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
= NULL
;
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(void *arg1
, void *arg2
, boolean_t byteswap
)
583 zvol_state_t
*zv
= arg1
;
584 lr_truncate_t
*lr
= arg2
;
585 uint64_t offset
, length
;
588 byteswap_uint64_array(lr
, sizeof (*lr
));
590 offset
= lr
->lr_offset
;
591 length
= lr
->lr_length
;
593 return (dmu_free_long_range(zv
->zv_objset
, ZVOL_OBJ
, offset
, length
));
597 * Replay a TX_WRITE ZIL transaction that didn't get committed
598 * after a system failure
601 zvol_replay_write(void *arg1
, void *arg2
, boolean_t byteswap
)
603 zvol_state_t
*zv
= arg1
;
604 lr_write_t
*lr
= arg2
;
605 objset_t
*os
= zv
->zv_objset
;
606 char *data
= (char *)(lr
+ 1); /* data follows lr_write_t */
607 uint64_t offset
, length
;
612 byteswap_uint64_array(lr
, sizeof (*lr
));
614 offset
= lr
->lr_offset
;
615 length
= lr
->lr_length
;
617 /* If it's a dmu_sync() block, write the whole block */
618 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
619 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
620 if (length
< blocksize
) {
621 offset
-= offset
% blocksize
;
626 tx
= dmu_tx_create(os
);
627 dmu_tx_hold_write(tx
, ZVOL_OBJ
, offset
, length
);
628 error
= dmu_tx_assign(tx
, TXG_WAIT
);
632 dmu_write(os
, ZVOL_OBJ
, offset
, length
, data
, tx
);
640 zvol_replay_err(void *arg1
, void *arg2
, boolean_t byteswap
)
642 return (SET_ERROR(ENOTSUP
));
646 * Callback vectors for replaying records.
647 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
649 zil_replay_func_t
*zvol_replay_vector
[TX_MAX_TYPE
] = {
650 zvol_replay_err
, /* no such transaction type */
651 zvol_replay_err
, /* TX_CREATE */
652 zvol_replay_err
, /* TX_MKDIR */
653 zvol_replay_err
, /* TX_MKXATTR */
654 zvol_replay_err
, /* TX_SYMLINK */
655 zvol_replay_err
, /* TX_REMOVE */
656 zvol_replay_err
, /* TX_RMDIR */
657 zvol_replay_err
, /* TX_LINK */
658 zvol_replay_err
, /* TX_RENAME */
659 zvol_replay_write
, /* TX_WRITE */
660 zvol_replay_truncate
, /* TX_TRUNCATE */
661 zvol_replay_err
, /* TX_SETATTR */
662 zvol_replay_err
, /* TX_ACL */
663 zvol_replay_err
, /* TX_CREATE_ATTR */
664 zvol_replay_err
, /* TX_CREATE_ACL_ATTR */
665 zvol_replay_err
, /* TX_MKDIR_ACL */
666 zvol_replay_err
, /* TX_MKDIR_ATTR */
667 zvol_replay_err
, /* TX_MKDIR_ACL_ATTR */
668 zvol_replay_err
, /* TX_WRITE2 */
672 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
674 * We store data in the log buffers if it's small enough.
675 * Otherwise we will later flush the data out via dmu_sync().
677 ssize_t zvol_immediate_write_sz
= 32768;
680 zvol_log_write(zvol_state_t
*zv
, dmu_tx_t
*tx
, uint64_t offset
,
681 uint64_t size
, int sync
)
683 uint32_t blocksize
= zv
->zv_volblocksize
;
684 zilog_t
*zilog
= zv
->zv_zilog
;
685 itx_wr_state_t write_state
;
687 if (zil_replaying(zilog
, tx
))
690 if (zilog
->zl_logbias
== ZFS_LOGBIAS_THROUGHPUT
)
691 write_state
= WR_INDIRECT
;
692 else if (!spa_has_slogs(zilog
->zl_spa
) &&
693 size
>= blocksize
&& blocksize
> zvol_immediate_write_sz
)
694 write_state
= WR_INDIRECT
;
696 write_state
= WR_COPIED
;
698 write_state
= WR_NEED_COPY
;
703 itx_wr_state_t wr_state
= write_state
;
706 if (wr_state
== WR_COPIED
&& size
> ZIL_MAX_COPIED_DATA
)
707 wr_state
= WR_NEED_COPY
;
708 else if (wr_state
== WR_INDIRECT
)
709 len
= MIN(blocksize
- P2PHASE(offset
, blocksize
), size
);
711 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
) +
712 (wr_state
== WR_COPIED
? len
: 0));
713 lr
= (lr_write_t
*)&itx
->itx_lr
;
714 if (wr_state
== WR_COPIED
&& dmu_read_by_dnode(zv
->zv_dn
,
715 offset
, len
, lr
+1, DMU_READ_NO_PREFETCH
) != 0) {
716 zil_itx_destroy(itx
);
717 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
718 lr
= (lr_write_t
*)&itx
->itx_lr
;
719 wr_state
= WR_NEED_COPY
;
722 itx
->itx_wr_state
= wr_state
;
723 lr
->lr_foid
= ZVOL_OBJ
;
724 lr
->lr_offset
= offset
;
727 BP_ZERO(&lr
->lr_blkptr
);
729 itx
->itx_private
= zv
;
730 itx
->itx_sync
= sync
;
732 (void) zil_itx_assign(zilog
, itx
, tx
);
739 typedef struct zv_request
{
746 uio_from_bio(uio_t
*uio
, struct bio
*bio
)
748 uio
->uio_bvec
= &bio
->bi_io_vec
[BIO_BI_IDX(bio
)];
749 uio
->uio_skip
= BIO_BI_SKIP(bio
);
750 uio
->uio_resid
= BIO_BI_SIZE(bio
);
751 uio
->uio_iovcnt
= bio
->bi_vcnt
- BIO_BI_IDX(bio
);
752 uio
->uio_loffset
= BIO_BI_SECTOR(bio
) << 9;
753 uio
->uio_limit
= MAXOFFSET_T
;
754 uio
->uio_segflg
= UIO_BVEC
;
758 zvol_write(void *arg
)
760 zv_request_t
*zvr
= arg
;
761 struct bio
*bio
= zvr
->bio
;
763 zvol_state_t
*zv
= zvr
->zv
;
764 uint64_t volsize
= zv
->zv_volsize
;
767 unsigned long start_jif
;
769 uio_from_bio(&uio
, bio
);
771 ASSERT(zv
&& zv
->zv_open_count
> 0);
774 blk_generic_start_io_acct(zv
->zv_queue
, WRITE
, bio_sectors(bio
),
775 &zv
->zv_disk
->part0
);
777 sync
= bio_is_fua(bio
) || zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
779 while (uio
.uio_resid
> 0 && uio
.uio_loffset
< volsize
) {
780 uint64_t bytes
= MIN(uio
.uio_resid
, DMU_MAX_ACCESS
>> 1);
781 uint64_t off
= uio
.uio_loffset
;
782 dmu_tx_t
*tx
= dmu_tx_create(zv
->zv_objset
);
784 if (bytes
> volsize
- off
) /* don't write past the end */
785 bytes
= volsize
- off
;
787 dmu_tx_hold_write(tx
, ZVOL_OBJ
, off
, bytes
);
789 /* This will only fail for ENOSPC */
790 error
= dmu_tx_assign(tx
, TXG_WAIT
);
795 error
= dmu_write_uio_dnode(zv
->zv_dn
, &uio
, bytes
, tx
);
797 zvol_log_write(zv
, tx
, off
, bytes
, sync
);
803 zfs_range_unlock(zvr
->rl
);
805 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
807 rw_exit(&zv
->zv_suspend_lock
);
808 blk_generic_end_io_acct(zv
->zv_queue
, WRITE
, &zv
->zv_disk
->part0
,
810 BIO_END_IO(bio
, -error
);
811 kmem_free(zvr
, sizeof (zv_request_t
));
815 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
818 zvol_log_truncate(zvol_state_t
*zv
, dmu_tx_t
*tx
, uint64_t off
, uint64_t len
,
823 zilog_t
*zilog
= zv
->zv_zilog
;
825 if (zil_replaying(zilog
, tx
))
828 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
829 lr
= (lr_truncate_t
*)&itx
->itx_lr
;
830 lr
->lr_foid
= ZVOL_OBJ
;
834 itx
->itx_sync
= sync
;
835 zil_itx_assign(zilog
, itx
, tx
);
839 zvol_discard(void *arg
)
841 zv_request_t
*zvr
= arg
;
842 struct bio
*bio
= zvr
->bio
;
843 zvol_state_t
*zv
= zvr
->zv
;
844 uint64_t start
= BIO_BI_SECTOR(bio
) << 9;
845 uint64_t size
= BIO_BI_SIZE(bio
);
846 uint64_t end
= start
+ size
;
850 unsigned long start_jif
;
852 ASSERT(zv
&& zv
->zv_open_count
> 0);
855 blk_generic_start_io_acct(zv
->zv_queue
, WRITE
, bio_sectors(bio
),
856 &zv
->zv_disk
->part0
);
858 sync
= bio_is_fua(bio
) || zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
860 if (end
> zv
->zv_volsize
) {
861 error
= SET_ERROR(EIO
);
866 * Align the request to volume block boundaries when a secure erase is
867 * not required. This will prevent dnode_free_range() from zeroing out
868 * the unaligned parts which is slow (read-modify-write) and useless
869 * since we are not freeing any space by doing so.
