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]
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, 2017 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.
71 * It is also worth keeping in mind that once add_disk() is called, the zvol is
72 * announced to the world, and zvol_open()/zvol_release() can be called at any
73 * time. Incidentally, add_disk() itself calls zvol_open()->zvol_first_open()
74 * and zvol_release()->zvol_last_close() directly as well.
78 #include <sys/dmu_traverse.h>
79 #include <sys/dsl_dataset.h>
80 #include <sys/dsl_prop.h>
81 #include <sys/dsl_dir.h>
83 #include <sys/zfeature.h>
84 #include <sys/zil_impl.h>
85 #include <sys/dmu_tx.h>
87 #include <sys/zfs_rlock.h>
88 #include <sys/zfs_znode.h>
89 #include <sys/spa_impl.h>
91 #include <linux/blkdev_compat.h>
93 unsigned int zvol_inhibit_dev
= 0;
94 unsigned int zvol_major
= ZVOL_MAJOR
;
95 unsigned int zvol_threads
= 32;
96 unsigned int zvol_request_sync
= 0;
97 unsigned int zvol_prefetch_bytes
= (128 * 1024);
98 unsigned long zvol_max_discard_blocks
= 16384;
99 unsigned int zvol_volmode
= ZFS_VOLMODE_GEOM
;
101 static taskq_t
*zvol_taskq
;
102 static kmutex_t zvol_state_lock
;
103 static list_t zvol_state_list
;
105 #define ZVOL_HT_SIZE 1024
106 static struct hlist_head
*zvol_htable
;
107 #define ZVOL_HT_HEAD(hash) (&zvol_htable[(hash) & (ZVOL_HT_SIZE-1)])
109 static struct ida zvol_ida
;
112 * The in-core state of each volume.
115 char zv_name
[MAXNAMELEN
]; /* name */
116 uint64_t zv_volsize
; /* advertised space */
117 uint64_t zv_volblocksize
; /* volume block size */
118 objset_t
*zv_objset
; /* objset handle */
119 uint32_t zv_flags
; /* ZVOL_* flags */
120 uint32_t zv_open_count
; /* open counts */
121 uint32_t zv_changed
; /* disk changed */
122 zilog_t
*zv_zilog
; /* ZIL handle */
123 zfs_rlock_t zv_range_lock
; /* range lock */
124 dnode_t
*zv_dn
; /* dnode hold */
125 dev_t zv_dev
; /* device id */
126 struct gendisk
*zv_disk
; /* generic disk */
127 struct request_queue
*zv_queue
; /* request queue */
128 list_node_t zv_next
; /* next zvol_state_t linkage */
129 uint64_t zv_hash
; /* name hash */
130 struct hlist_node zv_hlink
; /* hash link */
131 kmutex_t zv_state_lock
; /* protects zvol_state_t */
132 atomic_t zv_suspend_ref
; /* refcount for suspend */
133 krwlock_t zv_suspend_lock
; /* suspend lock */
137 ZVOL_ASYNC_CREATE_MINORS
,
138 ZVOL_ASYNC_REMOVE_MINORS
,
139 ZVOL_ASYNC_RENAME_MINORS
,
140 ZVOL_ASYNC_SET_SNAPDEV
,
141 ZVOL_ASYNC_SET_VOLMODE
,
147 char pool
[MAXNAMELEN
];
148 char name1
[MAXNAMELEN
];
149 char name2
[MAXNAMELEN
];
150 zprop_source_t source
;
154 #define ZVOL_RDONLY 0x1
157 zvol_name_hash(const char *name
)
160 uint64_t crc
= -1ULL;
161 uint8_t *p
= (uint8_t *)name
;
162 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
163 for (i
= 0; i
< MAXNAMELEN
- 1 && *p
; i
++, p
++) {
164 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (*p
)) & 0xFF];
170 * Find a zvol_state_t given the full major+minor dev_t. If found,
171 * return with zv_state_lock taken, otherwise, return (NULL) without
172 * taking zv_state_lock.
174 static zvol_state_t
*
175 zvol_find_by_dev(dev_t dev
)
179 mutex_enter(&zvol_state_lock
);
180 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
;
181 zv
= list_next(&zvol_state_list
, zv
)) {
182 mutex_enter(&zv
->zv_state_lock
);
183 if (zv
->zv_dev
== dev
) {
184 mutex_exit(&zvol_state_lock
);
187 mutex_exit(&zv
->zv_state_lock
);
189 mutex_exit(&zvol_state_lock
);
195 * Find a zvol_state_t given the name and hash generated by zvol_name_hash.
196 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
197 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
198 * before zv_state_lock. The mode argument indicates the mode (including none)
199 * for zv_suspend_lock to be taken.
201 static zvol_state_t
*
202 zvol_find_by_name_hash(const char *name
, uint64_t hash
, int mode
)
205 struct hlist_node
*p
;
207 mutex_enter(&zvol_state_lock
);
208 hlist_for_each(p
, ZVOL_HT_HEAD(hash
)) {
209 zv
= hlist_entry(p
, zvol_state_t
, zv_hlink
);
210 mutex_enter(&zv
->zv_state_lock
);
211 if (zv
->zv_hash
== hash
&&
212 strncmp(zv
->zv_name
, name
, MAXNAMELEN
) == 0) {
214 * this is the right zvol, take the locks in the
217 if (mode
!= RW_NONE
&&
218 !rw_tryenter(&zv
->zv_suspend_lock
, mode
)) {
219 mutex_exit(&zv
->zv_state_lock
);
220 rw_enter(&zv
->zv_suspend_lock
, mode
);
221 mutex_enter(&zv
->zv_state_lock
);
223 * zvol cannot be renamed as we continue
224 * to hold zvol_state_lock
226 ASSERT(zv
->zv_hash
== hash
&&
227 strncmp(zv
->zv_name
, name
, MAXNAMELEN
)
230 mutex_exit(&zvol_state_lock
);
233 mutex_exit(&zv
->zv_state_lock
);
235 mutex_exit(&zvol_state_lock
);
241 * Find a zvol_state_t given the name.
242 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
243 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
244 * before zv_state_lock. The mode argument indicates the mode (including none)
245 * for zv_suspend_lock to be taken.
247 static zvol_state_t
*
248 zvol_find_by_name(const char *name
, int mode
)
250 return (zvol_find_by_name_hash(name
, zvol_name_hash(name
), mode
));
255 * Given a path, return TRUE if path is a ZVOL.
258 zvol_is_zvol(const char *device
)
260 struct block_device
*bdev
;
263 bdev
= vdev_lookup_bdev(device
);
267 major
= MAJOR(bdev
->bd_dev
);
270 if (major
== zvol_major
)
277 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
280 zvol_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
282 zfs_creat_t
*zct
= arg
;
283 nvlist_t
*nvprops
= zct
->zct_props
;
285 uint64_t volblocksize
, volsize
;
287 VERIFY(nvlist_lookup_uint64(nvprops
,
288 zfs_prop_to_name(ZFS_PROP_VOLSIZE
), &volsize
) == 0);
289 if (nvlist_lookup_uint64(nvprops
,
290 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
), &volblocksize
) != 0)
291 volblocksize
= zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE
);
294 * These properties must be removed from the list so the generic
295 * property setting step won't apply to them.
297 VERIFY(nvlist_remove_all(nvprops
,
298 zfs_prop_to_name(ZFS_PROP_VOLSIZE
)) == 0);
299 (void) nvlist_remove_all(nvprops
,
300 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
));
302 error
= dmu_object_claim(os
, ZVOL_OBJ
, DMU_OT_ZVOL
, volblocksize
,
306 error
= zap_create_claim(os
, ZVOL_ZAP_OBJ
, DMU_OT_ZVOL_PROP
,
310 error
= zap_update(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
, tx
);
315 * ZFS_IOC_OBJSET_STATS entry point.
318 zvol_get_stats(objset_t
*os
, nvlist_t
*nv
)
321 dmu_object_info_t
*doi
;
324 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &val
);
326 return (SET_ERROR(error
));
328 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_VOLSIZE
, val
);
329 doi
= kmem_alloc(sizeof (dmu_object_info_t
), KM_SLEEP
);
330 error
= dmu_object_info(os
, ZVOL_OBJ
, doi
);
333 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_VOLBLOCKSIZE
,
334 doi
->doi_data_block_size
);
337 kmem_free(doi
, sizeof (dmu_object_info_t
));
339 return (SET_ERROR(error
));
343 zvol_size_changed(zvol_state_t
*zv
, uint64_t volsize
)
345 struct block_device
*bdev
;
347 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
349 bdev
= bdget_disk(zv
->zv_disk
, 0);
353 set_capacity(zv
->zv_disk
, volsize
>> 9);
354 zv
->zv_volsize
= volsize
;
355 check_disk_size_change(zv
->zv_disk
, bdev
);
361 * Sanity check volume size.
364 zvol_check_volsize(uint64_t volsize
, uint64_t blocksize
)
367 return (SET_ERROR(EINVAL
));
369 if (volsize
% blocksize
!= 0)
370 return (SET_ERROR(EINVAL
));
373 if (volsize
- 1 > SPEC_MAXOFFSET_T
)
374 return (SET_ERROR(EOVERFLOW
));
380 * Ensure the zap is flushed then inform the VFS of the capacity change.
383 zvol_update_volsize(uint64_t volsize
, objset_t
*os
)
389 tx
= dmu_tx_create(os
);
390 dmu_tx_hold_zap(tx
, ZVOL_ZAP_OBJ
, TRUE
, NULL
);
391 dmu_tx_mark_netfree(tx
);
392 error
= dmu_tx_assign(tx
, TXG_WAIT
);
395 return (SET_ERROR(error
));
397 txg
= dmu_tx_get_txg(tx
);
399 error
= zap_update(os
, ZVOL_ZAP_OBJ
, "size", 8, 1,
403 txg_wait_synced(dmu_objset_pool(os
), txg
);
406 error
= dmu_free_long_range(os
,
407 ZVOL_OBJ
, volsize
, DMU_OBJECT_END
);
413 zvol_update_live_volsize(zvol_state_t
*zv
, uint64_t volsize
)
415 zvol_size_changed(zv
, volsize
);
418 * We should post a event here describing the expansion. However,
419 * the zfs_ereport_post() interface doesn't nicely support posting
420 * events for zvols, it assumes events relate to vdevs or zios.
