4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (C) 2008-2010 Lawrence Livermore National Security, LLC.
23 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
24 * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>.
27 * ZFS volume emulation driver.
29 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
30 * Volumes are accessed through the symbolic links named:
32 * /dev/<pool_name>/<dataset_name>
34 * Volumes are persistent through reboot and module load. No user command
35 * needs to be run before opening and using a device.
37 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
38 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
39 * Copyright (c) 2012, 2019 by Delphix. All rights reserved.
43 * Note on locking of zvol state structures.
45 * These structures are used to maintain internal state used to emulate block
46 * devices on top of zvols. In particular, management of device minor number
47 * operations - create, remove, rename, and set_snapdev - involves access to
48 * these structures. The zvol_state_lock is primarily used to protect the
49 * zvol_state_list. The zv->zv_state_lock is used to protect the contents
50 * of the zvol_state_t structures, as well as to make sure that when the
51 * time comes to remove the structure from the list, it is not in use, and
52 * therefore, it can be taken off zvol_state_list and freed.
54 * The zv_suspend_lock was introduced to allow for suspending I/O to a zvol,
55 * e.g. for the duration of receive and rollback operations. This lock can be
56 * held for significant periods of time. Given that it is undesirable to hold
57 * mutexes for long periods of time, the following lock ordering applies:
58 * - take zvol_state_lock if necessary, to protect zvol_state_list
59 * - take zv_suspend_lock if necessary, by the code path in question
60 * - take zv_state_lock to protect zvol_state_t
62 * The minor operations are issued to spa->spa_zvol_taskq queues, that are
63 * single-threaded (to preserve order of minor operations), and are executed
64 * through the zvol_task_cb that dispatches the specific operations. Therefore,
65 * these operations are serialized per pool. Consequently, we can be certain
66 * that for a given zvol, there is only one operation at a time in progress.
67 * That is why one can be sure that first, zvol_state_t for a given zvol is
68 * allocated and placed on zvol_state_list, and then other minor operations
69 * for this zvol are going to proceed in the order of issue.
73 #include <sys/dataset_kstats.h>
75 #include <sys/dmu_traverse.h>
76 #include <sys/dsl_dataset.h>
77 #include <sys/dsl_prop.h>
78 #include <sys/dsl_dir.h>
80 #include <sys/zfeature.h>
81 #include <sys/zil_impl.h>
82 #include <sys/dmu_tx.h>
84 #include <sys/zfs_rlock.h>
85 #include <sys/spa_impl.h>
87 #include <sys/zvol_impl.h>
89 unsigned int zvol_inhibit_dev
= 0;
90 unsigned int zvol_volmode
= ZFS_VOLMODE_GEOM
;
92 struct hlist_head
*zvol_htable
;
93 static list_t zvol_state_list
;
94 krwlock_t zvol_state_lock
;
97 ZVOL_ASYNC_REMOVE_MINORS
,
98 ZVOL_ASYNC_RENAME_MINORS
,
99 ZVOL_ASYNC_SET_SNAPDEV
,
100 ZVOL_ASYNC_SET_VOLMODE
,
106 char name1
[MAXNAMELEN
];
107 char name2
[MAXNAMELEN
];
112 zvol_name_hash(const char *name
)
114 uint64_t crc
= -1ULL;
115 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
116 for (const uint8_t *p
= (const uint8_t *)name
; *p
!= 0; p
++)
117 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (*p
)) & 0xFF];
122 * Find a zvol_state_t given the name and hash generated by zvol_name_hash.
123 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
124 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
125 * before zv_state_lock. The mode argument indicates the mode (including none)
126 * for zv_suspend_lock to be taken.
129 zvol_find_by_name_hash(const char *name
, uint64_t hash
, int mode
)
132 struct hlist_node
*p
= NULL
;
134 rw_enter(&zvol_state_lock
, RW_READER
);
135 hlist_for_each(p
, ZVOL_HT_HEAD(hash
)) {
136 zv
= hlist_entry(p
, zvol_state_t
, zv_hlink
);
137 mutex_enter(&zv
->zv_state_lock
);
138 if (zv
->zv_hash
== hash
&& strcmp(zv
->zv_name
, name
) == 0) {
140 * this is the right zvol, take the locks in the
143 if (mode
!= RW_NONE
&&
144 !rw_tryenter(&zv
->zv_suspend_lock
, mode
)) {
145 mutex_exit(&zv
->zv_state_lock
);
146 rw_enter(&zv
->zv_suspend_lock
, mode
);
147 mutex_enter(&zv
->zv_state_lock
);
149 * zvol cannot be renamed as we continue
150 * to hold zvol_state_lock
152 ASSERT(zv
->zv_hash
== hash
&&
153 strcmp(zv
->zv_name
, name
) == 0);
155 rw_exit(&zvol_state_lock
);
158 mutex_exit(&zv
->zv_state_lock
);
160 rw_exit(&zvol_state_lock
);
166 * Find a zvol_state_t given the name.
167 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
168 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
169 * before zv_state_lock. The mode argument indicates the mode (including none)
170 * for zv_suspend_lock to be taken.
172 static zvol_state_t
*
173 zvol_find_by_name(const char *name
, int mode
)
175 return (zvol_find_by_name_hash(name
, zvol_name_hash(name
), mode
));
179 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
182 zvol_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
184 zfs_creat_t
*zct
= arg
;
185 nvlist_t
*nvprops
= zct
->zct_props
;
187 uint64_t volblocksize
, volsize
;
189 VERIFY(nvlist_lookup_uint64(nvprops
,
190 zfs_prop_to_name(ZFS_PROP_VOLSIZE
), &volsize
) == 0);
191 if (nvlist_lookup_uint64(nvprops
,
192 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
), &volblocksize
) != 0)
193 volblocksize
= zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE
);
196 * These properties must be removed from the list so the generic
197 * property setting step won't apply to them.