871 if (!bio_is_secure_erase(bio
)) {
872 start
= P2ROUNDUP(start
, zv
->zv_volblocksize
);
873 end
= P2ALIGN(end
, zv
->zv_volblocksize
);
880 tx
= dmu_tx_create(zv
->zv_objset
);
881 dmu_tx_mark_netfree(tx
);
882 error
= dmu_tx_assign(tx
, TXG_WAIT
);
886 zvol_log_truncate(zv
, tx
, start
, size
, B_TRUE
);
888 error
= dmu_free_long_range(zv
->zv_objset
,
889 ZVOL_OBJ
, start
, size
);
892 zfs_range_unlock(zvr
->rl
);
893 if (error
== 0 && sync
)
894 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
896 rw_exit(&zv
->zv_suspend_lock
);
897 blk_generic_end_io_acct(zv
->zv_queue
, WRITE
, &zv
->zv_disk
->part0
,
899 BIO_END_IO(bio
, -error
);
900 kmem_free(zvr
, sizeof (zv_request_t
));
906 zv_request_t
*zvr
= arg
;
907 struct bio
*bio
= zvr
->bio
;
909 zvol_state_t
*zv
= zvr
->zv
;
910 uint64_t volsize
= zv
->zv_volsize
;
912 unsigned long start_jif
;
914 uio_from_bio(&uio
, bio
);
916 ASSERT(zv
&& zv
->zv_open_count
> 0);
919 blk_generic_start_io_acct(zv
->zv_queue
, READ
, bio_sectors(bio
),
920 &zv
->zv_disk
->part0
);
922 while (uio
.uio_resid
> 0 && uio
.uio_loffset
< volsize
) {
923 uint64_t bytes
= MIN(uio
.uio_resid
, DMU_MAX_ACCESS
>> 1);
925 /* don't read past the end */
926 if (bytes
> volsize
- uio
.uio_loffset
)
927 bytes
= volsize
- uio
.uio_loffset
;
929 error
= dmu_read_uio_dnode(zv
->zv_dn
, &uio
, bytes
);
931 /* convert checksum errors into IO errors */
933 error
= SET_ERROR(EIO
);
937 zfs_range_unlock(zvr
->rl
);
939 rw_exit(&zv
->zv_suspend_lock
);
940 blk_generic_end_io_acct(zv
->zv_queue
, READ
, &zv
->zv_disk
->part0
,
942 BIO_END_IO(bio
, -error
);
943 kmem_free(zvr
, sizeof (zv_request_t
));
946 static MAKE_REQUEST_FN_RET
947 zvol_request(struct request_queue
*q
, struct bio
*bio
)
949 zvol_state_t
*zv
= q
->queuedata
;
950 fstrans_cookie_t cookie
= spl_fstrans_mark();
951 uint64_t offset
= BIO_BI_SECTOR(bio
) << 9;
952 uint64_t size
= BIO_BI_SIZE(bio
);
953 int rw
= bio_data_dir(bio
);
956 if (bio_has_data(bio
) && offset
+ size
> zv
->zv_volsize
) {
958 "%s: bad access: offset=%llu, size=%lu\n",
959 zv
->zv_disk
->disk_name
,
960 (long long unsigned)offset
,
961 (long unsigned)size
);
963 BIO_END_IO(bio
, -SET_ERROR(EIO
));
968 boolean_t need_sync
= B_FALSE
;
970 if (unlikely(zv
->zv_flags
& ZVOL_RDONLY
)) {
971 BIO_END_IO(bio
, -SET_ERROR(EROFS
));
976 * To be released in the I/O function. See the comment on
977 * zfs_range_lock below.
979 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
981 /* bio marked as FLUSH need to flush before write */
982 if (bio_is_flush(bio
))
983 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
985 /* Some requests are just for flush and nothing else. */
987 rw_exit(&zv
->zv_suspend_lock
);
992 zvr
= kmem_alloc(sizeof (zv_request_t
), KM_SLEEP
);
997 * To be released in the I/O function. Since the I/O functions
998 * are asynchronous, we take it here synchronously to make
999 * sure overlapped I/Os are properly ordered.
1001 zvr
->rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
1004 * Sync writes and discards execute zil_commit() which may need
1005 * to take a RL_READER lock on the whole block being modified
1006 * via its zillog->zl_get_data(): to avoid circular dependency
1007 * issues with taskq threads execute these requests
1008 * synchronously here in zvol_request().
1010 need_sync
= bio_is_fua(bio
) ||
1011 zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
1012 if (bio_is_discard(bio
) || bio_is_secure_erase(bio
)) {
1013 if (zvol_request_sync
|| need_sync
||
1014 taskq_dispatch(zvol_taskq
, zvol_discard
, zvr
,
1015 TQ_SLEEP
) == TASKQID_INVALID
)
1018 if (zvol_request_sync
|| need_sync
||
1019 taskq_dispatch(zvol_taskq
, zvol_write
, zvr
,
1020 TQ_SLEEP
) == TASKQID_INVALID
)
1024 zvr
= kmem_alloc(sizeof (zv_request_t
), KM_SLEEP
);
1028 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1030 zvr
->rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
1032 if (zvol_request_sync
|| taskq_dispatch(zvol_taskq
,
1033 zvol_read
, zvr
, TQ_SLEEP
) == TASKQID_INVALID
)
1038 spl_fstrans_unmark(cookie
);
1039 #ifdef HAVE_MAKE_REQUEST_FN_RET_INT
1041 #elif defined(HAVE_MAKE_REQUEST_FN_RET_QC)
1042 return (BLK_QC_T_NONE
);
1047 zvol_get_done(zgd_t
*zgd
, int error
)
1050 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1052 zfs_range_unlock(zgd
->zgd_rl
);
1054 if (error
== 0 && zgd
->zgd_bp
)
1055 zil_lwb_add_block(zgd
->zgd_lwb
, zgd
->zgd_bp
);
1057 kmem_free(zgd
, sizeof (zgd_t
));
1061 * Get data to generate a TX_WRITE intent log record.
1064 zvol_get_data(void *arg
, lr_write_t
*lr
, char *buf
, struct lwb
*lwb
, zio_t
*zio
)
1066 zvol_state_t
*zv
= arg
;
1067 uint64_t offset
= lr
->lr_offset
;
1068 uint64_t size
= lr
->lr_length
;
1073 ASSERT3P(lwb
, !=, NULL
);
1074 ASSERT3P(zio
, !=, NULL
);
1075 ASSERT3U(size
, !=, 0);
1077 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1081 * Write records come in two flavors: immediate and indirect.
1082 * For small writes it's cheaper to store the data with the
1083 * log record (immediate); for large writes it's cheaper to
1084 * sync the data and get a pointer to it (indirect) so that
1085 * we don't have to write the data twice.
1087 if (buf
!= NULL
) { /* immediate write */
1088 zgd
->zgd_rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
1090 error
= dmu_read_by_dnode(zv
->zv_dn
, offset
, size
, buf
,
1091 DMU_READ_NO_PREFETCH
);
1092 } else { /* indirect write */
1094 * Have to lock the whole block to ensure when it's written out
1095 * and its checksum is being calculated that no one can change
1096 * the data. Contrarily to zfs_get_data we need not re-check
1097 * blocksize after we get the lock because it cannot be changed.