427 * Set ZFS_PROP_VOLSIZE set entry point.
430 zvol_set_volsize(const char *name
, uint64_t volsize
)
432 zvol_state_t
*zv
= NULL
;
435 dmu_object_info_t
*doi
;
437 boolean_t owned
= B_FALSE
;
439 error
= dsl_prop_get_integer(name
,
440 zfs_prop_to_name(ZFS_PROP_READONLY
), &readonly
, NULL
);
442 return (SET_ERROR(error
));
444 return (SET_ERROR(EROFS
));
446 zv
= zvol_find_by_name(name
, RW_READER
);
448 ASSERT(zv
== NULL
|| (MUTEX_HELD(&zv
->zv_state_lock
) &&
449 RW_READ_HELD(&zv
->zv_suspend_lock
)));
451 if (zv
== NULL
|| zv
->zv_objset
== NULL
) {
453 rw_exit(&zv
->zv_suspend_lock
);
454 if ((error
= dmu_objset_own(name
, DMU_OST_ZVOL
, B_FALSE
, B_TRUE
,
457 mutex_exit(&zv
->zv_state_lock
);
458 return (SET_ERROR(error
));
467 doi
= kmem_alloc(sizeof (dmu_object_info_t
), KM_SLEEP
);
469 if ((error
= dmu_object_info(os
, ZVOL_OBJ
, doi
)) ||
470 (error
= zvol_check_volsize(volsize
, doi
->doi_data_block_size
)))
473 error
= zvol_update_volsize(volsize
, os
);
475 if (error
== 0 && zv
!= NULL
)
476 error
= zvol_update_live_volsize(zv
, volsize
);
478 kmem_free(doi
, sizeof (dmu_object_info_t
));
481 dmu_objset_disown(os
, B_TRUE
, FTAG
);
483 zv
->zv_objset
= NULL
;
485 rw_exit(&zv
->zv_suspend_lock
);
489 mutex_exit(&zv
->zv_state_lock
);
491 return (SET_ERROR(error
));
495 * Sanity check volume block size.
498 zvol_check_volblocksize(const char *name
, uint64_t volblocksize
)
500 /* Record sizes above 128k need the feature to be enabled */
501 if (volblocksize
> SPA_OLD_MAXBLOCKSIZE
) {
505 if ((error
= spa_open(name
, &spa
, FTAG
)) != 0)
508 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
509 spa_close(spa
, FTAG
);
510 return (SET_ERROR(ENOTSUP
));
514 * We don't allow setting the property above 1MB,
515 * unless the tunable has been changed.
517 if (volblocksize
> zfs_max_recordsize
)
518 return (SET_ERROR(EDOM
));
520 spa_close(spa
, FTAG
);
523 if (volblocksize
< SPA_MINBLOCKSIZE
||
524 volblocksize
> SPA_MAXBLOCKSIZE
||
526 return (SET_ERROR(EDOM
));
532 * Set ZFS_PROP_VOLBLOCKSIZE set entry point.
535 zvol_set_volblocksize(const char *name
, uint64_t volblocksize
)
541 zv
= zvol_find_by_name(name
, RW_READER
);
544 return (SET_ERROR(ENXIO
));
546 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
) &&
547 RW_READ_HELD(&zv
->zv_suspend_lock
));
549 if (zv
->zv_flags
& ZVOL_RDONLY
) {
550 mutex_exit(&zv
->zv_state_lock
);
551 rw_exit(&zv
->zv_suspend_lock
);
552 return (SET_ERROR(EROFS
));
555 tx
= dmu_tx_create(zv
->zv_objset
);
556 dmu_tx_hold_bonus(tx
, ZVOL_OBJ
);
557 error
= dmu_tx_assign(tx
, TXG_WAIT
);
561 error
= dmu_object_set_blocksize(zv
->zv_objset
, ZVOL_OBJ
,
562 volblocksize
, 0, tx
);
563 if (error
== ENOTSUP
)
564 error
= SET_ERROR(EBUSY
);
567 zv
->zv_volblocksize
= volblocksize
;
570 mutex_exit(&zv
->zv_state_lock
);
571 rw_exit(&zv
->zv_suspend_lock
);
573 return (SET_ERROR(error
));
577 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
578 * implement DKIOCFREE/free-long-range.
581 zvol_replay_truncate(zvol_state_t
*zv
, lr_truncate_t
*lr
, boolean_t byteswap
)
583 uint64_t offset
, length
;
586 byteswap_uint64_array(lr
, sizeof (*lr
));
588 offset
= lr
->lr_offset
;
589 length
= lr
->lr_length
;
591 return (dmu_free_long_range(zv
->zv_objset
, ZVOL_OBJ
, offset
, length
));
595 * Replay a TX_WRITE ZIL transaction that didn't get committed
596 * after a system failure
599 zvol_replay_write(zvol_state_t
*zv
, lr_write_t
*lr
, boolean_t byteswap
)
601 objset_t
*os
= zv
->zv_objset
;
602 char *data
= (char *)(lr
+ 1); /* data follows lr_write_t */
603 uint64_t offset
, length
;
608 byteswap_uint64_array(lr
, sizeof (*lr
));
610 offset
= lr
->lr_offset
;
611 length
= lr
->lr_length
;
613 /* If it's a dmu_sync() block, write the whole block */
614 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
615 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
616 if (length
< blocksize
) {
617 offset
-= offset
% blocksize
;
622 tx
= dmu_tx_create(os
);
623 dmu_tx_hold_write(tx
, ZVOL_OBJ
, offset
, length
);
624 error
= dmu_tx_assign(tx
, TXG_WAIT
);
628 dmu_write(os
, ZVOL_OBJ
, offset
, length
, data
, tx
);
636 zvol_replay_err(zvol_state_t
*zv
, lr_t
*lr
, boolean_t byteswap
)
638 return (SET_ERROR(ENOTSUP
));
642 * Callback vectors for replaying records.
643 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
645 zil_replay_func_t zvol_replay_vector
[TX_MAX_TYPE
] = {
646 (zil_replay_func_t
)zvol_replay_err
, /* no such transaction type */
647 (zil_replay_func_t
)zvol_replay_err
, /* TX_CREATE */
648 (zil_replay_func_t
)zvol_replay_err
, /* TX_MKDIR */
649 (zil_replay_func_t
)zvol_replay_err
, /* TX_MKXATTR */
650 (zil_replay_func_t
)zvol_replay_err
, /* TX_SYMLINK */
651 (zil_replay_func_t
)zvol_replay_err
, /* TX_REMOVE */
652 (zil_replay_func_t
)zvol_replay_err
, /* TX_RMDIR */
653 (zil_replay_func_t
)zvol_replay_err
, /* TX_LINK */
654 (zil_replay_func_t
)zvol_replay_err
, /* TX_RENAME */
655 (zil_replay_func_t
)zvol_replay_write
, /* TX_WRITE */
656 (zil_replay_func_t
)zvol_replay_truncate
, /* TX_TRUNCATE */
657 (zil_replay_func_t
)zvol_replay_err
, /* TX_SETATTR */
658 (zil_replay_func_t
)zvol_replay_err
, /* TX_ACL */
662 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
664 * We store data in the log buffers if it's small enough.
665 * Otherwise we will later flush the data out via dmu_sync().