199 VERIFY(nvlist_remove_all(nvprops
,
200 zfs_prop_to_name(ZFS_PROP_VOLSIZE
)) == 0);
201 (void) nvlist_remove_all(nvprops
,
202 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
));
204 error
= dmu_object_claim(os
, ZVOL_OBJ
, DMU_OT_ZVOL
, volblocksize
,
208 error
= zap_create_claim(os
, ZVOL_ZAP_OBJ
, DMU_OT_ZVOL_PROP
,
212 error
= zap_update(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
, tx
);
217 * ZFS_IOC_OBJSET_STATS entry point.
220 zvol_get_stats(objset_t
*os
, nvlist_t
*nv
)
223 dmu_object_info_t
*doi
;
226 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &val
);
228 return (SET_ERROR(error
));
230 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_VOLSIZE
, val
);
231 doi
= kmem_alloc(sizeof (dmu_object_info_t
), KM_SLEEP
);
232 error
= dmu_object_info(os
, ZVOL_OBJ
, doi
);
235 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_VOLBLOCKSIZE
,
236 doi
->doi_data_block_size
);
239 kmem_free(doi
, sizeof (dmu_object_info_t
));
241 return (SET_ERROR(error
));
245 * Sanity check volume size.
248 zvol_check_volsize(uint64_t volsize
, uint64_t blocksize
)
251 return (SET_ERROR(EINVAL
));
253 if (volsize
% blocksize
!= 0)
254 return (SET_ERROR(EINVAL
));
257 if (volsize
- 1 > SPEC_MAXOFFSET_T
)
258 return (SET_ERROR(EOVERFLOW
));
264 * Ensure the zap is flushed then inform the VFS of the capacity change.
267 zvol_update_volsize(uint64_t volsize
, objset_t
*os
)
273 tx
= dmu_tx_create(os
);
274 dmu_tx_hold_zap(tx
, ZVOL_ZAP_OBJ
, TRUE
, NULL
);
275 dmu_tx_mark_netfree(tx
);
276 error
= dmu_tx_assign(tx
, TXG_WAIT
);
279 return (SET_ERROR(error
));
281 txg
= dmu_tx_get_txg(tx
);
283 error
= zap_update(os
, ZVOL_ZAP_OBJ
, "size", 8, 1,
287 txg_wait_synced(dmu_objset_pool(os
), txg
);
290 error
= dmu_free_long_range(os
,
291 ZVOL_OBJ
, volsize
, DMU_OBJECT_END
);
297 * Set ZFS_PROP_VOLSIZE set entry point. Note that modifying the volume
298 * size will result in a udev "change" event being generated.
301 zvol_set_volsize(const char *name
, uint64_t volsize
)
306 boolean_t owned
= B_FALSE
;
308 error
= dsl_prop_get_integer(name
,
309 zfs_prop_to_name(ZFS_PROP_READONLY
), &readonly
, NULL
);
311 return (SET_ERROR(error
));
313 return (SET_ERROR(EROFS
));
315 zvol_state_t
*zv
= zvol_find_by_name(name
, RW_READER
);
317 ASSERT(zv
== NULL
|| (MUTEX_HELD(&zv
->zv_state_lock
) &&
318 RW_READ_HELD(&zv
->zv_suspend_lock
)));
320 if (zv
== NULL
|| zv
->zv_objset
== NULL
) {
322 rw_exit(&zv
->zv_suspend_lock
);
323 if ((error
= dmu_objset_own(name
, DMU_OST_ZVOL
, B_FALSE
, B_TRUE
,
326 mutex_exit(&zv
->zv_state_lock
);
327 return (SET_ERROR(error
));
336 dmu_object_info_t
*doi
= kmem_alloc(sizeof (*doi
), KM_SLEEP
);
338 if ((error
= dmu_object_info(os
, ZVOL_OBJ
, doi
)) ||
339 (error
= zvol_check_volsize(volsize
, doi
->doi_data_block_size
)))
342 error
= zvol_update_volsize(volsize
, os
);
343 if (error
== 0 && zv
!= NULL
) {
344 zv
->zv_volsize
= volsize
;
348 kmem_free(doi
, sizeof (dmu_object_info_t
));
351 dmu_objset_disown(os
, B_TRUE
, FTAG
);
353 zv
->zv_objset
= NULL
;
355 rw_exit(&zv
->zv_suspend_lock
);
359 mutex_exit(&zv
->zv_state_lock
);
361 if (error
== 0 && zv
!= NULL
)
362 zvol_os_update_volsize(zv
, volsize
);
364 return (SET_ERROR(error
));
368 * Update volthreading.
371 zvol_set_volthreading(const char *name
, boolean_t value
)
373 zvol_state_t
*zv
= zvol_find_by_name(name
, RW_NONE
);
376 zv
->zv_threading
= value
;
377 mutex_exit(&zv
->zv_state_lock
);
382 * Update zvol ro property.
385 zvol_set_ro(const char *name
, boolean_t value
)
387 zvol_state_t
*zv
= zvol_find_by_name(name
, RW_NONE
);
391 zvol_os_set_disk_ro(zv
, 1);
392 zv
->zv_flags
|= ZVOL_RDONLY
;
394 zvol_os_set_disk_ro(zv
, 0);
395 zv
->zv_flags
&= ~ZVOL_RDONLY
;
397 mutex_exit(&zv
->zv_state_lock
);
402 * Sanity check volume block size.
405 zvol_check_volblocksize(const char *name
, uint64_t volblocksize
)
407 /* Record sizes above 128k need the feature to be enabled */
408 if (volblocksize
> SPA_OLD_MAXBLOCKSIZE
) {
412 if ((error
= spa_open(name
, &spa
, FTAG
)) != 0)
415 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
416 spa_close(spa
, FTAG
);
417 return (SET_ERROR(ENOTSUP
));
421 * We don't allow setting the property above 1MB,
422 * unless the tunable has been changed.
424 if (volblocksize
> zfs_max_recordsize
)
425 return (SET_ERROR(EDOM
));
427 spa_close(spa
, FTAG
);
430 if (volblocksize
< SPA_MINBLOCKSIZE
||
431 volblocksize
> SPA_MAXBLOCKSIZE
||
433 return (SET_ERROR(EDOM
));
439 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
440 * implement DKIOCFREE/free-long-range.