1099 size
= zv
->zv_volblocksize
;
1100 offset
= P2ALIGN_TYPED(offset
, size
, uint64_t);
1101 zgd
->zgd_rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
1103 error
= dmu_buf_hold_by_dnode(zv
->zv_dn
, offset
, zgd
, &db
,
1104 DMU_READ_NO_PREFETCH
);
1106 blkptr_t
*bp
= &lr
->lr_blkptr
;
1112 ASSERT(db
->db_offset
== offset
);
1113 ASSERT(db
->db_size
== size
);
1115 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1116 zvol_get_done
, zgd
);
1123 zvol_get_done(zgd
, error
);
1125 return (SET_ERROR(error
));
1129 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable.
1132 zvol_insert(zvol_state_t
*zv
)
1134 ASSERT(MUTEX_HELD(&zvol_state_lock
));
1135 ASSERT3U(MINOR(zv
->zv_dev
) & ZVOL_MINOR_MASK
, ==, 0);
1136 list_insert_head(&zvol_state_list
, zv
);
1137 hlist_add_head(&zv
->zv_hlink
, ZVOL_HT_HEAD(zv
->zv_hash
));
1141 * Simply remove the zvol from to list of zvols.
1144 zvol_remove(zvol_state_t
*zv
)
1146 ASSERT(MUTEX_HELD(&zvol_state_lock
));
1147 list_remove(&zvol_state_list
, zv
);
1148 hlist_del(&zv
->zv_hlink
);
1152 * Setup zv after we just own the zv->objset
1155 zvol_setup_zv(zvol_state_t
*zv
)
1160 objset_t
*os
= zv
->zv_objset
;
1162 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
) &&
1163 RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1165 error
= dsl_prop_get_integer(zv
->zv_name
, "readonly", &ro
, NULL
);
1167 return (SET_ERROR(error
));
1169 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
1171 return (SET_ERROR(error
));
1173 error
= dnode_hold(os
, ZVOL_OBJ
, FTAG
, &zv
->zv_dn
);
1175 return (SET_ERROR(error
));
1177 set_capacity(zv
->zv_disk
, volsize
>> 9);
1178 zv
->zv_volsize
= volsize
;
1179 zv
->zv_zilog
= zil_open(os
, zvol_get_data
);
1181 if (ro
|| dmu_objset_is_snapshot(os
) ||
1182 !spa_writeable(dmu_objset_spa(os
))) {
1183 set_disk_ro(zv
->zv_disk
, 1);
1184 zv
->zv_flags
|= ZVOL_RDONLY
;
1186 set_disk_ro(zv
->zv_disk
, 0);
1187 zv
->zv_flags
&= ~ZVOL_RDONLY
;
1193 * Shutdown every zv_objset related stuff except zv_objset itself.
1194 * The is the reverse of zvol_setup_zv.
1197 zvol_shutdown_zv(zvol_state_t
*zv
)
1199 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
) &&
1200 RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1202 zil_close(zv
->zv_zilog
);
1203 zv
->zv_zilog
= NULL
;
1205 dnode_rele(zv
->zv_dn
, FTAG
);
1211 if (dsl_dataset_is_dirty(dmu_objset_ds(zv
->zv_objset
)) &&
1212 !(zv
->zv_flags
& ZVOL_RDONLY
))
1213 txg_wait_synced(dmu_objset_pool(zv
->zv_objset
), 0);
1214 (void) dmu_objset_evict_dbufs(zv
->zv_objset
);
1218 * return the proper tag for rollback and recv
1221 zvol_tag(zvol_state_t
*zv
)
1223 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1224 return (zv
->zv_open_count
> 0 ? zv
: NULL
);
1228 * Suspend the zvol for recv and rollback.
1231 zvol_suspend(const char *name
)
1235 zv
= zvol_find_by_name(name
, RW_WRITER
);
1240 /* block all I/O, release in zvol_resume. */
1241 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
) &&
1242 RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1244 atomic_inc(&zv
->zv_suspend_ref
);
1246 if (zv
->zv_open_count
> 0)
1247 zvol_shutdown_zv(zv
);
1250 * do not hold zv_state_lock across suspend/resume to
1251 * avoid locking up zvol lookups
1253 mutex_exit(&zv
->zv_state_lock
);
1255 /* zv_suspend_lock is released in zvol_resume() */
1260 zvol_resume(zvol_state_t
*zv
)
1264 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1266 mutex_enter(&zv
->zv_state_lock
);
1268 if (zv
->zv_open_count
> 0) {
1269 VERIFY0(dmu_objset_hold(zv
->zv_name
, zv
, &zv
->zv_objset
));
1270 VERIFY3P(zv
->zv_objset
->os_dsl_dataset
->ds_owner
, ==, zv
);
1271 VERIFY(dsl_dataset_long_held(zv
->zv_objset
->os_dsl_dataset
));
1272 dmu_objset_rele(zv
->zv_objset
, zv
);
1274 error
= zvol_setup_zv(zv
);
1277 mutex_exit(&zv
->zv_state_lock
);
1279 rw_exit(&zv
->zv_suspend_lock
);
1281 * We need this because we don't hold zvol_state_lock while releasing
1282 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check
1283 * zv_suspend_lock to determine it is safe to free because rwlock is
1284 * not inherent atomic.
1286 atomic_dec(&zv
->zv_suspend_ref
);
1288 return (SET_ERROR(error
));
1292 zvol_first_open(zvol_state_t
*zv
, boolean_t readonly
)
1295 int error
, locked
= 0;
1298 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
1299 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1302 * In all other cases the spa_namespace_lock is taken before the
1303 * bdev->bd_mutex lock. But in this case the Linux __blkdev_get()
1304 * function calls fops->open() with the bdev->bd_mutex lock held.
1305 * This deadlock can be easily observed with zvols used as vdevs.
1307 * To avoid a potential lock inversion deadlock we preemptively
1308 * try to take the spa_namespace_lock(). Normally it will not
1309 * be contended and this is safe because spa_open_common() handles
1310 * the case where the caller already holds the spa_namespace_lock.
1312 * When it is contended we risk a lock inversion if we were to
1313 * block waiting for the lock. Luckily, the __blkdev_get()
1314 * function allows us to return -ERESTARTSYS which will result in
1315 * bdev->bd_mutex being dropped, reacquired, and fops->open() being
1316 * called again. This process can be repeated safely until both
1317 * locks are acquired.
1319 if (!mutex_owned(&spa_namespace_lock
)) {
1320 locked
= mutex_tryenter(&spa_namespace_lock
);
1322 return (-SET_ERROR(ERESTARTSYS
));
1325 ro
= (readonly
|| (strchr(zv
->zv_name
, '@') != NULL
));
1326 error
= dmu_objset_own(zv
->zv_name
, DMU_OST_ZVOL
, ro
, B_TRUE
, zv
, &os
);
1332 error
= zvol_setup_zv(zv
);
1335 dmu_objset_disown(os
, 1, zv
);
1336 zv
->zv_objset
= NULL
;
1341 mutex_exit(&spa_namespace_lock
);
1342 return (SET_ERROR(-error
));
1346 zvol_last_close(zvol_state_t
*zv
)
1348 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
1349 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1351 zvol_shutdown_zv(zv
);
1353 dmu_objset_disown(zv
->zv_objset
, 1, zv
);
1354 zv
->zv_objset
= NULL
;
1358 zvol_open(struct block_device
*bdev
, fmode_t flag
)
1362 boolean_t drop_suspend
= B_TRUE
;
1364 ASSERT(!MUTEX_HELD(&zvol_state_lock
));
1366 mutex_enter(&zvol_state_lock
);
1368 * Obtain a copy of private_data under the zvol_state_lock to make
1369 * sure that either the result of zvol free code path setting
1370 * bdev->bd_disk->private_data to NULL is observed, or zvol_free()
1371 * is not called on this zv because of the positive zv_open_count.
