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 krwlock_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 rw_enter(&zvol_state_lock
, RW_READER
);
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 rw_exit(&zvol_state_lock
);
187 mutex_exit(&zv
->zv_state_lock
);
189 rw_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 rw_enter(&zvol_state_lock
, RW_READER
);
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 rw_exit(&zvol_state_lock
);
233 mutex_exit(&zv
->zv_state_lock
);
235 rw_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 * Sanity check volume size.
346 zvol_check_volsize(uint64_t volsize
, uint64_t blocksize
)
349 return (SET_ERROR(EINVAL
));
351 if (volsize
% blocksize
!= 0)
352 return (SET_ERROR(EINVAL
));
355 if (volsize
- 1 > SPEC_MAXOFFSET_T
)
356 return (SET_ERROR(EOVERFLOW
));
362 * Ensure the zap is flushed then inform the VFS of the capacity change.
365 zvol_update_volsize(uint64_t volsize
, objset_t
*os
)
371 tx
= dmu_tx_create(os
);
372 dmu_tx_hold_zap(tx
, ZVOL_ZAP_OBJ
, TRUE
, NULL
);
373 dmu_tx_mark_netfree(tx
);
374 error
= dmu_tx_assign(tx
, TXG_WAIT
);
377 return (SET_ERROR(error
));
379 txg
= dmu_tx_get_txg(tx
);
381 error
= zap_update(os
, ZVOL_ZAP_OBJ
, "size", 8, 1,
385 txg_wait_synced(dmu_objset_pool(os
), txg
);
388 error
= dmu_free_long_range(os
,
389 ZVOL_OBJ
, volsize
, DMU_OBJECT_END
);
395 * Set ZFS_PROP_VOLSIZE set entry point. Note that modifying the volume
396 * size will result in a udev "change" event being generated.
399 zvol_set_volsize(const char *name
, uint64_t volsize
)
402 struct gendisk
*disk
= NULL
;
405 boolean_t owned
= B_FALSE
;
407 error
= dsl_prop_get_integer(name
,
408 zfs_prop_to_name(ZFS_PROP_READONLY
), &readonly
, NULL
);
410 return (SET_ERROR(error
));
412 return (SET_ERROR(EROFS
));
414 zvol_state_t
*zv
= zvol_find_by_name(name
, RW_READER
);
416 ASSERT(zv
== NULL
|| (MUTEX_HELD(&zv
->zv_state_lock
) &&
417 RW_READ_HELD(&zv
->zv_suspend_lock
)));
419 if (zv
== NULL
|| zv
->zv_objset
== NULL
) {
421 rw_exit(&zv
->zv_suspend_lock
);
422 if ((error
= dmu_objset_own(name
, DMU_OST_ZVOL
, B_FALSE
,
425 mutex_exit(&zv
->zv_state_lock
);
426 return (SET_ERROR(error
));
435 dmu_object_info_t
*doi
= kmem_alloc(sizeof (*doi
), KM_SLEEP
);
437 if ((error
= dmu_object_info(os
, ZVOL_OBJ
, doi
)) ||
438 (error
= zvol_check_volsize(volsize
, doi
->doi_data_block_size
)))
441 error
= zvol_update_volsize(volsize
, os
);
442 if (error
== 0 && zv
!= NULL
) {
443 zv
->zv_volsize
= volsize
;
448 kmem_free(doi
, sizeof (dmu_object_info_t
));
451 dmu_objset_disown(os
, FTAG
);
453 zv
->zv_objset
= NULL
;
455 rw_exit(&zv
->zv_suspend_lock
);
459 mutex_exit(&zv
->zv_state_lock
);
462 revalidate_disk(disk
);
464 return (SET_ERROR(error
));
468 * Sanity check volume block size.
471 zvol_check_volblocksize(const char *name
, uint64_t volblocksize
)
473 /* Record sizes above 128k need the feature to be enabled */
474 if (volblocksize
> SPA_OLD_MAXBLOCKSIZE
) {
478 if ((error
= spa_open(name
, &spa
, FTAG
)) != 0)
481 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
482 spa_close(spa
, FTAG
);
483 return (SET_ERROR(ENOTSUP
));
487 * We don't allow setting the property above 1MB,
488 * unless the tunable has been changed.