443 zvol_replay_truncate(void *arg1
, void *arg2
, boolean_t byteswap
)
445 zvol_state_t
*zv
= arg1
;
446 lr_truncate_t
*lr
= arg2
;
447 uint64_t offset
, length
;
449 ASSERT3U(lr
->lr_common
.lrc_reclen
, >=, sizeof (*lr
));
452 byteswap_uint64_array(lr
, sizeof (*lr
));
454 offset
= lr
->lr_offset
;
455 length
= lr
->lr_length
;
457 dmu_tx_t
*tx
= dmu_tx_create(zv
->zv_objset
);
458 dmu_tx_mark_netfree(tx
);
459 int error
= dmu_tx_assign(tx
, TXG_WAIT
);
463 (void) zil_replaying(zv
->zv_zilog
, tx
);
465 error
= dmu_free_long_range(zv
->zv_objset
, ZVOL_OBJ
, offset
,
473 * Replay a TX_WRITE ZIL transaction that didn't get committed
474 * after a system failure
477 zvol_replay_write(void *arg1
, void *arg2
, boolean_t byteswap
)
479 zvol_state_t
*zv
= arg1
;
480 lr_write_t
*lr
= arg2
;
481 objset_t
*os
= zv
->zv_objset
;
482 char *data
= (char *)(lr
+ 1); /* data follows lr_write_t */
483 uint64_t offset
, length
;
487 ASSERT3U(lr
->lr_common
.lrc_reclen
, >=, sizeof (*lr
));
490 byteswap_uint64_array(lr
, sizeof (*lr
));
492 offset
= lr
->lr_offset
;
493 length
= lr
->lr_length
;
495 /* If it's a dmu_sync() block, write the whole block */
496 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
497 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
498 if (length
< blocksize
) {
499 offset
-= offset
% blocksize
;
504 tx
= dmu_tx_create(os
);
505 dmu_tx_hold_write(tx
, ZVOL_OBJ
, offset
, length
);
506 error
= dmu_tx_assign(tx
, TXG_WAIT
);
510 dmu_write(os
, ZVOL_OBJ
, offset
, length
, data
, tx
);
511 (void) zil_replaying(zv
->zv_zilog
, tx
);
519 zvol_replay_err(void *arg1
, void *arg2
, boolean_t byteswap
)
521 (void) arg1
, (void) arg2
, (void) byteswap
;
522 return (SET_ERROR(ENOTSUP
));
526 * Callback vectors for replaying records.
527 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
529 zil_replay_func_t
*const zvol_replay_vector
[TX_MAX_TYPE
] = {
530 zvol_replay_err
, /* no such transaction type */
531 zvol_replay_err
, /* TX_CREATE */
532 zvol_replay_err
, /* TX_MKDIR */
533 zvol_replay_err
, /* TX_MKXATTR */
534 zvol_replay_err
, /* TX_SYMLINK */
535 zvol_replay_err
, /* TX_REMOVE */
536 zvol_replay_err
, /* TX_RMDIR */
537 zvol_replay_err
, /* TX_LINK */
538 zvol_replay_err
, /* TX_RENAME */
539 zvol_replay_write
, /* TX_WRITE */
540 zvol_replay_truncate
, /* TX_TRUNCATE */
541 zvol_replay_err
, /* TX_SETATTR */
542 zvol_replay_err
, /* TX_ACL */
543 zvol_replay_err
, /* TX_CREATE_ATTR */
544 zvol_replay_err
, /* TX_CREATE_ACL_ATTR */
545 zvol_replay_err
, /* TX_MKDIR_ACL */
546 zvol_replay_err
, /* TX_MKDIR_ATTR */
547 zvol_replay_err
, /* TX_MKDIR_ACL_ATTR */
548 zvol_replay_err
, /* TX_WRITE2 */
549 zvol_replay_err
, /* TX_SETSAXATTR */
550 zvol_replay_err
, /* TX_RENAME_EXCHANGE */
551 zvol_replay_err
, /* TX_RENAME_WHITEOUT */
552 zvol_replay_err
, /* TX_CLONE_RANGE */
556 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
558 * We store data in the log buffers if it's small enough.
559 * Otherwise we will later flush the data out via dmu_sync().
561 static const ssize_t zvol_immediate_write_sz
= 32768;
564 zvol_log_write(zvol_state_t
*zv
, dmu_tx_t
*tx
, uint64_t offset
,
565 uint64_t size
, boolean_t commit
)
567 uint32_t blocksize
= zv
->zv_volblocksize
;
568 zilog_t
*zilog
= zv
->zv_zilog
;
569 itx_wr_state_t write_state
;
572 if (zil_replaying(zilog
, tx
))
575 if (zilog
->zl_logbias
== ZFS_LOGBIAS_THROUGHPUT
)
576 write_state
= WR_INDIRECT
;
577 else if (!spa_has_slogs(zilog
->zl_spa
) &&
578 size
>= blocksize
&& blocksize
> zvol_immediate_write_sz
)
579 write_state
= WR_INDIRECT
;
581 write_state
= WR_COPIED
;
583 write_state
= WR_NEED_COPY
;
588 itx_wr_state_t wr_state
= write_state
;
591 if (wr_state
== WR_COPIED
&& size
> zil_max_copied_data(zilog
))
592 wr_state
= WR_NEED_COPY
;
593 else if (wr_state
== WR_INDIRECT
)
594 len
= MIN(blocksize
- P2PHASE(offset
, blocksize
), size
);
596 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
) +
597 (wr_state
== WR_COPIED
? len
: 0));
598 lr
= (lr_write_t
*)&itx
->itx_lr
;
599 if (wr_state
== WR_COPIED
&& dmu_read_by_dnode(zv
->zv_dn
,
600 offset
, len
, lr
+1, DMU_READ_NO_PREFETCH
) != 0) {
601 zil_itx_destroy(itx
);
602 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
603 lr
= (lr_write_t
*)&itx
->itx_lr
;
604 wr_state
= WR_NEED_COPY
;
607 itx
->itx_wr_state
= wr_state
;
608 lr
->lr_foid
= ZVOL_OBJ
;
609 lr
->lr_offset
= offset
;
612 BP_ZERO(&lr
->lr_blkptr
);
614 itx
->itx_private
= zv
;
616 (void) zil_itx_assign(zilog
, itx
, tx
);
622 if (write_state
== WR_COPIED
|| write_state
== WR_NEED_COPY
) {
623 dsl_pool_wrlog_count(zilog
->zl_dmu_pool
, sz
, tx
->tx_txg
);
628 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
631 zvol_log_truncate(zvol_state_t
*zv
, dmu_tx_t
*tx
, uint64_t off
, uint64_t len
)
635 zilog_t
*zilog
= zv
->zv_zilog
;
637 if (zil_replaying(zilog
, tx
))
640 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
641 lr
= (lr_truncate_t
*)&itx
->itx_lr
;
642 lr
->lr_foid
= ZVOL_OBJ
;
646 zil_itx_assign(zilog
, itx
, tx
);
651 zvol_get_done(zgd_t
*zgd
, int error
)
655 dmu_buf_rele(zgd
->zgd_db
, zgd
);
657 zfs_rangelock_exit(zgd
->zgd_lr
);
659 kmem_free(zgd
, sizeof (zgd_t
));
663 * Get data to generate a TX_WRITE intent log record.