1373 zv
= bdev
->bd_disk
->private_data
;
1375 mutex_exit(&zvol_state_lock
);
1376 return (SET_ERROR(-ENXIO
));
1379 mutex_enter(&zv
->zv_state_lock
);
1381 * make sure zvol is not suspended during first open
1382 * (hold zv_suspend_lock) and respect proper lock acquisition
1383 * ordering - zv_suspend_lock before zv_state_lock
1385 if (zv
->zv_open_count
== 0) {
1386 if (!rw_tryenter(&zv
->zv_suspend_lock
, RW_READER
)) {
1387 mutex_exit(&zv
->zv_state_lock
);
1388 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1389 mutex_enter(&zv
->zv_state_lock
);
1390 /* check to see if zv_suspend_lock is needed */
1391 if (zv
->zv_open_count
!= 0) {
1392 rw_exit(&zv
->zv_suspend_lock
);
1393 drop_suspend
= B_FALSE
;
1397 drop_suspend
= B_FALSE
;
1399 mutex_exit(&zvol_state_lock
);
1401 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1402 ASSERT(zv
->zv_open_count
!= 0 || RW_READ_HELD(&zv
->zv_suspend_lock
));
1404 if (zv
->zv_open_count
== 0) {
1405 error
= zvol_first_open(zv
, !(flag
& FMODE_WRITE
));
1410 if ((flag
& FMODE_WRITE
) && (zv
->zv_flags
& ZVOL_RDONLY
)) {
1412 goto out_open_count
;
1415 zv
->zv_open_count
++;
1417 check_disk_change(bdev
);
1420 if (zv
->zv_open_count
== 0)
1421 zvol_last_close(zv
);
1423 mutex_exit(&zv
->zv_state_lock
);
1425 rw_exit(&zv
->zv_suspend_lock
);
1426 if (error
== -ERESTARTSYS
)
1429 return (SET_ERROR(error
));
1432 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1437 zvol_release(struct gendisk
*disk
, fmode_t mode
)
1440 boolean_t drop_suspend
= B_TRUE
;
1442 ASSERT(!MUTEX_HELD(&zvol_state_lock
));
1444 mutex_enter(&zvol_state_lock
);
1445 zv
= disk
->private_data
;
1447 mutex_enter(&zv
->zv_state_lock
);
1448 ASSERT(zv
->zv_open_count
> 0);
1450 * make sure zvol is not suspended during last close
1451 * (hold zv_suspend_lock) and respect proper lock acquisition
1452 * ordering - zv_suspend_lock before zv_state_lock
1454 if (zv
->zv_open_count
== 1) {
1455 if (!rw_tryenter(&zv
->zv_suspend_lock
, RW_READER
)) {
1456 mutex_exit(&zv
->zv_state_lock
);
1457 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1458 mutex_enter(&zv
->zv_state_lock
);
1459 /* check to see if zv_suspend_lock is needed */
1460 if (zv
->zv_open_count
!= 1) {
1461 rw_exit(&zv
->zv_suspend_lock
);
1462 drop_suspend
= B_FALSE
;
1466 drop_suspend
= B_FALSE
;
1468 mutex_exit(&zvol_state_lock
);
1470 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1471 ASSERT(zv
->zv_open_count
!= 1 || RW_READ_HELD(&zv
->zv_suspend_lock
));
1473 zv
->zv_open_count
--;
1474 if (zv
->zv_open_count
== 0)
1475 zvol_last_close(zv
);
1477 mutex_exit(&zv
->zv_state_lock
);
1480 rw_exit(&zv
->zv_suspend_lock
);
1482 #ifndef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1488 zvol_ioctl(struct block_device
*bdev
, fmode_t mode
,
1489 unsigned int cmd
, unsigned long arg
)
1491 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
1494 ASSERT(zv
&& zv
->zv_open_count
> 0);
1499 invalidate_bdev(bdev
);
1500 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1502 if (dsl_dataset_is_dirty(dmu_objset_ds(zv
->zv_objset
)) &&
1503 !(zv
->zv_flags
& ZVOL_RDONLY
))
1504 txg_wait_synced(dmu_objset_pool(zv
->zv_objset
), 0);
1506 rw_exit(&zv
->zv_suspend_lock
);
1510 mutex_enter(&zv
->zv_state_lock
);
1511 error
= copy_to_user((void *)arg
, zv
->zv_name
, MAXNAMELEN
);
1512 mutex_exit(&zv
->zv_state_lock
);
1520 return (SET_ERROR(error
));
1523 #ifdef CONFIG_COMPAT
1525 zvol_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1526 unsigned cmd
, unsigned long arg
)
1528 return (zvol_ioctl(bdev
, mode
, cmd
, arg
));
1531 #define zvol_compat_ioctl NULL
1534 static int zvol_media_changed(struct gendisk
*disk
)
1536 zvol_state_t
*zv
= disk
->private_data
;
1538 ASSERT(zv
&& zv
->zv_open_count
> 0);
1540 return (zv
->zv_changed
);
1543 static int zvol_revalidate_disk(struct gendisk
*disk
)
1545 zvol_state_t
*zv
= disk
->private_data
;
1547 ASSERT(zv
&& zv
->zv_open_count
> 0);
1550 set_capacity(zv
->zv_disk
, zv
->zv_volsize
>> 9);
1556 * Provide a simple virtual geometry for legacy compatibility. For devices
1557 * smaller than 1 MiB a small head and sector count is used to allow very
1558 * tiny devices. For devices over 1 Mib a standard head and sector count
1559 * is used to keep the cylinders count reasonable.
1562 zvol_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1564 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
1567 ASSERT(zv
&& zv
->zv_open_count
> 0);
1569 sectors
= get_capacity(zv
->zv_disk
);
1571 if (sectors
> 2048) {
1580 geo
->cylinders
= sectors
/ (geo
->heads
* geo
->sectors
);
1585 static struct kobject
*
1586 zvol_probe(dev_t dev
, int *part
, void *arg
)
1589 struct kobject
*kobj
;
1591 zv
= zvol_find_by_dev(dev
);
1592 kobj
= zv
? get_disk(zv
->zv_disk
) : NULL
;
1593 ASSERT(zv
== NULL
|| MUTEX_HELD(&zv
->zv_state_lock
));
1595 mutex_exit(&zv
->zv_state_lock
);
1600 #ifdef HAVE_BDEV_BLOCK_DEVICE_OPERATIONS
1601 static struct block_device_operations zvol_ops
= {
1603 .release
= zvol_release
,
1604 .ioctl
= zvol_ioctl
,
1605 .compat_ioctl
= zvol_compat_ioctl
,
1606 .media_changed
= zvol_media_changed
,
1607 .revalidate_disk
= zvol_revalidate_disk
,
1608 .getgeo
= zvol_getgeo
,
1609 .owner
= THIS_MODULE
,
1612 #else /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */
1615 zvol_open_by_inode(struct inode
*inode
, struct file
*file
)
1617 return (zvol_open(inode
->i_bdev
, file
->f_mode
));
1621 zvol_release_by_inode(struct inode
*inode
, struct file
*file
)
1623 return (zvol_release(inode
->i_bdev
->bd_disk
, file
->f_mode
));
1627 zvol_ioctl_by_inode(struct inode
*inode
, struct file
*file
,
1628 unsigned int cmd
, unsigned long arg
)
1630 if (file
== NULL
|| inode
== NULL
)
1631 return (SET_ERROR(-EINVAL
));
1633 return (zvol_ioctl(inode
->i_bdev
, file
->f_mode
, cmd
, arg
));
1636 #ifdef CONFIG_COMPAT
1638 zvol_compat_ioctl_by_inode(struct file
*file
,
1639 unsigned int cmd
, unsigned long arg
)
1642 return (SET_ERROR(-EINVAL
));
1644 return (zvol_compat_ioctl(file
->f_dentry
->d_inode
->i_bdev
,
1645 file
->f_mode
, cmd
, arg
));
1648 #define zvol_compat_ioctl_by_inode NULL
1651 static struct block_device_operations zvol_ops
= {
1652 .open
= zvol_open_by_inode
,
1653 .release
= zvol_release_by_inode
,
1654 .ioctl
= zvol_ioctl_by_inode
,
1655 .compat_ioctl
= zvol_compat_ioctl_by_inode
,
1656 .media_changed
= zvol_media_changed
,
1657 .revalidate_disk
= zvol_revalidate_disk
,
1658 .getgeo
= zvol_getgeo
,
1659 .owner
= THIS_MODULE
,
1661 #endif /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */
1664 * Allocate memory for a new zvol_state_t and setup the required
1665 * request queue and generic disk structures for the block device.