667 ssize_t zvol_immediate_write_sz
= 32768;
670 zvol_log_write(zvol_state_t
*zv
, dmu_tx_t
*tx
, uint64_t offset
,
671 uint64_t size
, int sync
)
673 uint32_t blocksize
= zv
->zv_volblocksize
;
674 zilog_t
*zilog
= zv
->zv_zilog
;
675 itx_wr_state_t write_state
;
677 if (zil_replaying(zilog
, tx
))
680 if (zilog
->zl_logbias
== ZFS_LOGBIAS_THROUGHPUT
)
681 write_state
= WR_INDIRECT
;
682 else if (!spa_has_slogs(zilog
->zl_spa
) &&
683 size
>= blocksize
&& blocksize
> zvol_immediate_write_sz
)
684 write_state
= WR_INDIRECT
;
686 write_state
= WR_COPIED
;
688 write_state
= WR_NEED_COPY
;
693 itx_wr_state_t wr_state
= write_state
;
696 if (wr_state
== WR_COPIED
&& size
> ZIL_MAX_COPIED_DATA
)
697 wr_state
= WR_NEED_COPY
;
698 else if (wr_state
== WR_INDIRECT
)
699 len
= MIN(blocksize
- P2PHASE(offset
, blocksize
), size
);
701 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
) +
702 (wr_state
== WR_COPIED
? len
: 0));
703 lr
= (lr_write_t
*)&itx
->itx_lr
;
704 if (wr_state
== WR_COPIED
&& dmu_read_by_dnode(zv
->zv_dn
,
705 offset
, len
, lr
+1, DMU_READ_NO_PREFETCH
) != 0) {
706 zil_itx_destroy(itx
);
707 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
708 lr
= (lr_write_t
*)&itx
->itx_lr
;
709 wr_state
= WR_NEED_COPY
;
712 itx
->itx_wr_state
= wr_state
;
713 lr
->lr_foid
= ZVOL_OBJ
;
714 lr
->lr_offset
= offset
;
717 BP_ZERO(&lr
->lr_blkptr
);
719 itx
->itx_private
= zv
;
720 itx
->itx_sync
= sync
;
722 (void) zil_itx_assign(zilog
, itx
, tx
);
729 typedef struct zv_request
{
736 uio_from_bio(uio_t
*uio
, struct bio
*bio
)
738 uio
->uio_bvec
= &bio
->bi_io_vec
[BIO_BI_IDX(bio
)];
739 uio
->uio_skip
= BIO_BI_SKIP(bio
);
740 uio
->uio_resid
= BIO_BI_SIZE(bio
);
741 uio
->uio_iovcnt
= bio
->bi_vcnt
- BIO_BI_IDX(bio
);
742 uio
->uio_loffset
= BIO_BI_SECTOR(bio
) << 9;
743 uio
->uio_limit
= MAXOFFSET_T
;
744 uio
->uio_segflg
= UIO_BVEC
;
748 zvol_write(void *arg
)
750 zv_request_t
*zvr
= arg
;
751 struct bio
*bio
= zvr
->bio
;
753 zvol_state_t
*zv
= zvr
->zv
;
754 uint64_t volsize
= zv
->zv_volsize
;
757 unsigned long start_jif
;
759 uio_from_bio(&uio
, bio
);
761 ASSERT(zv
&& zv
->zv_open_count
> 0);
764 generic_start_io_acct(WRITE
, bio_sectors(bio
), &zv
->zv_disk
->part0
);
766 sync
= bio_is_fua(bio
) || zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
768 while (uio
.uio_resid
> 0 && uio
.uio_loffset
< volsize
) {
769 uint64_t bytes
= MIN(uio
.uio_resid
, DMU_MAX_ACCESS
>> 1);
770 uint64_t off
= uio
.uio_loffset
;
771 dmu_tx_t
*tx
= dmu_tx_create(zv
->zv_objset
);
773 if (bytes
> volsize
- off
) /* don't write past the end */
774 bytes
= volsize
- off
;
776 dmu_tx_hold_write(tx
, ZVOL_OBJ
, off
, bytes
);
778 /* This will only fail for ENOSPC */
779 error
= dmu_tx_assign(tx
, TXG_WAIT
);
784 error
= dmu_write_uio_dnode(zv
->zv_dn
, &uio
, bytes
, tx
);
786 zvol_log_write(zv
, tx
, off
, bytes
, sync
);
792 zfs_range_unlock(zvr
->rl
);
794 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
796 rw_exit(&zv
->zv_suspend_lock
);
797 generic_end_io_acct(WRITE
, &zv
->zv_disk
->part0
, start_jif
);
798 BIO_END_IO(bio
, -error
);
799 kmem_free(zvr
, sizeof (zv_request_t
));
803 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
806 zvol_log_truncate(zvol_state_t
*zv
, dmu_tx_t
*tx
, uint64_t off
, uint64_t len
,
811 zilog_t
*zilog
= zv
->zv_zilog
;
813 if (zil_replaying(zilog
, tx
))
816 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
817 lr
= (lr_truncate_t
*)&itx
->itx_lr
;
818 lr
->lr_foid
= ZVOL_OBJ
;
822 itx
->itx_sync
= sync
;
823 zil_itx_assign(zilog
, itx
, tx
);
827 zvol_discard(void *arg
)
829 zv_request_t
*zvr
= arg
;
830 struct bio
*bio
= zvr
->bio
;
831 zvol_state_t
*zv
= zvr
->zv
;
832 uint64_t start
= BIO_BI_SECTOR(bio
) << 9;
833 uint64_t size
= BIO_BI_SIZE(bio
);
834 uint64_t end
= start
+ size
;
838 unsigned long start_jif
;
840 ASSERT(zv
&& zv
->zv_open_count
> 0);
843 generic_start_io_acct(WRITE
, bio_sectors(bio
), &zv
->zv_disk
->part0
);
845 sync
= bio_is_fua(bio
) || zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
847 if (end
> zv
->zv_volsize
) {
848 error
= SET_ERROR(EIO
);
853 * Align the request to volume block boundaries when a secure erase is
854 * not required. This will prevent dnode_free_range() from zeroing out
855 * the unaligned parts which is slow (read-modify-write) and useless
856 * since we are not freeing any space by doing so.
858 if (!bio_is_secure_erase(bio
)) {
859 start
= P2ROUNDUP(start
, zv
->zv_volblocksize
);
860 end
= P2ALIGN(end
, zv
->zv_volblocksize
);
867 tx
= dmu_tx_create(zv
->zv_objset
);
868 dmu_tx_mark_netfree(tx
);
869 error
= dmu_tx_assign(tx
, TXG_WAIT
);
873 zvol_log_truncate(zv
, tx
, start
, size
, B_TRUE
);
875 error
= dmu_free_long_range(zv
->zv_objset
,
876 ZVOL_OBJ
, start
, size
);
879 zfs_range_unlock(zvr
->rl
);
880 if (error
== 0 && sync
)
881 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
883 rw_exit(&zv
->zv_suspend_lock
);
884 generic_end_io_acct(WRITE
, &zv
->zv_disk
->part0
, start_jif
);
885 BIO_END_IO(bio
, -error
);
886 kmem_free(zvr
, sizeof (zv_request_t
));
892 zv_request_t
*zvr
= arg
;
893 struct bio
*bio
= zvr
->bio
;
895 zvol_state_t
*zv
= zvr
->zv
;
896 uint64_t volsize
= zv
->zv_volsize
;
898 unsigned long start_jif
;
900 uio_from_bio(&uio
, bio
);
902 ASSERT(zv
&& zv
->zv_open_count
> 0);
905 generic_start_io_acct(READ
, bio_sectors(bio
), &zv
->zv_disk
->part0
);
907 while (uio
.uio_resid
> 0 && uio
.uio_loffset
< volsize
) {
908 uint64_t bytes
= MIN(uio
.uio_resid
, DMU_MAX_ACCESS
>> 1);
910 /* don't read past the end */
911 if (bytes
> volsize
- uio
.uio_loffset
)
912 bytes
= volsize
- uio
.uio_loffset
;
914 error
= dmu_read_uio_dnode(zv
->zv_dn
, &uio
, bytes
);
916 /* convert checksum errors into IO errors */
918 error
= SET_ERROR(EIO
);
922 zfs_range_unlock(zvr
->rl
);
924 rw_exit(&zv
->zv_suspend_lock
);
925 generic_end_io_acct(READ
, &zv
->zv_disk
->part0
, start_jif
);
926 BIO_END_IO(bio
, -error
);
927 kmem_free(zvr
, sizeof (zv_request_t
));
930 static MAKE_REQUEST_FN_RET
931 zvol_request(struct request_queue
*q
, struct bio
*bio
)
933 zvol_state_t
*zv
= q
->queuedata
;
934 fstrans_cookie_t cookie
= spl_fstrans_mark();
935 uint64_t offset
= BIO_BI_SECTOR(bio
) << 9;
936 uint64_t size
= BIO_BI_SIZE(bio
);
937 int rw
= bio_data_dir(bio
);
940 if (bio_has_data(bio
) && offset
+ size
> zv
->zv_volsize
) {
942 "%s: bad access: offset=%llu, size=%lu\n",
943 zv
->zv_disk
->disk_name
,
944 (long long unsigned)offset
,
945 (long unsigned)size
);
947 BIO_END_IO(bio
, -SET_ERROR(EIO
));
952 boolean_t need_sync
= B_FALSE
;
954 if (unlikely(zv
->zv_flags
& ZVOL_RDONLY
)) {
955 BIO_END_IO(bio
, -SET_ERROR(EROFS
));
960 * To be released in the I/O function. See the comment on
961 * zfs_range_lock below.
963 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
965 /* bio marked as FLUSH need to flush before write */
966 if (bio_is_flush(bio
))
967 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
969 /* Some requests are just for flush and nothing else. */
971 rw_exit(&zv
->zv_suspend_lock
);
976 zvr
= kmem_alloc(sizeof (zv_request_t
), KM_SLEEP
);
981 * To be released in the I/O function. Since the I/O functions
982 * are asynchronous, we take it here synchronously to make
983 * sure overlapped I/Os are properly ordered.
985 zvr
->rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
988 * Sync writes and discards execute zil_commit() which may need
989 * to take a RL_READER lock on the whole block being modified
990 * via its zillog->zl_get_data(): to avoid circular dependency
991 * issues with taskq threads execute these requests
992 * synchronously here in zvol_request().
994 need_sync
= bio_is_fua(bio
) ||
995 zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
996 if (bio_is_discard(bio
) || bio_is_secure_erase(bio
)) {
997 if (zvol_request_sync
|| need_sync
||
998 taskq_dispatch(zvol_taskq
, zvol_discard
, zvr
,
999 TQ_SLEEP
) == TASKQID_INVALID
)
1002 if (zvol_request_sync
|| need_sync
||
1003 taskq_dispatch(zvol_taskq
, zvol_write
, zvr
,
1004 TQ_SLEEP
) == TASKQID_INVALID
)
1008 zvr
= kmem_alloc(sizeof (zv_request_t
), KM_SLEEP
);
1012 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1014 zvr
->rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
1016 if (zvol_request_sync
|| taskq_dispatch(zvol_taskq
,
1017 zvol_read
, zvr
, TQ_SLEEP
) == TASKQID_INVALID
)
1022 spl_fstrans_unmark(cookie
);
1023 #ifdef HAVE_MAKE_REQUEST_FN_RET_INT
1025 #elif defined(HAVE_MAKE_REQUEST_FN_RET_QC)
1026 return (BLK_QC_T_NONE
);
1031 zvol_get_done(zgd_t
*zgd
, int error
)
1034 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1036 zfs_range_unlock(zgd
->zgd_rl
);
1038 if (error
== 0 && zgd
->zgd_bp
)
1039 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
1041 kmem_free(zgd
, sizeof (zgd_t
));
1045 * Get data to generate a TX_WRITE intent log record.
1048 zvol_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
1050 zvol_state_t
*zv
= arg
;
1051 uint64_t offset
= lr
->lr_offset
;
1052 uint64_t size
= lr
->lr_length
;
1057 ASSERT(zio
!= NULL
);
1060 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1061 zgd
->zgd_zilog
= zv
->zv_zilog
;
1064 * Write records come in two flavors: immediate and indirect.
1065 * For small writes it's cheaper to store the data with the
1066 * log record (immediate); for large writes it's cheaper to
1067 * sync the data and get a pointer to it (indirect) so that
1068 * we don't have to write the data twice.