666 zvol_get_data(void *arg
, uint64_t arg2
, lr_write_t
*lr
, char *buf
,
667 struct lwb
*lwb
, zio_t
*zio
)
669 zvol_state_t
*zv
= arg
;
670 uint64_t offset
= lr
->lr_offset
;
671 uint64_t size
= lr
->lr_length
;
676 ASSERT3P(lwb
, !=, NULL
);
677 ASSERT3U(size
, !=, 0);
679 zgd
= kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
683 * Write records come in two flavors: immediate and indirect.
684 * For small writes it's cheaper to store the data with the
685 * log record (immediate); for large writes it's cheaper to
686 * sync the data and get a pointer to it (indirect) so that
687 * we don't have to write the data twice.
689 if (buf
!= NULL
) { /* immediate write */
690 zgd
->zgd_lr
= zfs_rangelock_enter(&zv
->zv_rangelock
, offset
,
692 error
= dmu_read_by_dnode(zv
->zv_dn
, offset
, size
, buf
,
693 DMU_READ_NO_PREFETCH
);
694 } else { /* indirect write */
695 ASSERT3P(zio
, !=, NULL
);
697 * Have to lock the whole block to ensure when it's written out
698 * and its checksum is being calculated that no one can change
699 * the data. Contrarily to zfs_get_data we need not re-check
700 * blocksize after we get the lock because it cannot be changed.
702 size
= zv
->zv_volblocksize
;
703 offset
= P2ALIGN_TYPED(offset
, size
, uint64_t);
704 zgd
->zgd_lr
= zfs_rangelock_enter(&zv
->zv_rangelock
, offset
,
706 error
= dmu_buf_hold_noread_by_dnode(zv
->zv_dn
, offset
, zgd
,
709 blkptr_t
*bp
= &lr
->lr_blkptr
;
715 ASSERT(db
->db_offset
== offset
);
716 ASSERT(db
->db_size
== size
);
718 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
726 zvol_get_done(zgd
, error
);
728 return (SET_ERROR(error
));
732 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable.
736 zvol_insert(zvol_state_t
*zv
)
738 ASSERT(RW_WRITE_HELD(&zvol_state_lock
));
739 list_insert_head(&zvol_state_list
, zv
);
740 hlist_add_head(&zv
->zv_hlink
, ZVOL_HT_HEAD(zv
->zv_hash
));
744 * Simply remove the zvol from to list of zvols.
747 zvol_remove(zvol_state_t
*zv
)
749 ASSERT(RW_WRITE_HELD(&zvol_state_lock
));
750 list_remove(&zvol_state_list
, zv
);
751 hlist_del(&zv
->zv_hlink
);
755 * Setup zv after we just own the zv->objset
758 zvol_setup_zv(zvol_state_t
*zv
)
763 objset_t
*os
= zv
->zv_objset
;
765 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
766 ASSERT(RW_LOCK_HELD(&zv
->zv_suspend_lock
));
769 zv
->zv_flags
&= ~ZVOL_WRITTEN_TO
;
771 error
= dsl_prop_get_integer(zv
->zv_name
, "readonly", &ro
, NULL
);
773 return (SET_ERROR(error
));
775 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
777 return (SET_ERROR(error
));
779 error
= dnode_hold(os
, ZVOL_OBJ
, zv
, &zv
->zv_dn
);
781 return (SET_ERROR(error
));
783 zvol_os_set_capacity(zv
, volsize
>> 9);
784 zv
->zv_volsize
= volsize
;
786 if (ro
|| dmu_objset_is_snapshot(os
) ||
787 !spa_writeable(dmu_objset_spa(os
))) {
788 zvol_os_set_disk_ro(zv
, 1);
789 zv
->zv_flags
|= ZVOL_RDONLY
;
791 zvol_os_set_disk_ro(zv
, 0);
792 zv
->zv_flags
&= ~ZVOL_RDONLY
;
798 * Shutdown every zv_objset related stuff except zv_objset itself.
799 * The is the reverse of zvol_setup_zv.
802 zvol_shutdown_zv(zvol_state_t
*zv
)
804 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
) &&
805 RW_LOCK_HELD(&zv
->zv_suspend_lock
));
807 if (zv
->zv_flags
& ZVOL_WRITTEN_TO
) {
808 ASSERT(zv
->zv_zilog
!= NULL
);
809 zil_close(zv
->zv_zilog
);
814 dnode_rele(zv
->zv_dn
, zv
);
818 * Evict cached data. We must write out any dirty data before
819 * disowning the dataset.
821 if (zv
->zv_flags
& ZVOL_WRITTEN_TO
)
822 txg_wait_synced(dmu_objset_pool(zv
->zv_objset
), 0);
823 (void) dmu_objset_evict_dbufs(zv
->zv_objset
);
827 * return the proper tag for rollback and recv
830 zvol_tag(zvol_state_t
*zv
)
832 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
833 return (zv
->zv_open_count
> 0 ? zv
: NULL
);
837 * Suspend the zvol for recv and rollback.