1667 static zvol_state_t
*
1668 zvol_alloc(dev_t dev
, const char *name
)
1673 if (dsl_prop_get_integer(name
, "volmode", &volmode
, NULL
) != 0)
1676 if (volmode
== ZFS_VOLMODE_DEFAULT
)
1677 volmode
= zvol_volmode
;
1679 if (volmode
== ZFS_VOLMODE_NONE
)
1682 zv
= kmem_zalloc(sizeof (zvol_state_t
), KM_SLEEP
);
1684 list_link_init(&zv
->zv_next
);
1686 mutex_init(&zv
->zv_state_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1688 zv
->zv_queue
= blk_alloc_queue(GFP_ATOMIC
);
1689 if (zv
->zv_queue
== NULL
)
1692 blk_queue_make_request(zv
->zv_queue
, zvol_request
);
1693 blk_queue_set_write_cache(zv
->zv_queue
, B_TRUE
, B_TRUE
);
1695 /* Limit read-ahead to a single page to prevent over-prefetching. */
1696 blk_queue_set_read_ahead(zv
->zv_queue
, 1);
1698 /* Disable write merging in favor of the ZIO pipeline. */
1699 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES
, zv
->zv_queue
);
1701 zv
->zv_disk
= alloc_disk(ZVOL_MINORS
);
1702 if (zv
->zv_disk
== NULL
)
1705 zv
->zv_queue
->queuedata
= zv
;
1707 zv
->zv_open_count
= 0;
1708 strlcpy(zv
->zv_name
, name
, MAXNAMELEN
);
1710 zfs_rlock_init(&zv
->zv_range_lock
);
1711 rw_init(&zv
->zv_suspend_lock
, NULL
, RW_DEFAULT
, NULL
);
1713 zv
->zv_disk
->major
= zvol_major
;
1714 if (volmode
== ZFS_VOLMODE_DEV
) {
1716 * ZFS_VOLMODE_DEV disable partitioning on ZVOL devices: set
1717 * gendisk->minors = 1 as noted in include/linux/genhd.h.
1718 * Also disable extended partition numbers (GENHD_FL_EXT_DEVT)
1719 * and suppresses partition scanning (GENHD_FL_NO_PART_SCAN)
1720 * setting gendisk->flags accordingly.
1722 zv
->zv_disk
->minors
= 1;
1723 #if defined(GENHD_FL_EXT_DEVT)
1724 zv
->zv_disk
->flags
&= ~GENHD_FL_EXT_DEVT
;
1726 #if defined(GENHD_FL_NO_PART_SCAN)
1727 zv
->zv_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1730 zv
->zv_disk
->first_minor
= (dev
& MINORMASK
);
1731 zv
->zv_disk
->fops
= &zvol_ops
;
1732 zv
->zv_disk
->private_data
= zv
;
1733 zv
->zv_disk
->queue
= zv
->zv_queue
;
1734 snprintf(zv
->zv_disk
->disk_name
, DISK_NAME_LEN
, "%s%d",
1735 ZVOL_DEV_NAME
, (dev
& MINORMASK
));
1740 blk_cleanup_queue(zv
->zv_queue
);
1742 kmem_free(zv
, sizeof (zvol_state_t
));
1748 * Cleanup then free a zvol_state_t which was created by zvol_alloc().
1749 * At this time, the structure is not opened by anyone, is taken off
1750 * the zvol_state_list, and has its private data set to NULL.
1751 * The zvol_state_lock is dropped.
1754 zvol_free(void *arg
)
1756 zvol_state_t
*zv
= arg
;
1758 ASSERT(!MUTEX_HELD(&zvol_state_lock
));
1759 ASSERT(!RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1760 ASSERT(!MUTEX_HELD(&zv
->zv_state_lock
));
1761 ASSERT(zv
->zv_open_count
== 0);
1762 ASSERT(zv
->zv_disk
->private_data
== NULL
);
1764 rw_destroy(&zv
->zv_suspend_lock
);
1765 zfs_rlock_destroy(&zv
->zv_range_lock
);
1767 del_gendisk(zv
->zv_disk
);
1768 blk_cleanup_queue(zv
->zv_queue
);
1769 put_disk(zv
->zv_disk
);
1771 ida_simple_remove(&zvol_ida
, MINOR(zv
->zv_dev
) >> ZVOL_MINOR_BITS
);
1773 mutex_destroy(&zv
->zv_state_lock
);
1775 kmem_free(zv
, sizeof (zvol_state_t
));
1779 * Create a block device minor node and setup the linkage between it
1780 * and the specified volume. Once this function returns the block
1781 * device is live and ready for use.
1784 zvol_create_minor_impl(const char *name
)
1788 dmu_object_info_t
*doi
;
1794 uint64_t hash
= zvol_name_hash(name
);
1796 if (zvol_inhibit_dev
)
1799 idx
= ida_simple_get(&zvol_ida
, 0, 0, kmem_flags_convert(KM_SLEEP
));
1801 return (SET_ERROR(-idx
));
1802 minor
= idx
<< ZVOL_MINOR_BITS
;
1804 zv
= zvol_find_by_name_hash(name
, hash
, RW_NONE
);
1806 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1807 mutex_exit(&zv
->zv_state_lock
);
1808 ida_simple_remove(&zvol_ida
, idx
);
1809 return (SET_ERROR(EEXIST
));
1812 doi
= kmem_alloc(sizeof (dmu_object_info_t
), KM_SLEEP
);
1814 error
= dmu_objset_own(name
, DMU_OST_ZVOL
, B_TRUE
, B_TRUE
, FTAG
, &os
);
1818 error
= dmu_object_info(os
, ZVOL_OBJ
, doi
);
1820 goto out_dmu_objset_disown
;
1822 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
1824 goto out_dmu_objset_disown
;
1826 zv
= zvol_alloc(MKDEV(zvol_major
, minor
), name
);
1828 error
= SET_ERROR(EAGAIN
);
1829 goto out_dmu_objset_disown
;
1833 if (dmu_objset_is_snapshot(os
))
1834 zv
->zv_flags
|= ZVOL_RDONLY
;
1836 zv
->zv_volblocksize
= doi
->doi_data_block_size
;
1837 zv
->zv_volsize
= volsize
;
1840 set_capacity(zv
->zv_disk
, zv
->zv_volsize
>> 9);
1842 blk_queue_max_hw_sectors(zv
->zv_queue
, (DMU_MAX_ACCESS
/ 4) >> 9);
1843 blk_queue_max_segments(zv
->zv_queue
, UINT16_MAX
);
1844 blk_queue_max_segment_size(zv
->zv_queue
, UINT_MAX
);
1845 blk_queue_physical_block_size(zv
->zv_queue
, zv
->zv_volblocksize
);
1846 blk_queue_io_opt(zv
->zv_queue
, zv
->zv_volblocksize
);
1847 blk_queue_max_discard_sectors(zv
->zv_queue
,
1848 (zvol_max_discard_blocks
* zv
->zv_volblocksize
) >> 9);
1849 blk_queue_discard_granularity(zv
->zv_queue
, zv
->zv_volblocksize
);
1850 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, zv
->zv_queue
);
1851 #ifdef QUEUE_FLAG_NONROT
1852 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, zv
->zv_queue
);
1854 #ifdef QUEUE_FLAG_ADD_RANDOM
1855 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, zv
->zv_queue
);
1858 if (spa_writeable(dmu_objset_spa(os
))) {
1859 if (zil_replay_disable
)
1860 zil_destroy(dmu_objset_zil(os
), B_FALSE
);
1862 zil_replay(os
, zv
, zvol_replay_vector
);
1866 * When udev detects the addition of the device it will immediately
1867 * invoke blkid(8) to determine the type of content on the device.
1868 * Prefetching the blocks commonly scanned by blkid(8) will speed
1871 len
= MIN(MAX(zvol_prefetch_bytes
, 0), SPA_MAXBLOCKSIZE
);
1873 dmu_prefetch(os
, ZVOL_OBJ
, 0, 0, len
, ZIO_PRIORITY_SYNC_READ
);
1874 dmu_prefetch(os
, ZVOL_OBJ
, 0, volsize
- len
, len
,
1875 ZIO_PRIORITY_SYNC_READ
);
1878 zv
->zv_objset
= NULL
;
1879 out_dmu_objset_disown
:
1880 dmu_objset_disown(os
, B_TRUE
, FTAG
);
1882 kmem_free(doi
, sizeof (dmu_object_info_t
));
1885 mutex_enter(&zvol_state_lock
);
1887 mutex_exit(&zvol_state_lock
);
1888 add_disk(zv
->zv_disk
);
1890 ida_simple_remove(&zvol_ida
, idx
);
1893 return (SET_ERROR(error
));
1897 * Rename a block device minor mode for the specified volume.