490 if (volblocksize
> zfs_max_recordsize
)
491 return (SET_ERROR(EDOM
));
493 spa_close(spa
, FTAG
);
496 if (volblocksize
< SPA_MINBLOCKSIZE
||
497 volblocksize
> SPA_MAXBLOCKSIZE
||
499 return (SET_ERROR(EDOM
));
505 * Set ZFS_PROP_VOLBLOCKSIZE set entry point.
508 zvol_set_volblocksize(const char *name
, uint64_t volblocksize
)
514 zv
= zvol_find_by_name(name
, RW_READER
);
517 return (SET_ERROR(ENXIO
));
519 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
520 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
522 if (zv
->zv_flags
& ZVOL_RDONLY
) {
523 mutex_exit(&zv
->zv_state_lock
);
524 rw_exit(&zv
->zv_suspend_lock
);
525 return (SET_ERROR(EROFS
));
528 tx
= dmu_tx_create(zv
->zv_objset
);
529 dmu_tx_hold_bonus(tx
, ZVOL_OBJ
);
530 error
= dmu_tx_assign(tx
, TXG_WAIT
);
534 error
= dmu_object_set_blocksize(zv
->zv_objset
, ZVOL_OBJ
,
535 volblocksize
, 0, tx
);
536 if (error
== ENOTSUP
)
537 error
= SET_ERROR(EBUSY
);
540 zv
->zv_volblocksize
= volblocksize
;
543 mutex_exit(&zv
->zv_state_lock
);
544 rw_exit(&zv
->zv_suspend_lock
);
546 return (SET_ERROR(error
));
550 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
551 * implement DKIOCFREE/free-long-range.
554 zvol_replay_truncate(zvol_state_t
*zv
, lr_truncate_t
*lr
, boolean_t byteswap
)
556 uint64_t offset
, length
;
559 byteswap_uint64_array(lr
, sizeof (*lr
));
561 offset
= lr
->lr_offset
;
562 length
= lr
->lr_length
;
564 return (dmu_free_long_range(zv
->zv_objset
, ZVOL_OBJ
, offset
, length
));
568 * Replay a TX_WRITE ZIL transaction that didn't get committed
569 * after a system failure
572 zvol_replay_write(zvol_state_t
*zv
, lr_write_t
*lr
, boolean_t byteswap
)
574 objset_t
*os
= zv
->zv_objset
;
575 char *data
= (char *)(lr
+ 1); /* data follows lr_write_t */
576 uint64_t offset
, length
;
581 byteswap_uint64_array(lr
, sizeof (*lr
));
583 offset
= lr
->lr_offset
;
584 length
= lr
->lr_length
;
586 /* If it's a dmu_sync() block, write the whole block */
587 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
588 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
589 if (length
< blocksize
) {
590 offset
-= offset
% blocksize
;
595 tx
= dmu_tx_create(os
);
596 dmu_tx_hold_write(tx
, ZVOL_OBJ
, offset
, length
);
597 error
= dmu_tx_assign(tx
, TXG_WAIT
);
601 dmu_write(os
, ZVOL_OBJ
, offset
, length
, data
, tx
);
609 zvol_replay_err(zvol_state_t
*zv
, lr_t
*lr
, boolean_t byteswap
)
611 return (SET_ERROR(ENOTSUP
));
615 * Callback vectors for replaying records.
616 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
618 zil_replay_func_t zvol_replay_vector
[TX_MAX_TYPE
] = {
619 (zil_replay_func_t
)zvol_replay_err
, /* no such transaction type */
620 (zil_replay_func_t
)zvol_replay_err
, /* TX_CREATE */
621 (zil_replay_func_t
)zvol_replay_err
, /* TX_MKDIR */
622 (zil_replay_func_t
)zvol_replay_err
, /* TX_MKXATTR */
623 (zil_replay_func_t
)zvol_replay_err
, /* TX_SYMLINK */
624 (zil_replay_func_t
)zvol_replay_err
, /* TX_REMOVE */
625 (zil_replay_func_t
)zvol_replay_err
, /* TX_RMDIR */
626 (zil_replay_func_t
)zvol_replay_err
, /* TX_LINK */
627 (zil_replay_func_t
)zvol_replay_err
, /* TX_RENAME */
628 (zil_replay_func_t
)zvol_replay_write
, /* TX_WRITE */
629 (zil_replay_func_t
)zvol_replay_truncate
, /* TX_TRUNCATE */
630 (zil_replay_func_t
)zvol_replay_err
, /* TX_SETATTR */
631 (zil_replay_func_t
)zvol_replay_err
, /* TX_ACL */
635 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
637 * We store data in the log buffers if it's small enough.