840 zvol_suspend(const char *name
)
844 zv
= zvol_find_by_name(name
, RW_WRITER
);
849 /* block all I/O, release in zvol_resume. */
850 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
851 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
853 atomic_inc(&zv
->zv_suspend_ref
);
855 if (zv
->zv_open_count
> 0)
856 zvol_shutdown_zv(zv
);
859 * do not hold zv_state_lock across suspend/resume to
860 * avoid locking up zvol lookups
862 mutex_exit(&zv
->zv_state_lock
);
864 /* zv_suspend_lock is released in zvol_resume() */
869 zvol_resume(zvol_state_t
*zv
)
873 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
875 mutex_enter(&zv
->zv_state_lock
);
877 if (zv
->zv_open_count
> 0) {
878 VERIFY0(dmu_objset_hold(zv
->zv_name
, zv
, &zv
->zv_objset
));
879 VERIFY3P(zv
->zv_objset
->os_dsl_dataset
->ds_owner
, ==, zv
);
880 VERIFY(dsl_dataset_long_held(zv
->zv_objset
->os_dsl_dataset
));
881 dmu_objset_rele(zv
->zv_objset
, zv
);
883 error
= zvol_setup_zv(zv
);
886 mutex_exit(&zv
->zv_state_lock
);
888 rw_exit(&zv
->zv_suspend_lock
);
890 * We need this because we don't hold zvol_state_lock while releasing
891 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check
892 * zv_suspend_lock to determine it is safe to free because rwlock is
893 * not inherent atomic.
895 atomic_dec(&zv
->zv_suspend_ref
);
897 return (SET_ERROR(error
));
901 zvol_first_open(zvol_state_t
*zv
, boolean_t readonly
)
906 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
907 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
908 ASSERT(mutex_owned(&spa_namespace_lock
));
910 boolean_t ro
= (readonly
|| (strchr(zv
->zv_name
, '@') != NULL
));
911 error
= dmu_objset_own(zv
->zv_name
, DMU_OST_ZVOL
, ro
, B_TRUE
, zv
, &os
);
913 return (SET_ERROR(error
));
917 error
= zvol_setup_zv(zv
);
919 dmu_objset_disown(os
, 1, zv
);
920 zv
->zv_objset
= NULL
;
927 zvol_last_close(zvol_state_t
*zv
)
929 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
930 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
932 zvol_shutdown_zv(zv
);
934 dmu_objset_disown(zv
->zv_objset
, 1, zv
);
935 zv
->zv_objset
= NULL
;
938 typedef struct minors_job
{
948 * Prefetch zvol dnodes for the minors_job
951 zvol_prefetch_minors_impl(void *arg
)
953 minors_job_t
*job
= arg
;
954 char *dsname
= job
->name
;
957 job
->error
= dmu_objset_own(dsname
, DMU_OST_ZVOL
, B_TRUE
, B_TRUE
,
959 if (job
->error
== 0) {
960 dmu_prefetch_dnode(os
, ZVOL_OBJ
, ZIO_PRIORITY_SYNC_READ
);
961 dmu_objset_disown(os
, B_TRUE
, FTAG
);
966 * Mask errors to continue dmu_objset_find() traversal
969 zvol_create_snap_minor_cb(const char *dsname
, void *arg
)
971 minors_job_t
*j
= arg
;
972 list_t
*minors_list
= j
->list
;
973 const char *name
= j
->name
;
975 ASSERT0(MUTEX_HELD(&spa_namespace_lock
));
977 /* skip the designated dataset */
978 if (name
&& strcmp(dsname
, name
) == 0)
981 /* at this point, the dsname should name a snapshot */
982 if (strchr(dsname
, '@') == 0) {
983 dprintf("zvol_create_snap_minor_cb(): "
984 "%s is not a snapshot name\n", dsname
);
987 char *n
= kmem_strdup(dsname
);
991 job
= kmem_alloc(sizeof (minors_job_t
), KM_SLEEP
);
993 job
->list
= minors_list
;
995 list_insert_tail(minors_list
, job
);
996 /* don't care if dispatch fails, because job->error is 0 */
997 taskq_dispatch(system_taskq
, zvol_prefetch_minors_impl
, job
,
1005 * If spa_keystore_load_wkey() is called for an encrypted zvol,
1006 * we need to look for any clones also using the key. This function
1007 * is "best effort" - so we just skip over it if there are failures.
1010 zvol_add_clones(const char *dsname
, list_t
*minors_list
)
1012 /* Also check if it has clones */
1013 dsl_dir_t
*dd
= NULL
;
1014 dsl_pool_t
*dp
= NULL
;
1016 if (dsl_pool_hold(dsname
, FTAG
, &dp
) != 0)
1019 if (!spa_feature_is_enabled(dp
->dp_spa
,
1020 SPA_FEATURE_ENCRYPTION
))
1023 if (dsl_dir_hold(dp
, dsname
, FTAG
, &dd
, NULL
) != 0)
1026 if (dsl_dir_phys(dd
)->dd_clones
== 0)
1029 zap_cursor_t
*zc
= kmem_alloc(sizeof (zap_cursor_t
), KM_SLEEP
);
1030 zap_attribute_t
*za
= kmem_alloc(sizeof (zap_attribute_t
), KM_SLEEP
);
1031 objset_t
*mos
= dd
->dd_pool
->dp_meta_objset
;
1033 for (zap_cursor_init(zc
, mos
, dsl_dir_phys(dd
)->dd_clones
);
1034 zap_cursor_retrieve(zc
, za
) == 0;
1035 zap_cursor_advance(zc
)) {
1036 dsl_dataset_t
*clone
;
1039 if (dsl_dataset_hold_obj(dd
->dd_pool
,
1040 za
->za_first_integer
, FTAG
, &clone
) == 0) {
1042 char name
[ZFS_MAX_DATASET_NAME_LEN
];
1043 dsl_dataset_name(clone
, name
);
1045 char *n
= kmem_strdup(name
);
1046 job
= kmem_alloc(sizeof (minors_job_t
), KM_SLEEP
);
1048 job
->list
= minors_list
;
1050 list_insert_tail(minors_list
, job
);
1052 dsl_dataset_rele(clone
, FTAG
);
1055 zap_cursor_fini(zc
);
1056 kmem_free(za
, sizeof (zap_attribute_t
));
1057 kmem_free(zc
, sizeof (zap_cursor_t
));
1061 dsl_dir_rele(dd
, FTAG
);
1062 dsl_pool_rele(dp
, FTAG
);
1066 * Mask errors to continue dmu_objset_find() traversal
1069 zvol_create_minors_cb(const char *dsname
, void *arg
)
1073 list_t
*minors_list
= arg
;
1075 ASSERT0(MUTEX_HELD(&spa_namespace_lock
));
1077 error
= dsl_prop_get_integer(dsname
, "snapdev", &snapdev
, NULL
);
1082 * Given the name and the 'snapdev' property, create device minor nodes
1083 * with the linkages to zvols/snapshots as needed.