1900 zvol_rename_minor(zvol_state_t
*zv
, const char *newname
)
1902 int readonly
= get_disk_ro(zv
->zv_disk
);
1904 ASSERT(MUTEX_HELD(&zvol_state_lock
));
1905 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1907 strlcpy(zv
->zv_name
, newname
, sizeof (zv
->zv_name
));
1909 /* move to new hashtable entry */
1910 zv
->zv_hash
= zvol_name_hash(zv
->zv_name
);
1911 hlist_del(&zv
->zv_hlink
);
1912 hlist_add_head(&zv
->zv_hlink
, ZVOL_HT_HEAD(zv
->zv_hash
));
1915 * The block device's read-only state is briefly changed causing
1916 * a KOBJ_CHANGE uevent to be issued. This ensures udev detects
1917 * the name change and fixes the symlinks. This does not change
1918 * ZVOL_RDONLY in zv->zv_flags so the actual read-only state never
1919 * changes. This would normally be done using kobject_uevent() but
1920 * that is a GPL-only symbol which is why we need this workaround.
1922 set_disk_ro(zv
->zv_disk
, !readonly
);
1923 set_disk_ro(zv
->zv_disk
, readonly
);
1926 typedef struct minors_job
{
1936 * Prefetch zvol dnodes for the minors_job
1939 zvol_prefetch_minors_impl(void *arg
)
1941 minors_job_t
*job
= arg
;
1942 char *dsname
= job
->name
;
1943 objset_t
*os
= NULL
;
1945 job
->error
= dmu_objset_own(dsname
, DMU_OST_ZVOL
, B_TRUE
, B_TRUE
,
1947 if (job
->error
== 0) {
1948 dmu_prefetch(os
, ZVOL_OBJ
, 0, 0, 0, ZIO_PRIORITY_SYNC_READ
);
1949 dmu_objset_disown(os
, B_TRUE
, FTAG
);
1954 * Mask errors to continue dmu_objset_find() traversal
1957 zvol_create_snap_minor_cb(const char *dsname
, void *arg
)
1959 minors_job_t
*j
= arg
;
1960 list_t
*minors_list
= j
->list
;
1961 const char *name
= j
->name
;
1963 ASSERT0(MUTEX_HELD(&spa_namespace_lock
));
1965 /* skip the designated dataset */
1966 if (name
&& strcmp(dsname
, name
) == 0)
1969 /* at this point, the dsname should name a snapshot */
1970 if (strchr(dsname
, '@') == 0) {
1971 dprintf("zvol_create_snap_minor_cb(): "
1972 "%s is not a shapshot name\n", dsname
);
1975 char *n
= strdup(dsname
);
1979 job
= kmem_alloc(sizeof (minors_job_t
), KM_SLEEP
);
1981 job
->list
= minors_list
;
1983 list_insert_tail(minors_list
, job
);
1984 /* don't care if dispatch fails, because job->error is 0 */
1985 taskq_dispatch(system_taskq
, zvol_prefetch_minors_impl
, job
,
1993 * Mask errors to continue dmu_objset_find() traversal
1996 zvol_create_minors_cb(const char *dsname
, void *arg
)
2000 list_t
*minors_list
= arg
;
2002 ASSERT0(MUTEX_HELD(&spa_namespace_lock
));
2004 error
= dsl_prop_get_integer(dsname
, "snapdev", &snapdev
, NULL
);
2009 * Given the name and the 'snapdev' property, create device minor nodes
2010 * with the linkages to zvols/snapshots as needed.
2011 * If the name represents a zvol, create a minor node for the zvol, then
2012 * check if its snapshots are 'visible', and if so, iterate over the
2013 * snapshots and create device minor nodes for those.
2015 if (strchr(dsname
, '@') == 0) {
2017 char *n
= strdup(dsname
);
2021 job
= kmem_alloc(sizeof (minors_job_t
), KM_SLEEP
);
2023 job
->list
= minors_list
;
2025 list_insert_tail(minors_list
, job
);
2026 /* don't care if dispatch fails, because job->error is 0 */
2027 taskq_dispatch(system_taskq
, zvol_prefetch_minors_impl
, job
,
2030 if (snapdev
== ZFS_SNAPDEV_VISIBLE
) {
2032 * traverse snapshots only, do not traverse children,
2033 * and skip the 'dsname'
2035 error
= dmu_objset_find((char *)dsname
,
2036 zvol_create_snap_minor_cb
, (void *)job
,
2040 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n",
2048 * Create minors for the specified dataset, including children and snapshots.
2049 * Pay attention to the 'snapdev' property and iterate over the snapshots
2050 * only if they are 'visible'. This approach allows one to assure that the
2051 * snapshot metadata is read from disk only if it is needed.
2053 * The name can represent a dataset to be recursively scanned for zvols and
2054 * their snapshots, or a single zvol snapshot. If the name represents a
2055 * dataset, the scan is performed in two nested stages:
2056 * - scan the dataset for zvols, and
2057 * - for each zvol, create a minor node, then check if the zvol's snapshots
2058 * are 'visible', and only then iterate over the snapshots if needed
2060 * If the name represents a snapshot, a check is performed if the snapshot is
2061 * 'visible' (which also verifies that the parent is a zvol), and if so,
2062 * a minor node for that snapshot is created.
2065 zvol_create_minors_impl(const char *name
)
2068 fstrans_cookie_t cookie
;
2073 if (zvol_inhibit_dev
)
2077 * This is the list for prefetch jobs. Whenever we found a match
2078 * during dmu_objset_find, we insert a minors_job to the list and do
2079 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need
2080 * any lock because all list operation is done on the current thread.
2082 * We will use this list to do zvol_create_minor_impl after prefetch
2083 * so we don't have to traverse using dmu_objset_find again.
2085 list_create(&minors_list
, sizeof (minors_job_t
),
2086 offsetof(minors_job_t
, link
));
2088 parent
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
2089 (void) strlcpy(parent
, name
, MAXPATHLEN
);
2091 if ((atp
= strrchr(parent
, '@')) != NULL
) {
2095 error
= dsl_prop_get_integer(parent
, "snapdev",
2098 if (error
== 0 && snapdev
== ZFS_SNAPDEV_VISIBLE
)
2099 error
= zvol_create_minor_impl(name
);
2101 cookie
= spl_fstrans_mark();
2102 error
= dmu_objset_find(parent
, zvol_create_minors_cb
,
2103 &minors_list
, DS_FIND_CHILDREN
);
2104 spl_fstrans_unmark(cookie
);
2107 kmem_free(parent
, MAXPATHLEN
);
2108 taskq_wait_outstanding(system_taskq
, 0);
2111 * Prefetch is completed, we can do zvol_create_minor_impl
2114 while ((job
= list_head(&minors_list
)) != NULL
) {
2115 list_remove(&minors_list
, job
);
2117 zvol_create_minor_impl(job
->name
);
2119 kmem_free(job
, sizeof (minors_job_t
));
2122 list_destroy(&minors_list
);
2124 return (SET_ERROR(error
));
2128 * Remove minors for specified dataset including children and snapshots.