1070 if (buf
!= NULL
) { /* immediate write */
1071 zgd
->zgd_rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
1073 error
= dmu_read_by_dnode(zv
->zv_dn
, offset
, size
, buf
,
1074 DMU_READ_NO_PREFETCH
);
1075 } else { /* indirect write */
1077 * Have to lock the whole block to ensure when it's written out
1078 * and its checksum is being calculated that no one can change
1079 * the data. Contrarily to zfs_get_data we need not re-check
1080 * blocksize after we get the lock because it cannot be changed.
1082 size
= zv
->zv_volblocksize
;
1083 offset
= P2ALIGN_TYPED(offset
, size
, uint64_t);
1084 zgd
->zgd_rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
1086 error
= dmu_buf_hold_by_dnode(zv
->zv_dn
, offset
, zgd
, &db
,
1087 DMU_READ_NO_PREFETCH
);
1089 blkptr_t
*bp
= &lr
->lr_blkptr
;
1095 ASSERT(db
->db_offset
== offset
);
1096 ASSERT(db
->db_size
== size
);
1098 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1099 zvol_get_done
, zgd
);
1106 zvol_get_done(zgd
, error
);
1108 return (SET_ERROR(error
));
1112 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable.
1115 zvol_insert(zvol_state_t
*zv
)
1117 ASSERT(MUTEX_HELD(&zvol_state_lock
));
1118 ASSERT3U(MINOR(zv
->zv_dev
) & ZVOL_MINOR_MASK
, ==, 0);
1119 list_insert_head(&zvol_state_list
, zv
);
1120 hlist_add_head(&zv
->zv_hlink
, ZVOL_HT_HEAD(zv
->zv_hash
));
1124 * Simply remove the zvol from to list of zvols.
1127 zvol_remove(zvol_state_t
*zv
)
1129 ASSERT(MUTEX_HELD(&zvol_state_lock
));
1130 list_remove(&zvol_state_list
, zv
);
1131 hlist_del(&zv
->zv_hlink
);
1135 * Setup zv after we just own the zv->objset
1138 zvol_setup_zv(zvol_state_t
*zv
)
1143 objset_t
*os
= zv
->zv_objset
;
1145 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
) &&
1146 RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1148 error
= dsl_prop_get_integer(zv
->zv_name
, "readonly", &ro
, NULL
);
1150 return (SET_ERROR(error
));
1152 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
1154 return (SET_ERROR(error
));
1156 error
= dnode_hold(os
, ZVOL_OBJ
, FTAG
, &zv
->zv_dn
);
1158 return (SET_ERROR(error
));
1160 set_capacity(zv
->zv_disk
, volsize
>> 9);
1161 zv
->zv_volsize
= volsize
;
1162 zv
->zv_zilog
= zil_open(os
, zvol_get_data
);
1164 if (ro
|| dmu_objset_is_snapshot(os
) ||
1165 !spa_writeable(dmu_objset_spa(os
))) {
1166 set_disk_ro(zv
->zv_disk
, 1);
1167 zv
->zv_flags
|= ZVOL_RDONLY
;
1169 set_disk_ro(zv
->zv_disk
, 0);
1170 zv
->zv_flags
&= ~ZVOL_RDONLY
;
1176 * Shutdown every zv_objset related stuff except zv_objset itself.
1177 * The is the reverse of zvol_setup_zv.
1180 zvol_shutdown_zv(zvol_state_t
*zv
)
1182 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
) &&
1183 RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1185 zil_close(zv
->zv_zilog
);
1186 zv
->zv_zilog
= NULL
;
1188 dnode_rele(zv
->zv_dn
, FTAG
);
1194 if (dsl_dataset_is_dirty(dmu_objset_ds(zv
->zv_objset
)) &&
1195 !(zv
->zv_flags
& ZVOL_RDONLY
))
1196 txg_wait_synced(dmu_objset_pool(zv
->zv_objset
), 0);
1197 (void) dmu_objset_evict_dbufs(zv
->zv_objset
);
1201 * return the proper tag for rollback and recv
1204 zvol_tag(zvol_state_t
*zv
)
1206 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1207 return (zv
->zv_open_count
> 0 ? zv
: NULL
);
1211 * Suspend the zvol for recv and rollback.
1214 zvol_suspend(const char *name
)
1218 zv
= zvol_find_by_name(name
, RW_WRITER
);
1223 /* block all I/O, release in zvol_resume. */
1224 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
) &&
1225 RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1227 atomic_inc(&zv
->zv_suspend_ref
);
1229 if (zv
->zv_open_count
> 0)
1230 zvol_shutdown_zv(zv
);
1233 * do not hold zv_state_lock across suspend/resume to
1234 * avoid locking up zvol lookups
1236 mutex_exit(&zv
->zv_state_lock
);
1238 /* zv_suspend_lock is released in zvol_resume() */
1243 zvol_resume(zvol_state_t
*zv
)
1247 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1249 mutex_enter(&zv
->zv_state_lock
);
1251 if (zv
->zv_open_count
> 0) {
1252 VERIFY0(dmu_objset_hold(zv
->zv_name
, zv
, &zv
->zv_objset
));
1253 VERIFY3P(zv
->zv_objset
->os_dsl_dataset
->ds_owner
, ==, zv
);
1254 VERIFY(dsl_dataset_long_held(zv
->zv_objset
->os_dsl_dataset
));
1255 dmu_objset_rele(zv
->zv_objset
, zv
);
1257 error
= zvol_setup_zv(zv
);
1260 mutex_exit(&zv
->zv_state_lock
);
1262 rw_exit(&zv
->zv_suspend_lock
);
1264 * We need this because we don't hold zvol_state_lock while releasing
1265 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check
1266 * zv_suspend_lock to determine it is safe to free because rwlock is
1267 * not inherent atomic.
1269 atomic_dec(&zv
->zv_suspend_ref
);
1271 return (SET_ERROR(error
));
1275 zvol_first_open(zvol_state_t
*zv
)
1278 int error
, locked
= 0;
1280 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
1281 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1284 * In all other cases the spa_namespace_lock is taken before the
1285 * bdev->bd_mutex lock. But in this case the Linux __blkdev_get()
1286 * function calls fops->open() with the bdev->bd_mutex lock held.
1287 * This deadlock can be easily observed with zvols used as vdevs.
1289 * To avoid a potential lock inversion deadlock we preemptively
1290 * try to take the spa_namespace_lock(). Normally it will not
1291 * be contended and this is safe because spa_open_common() handles
1292 * the case where the caller already holds the spa_namespace_lock.
1294 * When it is contended we risk a lock inversion if we were to
1295 * block waiting for the lock. Luckily, the __blkdev_get()
1296 * function allows us to return -ERESTARTSYS which will result in
1297 * bdev->bd_mutex being dropped, reacquired, and fops->open() being
1298 * called again. This process can be repeated safely until both
1299 * locks are acquired.
1301 if (!mutex_owned(&spa_namespace_lock
)) {
1302 locked
= mutex_tryenter(&spa_namespace_lock
);
1304 return (-SET_ERROR(ERESTARTSYS
));
1307 /* lie and say we're read-only */
1308 error
= dmu_objset_own(zv
->zv_name
, DMU_OST_ZVOL
, 1, 1, zv
, &os
);
1314 error
= zvol_setup_zv(zv
);
1317 dmu_objset_disown(os
, 1, zv
);
1318 zv
->zv_objset
= NULL
;
1323 mutex_exit(&spa_namespace_lock
);
1324 return (SET_ERROR(-error
));
1328 zvol_last_close(zvol_state_t
*zv
)
1330 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
1331 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1333 zvol_shutdown_zv(zv
);
1335 dmu_objset_disown(zv
->zv_objset
, 1, zv
);
1336 zv
->zv_objset
= NULL
;
1340 zvol_open(struct block_device
*bdev
, fmode_t flag
)
1344 boolean_t drop_suspend
= B_TRUE
;
1346 ASSERT(!MUTEX_HELD(&zvol_state_lock
));
1348 mutex_enter(&zvol_state_lock
);
1350 * Obtain a copy of private_data under the zvol_state_lock to make
1351 * sure that either the result of zvol free code path setting
1352 * bdev->bd_disk->private_data to NULL is observed, or zvol_free()
1353 * is not called on this zv because of the positive zv_open_count.