1084 * If the name represents a zvol, create a minor node for the zvol, then
1085 * check if its snapshots are 'visible', and if so, iterate over the
1086 * snapshots and create device minor nodes for those.
1088 if (strchr(dsname
, '@') == 0) {
1090 char *n
= kmem_strdup(dsname
);
1094 job
= kmem_alloc(sizeof (minors_job_t
), KM_SLEEP
);
1096 job
->list
= minors_list
;
1098 list_insert_tail(minors_list
, job
);
1099 /* don't care if dispatch fails, because job->error is 0 */
1100 taskq_dispatch(system_taskq
, zvol_prefetch_minors_impl
, job
,
1103 zvol_add_clones(dsname
, minors_list
);
1105 if (snapdev
== ZFS_SNAPDEV_VISIBLE
) {
1107 * traverse snapshots only, do not traverse children,
1108 * and skip the 'dsname'
1110 (void) dmu_objset_find(dsname
,
1111 zvol_create_snap_minor_cb
, (void *)job
,
1115 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n",
1123 * Create minors for the specified dataset, including children and snapshots.
1124 * Pay attention to the 'snapdev' property and iterate over the snapshots
1125 * only if they are 'visible'. This approach allows one to assure that the
1126 * snapshot metadata is read from disk only if it is needed.
1128 * The name can represent a dataset to be recursively scanned for zvols and
1129 * their snapshots, or a single zvol snapshot. If the name represents a
1130 * dataset, the scan is performed in two nested stages:
1131 * - scan the dataset for zvols, and
1132 * - for each zvol, create a minor node, then check if the zvol's snapshots
1133 * are 'visible', and only then iterate over the snapshots if needed
1135 * If the name represents a snapshot, a check is performed if the snapshot is
1136 * 'visible' (which also verifies that the parent is a zvol), and if so,
1137 * a minor node for that snapshot is created.
1140 zvol_create_minors_recursive(const char *name
)
1145 if (zvol_inhibit_dev
)
1149 * This is the list for prefetch jobs. Whenever we found a match
1150 * during dmu_objset_find, we insert a minors_job to the list and do
1151 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need
1152 * any lock because all list operation is done on the current thread.
1154 * We will use this list to do zvol_os_create_minor after prefetch
1155 * so we don't have to traverse using dmu_objset_find again.
1157 list_create(&minors_list
, sizeof (minors_job_t
),
1158 offsetof(minors_job_t
, link
));
1161 if (strchr(name
, '@') != NULL
) {
1164 int error
= dsl_prop_get_integer(name
, "snapdev",
1167 if (error
== 0 && snapdev
== ZFS_SNAPDEV_VISIBLE
)
1168 (void) zvol_os_create_minor(name
);
1170 fstrans_cookie_t cookie
= spl_fstrans_mark();
1171 (void) dmu_objset_find(name
, zvol_create_minors_cb
,
1172 &minors_list
, DS_FIND_CHILDREN
);
1173 spl_fstrans_unmark(cookie
);
1176 taskq_wait_outstanding(system_taskq
, 0);
1179 * Prefetch is completed, we can do zvol_os_create_minor
1182 while ((job
= list_remove_head(&minors_list
)) != NULL
) {
1184 (void) zvol_os_create_minor(job
->name
);
1185 kmem_strfree(job
->name
);
1186 kmem_free(job
, sizeof (minors_job_t
));
1189 list_destroy(&minors_list
);
1193 zvol_create_minor(const char *name
)
1196 * Note: the dsl_pool_config_lock must not be held.
1197 * Minor node creation needs to obtain the zvol_state_lock.
1198 * zvol_open() obtains the zvol_state_lock and then the dsl pool
1199 * config lock. Therefore, we can't have the config lock now if
1200 * we are going to wait for the zvol_state_lock, because it
1201 * would be a lock order inversion which could lead to deadlock.
1204 if (zvol_inhibit_dev
)
1207 if (strchr(name
, '@') != NULL
) {
1210 int error
= dsl_prop_get_integer(name
,
1211 "snapdev", &snapdev
, NULL
);
1213 if (error
== 0 && snapdev
== ZFS_SNAPDEV_VISIBLE
)
1214 (void) zvol_os_create_minor(name
);
1216 (void) zvol_os_create_minor(name
);
1221 * Remove minors for specified dataset including children and snapshots.
1225 zvol_free_task(void *arg
)
1231 zvol_remove_minors_impl(const char *name
)
1233 zvol_state_t
*zv
, *zv_next
;
1234 int namelen
= ((name
) ? strlen(name
) : 0);
1238 if (zvol_inhibit_dev
)
1241 list_create(&free_list
, sizeof (zvol_state_t
),
1242 offsetof(zvol_state_t
, zv_next
));
1244 rw_enter(&zvol_state_lock
, RW_WRITER
);
1246 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
1247 zv_next
= list_next(&zvol_state_list
, zv
);
1249 mutex_enter(&zv
->zv_state_lock
);
1250 if (name
== NULL
|| strcmp(zv
->zv_name
, name
) == 0 ||
1251 (strncmp(zv
->zv_name
, name
, namelen
) == 0 &&
1252 (zv
->zv_name
[namelen
] == '/' ||
1253 zv
->zv_name
[namelen
] == '@'))) {
1255 * By holding zv_state_lock here, we guarantee that no
1256 * one is currently using this zv
1259 /* If in use, leave alone */
1260 if (zv
->zv_open_count
> 0 ||
1261 atomic_read(&zv
->zv_suspend_ref
)) {
1262 mutex_exit(&zv
->zv_state_lock
);
1269 * Cleared while holding zvol_state_lock as a writer
1270 * which will prevent zvol_open() from opening it.
