio interface gives us priority over a pending writer.
2405 dsl_pool_config_enter_prio(dp
, FTAG
);
2407 dmu_objset_find_dp_impl(dcp
);
2409 dsl_pool_config_exit(dp
, FTAG
);
2413 * Find objsets under and including ddobj, call func(ds) on each.
2414 * The order for the enumeration is completely undefined.
2415 * func is called with dsl_pool_config held.
2418 dmu_objset_find_dp(dsl_pool_t
*dp
, uint64_t ddobj
,
2419 int func(dsl_pool_t
*, dsl_dataset_t
*, void *), void *arg
, int flags
)
2424 dmu_objset_find_ctx_t
*dcp
;
2427 mutex_init(&err_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2428 dcp
= kmem_alloc(sizeof (*dcp
), KM_SLEEP
);
2431 dcp
->dc_ddobj
= ddobj
;
2432 dcp
->dc_ddname
= NULL
;
2433 dcp
->dc_func
= func
;
2435 dcp
->dc_flags
= flags
;
2436 dcp
->dc_error_lock
= &err_lock
;
2437 dcp
->dc_error
= &error
;
2439 if ((flags
& DS_FIND_SERIALIZE
) || dsl_pool_config_held_writer(dp
)) {
2441 * In case a write lock is held we can't make use of
2442 * parallelism, as down the stack of the worker threads
2443 * the lock is asserted via dsl_pool_config_held.
2444 * In case of a read lock this is solved by getting a read
2445 * lock in each worker thread, which isn't possible in case
2446 * of a writer lock. So we fall back to the synchronous path
2448 * In the future it might be possible to get some magic into
2449 * dsl_pool_config_held in a way that it returns true for
2450 * the worker threads so that a single lock held from this
2451 * thread suffices. For now, stay single threaded.
2453 dmu_objset_find_dp_impl(dcp
);
2454 mutex_destroy(&err_lock
);
2459 ntasks
= dmu_find_threads
;
2461 ntasks
= vdev_count_leaves(dp
->dp_spa
) * 4;
2462 tq
= taskq_create("dmu_objset_find", ntasks
, maxclsyspri
, ntasks
,
2465 kmem_free(dcp
, sizeof (*dcp
));
2466 mutex_destroy(&err_lock
);
2468 return (SET_ERROR(ENOMEM
));
2472 /* dcp will be freed by task */
2473 (void) taskq_dispatch(tq
, dmu_objset_find_dp_cb
, dcp
, TQ_SLEEP
);
2476 * PORTING: this code relies on the property of taskq_wait to wait
2477 * until no more tasks are queued and no more tasks are active. As
2478 * we always queue new tasks from within other tasks, task_wait
2479 * reliably waits for the full recursion to finish, even though we
2480 * enqueue new tasks after taskq_wait has been called.
2481 * On platforms other than illumos, taskq_wait may not have this
2486 mutex_destroy(&err_lock
);
2492 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2493 * The dp_config_rwlock must not be held when this is called, and it
2494 * will not be held when the callback is called.
2495 * Therefore this function should only be used when the pool is not changing
2496 * (e.g. in syncing context), or the callback can deal with the possible races.
2499 dmu_objset_find_impl(spa_t
*spa
, const char *name
,
2500 int func(const char *, void *), void *arg
, int flags
)
2503 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2506 zap_attribute_t
*attr
;
2511 dsl_pool_config_enter(dp
, FTAG
);
2513 err
= dsl_dir_hold(dp
, name
, FTAG
, &dd
, NULL
);
2515 dsl_pool_config_exit(dp
, FTAG
);
2519 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2520 if (dd
->dd_myname
[0] == '$') {
2521 dsl_dir_rele(dd
, FTAG
);
2522 dsl_pool_config_exit(dp
, FTAG
);
2526 thisobj
= dsl_dir_phys(dd
)->dd_head_dataset_obj
;
2527 attr
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
2530 * Iterate over all children.
2532 if (flags
& DS_FIND_CHILDREN
) {
2533 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
,
2534 dsl_dir_phys(dd
)->dd_child_dir_zapobj
);
2535 zap_cursor_retrieve(&zc
, attr
) == 0;
2536 (void) zap_cursor_advance(&zc
)) {
2537 ASSERT3U(attr
->za_integer_length
, ==,
2539 ASSERT3U(attr
->za_num_integers
, ==, 1);
2541 child
= kmem_asprintf("%s/%s", name
, attr
->za_name
);
2542 dsl_pool_config_exit(dp
, FTAG
);
2543 err
= dmu_objset_find_impl(spa
, child
,
2545 dsl_pool_config_enter(dp
, FTAG
);
2550 zap_cursor_fini(&zc
);
2553 dsl_dir_rele(dd
, FTAG
);
2554 dsl_pool_config_exit(dp
, FTAG
);
2555 kmem_free(attr
, sizeof (zap_attribute_t
));
2561 * Iterate over all snapshots.