638 * Otherwise we will later flush the data out via dmu_sync().
640 ssize_t zvol_immediate_write_sz
= 32768;
643 zvol_log_write(zvol_state_t
*zv
, dmu_tx_t
*tx
, uint64_t offset
,
644 uint64_t size
, int sync
)
646 uint32_t blocksize
= zv
->zv_volblocksize
;
647 zilog_t
*zilog
= zv
->zv_zilog
;
648 itx_wr_state_t write_state
;
650 if (zil_replaying(zilog
, tx
))
653 if (zilog
->zl_logbias
== ZFS_LOGBIAS_THROUGHPUT
)
654 write_state
= WR_INDIRECT
;
655 else if (!spa_has_slogs(zilog
->zl_spa
) &&
656 size
>= blocksize
&& blocksize
> zvol_immediate_write_sz
)
657 write_state
= WR_INDIRECT
;
659 write_state
= WR_COPIED
;
661 write_state
= WR_NEED_COPY
;
666 itx_wr_state_t wr_state
= write_state
;
669 if (wr_state
== WR_COPIED
&& size
> ZIL_MAX_COPIED_DATA
)
670 wr_state
= WR_NEED_COPY
;
671 else if (wr_state
== WR_INDIRECT
)
672 len
= MIN(blocksize
- P2PHASE(offset
, blocksize
), size
);
674 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
) +
675 (wr_state
== WR_COPIED
? len
: 0));
676 lr
= (lr_write_t
*)&itx
->itx_lr
;
677 if (wr_state
== WR_COPIED
&& dmu_read_by_dnode(zv
->zv_dn
,
678 offset
, len
, lr
+1, DMU_READ_NO_PREFETCH
) != 0) {
679 zil_itx_destroy(itx
);
680 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
681 lr
= (lr_write_t
*)&itx
->itx_lr
;
682 wr_state
= WR_NEED_COPY
;
685 itx
->itx_wr_state
= wr_state
;
686 lr
->lr_foid
= ZVOL_OBJ
;
687 lr
->lr_offset
= offset
;
690 BP_ZERO(&lr
->lr_blkptr
);
692 itx
->itx_private
= zv
;
693 itx
->itx_sync
= sync
;
695 (void) zil_itx_assign(zilog
, itx
, tx
);
702 typedef struct zv_request
{
709 uio_from_bio(uio_t
*uio
, struct bio
*bio
)
711 uio
->uio_bvec
= &bio
->bi_io_vec
[BIO_BI_IDX(bio
)];
712 uio
->uio_skip
= BIO_BI_SKIP(bio
);
713 uio
->uio_resid
= BIO_BI_SIZE(bio
);
714 uio
->uio_iovcnt
= bio
->bi_vcnt
- BIO_BI_IDX(bio
);
715 uio
->uio_loffset
= BIO_BI_SECTOR(bio
) << 9;
716 uio
->uio_limit
= MAXOFFSET_T
;
717 uio
->uio_segflg
= UIO_BVEC
;
721 zvol_write(void *arg
)
723 zv_request_t
*zvr
= arg
;
724 struct bio
*bio
= zvr
->bio
;
726 zvol_state_t
*zv
= zvr
->zv
;
727 uint64_t volsize
= zv
->zv_volsize
;
730 unsigned long start_jif
;
732 uio_from_bio(&uio
, bio
);
734 ASSERT(zv
&& zv
->zv_open_count
> 0);
737 blk_generic_start_io_acct(zv
->zv_queue
, WRITE
, bio_sectors(bio
),
738 &zv
->zv_disk
->part0
);
740 sync
= bio_is_fua(bio
) || zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
742 while (uio
.uio_resid
> 0 && uio
.uio_loffset
< volsize
) {
743 uint64_t bytes
= MIN(uio
.uio_resid
, DMU_MAX_ACCESS
>> 1);
744 uint64_t off
= uio
.uio_loffset
;
745 dmu_tx_t
*tx
= dmu_tx_create(zv
->zv_objset
);
747 if (bytes
> volsize
- off
) /* don't write past the end */
748 bytes
= volsize
- off
;
750 dmu_tx_hold_write(tx
, ZVOL_OBJ
, off
, bytes
);
752 /* This will only fail for ENOSPC */
753 error
= dmu_tx_assign(tx
, TXG_WAIT
);
758 error
= dmu_write_uio_dnode(zv
->zv_dn
, &uio