1272 zvol_os_clear_private(zv
);
1274 /* Drop zv_state_lock before zvol_free() */
1275 mutex_exit(&zv
->zv_state_lock
);
1277 /* Try parallel zv_free, if failed do it in place */
1278 t
= taskq_dispatch(system_taskq
, zvol_free_task
, zv
,
1280 if (t
== TASKQID_INVALID
)
1281 list_insert_head(&free_list
, zv
);
1283 mutex_exit(&zv
->zv_state_lock
);
1286 rw_exit(&zvol_state_lock
);
1288 /* Drop zvol_state_lock before calling zvol_free() */
1289 while ((zv
= list_remove_head(&free_list
)) != NULL
)
1293 /* Remove minor for this specific volume only */
1295 zvol_remove_minor_impl(const char *name
)
1297 zvol_state_t
*zv
= NULL
, *zv_next
;
1299 if (zvol_inhibit_dev
)
1302 rw_enter(&zvol_state_lock
, RW_WRITER
);
1304 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
1305 zv_next
= list_next(&zvol_state_list
, zv
);
1307 mutex_enter(&zv
->zv_state_lock
);
1308 if (strcmp(zv
->zv_name
, name
) == 0) {
1310 * By holding zv_state_lock here, we guarantee that no
1311 * one is currently using this zv
1314 /* If in use, leave alone */
1315 if (zv
->zv_open_count
> 0 ||
1316 atomic_read(&zv
->zv_suspend_ref
)) {
1317 mutex_exit(&zv
->zv_state_lock
);
1322 zvol_os_clear_private(zv
);
1323 mutex_exit(&zv
->zv_state_lock
);
1326 mutex_exit(&zv
->zv_state_lock
);
1330 /* Drop zvol_state_lock before calling zvol_free() */
1331 rw_exit(&zvol_state_lock
);
1338 * Rename minors for specified dataset including children and snapshots.
1341 zvol_rename_minors_impl(const char *oldname
, const char *newname
)
1343 zvol_state_t
*zv
, *zv_next
;
1346 if (zvol_inhibit_dev
)
1349 oldnamelen
= strlen(oldname
);
1351 rw_enter(&zvol_state_lock
, RW_READER
);
1353 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
1354 zv_next
= list_next(&zvol_state_list
, zv
);
1356 mutex_enter(&zv
->zv_state_lock
);
1358 if (strcmp(zv
->zv_name
, oldname
) == 0) {
1359 zvol_os_rename_minor(zv
, newname
);
1360 } else if (strncmp(zv
->zv_name
, oldname
, oldnamelen
) == 0 &&
1361 (zv
->zv_name
[oldnamelen
] == '/' ||
1362 zv
->zv_name
[oldnamelen
] == '@')) {
1363 char *name
= kmem_asprintf("%s%c%s", newname
,
1364 zv
->zv_name
[oldnamelen
],
1365 zv
->zv_name
+ oldnamelen
+ 1);
1366 zvol_os_rename_minor(zv
, name
);
1370 mutex_exit(&zv
->zv_state_lock
);
1373 rw_exit(&zvol_state_lock
);
1376 typedef struct zvol_snapdev_cb_arg
{
1378 } zvol_snapdev_cb_arg_t
;
1381 zvol_set_snapdev_cb(const char *dsname
, void *param
)
1383 zvol_snapdev_cb_arg_t
*arg
= param
;
1385 if (strchr(dsname
, '@') == NULL
)
1388 switch (arg
->snapdev
) {
1389 case ZFS_SNAPDEV_VISIBLE
:
1390 (void) zvol_os_create_minor(dsname
);
1392 case ZFS_SNAPDEV_HIDDEN
:
1393 (void) zvol_remove_minor_impl(dsname
);
1401 zvol_set_snapdev_impl(char *name
, uint64_t snapdev
)
1403 zvol_snapdev_cb_arg_t arg
= {snapdev
};
1404 fstrans_cookie_t cookie
= spl_fstrans_mark();
1406 * The zvol_set_snapdev_sync() sets snapdev appropriately
1407 * in the dataset hierarchy. Here, we only scan snapshots.
1409 dmu_objset_find(name
, zvol_set_snapdev_cb
, &arg
, DS_FIND_SNAPSHOTS
);
1410 spl_fstrans_unmark(cookie
);
1414 zvol_set_volmode_impl(char *name
, uint64_t volmode
)
1416 fstrans_cookie_t cookie
;
1417 uint64_t old_volmode
;
1420 if (strchr(name
, '@') != NULL
)
1424 * It's unfortunate we need to remove minors before we create new ones:
1425 * this is necessary because our backing gendisk (zvol_state->zv_disk)
1426 * could be different when we set, for instance, volmode from "geom"
1427 * to "dev" (or vice versa).
1429 zv
= zvol_find_by_name(name
, RW_NONE
);
1430 if (zv
== NULL
&& volmode
== ZFS_VOLMODE_NONE
)
1433 old_volmode
= zv
->zv_volmode
;
1434 mutex_exit(&zv
->zv_state_lock
);
1435 if (old_volmode
== volmode
)
1437 zvol_wait_close(zv
);
1439 cookie
= spl_fstrans_mark();
1441 case ZFS_VOLMODE_NONE
:
1442 (void) zvol_remove_minor_impl(name
);
1444 case ZFS_VOLMODE_GEOM
:
1445 case ZFS_VOLMODE_DEV
:
1446 (void) zvol_remove_minor_impl(name
);
1447 (void) zvol_os_create_minor(name
);
1449 case ZFS_VOLMODE_DEFAULT
:
1450 (void) zvol_remove_minor_impl(name
);
1451 if (zvol_volmode
== ZFS_VOLMODE_NONE
)
1453 else /* if zvol_volmode is invalid defaults to "geom" */
1454 (void) zvol_os_create_minor(name
);
1457 spl_fstrans_unmark(cookie
);
1460 static zvol_task_t
*
1461 zvol_task_alloc(zvol_async_op_t op
, const char *name1
, const char *name2
,
1466 /* Never allow tasks on hidden names. */
1467 if (name1
[0] == '$')
1470 task
= kmem_zalloc(sizeof (zvol_task_t
), KM_SLEEP
);
1472 task
->value
= value
;
1474 strlcpy(task
->name1
, name1
, sizeof (task
->name1
));
1476 strlcpy(task
->name2
, name2
, sizeof (task
->name2
));
1482 zvol_task_free(zvol_task_t
*task
)
1484 kmem_free(task
, sizeof (zvol_task_t
));
1488 * The worker thread function performed asynchronously.