2563 if (flags
& DS_FIND_SNAPSHOTS
) {
2564 err
= dsl_dataset_hold_obj(dp
, thisobj
, FTAG
, &ds
);
2569 snapobj
= dsl_dataset_phys(ds
)->ds_snapnames_zapobj
;
2570 dsl_dataset_rele(ds
, FTAG
);
2572 for (zap_cursor_init(&zc
, dp
->dp_meta_objset
, snapobj
);
2573 zap_cursor_retrieve(&zc
, attr
) == 0;
2574 (void) zap_cursor_advance(&zc
)) {
2575 ASSERT3U(attr
->za_integer_length
, ==,
2577 ASSERT3U(attr
->za_num_integers
, ==, 1);
2579 child
= kmem_asprintf("%s@%s",
2580 name
, attr
->za_name
);
2581 dsl_pool_config_exit(dp
, FTAG
);
2582 err
= func(child
, arg
);
2583 dsl_pool_config_enter(dp
, FTAG
);
2588 zap_cursor_fini(&zc
);
2592 dsl_dir_rele(dd
, FTAG
);
2593 kmem_free(attr
, sizeof (zap_attribute_t
));
2594 dsl_pool_config_exit(dp
, FTAG
);
2599 /* Apply to self. */
2600 return (func(name
, arg
));
2604 * See comment above dmu_objset_find_impl().
2607 dmu_objset_find(char *name
, int func(const char *, void *), void *arg
,
2613 error
= spa_open(name
, &spa
, FTAG
);
2616 error
= dmu_objset_find_impl(spa
, name
, func
, arg
, flags
);
2617 spa_close(spa
, FTAG
);
2622 dmu_objset_set_user(objset_t
*os
, void *user_ptr
)
2624 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2625 os
->os_user_ptr
= user_ptr
;
2629 dmu_objset_get_user(objset_t
*os
)
2631 ASSERT(MUTEX_HELD(&os
->os_user_ptr_lock
));
2632 return (os
->os_user_ptr
);
2636 * Determine name of filesystem, given name of snapshot.
2637 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2640 dmu_fsname(const char *snapname
, char *buf
)
2642 char *atp
= strchr(snapname
, '@');
2644 return (SET_ERROR(EINVAL
));
2645 if (atp
- snapname
>= ZFS_MAX_DATASET_NAME_LEN
)
2646 return (SET_ERROR(ENAMETOOLONG
));
2647 (void) strlcpy(buf
, snapname
, atp
- snapname
+ 1);
2652 * Call when we think we're going to write/free space in open context to track
2653 * the amount of dirty data in the open txg, which is also the amount
2654 * of memory that can not be evicted until this txg syncs.
2657 dmu_objset_willuse_space(objset_t
*os
, int64_t space
, dmu_tx_t
*tx
)
2659 dsl_dataset_t
*ds
= os
->os_dsl_dataset
;
2660 int64_t aspace
= spa_get_worst_case_asize(os
->os_spa
, space
);
2663 dsl_dir_willuse_space(ds
->ds_dir
, aspace
, tx
);
2664 dsl_pool_dirty_space(dmu_tx_pool(tx
), space
, tx
);
2668 #if defined(_KERNEL) && defined(HAVE_SPL)
2669 EXPORT_SYMBOL(dmu_objset_zil
);
2670 EXPORT_SYMBOL(dmu_objset_pool
);
2671 EXPORT_SYMBOL(dmu_objset_ds
);
2672 EXPORT_SYMBOL(dmu_objset_type
);
2673 EXPORT_SYMBOL(dmu_objset_name
);
2674 EXPORT_SYMBOL(dmu_objset_hold
);
2675 EXPORT_SYMBOL(dmu_objset_hold_flags
);
2676 EXPORT_SYMBOL(dmu_objset_own
);
2677 EXPORT_SYMBOL(dmu_objset_rele
);
2678 EXPORT_SYMBOL(dmu_objset_rele_flags
);
2679 EXPORT_SYMBOL(dmu_objset_disown
);
2680 EXPORT_SYMBOL(dmu_objset_from_ds
);
2681 EXPORT_SYMBOL(dmu_objset_create
);
2682 EXPORT_SYMBOL(dmu_objset_clone
);
2683 EXPORT_SYMBOL(dmu_objset_stats
);
2684 EXPORT_SYMBOL(dmu_objset_fast_stat
);
2685 EXPORT_SYMBOL(dmu_objset_spa
);
2686 EXPORT_SYMBOL(dmu_objset_space
);
2687 EXPORT_SYMBOL(dmu_objset_fsid_guid
);
2688 EXPORT_SYMBOL(dmu_objset_find
);
2689 EXPORT_SYMBOL(dmu_objset_byteswap
);
2690 EXPORT_SYMBOL(dmu_objset_evict_dbufs
);
2691 EXPORT_SYMBOL(dmu_objset_snap_cmtime
);
2692 EXPORT_SYMBOL(dmu_objset_dnodesize
);
2694 EXPORT_SYMBOL(dmu_objset_sync
);
2695 EXPORT_SYMBOL(dmu_objset_is_dirty
);
2696 EXPORT_SYMBOL(dmu_objset_create_impl_dnstats
);
2697 EXPORT_SYMBOL(dmu_objset_create_impl
);
2698 EXPORT_SYMBOL(dmu_objset_open_impl
);
2699 EXPORT_SYMBOL(dmu_objset_evict
);
2700 EXPORT_SYMBOL(dmu_objset_register_type
);
2701 EXPORT_SYMBOL(dmu_objset_do_userquota_updates
);
2702 EXPORT_SYMBOL(dmu_objset_userquota_get_ids
);
2703 EXPORT_SYMBOL(dmu_objset_userused_enabled
);
2704 EXPORT_SYMBOL(dmu_objset_userspace_upgrade
);
2705 EXPORT_SYMBOL(dmu_objset_userspace_present
);
2706 EXPORT_SYMBOL(dmu_objset_userobjused_enabled
);
2707 EXPORT_SYMBOL(dmu_objset_userobjspace_upgrade
);
2708 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable
);
2709 EXPORT_SYMBOL(dmu_objset_userobjspace_present
);