1355 zv
= bdev
->bd_disk
->private_data
;
1357 mutex_exit(&zvol_state_lock
);
1358 return (SET_ERROR(-ENXIO
));
1361 mutex_enter(&zv
->zv_state_lock
);
1363 * make sure zvol is not suspended during first open
1364 * (hold zv_suspend_lock) and respect proper lock acquisition
1365 * ordering - zv_suspend_lock before zv_state_lock
1367 if (zv
->zv_open_count
== 0) {
1368 if (!rw_tryenter(&zv
->zv_suspend_lock
, RW_READER
)) {
1369 mutex_exit(&zv
->zv_state_lock
);
1370 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1371 mutex_enter(&zv
->zv_state_lock
);
1372 /* check to see if zv_suspend_lock is needed */
1373 if (zv
->zv_open_count
!= 0) {
1374 rw_exit(&zv
->zv_suspend_lock
);
1375 drop_suspend
= B_FALSE
;
1379 drop_suspend
= B_FALSE
;
1381 mutex_exit(&zvol_state_lock
);
1383 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1384 ASSERT(zv
->zv_open_count
!= 0 || RW_READ_HELD(&zv
->zv_suspend_lock
));
1386 if (zv
->zv_open_count
== 0) {
1387 error
= zvol_first_open(zv
);
1392 if ((flag
& FMODE_WRITE
) && (zv
->zv_flags
& ZVOL_RDONLY
)) {
1394 goto out_open_count
;
1397 zv
->zv_open_count
++;
1399 check_disk_change(bdev
);
1402 if (zv
->zv_open_count
== 0)
1403 zvol_last_close(zv
);
1405 mutex_exit(&zv
->zv_state_lock
);
1407 rw_exit(&zv
->zv_suspend_lock
);
1408 if (error
== -ERESTARTSYS
)
1411 return (SET_ERROR(error
));
1414 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1419 zvol_release(struct gendisk
*disk
, fmode_t mode
)
1422 boolean_t drop_suspend
= B_TRUE
;
1424 ASSERT(!MUTEX_HELD(&zvol_state_lock
));
1426 mutex_enter(&zvol_state_lock
);
1427 zv
= disk
->private_data
;
1429 mutex_enter(&zv
->zv_state_lock
);
1430 ASSERT(zv
->zv_open_count
> 0);
1432 * make sure zvol is not suspended during last close
1433 * (hold zv_suspend_lock) and respect proper lock acquisition
1434 * ordering - zv_suspend_lock before zv_state_lock
1436 if (zv
->zv_open_count
== 1) {
1437 if (!rw_tryenter(&zv
->zv_suspend_lock
, RW_READER
)) {
1438 mutex_exit(&zv
->zv_state_lock
);
1439 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1440 mutex_enter(&zv
->zv_state_lock
);
1441 /* check to see if zv_suspend_lock is needed */
1442 if (zv
->zv_open_count
!= 1) {
1443 rw_exit(&zv
->zv_suspend_lock
);
1444 drop_suspend
= B_FALSE
;
1448 drop_suspend
= B_FALSE
;
1450 mutex_exit(&zvol_state_lock
);
1452 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1453 ASSERT(zv
->zv_open_count
!= 1 || RW_READ_HELD(&zv
->zv_suspend_lock
));
1455 zv
->zv_open_count
--;
1456 if (zv
->zv_open_count
== 0)
1457 zvol_last_close(zv
);
1459 mutex_exit(&zv
->zv_state_lock
);
1462 rw_exit(&zv
->zv_suspend_lock
);
1464 #ifndef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1470 zvol_ioctl(struct block_device
*bdev
, fmode_t mode
,
1471 unsigned int cmd
, unsigned long arg
)
1473 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
1476 ASSERT(zv
&& zv
->zv_open_count
> 0);
1481 invalidate_bdev(bdev
);
1482 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1484 if (dsl_dataset_is_dirty(dmu_objset_ds(zv
->zv_objset
)) &&
1485 !(zv
->zv_flags
& ZVOL_RDONLY
))
1486 txg_wait_synced(dmu_objset_pool(zv
->zv_objset
), 0);
1488 rw_exit(&zv
->zv_suspend_lock
);
1492 mutex_enter(&zv
->zv_state_lock
);
1493 error
= copy_to_user((void *)arg
, zv
->zv_name
, MAXNAMELEN
);
1494 mutex_exit(&zv
->zv_state_lock
);
1502 return (SET_ERROR(error
));
1505 #ifdef CONFIG_COMPAT
1507 zvol_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1508 unsigned cmd
, unsigned long arg
)
1510 return (zvol_ioctl(bdev
, mode
, cmd
, arg
));
1513 #define zvol_compat_ioctl NULL
1516 static int zvol_media_changed(struct gendisk
*disk
)
1518 zvol_state_t
*zv
= disk
->private_data
;
1520 ASSERT(zv
&& zv
->zv_open_count
> 0);
1522 return (zv
->zv_changed
);
1525 static int zvol_revalidate_disk(struct gendisk
*disk
)
1527 zvol_state_t
*zv
= disk
->private_data
;
1529 ASSERT(zv
&& zv
->zv_open_count
> 0);
1532 set_capacity(zv
->zv_disk
, zv
->zv_volsize
>> 9);
1538 * Provide a simple virtual geometry for legacy compatibility. For devices
1539 * smaller than 1 MiB a small head and sector count is used to allow very
1540 * tiny devices. For devices over 1 Mib a standard head and sector count
1541 * is used to keep the cylinders count reasonable.
1544 zvol_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1546 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
1549 ASSERT(zv
&& zv
->zv_open_count
> 0);
1551 sectors
= get_capacity(zv
->zv_disk
);
1553 if (sectors
> 2048) {
1562 geo
->cylinders
= sectors
/ (geo
->heads
* geo
->sectors
);
1567 static struct kobject
*
1568 zvol_probe(dev_t dev
, int *part
, void *arg
)
1571 struct kobject
*kobj
;
1573 zv
= zvol_find_by_dev(dev
);
1574 kobj
= zv
? get_disk(zv
->zv_disk
) : NULL
;
1575 ASSERT(zv
== NULL
|| MUTEX_HELD(&zv
->zv_state_lock
));
1577 mutex_exit(&zv
->zv_state_lock
);
1582 #ifdef HAVE_BDEV_BLOCK_DEVICE_OPERATIONS
1583 static struct block_device_operations zvol_ops
= {
1585 .release
= zvol_release
,
1586 .ioctl
= zvol_ioctl
,
1587 .compat_ioctl
= zvol_compat_ioctl
,
1588 .media_changed
= zvol_media_changed
,
1589 .revalidate_disk
= zvol_revalidate_disk
,
1590 .getgeo
= zvol_getgeo
,
1591 .owner
= THIS_MODULE
,
1594 #else /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */
1597 zvol_open_by_inode(struct inode
*inode
, struct file
*file
)
1599 return (zvol_open(inode
->i_bdev
, file
->f_mode
));
1603 zvol_release_by_inode(struct inode
*inode
, struct file
*file
)
1605 return (zvol_release(inode
->i_bdev
->bd_disk
, file
->f_mode
));
1609 zvol_ioctl_by_inode(struct inode
*inode
, struct file
*file
,
1610 unsigned int cmd
, unsigned long arg
)
1612 if (file
== NULL
|| inode
== NULL
)
1613 return (SET_ERROR(-EINVAL
));
1615 return (zvol_ioctl(inode
->i_bdev
, file
->f_mode
, cmd
, arg
));
1618 #ifdef CONFIG_COMPAT
1620 zvol_compat_ioctl_by_inode(struct file
*file
,
1621 unsigned int cmd
, unsigned long arg
)
1624 return (SET_ERROR(-EINVAL
));
1626 return (zvol_compat_ioctl(file
->f_dentry
->d_inode
->i_bdev
,
1627 file
->f_mode
, cmd
, arg
));
1630 #define zvol_compat_ioctl_by_inode NULL
1633 static struct block_device_operations zvol_ops
= {
1634 .open
= zvol_open_by_inode
,
1635 .release
= zvol_release_by_inode
,
1636 .ioctl
= zvol_ioctl_by_inode
,
1637 .compat_ioctl
= zvol_compat_ioctl_by_inode
,
1638 .media_changed
= zvol_media_changed
,
1639 .revalidate_disk
= zvol_revalidate_disk
,
1640 .getgeo
= zvol_getgeo
,
1641 .owner
= THIS_MODULE
,
1643 #endif /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */
1646 * Allocate memory for a new zvol_state_t and setup the required
1647 * request queue and generic disk structures for the block device.
1649 static zvol_state_t
*
1650 zvol_alloc(dev_t dev
, const char *name
)
1655 if (dsl_prop_get_integer(name
, "volmode", &volmode
, NULL
) != 0)
1658 if (volmode
== ZFS_VOLMODE_DEFAULT
)
1659 volmode
= zvol_volmode
;
1661 if (volmode
== ZFS_VOLMODE_NONE
)
1664 zv
= kmem_zalloc(sizeof (zvol_state_t
), KM_SLEEP
);
1666 list_link_init(&zv
->zv_next
);
1668 mutex_init(&zv
->zv_state_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1670 zv
->zv_queue
= blk_alloc_queue(GFP_ATOMIC
);
1671 if (zv
->zv_queue
== NULL
)
1674 blk_queue_make_request(zv
->zv_queue
, zvol_request
);
1675 blk_queue_set_write_cache(zv
->zv_queue
, B_TRUE
, B_TRUE
);
1677 /* Limit read-ahead to a single page to prevent over-prefetching. */
1678 blk_queue_set_read_ahead(zv
->zv_queue
, 1);
1680 /* Disable write merging in favor of the ZIO pipeline. */
1681 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES
, zv
->zv_queue
);
1683 zv
->zv_disk
= alloc_disk(ZVOL_MINORS
);
1684 if (zv
->zv_disk
== NULL
)
1687 zv
->zv_queue
->queuedata
= zv
;
1689 zv
->zv_open_count
= 0;
1690 strlcpy(zv
->zv_name
, name
, MAXNAMELEN
);
1692 zfs_rlock_init(&zv
->zv_range_lock
);
1693 rw_init(&zv
->zv_suspend_lock
, NULL
, RW_DEFAULT
, NULL
);
1695 zv
->zv_disk
->major
= zvol_major
;
1696 if (volmode
== ZFS_VOLMODE_DEV
) {
1698 * ZFS_VOLMODE_DEV disable partitioning on ZVOL devices: set
1699 * gendisk->minors = 1 as noted in include/linux/genhd.h.
1700 * Also disable extended partition numbers (GENHD_FL_EXT_DEVT)
1701 * and suppresses partition scanning (GENHD_FL_NO_PART_SCAN)
1702 * setting gendisk->flags accordingly.
1704 zv
->zv_disk
->minors
= 1;
1705 #if defined(GENHD_FL_EXT_DEVT)
1706 zv
->zv_disk
->flags
&= ~GENHD_FL_EXT_DEVT
;
1708 #if defined(GENHD_FL_NO_PART_SCAN)
1709 zv
->zv_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1712 zv
->zv_disk
->first_minor
= (dev
& MINORMASK
);
1713 zv
->zv_disk
->fops
= &zvol_ops
;
1714 zv
->zv_disk
->private_data
= zv
;
1715 zv
->zv_disk
->queue
= zv
->zv_queue
;
1716 snprintf(zv
->zv_disk
->disk_name
, DISK_NAME_LEN
, "%s%d",
1717 ZVOL_DEV_NAME
, (dev
& MINORMASK
));
1722 blk_cleanup_queue(zv
->zv_queue
);
1724 kmem_free(zv
, sizeof (zvol_state_t
));
1730 * Cleanup then free a zvol_state_t which was created by zvol_alloc().
1731 * At this time, the structure is not opened by anyone, is taken off
1732 * the zvol_state_list, and has its private data set to NULL.
1733 * The zvol_state_lock is dropped.