1491 zvol_task_cb(void *arg
)
1493 zvol_task_t
*task
= arg
;
1496 case ZVOL_ASYNC_REMOVE_MINORS
:
1497 zvol_remove_minors_impl(task
->name1
);
1499 case ZVOL_ASYNC_RENAME_MINORS
:
1500 zvol_rename_minors_impl(task
->name1
, task
->name2
);
1502 case ZVOL_ASYNC_SET_SNAPDEV
:
1503 zvol_set_snapdev_impl(task
->name1
, task
->value
);
1505 case ZVOL_ASYNC_SET_VOLMODE
:
1506 zvol_set_volmode_impl(task
->name1
, task
->value
);
1513 zvol_task_free(task
);
1516 typedef struct zvol_set_prop_int_arg
{
1517 const char *zsda_name
;
1518 uint64_t zsda_value
;
1519 zprop_source_t zsda_source
;
1520 zfs_prop_t zsda_prop
;
1521 } zvol_set_prop_int_arg_t
;
1524 * Sanity check the dataset for safe use by the sync task. No additional
1525 * conditions are imposed.
1528 zvol_set_common_check(void *arg
, dmu_tx_t
*tx
)
1530 zvol_set_prop_int_arg_t
*zsda
= arg
;
1531 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1535 error
= dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
);
1539 dsl_dir_rele(dd
, FTAG
);
1545 zvol_set_common_sync_cb(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
1547 zvol_set_prop_int_arg_t
*zsda
= arg
;
1548 char dsname
[ZFS_MAX_DATASET_NAME_LEN
];
1552 const char *prop_name
= zfs_prop_to_name(zsda
->zsda_prop
);
1553 dsl_dataset_name(ds
, dsname
);
1555 if (dsl_prop_get_int_ds(ds
, prop_name
, &prop
) != 0)
1558 switch (zsda
->zsda_prop
) {
1559 case ZFS_PROP_VOLMODE
:
1560 task
= zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE
, dsname
,
1563 case ZFS_PROP_SNAPDEV
:
1564 task
= zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV
, dsname
,
1575 (void) taskq_dispatch(dp
->dp_spa
->spa_zvol_taskq
, zvol_task_cb
,
1581 * Traverse all child datasets and apply the property appropriately.
1582 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
1583 * dataset and read the effective "property" on every child in the callback
1584 * function: this is because the value is not guaranteed to be the same in the
1585 * whole dataset hierarchy.
1588 zvol_set_common_sync(void *arg
, dmu_tx_t
*tx
)
1590 zvol_set_prop_int_arg_t
*zsda
= arg
;
1591 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
1596 VERIFY0(dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
));
1598 error
= dsl_dataset_hold(dp
, zsda
->zsda_name
, FTAG
, &ds
);
1600 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(zsda
->zsda_prop
),
1601 zsda
->zsda_source
, sizeof (zsda
->zsda_value
), 1,
1602 &zsda
->zsda_value
, tx
);
1603 dsl_dataset_rele(ds
, FTAG
);
1606 dmu_objset_find_dp(dp
, dd
->dd_object
, zvol_set_common_sync_cb
,
1607 zsda
, DS_FIND_CHILDREN
);
1609 dsl_dir_rele(dd
, FTAG
);
1613 zvol_set_common(const char *ddname
, zfs_prop_t prop
, zprop_source_t source
,
1616 zvol_set_prop_int_arg_t zsda
;
1618 zsda
.zsda_name
= ddname
;
1619 zsda
.zsda_source
= source
;
1620 zsda
.zsda_value
= val
;
1621 zsda
.zsda_prop
= prop
;
1623 return (dsl_sync_task(ddname
, zvol_set_common_check
,
1624 zvol_set_common_sync
, &zsda
, 0, ZFS_SPACE_CHECK_NONE
));
1628 zvol_remove_minors(spa_t
*spa
, const char *name
, boolean_t async
)
1633 task
= zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS
, name
, NULL
, ~0ULL);
1637 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
1638 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
1639 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
1643 zvol_rename_minors(spa_t
*spa
, const char *name1
, const char *name2
,
1649 task
= zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS
, name1
, name2
, ~0ULL);
1653 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
1654 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
1655 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
1659 zvol_is_zvol(const char *name
)
1662 return (zvol_os_is_zvol(name
));
1666 zvol_init_impl(void)
1670 list_create(&zvol_state_list
, sizeof (zvol_state_t
),
1671 offsetof(zvol_state_t
, zv_next
));
1672 rw_init(&zvol_state_lock
, NULL
, RW_DEFAULT
, NULL
);
1674 zvol_htable
= kmem_alloc(ZVOL_HT_SIZE
* sizeof (struct hlist_head
),
1676 for (i
= 0; i
< ZVOL_HT_SIZE
; i
++)
1677 INIT_HLIST_HEAD(&zvol_htable
[i
]);
1683 zvol_fini_impl(void)
1685 zvol_remove_minors_impl(NULL
);
1688 * The call to "zvol_remove_minors_impl" may dispatch entries to
1689 * the system_taskq, but it doesn't wait for those entries to
1690 * complete before it returns. Thus, we must wait for all of the
1691 * removals to finish, before we can continue.
1693 taskq_wait_outstanding(system_taskq
, 0);
1695 kmem_free(zvol_htable
, ZVOL_HT_SIZE
* sizeof (struct hlist_head
));
1696 list_destroy(&zvol_state_list
);
1697 rw_destroy(&zvol_state_lock
);