2131 zvol_remove_minors_impl(const char *name
)
2133 zvol_state_t
*zv
, *zv_next
;
2134 int namelen
= ((name
) ? strlen(name
) : 0);
2135 taskqid_t t
, tid
= TASKQID_INVALID
;
2138 if (zvol_inhibit_dev
)
2141 list_create(&free_list
, sizeof (zvol_state_t
),
2142 offsetof(zvol_state_t
, zv_next
));
2144 mutex_enter(&zvol_state_lock
);
2146 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2147 zv_next
= list_next(&zvol_state_list
, zv
);
2149 mutex_enter(&zv
->zv_state_lock
);
2150 if (name
== NULL
|| strcmp(zv
->zv_name
, name
) == 0 ||
2151 (strncmp(zv
->zv_name
, name
, namelen
) == 0 &&
2152 (zv
->zv_name
[namelen
] == '/' ||
2153 zv
->zv_name
[namelen
] == '@'))) {
2155 * By holding zv_state_lock here, we guarantee that no
2156 * one is currently using this zv
2159 /* If in use, leave alone */
2160 if (zv
->zv_open_count
> 0 ||
2161 atomic_read(&zv
->zv_suspend_ref
)) {
2162 mutex_exit(&zv
->zv_state_lock
);
2169 * clear this while holding zvol_state_lock so
2170 * zvol_open won't open it
2172 zv
->zv_disk
->private_data
= NULL
;
2174 /* Drop zv_state_lock before zvol_free() */
2175 mutex_exit(&zv
->zv_state_lock
);
2177 /* try parallel zv_free, if failed do it in place */
2178 t
= taskq_dispatch(system_taskq
, zvol_free
, zv
,
2180 if (t
== TASKQID_INVALID
)
2181 list_insert_head(&free_list
, zv
);
2185 mutex_exit(&zv
->zv_state_lock
);
2188 mutex_exit(&zvol_state_lock
);
2191 * Drop zvol_state_lock before calling zvol_free()
2193 while ((zv
= list_head(&free_list
)) != NULL
) {
2194 list_remove(&free_list
, zv
);
2198 if (tid
!= TASKQID_INVALID
)
2199 taskq_wait_outstanding(system_taskq
, tid
);
2202 /* Remove minor for this specific volume only */
2204 zvol_remove_minor_impl(const char *name
)
2206 zvol_state_t
*zv
= NULL
, *zv_next
;
2208 if (zvol_inhibit_dev
)
2211 mutex_enter(&zvol_state_lock
);
2213 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2214 zv_next
= list_next(&zvol_state_list
, zv
);
2216 mutex_enter(&zv
->zv_state_lock
);
2217 if (strcmp(zv
->zv_name
, name
) == 0) {
2219 * By holding zv_state_lock here, we guarantee that no
2220 * one is currently using this zv
2223 /* If in use, leave alone */
2224 if (zv
->zv_open_count
> 0 ||
2225 atomic_read(&zv
->zv_suspend_ref
)) {
2226 mutex_exit(&zv
->zv_state_lock
);
2231 /* clear this so zvol_open won't open it */
2232 zv
->zv_disk
->private_data
= NULL
;
2234 mutex_exit(&zv
->zv_state_lock
);
2237 mutex_exit(&zv
->zv_state_lock
);
2241 /* Drop zvol_state_lock before calling zvol_free() */
2242 mutex_exit(&zvol_state_lock
);
2249 * Rename minors for specified dataset including children and snapshots.
2252 zvol_rename_minors_impl(const char *oldname
, const char *newname
)
2254 zvol_state_t
*zv
, *zv_next
;
2255 int oldnamelen
, newnamelen
;
2257 if (zvol_inhibit_dev
)
2260 oldnamelen
= strlen(oldname
);
2261 newnamelen
= strlen(newname
);
2263 mutex_enter(&zvol_state_lock
);
2265 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2266 zv_next
= list_next(&zvol_state_list
, zv
);
2268 mutex_enter(&zv
->zv_state_lock
);
2270 /* If in use, leave alone */
2271 if (zv
->zv_open_count
> 0) {
2272 mutex_exit(&zv
->zv_state_lock
);
2276 if (strcmp(zv
->zv_name
, oldname
) == 0) {
2277 zvol_rename_minor(zv
, newname
);
2278 } else if (strncmp(zv
->zv_name
, oldname
, oldnamelen
) == 0 &&
2279 (zv
->zv_name
[oldnamelen
] == '/' ||
2280 zv
->zv_name
[oldnamelen
] == '@')) {
2281 char *name
= kmem_asprintf("%s%c%s", newname
,
2282 zv
->zv_name
[oldnamelen
],
2283 zv
->zv_name
+ oldnamelen
+ 1);
2284 zvol_rename_minor(zv
, name
);
2285 kmem_free(name
, strlen(name
+ 1));
2288 mutex_exit(&zv
->zv_state_lock
);
2291 mutex_exit(&zvol_state_lock
);
2294 typedef struct zvol_snapdev_cb_arg
{
2296 } zvol_snapdev_cb_arg_t
;
2299 zvol_set_snapdev_cb(const char *dsname
, void *param
)
2301 zvol_snapdev_cb_arg_t
*arg
= param
;
2303 if (strchr(dsname
, '@') == NULL
)
2306 switch (arg
->snapdev
) {
2307 case ZFS_SNAPDEV_VISIBLE
:
2308 (void) zvol_create_minor_impl(dsname
);
2310 case ZFS_SNAPDEV_HIDDEN
:
2311 (void) zvol_remove_minor_impl(dsname
);
2319 zvol_set_snapdev_impl(char *name
, uint64_t snapdev
)
2321 zvol_snapdev_cb_arg_t arg
= {snapdev
};
2322 fstrans_cookie_t cookie
= spl_fstrans_mark();
2324 * The zvol_set_snapdev_sync() sets snapdev appropriately
2325 * in the dataset hierarchy. Here, we only scan snapshots.
2327 dmu_objset_find(name
, zvol_set_snapdev_cb
, &arg
, DS_FIND_SNAPSHOTS
);
2328 spl_fstrans_unmark(cookie
);
2331 typedef struct zvol_volmode_cb_arg
{
2333 } zvol_volmode_cb_arg_t
;
2336 zvol_set_volmode_impl(char *name
, uint64_t volmode
)
2338 fstrans_cookie_t cookie
= spl_fstrans_mark();
2340 if (strchr(name
, '@') != NULL
)
2344 * It's unfortunate we need to remove minors before we create new ones:
2345 * this is necessary because our backing gendisk (zvol_state->zv_disk)
2346 * coule be different when we set, for instance, volmode from "geom"
2347 * to "dev" (or vice versa).
2348 * A possible optimization is to modify our consumers so we don't get
2349 * called when "volmode" does not change.
2352 case ZFS_VOLMODE_NONE
:
2353 (void) zvol_remove_minor_impl(name
);
2355 case ZFS_VOLMODE_GEOM
:
2356 case ZFS_VOLMODE_DEV
:
2357 (void) zvol_remove_minor_impl(name
);
2358 (void) zvol_create_minor_impl(name
);
2360 case ZFS_VOLMODE_DEFAULT
:
2361 (void) zvol_remove_minor_impl(name
);
2362 if (zvol_volmode
== ZFS_VOLMODE_NONE
)
2364 else /* if zvol_volmode is invalid defaults to "geom" */
2365 (void) zvol_create_minor_impl(name
);
2369 spl_fstrans_unmark(cookie
);
2372 static zvol_task_t
*
2373 zvol_task_alloc(zvol_async_op_t op
, const char *name1
, const char *name2
,
2379 /* Never allow tasks on hidden names. */
2380 if (name1
[0] == '$')
2383 task
= kmem_zalloc(sizeof (zvol_task_t
), KM_SLEEP
);
2385 task
->value
= value
;
2386 delim
= strchr(name1
, '/');
2387 strlcpy(task
->pool
, name1
, delim
? (delim
- name1
+ 1) : MAXNAMELEN
);
2389 strlcpy(task
->name1
, name1
, MAXNAMELEN
);
2391 strlcpy(task
->name2
, name2
, MAXNAMELEN
);
2397 zvol_task_free(zvol_task_t
*task
)
2399 kmem_free(task
, sizeof (zvol_task_t
));
2403 * The worker thread function performed asynchronously.
2406 zvol_task_cb(void *param
)
2408 zvol_task_t
*task
= (zvol_task_t
*)param
;
2411 case ZVOL_ASYNC_CREATE_MINORS
:
2412 (void) zvol_create_minors_impl(task
->name1
);
2414 case ZVOL_ASYNC_REMOVE_MINORS
:
2415 zvol_remove_minors_impl(task
->name1
);
2417 case ZVOL_ASYNC_RENAME_MINORS
:
2418 zvol_rename_minors_impl(task
->name1
, task
->name2
);
2420 case ZVOL_ASYNC_SET_SNAPDEV
:
2421 zvol_set_snapdev_impl(task
->name1
, task
->value
);
2423 case ZVOL_ASYNC_SET_VOLMODE
:
2424 zvol_set_volmode_impl(task
->name1
, task
->value
);
2431 zvol_task_free(task
);
2434 typedef struct zvol_set_prop_int_arg
{
2435 const char *zsda_name
;
2436 uint64_t zsda_value
;
2437 zprop_source_t zsda_source
;
2439 } zvol_set_prop_int_arg_t
;
2442 * Sanity check the dataset for safe use by the sync task. No additional
2443 * conditions are imposed.