1736 zvol_free(void *arg
)
1738 zvol_state_t
*zv
= arg
;
1740 ASSERT(!MUTEX_HELD(&zvol_state_lock
));
1741 ASSERT(!RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1742 ASSERT(!MUTEX_HELD(&zv
->zv_state_lock
));
1743 ASSERT(zv
->zv_open_count
== 0);
1744 ASSERT(zv
->zv_disk
->private_data
== NULL
);
1746 rw_destroy(&zv
->zv_suspend_lock
);
1747 zfs_rlock_destroy(&zv
->zv_range_lock
);
1749 del_gendisk(zv
->zv_disk
);
1750 blk_cleanup_queue(zv
->zv_queue
);
1751 put_disk(zv
->zv_disk
);
1753 ida_simple_remove(&zvol_ida
, MINOR(zv
->zv_dev
) >> ZVOL_MINOR_BITS
);
1755 mutex_destroy(&zv
->zv_state_lock
);
1757 kmem_free(zv
, sizeof (zvol_state_t
));
1761 * Create a block device minor node and setup the linkage between it
1762 * and the specified volume. Once this function returns the block
1763 * device is live and ready for use.
1766 zvol_create_minor_impl(const char *name
)
1770 dmu_object_info_t
*doi
;
1776 uint64_t hash
= zvol_name_hash(name
);
1778 if (zvol_inhibit_dev
)
1781 idx
= ida_simple_get(&zvol_ida
, 0, 0, kmem_flags_convert(KM_SLEEP
));
1783 return (SET_ERROR(-idx
));
1784 minor
= idx
<< ZVOL_MINOR_BITS
;
1786 zv
= zvol_find_by_name_hash(name
, hash
, RW_NONE
);
1788 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1789 mutex_exit(&zv
->zv_state_lock
);
1790 ida_simple_remove(&zvol_ida
, idx
);
1791 return (SET_ERROR(EEXIST
));
1794 doi
= kmem_alloc(sizeof (dmu_object_info_t
), KM_SLEEP
);
1796 error
= dmu_objset_own(name
, DMU_OST_ZVOL
, B_TRUE
, B_TRUE
, FTAG
, &os
);
1800 error
= dmu_object_info(os
, ZVOL_OBJ
, doi
);
1802 goto out_dmu_objset_disown
;
1804 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
1806 goto out_dmu_objset_disown
;
1808 zv
= zvol_alloc(MKDEV(zvol_major
, minor
), name
);
1810 error
= SET_ERROR(EAGAIN
);
1811 goto out_dmu_objset_disown
;
1815 if (dmu_objset_is_snapshot(os
))
1816 zv
->zv_flags
|= ZVOL_RDONLY
;
1818 zv
->zv_volblocksize
= doi
->doi_data_block_size
;
1819 zv
->zv_volsize
= volsize
;
1822 set_capacity(zv
->zv_disk
, zv
->zv_volsize
>> 9);
1824 blk_queue_max_hw_sectors(zv
->zv_queue
, (DMU_MAX_ACCESS
/ 4) >> 9);
1825 blk_queue_max_segments(zv
->zv_queue
, UINT16_MAX
);
1826 blk_queue_max_segment_size(zv
->zv_queue
, UINT_MAX
);
1827 blk_queue_physical_block_size(zv
->zv_queue
, zv
->zv_volblocksize
);
1828 blk_queue_io_opt(zv
->zv_queue
, zv
->zv_volblocksize
);
1829 blk_queue_max_discard_sectors(zv
->zv_queue
,
1830 (zvol_max_discard_blocks
* zv
->zv_volblocksize
) >> 9);
1831 blk_queue_discard_granularity(zv
->zv_queue
, zv
->zv_volblocksize
);
1832 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, zv
->zv_queue
);
1833 #ifdef QUEUE_FLAG_NONROT
1834 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, zv
->zv_queue
);
1836 #ifdef QUEUE_FLAG_ADD_RANDOM
1837 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, zv
->zv_queue
);
1840 if (spa_writeable(dmu_objset_spa(os
))) {
1841 if (zil_replay_disable
)
1842 zil_destroy(dmu_objset_zil(os
), B_FALSE
);
1844 zil_replay(os
, zv
, zvol_replay_vector
);
1848 * When udev detects the addition of the device it will immediately
1849 * invoke blkid(8) to determine the type of content on the device.
1850 * Prefetching the blocks commonly scanned by blkid(8) will speed
1853 len
= MIN(MAX(zvol_prefetch_bytes
, 0), SPA_MAXBLOCKSIZE
);
1855 dmu_prefetch(os
, ZVOL_OBJ
, 0, 0, len
, ZIO_PRIORITY_SYNC_READ
);
1856 dmu_prefetch(os
, ZVOL_OBJ
, 0, volsize
- len
, len
,
1857 ZIO_PRIORITY_SYNC_READ
);
1860 zv
->zv_objset
= NULL
;
1861 out_dmu_objset_disown
:
1862 dmu_objset_disown(os
, B_TRUE
, FTAG
);
1864 kmem_free(doi
, sizeof (dmu_object_info_t
));
1867 mutex_enter(&zvol_state_lock
);
1869 mutex_exit(&zvol_state_lock
);
1870 add_disk(zv
->zv_disk
);
1872 ida_simple_remove(&zvol_ida
, idx
);
1875 return (SET_ERROR(error
));
1879 * Rename a block device minor mode for the specified volume.
1882 zvol_rename_minor(zvol_state_t
*zv
, const char *newname
)
1884 int readonly
= get_disk_ro(zv
->zv_disk
);
1886 ASSERT(MUTEX_HELD(&zvol_state_lock
));
1887 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1889 strlcpy(zv
->zv_name
, newname
, sizeof (zv
->zv_name
));
1891 /* move to new hashtable entry */
1892 zv
->zv_hash
= zvol_name_hash(zv
->zv_name
);
1893 hlist_del(&zv
->zv_hlink
);
1894 hlist_add_head(&zv
->zv_hlink
, ZVOL_HT_HEAD(zv
->zv_hash
));
1897 * The block device's read-only state is briefly changed causing
1898 * a KOBJ_CHANGE uevent to be issued. This ensures udev detects
1899 * the name change and fixes the symlinks. This does not change
1900 * ZVOL_RDONLY in zv->zv_flags so the actual read-only state never
1901 * changes. This would normally be done using kobject_uevent() but
1902 * that is a GPL-only symbol which is why we need this workaround.
1904 set_disk_ro(zv
->zv_disk
, !readonly
);
1905 set_disk_ro(zv
->zv_disk
, readonly
);
1908 typedef struct minors_job
{
1918 * Prefetch zvol dnodes for the minors_job
1921 zvol_prefetch_minors_impl(void *arg
)
1923 minors_job_t
*job
= arg
;
1924 char *dsname
= job
->name
;
1925 objset_t
*os
= NULL
;
1927 job
->error
= dmu_objset_own(dsname
, DMU_OST_ZVOL
, B_TRUE
, B_TRUE
,
1929 if (job
->error
== 0) {
1930 dmu_prefetch(os
, ZVOL_OBJ
, 0, 0, 0, ZIO_PRIORITY_SYNC_READ
);
1931 dmu_objset_disown(os
, B_TRUE
, FTAG
);
1936 * Mask errors to continue dmu_objset_find() traversal
1939 zvol_create_snap_minor_cb(const char *dsname
, void *arg
)
1941 minors_job_t
*j
= arg
;
1942 list_t
*minors_list
= j
->list
;
1943 const char *name
= j
->name
;
1945 ASSERT0(MUTEX_HELD(&spa_namespace_lock
));
1947 /* skip the designated dataset */
1948 if (name
&& strcmp(dsname
, name
) == 0)
1951 /* at this point, the dsname should name a snapshot */
1952 if (strchr(dsname
, '@') == 0) {
1953 dprintf("zvol_create_snap_minor_cb(): "
1954 "%s is not a shapshot name\n", dsname
);
1957 char *n
= strdup(dsname
);
1961 job
= kmem_alloc(sizeof (minors_job_t
), KM_SLEEP
);
1963 job
->list
= minors_list
;
1965 list_insert_tail(minors_list
, job
);
1966 /* don't care if dispatch fails, because job->error is 0 */
1967 taskq_dispatch(system_taskq
, zvol_prefetch_minors_impl
, job
,
1975 * Mask errors to continue dmu_objset_find() traversal
1978 zvol_create_minors_cb(const char *dsname
, void *arg
)
1982 list_t
*minors_list
= arg
;
1984 ASSERT0(MUTEX_HELD(&spa_namespace_lock
));
1986 error
= dsl_prop_get_integer(dsname
, "snapdev", &snapdev
, NULL
);
1991 * Given the name and the 'snapdev' property, create device minor nodes
1992 * with the linkages to zvols/snapshots as needed.
1993 * If the name represents a zvol, create a minor node for the zvol, then
1994 * check if its snapshots are 'visible', and if so, iterate over the
1995 * snapshots and create device minor nodes for those.
1997 if (strchr(dsname
, '@') == 0) {
1999 char *n
= strdup(dsname
);
2003 job
= kmem_alloc(sizeof (minors_job_t
), KM_SLEEP
);
2005 job
->list
= minors_list
;
2007 list_insert_tail(minors_list
, job
);
2008 /* don't care if dispatch fails, because job->error is 0 */
2009 taskq_dispatch(system_taskq
, zvol_prefetch_minors_impl
, job
,
2012 if (snapdev
== ZFS_SNAPDEV_VISIBLE
) {
2014 * traverse snapshots only, do not traverse children,
2015 * and skip the 'dsname'
2017 error
= dmu_objset_find((char *)dsname
,
2018 zvol_create_snap_minor_cb
, (void *)job
,
2022 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n",
2030 * Create minors for the specified dataset, including children and snapshots.
2031 * Pay attention to the 'snapdev' property and iterate over the snapshots
2032 * only if they are 'visible'. This approach allows one to assure that the
2033 * snapshot metadata is read from disk only if it is needed.
2035 * The name can represent a dataset to be recursively scanned for zvols and
2036 * their snapshots, or a single zvol snapshot. If the name represents a
2037 * dataset, the scan is performed in two nested stages:
2038 * - scan the dataset for zvols, and
2039 * - for each zvol, create a minor node, then check if the zvol's snapshots
2040 * are 'visible', and only then iterate over the snapshots if needed
2042 * If the name represents a snapshot, a check is performed if the snapshot is
2043 * 'visible' (which also verifies that the parent is a zvol), and if so,
2044 * a minor node for that snapshot is created.
2047 zvol_create_minors_impl(const char *name
)
2050 fstrans_cookie_t cookie
;
2055 if (zvol_inhibit_dev
)
2059 * This is the list for prefetch jobs. Whenever we found a match
2060 * during dmu_objset_find, we insert a minors_job to the list and do
2061 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need
2062 * any lock because all list operation is done on the current thread.
2064 * We will use this list to do zvol_create_minor_impl after prefetch
2065 * so we don't have to traverse using dmu_objset_find again.
2067 list_create(&minors_list
, sizeof (minors_job_t
),
2068 offsetof(minors_job_t
, link
));
2070 parent
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
2071 (void) strlcpy(parent
, name
, MAXPATHLEN
);
2073 if ((atp
= strrchr(parent
, '@')) != NULL
) {
2077 error
= dsl_prop_get_integer(parent
, "snapdev",
2080 if (error
== 0 && snapdev
== ZFS_SNAPDEV_VISIBLE
)
2081 error
= zvol_create_minor_impl(name
);
2083 cookie
= spl_fstrans_mark();
2084 error
= dmu_objset_find(parent
, zvol_create_minors_cb
,
2085 &minors_list
, DS_FIND_CHILDREN
);
2086 spl_fstrans_unmark(cookie
);
2089 kmem_free(parent
, MAXPATHLEN
);
2090 taskq_wait_outstanding(system_taskq
, 0);
2093 * Prefetch is completed, we can do zvol_create_minor_impl
2096 while ((job
= list_head(&minors_list
)) != NULL
) {
2097 list_remove(&minors_list
, job
);
2099 zvol_create_minor_impl(job
->name
);
2101 kmem_free(job
, sizeof (minors_job_t
));
2104 list_destroy(&minors_list
);
2106 return (SET_ERROR(error
));
2110 * Remove minors for specified dataset including children and snapshots.
2113 zvol_remove_minors_impl(const char *name
)
2115 zvol_state_t
*zv
, *zv_next
;
2116 int namelen
= ((name
) ? strlen(name
) : 0);
2117 taskqid_t t
, tid
= TASKQID_INVALID
;
2120 if (zvol_inhibit_dev
)
2123 list_create(&free_list
, sizeof (zvol_state_t
),
2124 offsetof(zvol_state_t
, zv_next
));
2126 mutex_enter(&zvol_state_lock
);
2128 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2129 zv_next
= list_next(&zvol_state_list
, zv
);
2131 mutex_enter(&zv
->zv_state_lock
);
2132 if (name
== NULL
|| strcmp(zv
->zv_name
, name
) == 0 ||
2133 (strncmp(zv
->zv_name
, name
, namelen
) == 0 &&
2134 (zv
->zv_name
[namelen
] == '/' ||
2135 zv
->zv_name
[namelen
] == '@'))) {
2137 * By holding zv_state_lock here, we guarantee that no
2138 * one is currently using this zv
2141 /* If in use, leave alone */
2142 if (zv
->zv_open_count
> 0 ||
2143 atomic_read(&zv
->zv_suspend_ref
)) {
2144 mutex_exit(&zv
->zv_state_lock
);
2151 * clear this while holding zvol_state_lock so
2152 * zvol_open won't open it
2154 zv
->zv_disk
->private_data
= NULL
;
2156 /* Drop zv_state_lock before zvol_free() */
2157 mutex_exit(&zv
->zv_state_lock
);
2159 /* try parallel zv_free, if failed do it in place */
2160 t
= taskq_dispatch(system_taskq
, zvol_free
, zv
,
2162 if (t
== TASKQID_INVALID
)
2163 list_insert_head(&free_list
, zv
);
2167 mutex_exit(&zv
->zv_state_lock
);
2170 mutex_exit(&zvol_state_lock
);
2173 * Drop zvol_state_lock before calling zvol_free()
2175 while ((zv
= list_head(&free_list
)) != NULL
) {
2176 list_remove(&free_list
, zv
);
2180 if (tid
!= TASKQID_INVALID
)
2181 taskq_wait_outstanding(system_taskq
, tid
);
2184 /* Remove minor for this specific volume only */
2186 zvol_remove_minor_impl(const char *name
)
2188 zvol_state_t
*zv
= NULL
, *zv_next
;
2190 if (zvol_inhibit_dev
)
2193 mutex_enter(&zvol_state_lock
);
2195 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2196 zv_next
= list_next(&zvol_state_list
, zv
);
2198 mutex_enter(&zv
->zv_state_lock
);
2199 if (strcmp(zv
->zv_name
, name
) == 0) {
2201 * By holding zv_state_lock here, we guarantee that no
2202 * one is currently using this zv
2205 /* If in use, leave alone */
2206 if (zv
->zv_open_count
> 0 ||
2207 atomic_read(&zv
->zv_suspend_ref
)) {
2208 mutex_exit(&zv
->zv_state_lock
);
2213 /* clear this so zvol_open won't open it */
2214 zv
->zv_disk
->private_data
= NULL
;
2216 mutex_exit(&zv
->zv_state_lock
);
2219 mutex_exit(&zv
->zv_state_lock
);
2223 /* Drop zvol_state_lock before calling zvol_free() */
2224 mutex_exit(&zvol_state_lock
);
2231 * Rename minors for specified dataset including children and snapshots.
2234 zvol_rename_minors_impl(const char *oldname
, const char *newname
)
2236 zvol_state_t
*zv
, *zv_next
;
2237 int oldnamelen
, newnamelen
;
2239 if (zvol_inhibit_dev
)
2242 oldnamelen
= strlen(oldname
);
2243 newnamelen
= strlen(newname
);
2245 mutex_enter(&zvol_state_lock
);
2247 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2248 zv_next
= list_next(&zvol_state_list
, zv
);
2250 mutex_enter(&zv
->zv_state_lock
);
2252 /* If in use, leave alone */
2253 if (zv
->zv_open_count
> 0) {
2254 mutex_exit(&zv
->zv_state_lock
);
2258 if (strcmp(zv
->zv_name
, oldname
) == 0) {
2259 zvol_rename_minor(zv
, newname
);
2260 } else if (strncmp(zv
->zv_name
, oldname
, oldnamelen
) == 0 &&
2261 (zv
->zv_name
[oldnamelen
] == '/' ||
2262 zv
->zv_name
[oldnamelen
] == '@')) {
2263 char *name
= kmem_asprintf("%s%c%s", newname
,
2264 zv
->zv_name
[oldnamelen
],
2265 zv
->zv_name
+ oldnamelen
+ 1);
2266 zvol_rename_minor(zv
, name
);
2267 kmem_free(name
, strlen(name
+ 1));
2270 mutex_exit(&zv
->zv_state_lock
);
2273 mutex_exit(&zvol_state_lock
);
2276 typedef struct zvol_snapdev_cb_arg
{
2278 } zvol_snapdev_cb_arg_t
;
2281 zvol_set_snapdev_cb(const char *dsname
, void *param
)
2283 zvol_snapdev_cb_arg_t
*arg
= param
;
2285 if (strchr(dsname
, '@') == NULL
)
2288 switch (arg
->snapdev
) {
2289 case ZFS_SNAPDEV_VISIBLE
:
2290 (void) zvol_create_minor_impl(dsname
);
2292 case ZFS_SNAPDEV_HIDDEN
:
2293 (void) zvol_remove_minor_impl(dsname
);
2301 zvol_set_snapdev_impl(char *name
, uint64_t snapdev
)
2303 zvol_snapdev_cb_arg_t arg
= {snapdev
};
2304 fstrans_cookie_t cookie
= spl_fstrans_mark();
2306 * The zvol_set_snapdev_sync() sets snapdev appropriately
2307 * in the dataset hierarchy. Here, we only scan snapshots.
2309 dmu_objset_find(name
, zvol_set_snapdev_cb
, &arg
, DS_FIND_SNAPSHOTS
);
2310 spl_fstrans_unmark(cookie
);
2313 typedef struct zvol_volmode_cb_arg
{
2315 } zvol_volmode_cb_arg_t
;
2318 zvol_set_volmode_impl(char *name
, uint64_t volmode
)
2320 fstrans_cookie_t cookie
= spl_fstrans_mark();
2322 if (strchr(name
, '@') != NULL
)
2326 * It's unfortunate we need to remove minors before we create new ones:
2327 * this is necessary because our backing gendisk (zvol_state->zv_disk)
2328 * coule be different when we set, for instance, volmode from "geom"
2329 * to "dev" (or vice versa).
2330 * A possible optimization is to modify our consumers so we don't get
2331 * called when "volmode" does not change.
2334 case ZFS_VOLMODE_NONE
:
2335 (void) zvol_remove_minor_impl(name
);
2337 case ZFS_VOLMODE_GEOM
:
2338 case ZFS_VOLMODE_DEV
:
2339 (void) zvol_remove_minor_impl(name
);
2340 (void) zvol_create_minor_impl(name
);
2342 case ZFS_VOLMODE_DEFAULT
:
2343 (void) zvol_remove_minor_impl(name
);
2344 if (zvol_volmode
== ZFS_VOLMODE_NONE
)
2346 else /* if zvol_volmode is invalid defaults to "geom" */
2347 (void) zvol_create_minor_impl(name
);
2351 spl_fstrans_unmark(cookie
);
2354 static zvol_task_t
*
2355 zvol_task_alloc(zvol_async_op_t op
, const char *name1
, const char *name2
,
2361 /* Never allow tasks on hidden names. */
2362 if (name1
[0] == '$')
2365 task
= kmem_zalloc(sizeof (zvol_task_t
), KM_SLEEP
);
2367 task
->value
= value
;
2368 delim
= strchr(name1
, '/');
2369 strlcpy(task
->pool
, name1
, delim
? (delim
- name1
+ 1) : MAXNAMELEN
);
2371 strlcpy(task
->name1
, name1
, MAXNAMELEN
);
2373 strlcpy(task
->name2
, name2
, MAXNAMELEN
);
2379 zvol_task_free(zvol_task_t
*task
)
2381 kmem_free(task
, sizeof (zvol_task_t
));
2385 * The worker thread function performed asynchronously.