2446 zvol_set_snapdev_check(void *arg
, dmu_tx_t
*tx
)
2448 zvol_set_prop_int_arg_t
*zsda
= arg
;
2449 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2453 error
= dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
);
2457 dsl_dir_rele(dd
, FTAG
);
2464 zvol_set_snapdev_sync_cb(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2466 char dsname
[MAXNAMELEN
];
2470 dsl_dataset_name(ds
, dsname
);
2471 if (dsl_prop_get_int_ds(ds
, "snapdev", &snapdev
) != 0)
2473 task
= zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV
, dsname
, NULL
, snapdev
);
2477 (void) taskq_dispatch(dp
->dp_spa
->spa_zvol_taskq
, zvol_task_cb
,
2483 * Traverse all child datasets and apply snapdev appropriately.
2484 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
2485 * dataset and read the effective "snapdev" on every child in the callback
2486 * function: this is because the value is not guaranteed to be the same in the
2487 * whole dataset hierarchy.
2490 zvol_set_snapdev_sync(void *arg
, dmu_tx_t
*tx
)
2492 zvol_set_prop_int_arg_t
*zsda
= arg
;
2493 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2498 VERIFY0(dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
));
2501 error
= dsl_dataset_hold(dp
, zsda
->zsda_name
, FTAG
, &ds
);
2503 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_SNAPDEV
),
2504 zsda
->zsda_source
, sizeof (zsda
->zsda_value
), 1,
2505 &zsda
->zsda_value
, zsda
->zsda_tx
);
2506 dsl_dataset_rele(ds
, FTAG
);
2508 dmu_objset_find_dp(dp
, dd
->dd_object
, zvol_set_snapdev_sync_cb
,
2509 zsda
, DS_FIND_CHILDREN
);
2511 dsl_dir_rele(dd
, FTAG
);
2515 zvol_set_snapdev(const char *ddname
, zprop_source_t source
, uint64_t snapdev
)
2517 zvol_set_prop_int_arg_t zsda
;
2519 zsda
.zsda_name
= ddname
;
2520 zsda
.zsda_source
= source
;
2521 zsda
.zsda_value
= snapdev
;
2523 return (dsl_sync_task(ddname
, zvol_set_snapdev_check
,
2524 zvol_set_snapdev_sync
, &zsda
, 0, ZFS_SPACE_CHECK_NONE
));
2528 * Sanity check the dataset for safe use by the sync task. No additional
2529 * conditions are imposed.
2532 zvol_set_volmode_check(void *arg
, dmu_tx_t
*tx
)
2534 zvol_set_prop_int_arg_t
*zsda
= arg
;
2535 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2539 error
= dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
);
2543 dsl_dir_rele(dd
, FTAG
);
2550 zvol_set_volmode_sync_cb(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2552 char dsname
[MAXNAMELEN
];
2556 dsl_dataset_name(ds
, dsname
);
2557 if (dsl_prop_get_int_ds(ds
, "volmode", &volmode
) != 0)
2559 task
= zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE
, dsname
, NULL
, volmode
);
2563 (void) taskq_dispatch(dp
->dp_spa
->spa_zvol_taskq
, zvol_task_cb
,
2569 * Traverse all child datasets and apply volmode appropriately.
2570 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
2571 * dataset and read the effective "volmode" on every child in the callback
2572 * function: this is because the value is not guaranteed to be the same in the
2573 * whole dataset hierarchy.
2576 zvol_set_volmode_sync(void *arg
, dmu_tx_t
*tx
)
2578 zvol_set_prop_int_arg_t
*zsda
= arg
;
2579 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2584 VERIFY0(dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
));
2587 error
= dsl_dataset_hold(dp
, zsda
->zsda_name
, FTAG
, &ds
);
2589 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_VOLMODE
),
2590 zsda
->zsda_source
, sizeof (zsda
->zsda_value
), 1,
2591 &zsda
->zsda_value
, zsda
->zsda_tx
);
2592 dsl_dataset_rele(ds
, FTAG
);
2595 dmu_objset_find_dp(dp
, dd
->dd_object
, zvol_set_volmode_sync_cb
,
2596 zsda
, DS_FIND_CHILDREN
);
2598 dsl_dir_rele(dd
, FTAG
);
2602 zvol_set_volmode(const char *ddname
, zprop_source_t source
, uint64_t volmode
)
2604 zvol_set_prop_int_arg_t zsda
;
2606 zsda
.zsda_name
= ddname
;
2607 zsda
.zsda_source
= source
;
2608 zsda
.zsda_value
= volmode
;
2610 return (dsl_sync_task(ddname
, zvol_set_volmode_check
,
2611 zvol_set_volmode_sync
, &zsda
, 0, ZFS_SPACE_CHECK_NONE
));
2615 zvol_create_minors(spa_t
*spa
, const char *name
, boolean_t async
)
2620 task
= zvol_task_alloc(ZVOL_ASYNC_CREATE_MINORS
, name
, NULL
, ~0ULL);
2624 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2625 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2626 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2630 zvol_remove_minors(spa_t
*spa
, const char *name
, boolean_t async
)
2635 task
= zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS
, name
, NULL
, ~0ULL);
2639 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2640 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2641 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2645 zvol_rename_minors(spa_t
*spa
, const char *name1
, const char *name2
,
2651 task
= zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS
, name1
, name2
, ~0ULL);
2655 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2656 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2657 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2663 int threads
= MIN(MAX(zvol_threads
, 1), 1024);
2666 list_create(&zvol_state_list
, sizeof (zvol_state_t
),
2667 offsetof(zvol_state_t
, zv_next
));
2668 mutex_init(&zvol_state_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2669 ida_init(&zvol_ida
);
2671 zvol_taskq
= taskq_create(ZVOL_DRIVER
, threads
, maxclsyspri
,
2672 threads
* 2, INT_MAX
, TASKQ_PREPOPULATE
| TASKQ_DYNAMIC
);
2673 if (zvol_taskq
== NULL
) {
2674 printk(KERN_INFO
"ZFS: taskq_create() failed\n");
2679 zvol_htable
= kmem_alloc(ZVOL_HT_SIZE
* sizeof (struct hlist_head
),
2685 for (i
= 0; i
< ZVOL_HT_SIZE
; i
++)
2686 INIT_HLIST_HEAD(&zvol_htable
[i
]);
2688 error
= register_blkdev(zvol_major
, ZVOL_DRIVER
);
2690 printk(KERN_INFO
"ZFS: register_blkdev() failed %d\n", error
);
2694 blk_register_region(MKDEV(zvol_major
, 0), 1UL << MINORBITS
,
2695 THIS_MODULE
, zvol_probe
, NULL
, NULL
);
2700 kmem_free(zvol_htable
, ZVOL_HT_SIZE
* sizeof (struct hlist_head
));
2702 taskq_destroy(zvol_taskq
);
2704 ida_destroy(&zvol_ida
);
2705 mutex_destroy(&zvol_state_lock
);
2706 list_destroy(&zvol_state_list
);
2708 return (SET_ERROR(error
));
2714 zvol_remove_minors_impl(NULL
);
2716 blk_unregister_region(MKDEV(zvol_major
, 0), 1UL << MINORBITS
);
2717 unregister_blkdev(zvol_major
, ZVOL_DRIVER
);
2718 kmem_free(zvol_htable
, ZVOL_HT_SIZE
* sizeof (struct hlist_head
));
2720 taskq_destroy(zvol_taskq
);
2721 list_destroy(&zvol_state_list
);
2722 mutex_destroy(&zvol_state_lock
);
2724 ida_destroy(&zvol_ida
);
2728 module_param(zvol_inhibit_dev
, uint
, 0644);
2729 MODULE_PARM_DESC(zvol_inhibit_dev
, "Do not create zvol device nodes");
2731 module_param(zvol_major
, uint
, 0444);
2732 MODULE_PARM_DESC(zvol_major
, "Major number for zvol device");
2734 module_param(zvol_threads
, uint
, 0444);
2735 MODULE_PARM_DESC(zvol_threads
, "Max number of threads to handle I/O requests");
2737 module_param(zvol_request_sync
, uint
, 0644);
2738 MODULE_PARM_DESC(zvol_request_sync
, "Synchronously handle bio requests");
2740 module_param(zvol_max_discard_blocks
, ulong
, 0444);
2741 MODULE_PARM_DESC(zvol_max_discard_blocks
, "Max number of blocks to discard");
2743 module_param(zvol_prefetch_bytes
, uint
, 0644);
2744 MODULE_PARM_DESC(zvol_prefetch_bytes
, "Prefetch N bytes at zvol start+end");
2746 module_param(zvol_volmode
, uint
, 0644);
2747 MODULE_PARM_DESC(zvol_volmode
, "Default volmode property value");