2388 zvol_task_cb(void *param
)
2390 zvol_task_t
*task
= (zvol_task_t
*)param
;
2393 case ZVOL_ASYNC_CREATE_MINORS
:
2394 (void) zvol_create_minors_impl(task
->name1
);
2396 case ZVOL_ASYNC_REMOVE_MINORS
:
2397 zvol_remove_minors_impl(task
->name1
);
2399 case ZVOL_ASYNC_RENAME_MINORS
:
2400 zvol_rename_minors_impl(task
->name1
, task
->name2
);
2402 case ZVOL_ASYNC_SET_SNAPDEV
:
2403 zvol_set_snapdev_impl(task
->name1
, task
->value
);
2405 case ZVOL_ASYNC_SET_VOLMODE
:
2406 zvol_set_volmode_impl(task
->name1
, task
->value
);
2413 zvol_task_free(task
);
2416 typedef struct zvol_set_prop_int_arg
{
2417 const char *zsda_name
;
2418 uint64_t zsda_value
;
2419 zprop_source_t zsda_source
;
2421 } zvol_set_prop_int_arg_t
;
2424 * Sanity check the dataset for safe use by the sync task. No additional
2425 * conditions are imposed.
2428 zvol_set_snapdev_check(void *arg
, dmu_tx_t
*tx
)
2430 zvol_set_prop_int_arg_t
*zsda
= arg
;
2431 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2435 error
= dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
);
2439 dsl_dir_rele(dd
, FTAG
);
2446 zvol_set_snapdev_sync_cb(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2448 char dsname
[MAXNAMELEN
];
2452 dsl_dataset_name(ds
, dsname
);
2453 if (dsl_prop_get_int_ds(ds
, "snapdev", &snapdev
) != 0)
2455 task
= zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV
, dsname
, NULL
, snapdev
);
2459 (void) taskq_dispatch(dp
->dp_spa
->spa_zvol_taskq
, zvol_task_cb
,
2465 * Traverse all child datasets and apply snapdev appropriately.
2466 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
2467 * dataset and read the effective "snapdev" on every child in the callback
2468 * function: this is because the value is not guaranteed to be the same in the
2469 * whole dataset hierarchy.
2472 zvol_set_snapdev_sync(void *arg
, dmu_tx_t
*tx
)
2474 zvol_set_prop_int_arg_t
*zsda
= arg
;
2475 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2480 VERIFY0(dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
));
2483 error
= dsl_dataset_hold(dp
, zsda
->zsda_name
, FTAG
, &ds
);
2485 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_SNAPDEV
),
2486 zsda
->zsda_source
, sizeof (zsda
->zsda_value
), 1,
2487 &zsda
->zsda_value
, zsda
->zsda_tx
);
2488 dsl_dataset_rele(ds
, FTAG
);
2490 dmu_objset_find_dp(dp
, dd
->dd_object
, zvol_set_snapdev_sync_cb
,
2491 zsda
, DS_FIND_CHILDREN
);
2493 dsl_dir_rele(dd
, FTAG
);
2497 zvol_set_snapdev(const char *ddname
, zprop_source_t source
, uint64_t snapdev
)
2499 zvol_set_prop_int_arg_t zsda
;
2501 zsda
.zsda_name
= ddname
;
2502 zsda
.zsda_source
= source
;
2503 zsda
.zsda_value
= snapdev
;
2505 return (dsl_sync_task(ddname
, zvol_set_snapdev_check
,
2506 zvol_set_snapdev_sync
, &zsda
, 0, ZFS_SPACE_CHECK_NONE
));
2510 * Sanity check the dataset for safe use by the sync task. No additional
2511 * conditions are imposed.
2514 zvol_set_volmode_check(void *arg
, dmu_tx_t
*tx
)
2516 zvol_set_prop_int_arg_t
*zsda
= arg
;
2517 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2521 error
= dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
);
2525 dsl_dir_rele(dd
, FTAG
);
2532 zvol_set_volmode_sync_cb(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2534 char dsname
[MAXNAMELEN
];
2538 dsl_dataset_name(ds
, dsname
);
2539 if (dsl_prop_get_int_ds(ds
, "volmode", &volmode
) != 0)
2541 task
= zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE
, dsname
, NULL
, volmode
);
2545 (void) taskq_dispatch(dp
->dp_spa
->spa_zvol_taskq
, zvol_task_cb
,
2551 * Traverse all child datasets and apply volmode appropriately.
2552 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
2553 * dataset and read the effective "volmode" on every child in the callback
2554 * function: this is because the value is not guaranteed to be the same in the
2555 * whole dataset hierarchy.
2558 zvol_set_volmode_sync(void *arg
, dmu_tx_t
*tx
)
2560 zvol_set_prop_int_arg_t
*zsda
= arg
;
2561 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2566 VERIFY0(dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
));
2569 error
= dsl_dataset_hold(dp
, zsda
->zsda_name
, FTAG
, &ds
);
2571 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_VOLMODE
),
2572 zsda
->zsda_source
, sizeof (zsda
->zsda_value
), 1,
2573 &zsda
->zsda_value
, zsda
->zsda_tx
);
2574 dsl_dataset_rele(ds
, FTAG
);
2577 dmu_objset_find_dp(dp
, dd
->dd_object
, zvol_set_volmode_sync_cb
,
2578 zsda
, DS_FIND_CHILDREN
);
2580 dsl_dir_rele(dd
, FTAG
);
2584 zvol_set_volmode(const char *ddname
, zprop_source_t source
, uint64_t volmode
)
2586 zvol_set_prop_int_arg_t zsda
;
2588 zsda
.zsda_name
= ddname
;
2589 zsda
.zsda_source
= source
;
2590 zsda
.zsda_value
= volmode
;
2592 return (dsl_sync_task(ddname
, zvol_set_volmode_check
,
2593 zvol_set_volmode_sync
, &zsda
, 0, ZFS_SPACE_CHECK_NONE
));
2597 zvol_create_minors(spa_t
*spa
, const char *name
, boolean_t async
)
2602 task
= zvol_task_alloc(ZVOL_ASYNC_CREATE_MINORS
, name
, NULL
, ~0ULL);
2606 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2607 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2608 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2612 zvol_remove_minors(spa_t
*spa
, const char *name
, boolean_t async
)
2617 task
= zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS
, name
, NULL
, ~0ULL);
2621 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2622 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2623 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2627 zvol_rename_minors(spa_t
*spa
, const char *name1
, const char *name2
,
2633 task
= zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS
, name1
, name2
, ~0ULL);
2637 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2638 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2639 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2645 int threads
= MIN(MAX(zvol_threads
, 1), 1024);
2648 list_create(&zvol_state_list
, sizeof (zvol_state_t
),
2649 offsetof(zvol_state_t
, zv_next
));
2650 mutex_init(&zvol_state_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2651 ida_init(&zvol_ida
);
2653 zvol_taskq
= taskq_create(ZVOL_DRIVER
, threads
, maxclsyspri
,
2654 threads
* 2, INT_MAX
, TASKQ_PREPOPULATE
| TASKQ_DYNAMIC
);
2655 if (zvol_taskq
== NULL
) {
2656 printk(KERN_INFO
"ZFS: taskq_create() failed\n");
2661 zvol_htable
= kmem_alloc(ZVOL_HT_SIZE
* sizeof (struct hlist_head
),
2667 for (i
= 0; i
< ZVOL_HT_SIZE
; i
++)
2668 INIT_HLIST_HEAD(&zvol_htable
[i
]);
2670 error
= register_blkdev(zvol_major
, ZVOL_DRIVER
);
2672 printk(KERN_INFO
"ZFS: register_blkdev() failed %d\n", error
);
2676 blk_register_region(MKDEV(zvol_major
, 0), 1UL << MINORBITS
,
2677 THIS_MODULE
, zvol_probe
, NULL
, NULL
);
2682 kmem_free(zvol_htable
, ZVOL_HT_SIZE
* sizeof (struct hlist_head
));
2684 taskq_destroy(zvol_taskq
);
2686 ida_destroy(&zvol_ida
);
2687 mutex_destroy(&zvol_state_lock
);
2688 list_destroy(&zvol_state_list
);
2690 return (SET_ERROR(error
));
2696 zvol_remove_minors_impl(NULL
);
2698 blk_unregister_region(MKDEV(zvol_major
, 0), 1UL << MINORBITS
);
2699 unregister_blkdev(zvol_major
, ZVOL_DRIVER
);
2700 kmem_free(zvol_htable
, ZVOL_HT_SIZE
* sizeof (struct hlist_head
));
2702 taskq_destroy(zvol_taskq
);
2703 list_destroy(&zvol_state_list
);
2704 mutex_destroy(&zvol_state_lock
);
2706 ida_destroy(&zvol_ida
);
2710 module_param(zvol_inhibit_dev
, uint
, 0644);
2711 MODULE_PARM_DESC(zvol_inhibit_dev
, "Do not create zvol device nodes");
2713 module_param(zvol_major
, uint
, 0444);
2714 MODULE_PARM_DESC(zvol_major
, "Major number for zvol device");
2716 module_param(zvol_threads
, uint
, 0444);
2717 MODULE_PARM_DESC(zvol_threads
, "Max number of threads to handle I/O requests");
2719 module_param(zvol_request_sync
, uint
, 0644);
2720 MODULE_PARM_DESC(zvol_request_sync
, "Synchronously handle bio requests");
2722 module_param(zvol_max_discard_blocks
, ulong
, 0444);
2723 MODULE_PARM_DESC(zvol_max_discard_blocks
, "Max number of blocks to discard");
2725 module_param(zvol_prefetch_bytes
, uint
, 0644);
2726 MODULE_PARM_DESC(zvol_prefetch_bytes
, "Prefetch N bytes at zvol start+end");
2728 module_param(zvol_volmode
, uint
, 0644);
2729 MODULE_PARM_DESC(zvol_volmode
, "Default volmode property value");