, bytes
, tx
);
760 zvol_log_write(zv
, tx
, off
, bytes
, sync
);
766 zfs_range_unlock(zvr
->rl
);
768 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
770 rw_exit(&zv
->zv_suspend_lock
);
771 blk_generic_end_io_acct(zv
->zv_queue
, WRITE
, &zv
->zv_disk
->part0
,
773 BIO_END_IO(bio
, -error
);
774 kmem_free(zvr
, sizeof (zv_request_t
));
778 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
781 zvol_log_truncate(zvol_state_t
*zv
, dmu_tx_t
*tx
, uint64_t off
, uint64_t len
,
786 zilog_t
*zilog
= zv
->zv_zilog
;
788 if (zil_replaying(zilog
, tx
))
791 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
792 lr
= (lr_truncate_t
*)&itx
->itx_lr
;
793 lr
->lr_foid
= ZVOL_OBJ
;
797 itx
->itx_sync
= sync
;
798 zil_itx_assign(zilog
, itx
, tx
);
802 zvol_discard(void *arg
)
804 zv_request_t
*zvr
= arg
;
805 struct bio
*bio
= zvr
->bio
;
806 zvol_state_t
*zv
= zvr
->zv
;
807 uint64_t start
= BIO_BI_SECTOR(bio
) << 9;
808 uint64_t size
= BIO_BI_SIZE(bio
);
809 uint64_t end
= start
+ size
;
813 unsigned long start_jif
;
815 ASSERT(zv
&& zv
->zv_open_count
> 0);
818 blk_generic_start_io_acct(zv
->zv_queue
, WRITE
, bio_sectors(bio
),
819 &zv
->zv_disk
->part0
);
821 sync
= bio_is_fua(bio
) || zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
823 if (end
> zv
->zv_volsize
) {
824 error
= SET_ERROR(EIO
);
829 * Align the request to volume block boundaries when a secure erase is
830 * not required. This will prevent dnode_free_range() from zeroing out
831 * the unaligned parts which is slow (read-modify-write) and useless
832 * since we are not freeing any space by doing so.
834 if (!bio_is_secure_erase(bio
)) {
835 start
= P2ROUNDUP(start
, zv
->zv_volblocksize
);
836 end
= P2ALIGN(end
, zv
->zv_volblocksize
);
843 tx
= dmu_tx_create(zv
->zv_objset
);
844 dmu_tx_mark_netfree(tx
);
845 error
= dmu_tx_assign(tx
, TXG_WAIT
);
849 zvol_log_truncate(zv
, tx
, start
, size
, B_TRUE
);
851 error
= dmu_free_long_range(zv
->zv_objset
,
852 ZVOL_OBJ
, start
, size
);
855 zfs_range_unlock(zvr
->rl
);
856 if (error
== 0 && sync
)
857 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
859 rw_exit(&zv
->zv_suspend_lock
);
860 blk_generic_end_io_acct(zv
->zv_queue
, WRITE
, &zv
->zv_disk
->part0
,
862 BIO_END_IO(bio
, -error
);
863 kmem_free(zvr
, sizeof (zv_request_t
));
869 zv_request_t
*zvr
= arg
;
870 struct bio
*bio
= zvr
->bio
;
872 zvol_state_t
*zv
= zvr
->zv
;
873 uint64_t volsize
= zv
->zv_volsize
;
875 unsigned long start_jif
;
877 uio_from_bio(&uio
, bio
);
879 ASSERT(zv
&& zv
->zv_open_count
> 0);
882 blk_generic_start_io_acct(zv
->zv_queue
, READ
, bio_sectors(bio
),
883 &zv
->zv_disk
->part0
);
885 while (uio
.uio_resid
> 0 && uio
.uio_loffset
< volsize
) {
886 uint64_t bytes
= MIN(uio
.uio_resid
, DMU_MAX_ACCESS
>> 1);
888 /* don't read past the end */
889 if (bytes
> volsize
- uio
.uio_loffset
)
890 bytes
= volsize
- uio
.uio_loffset
;
892 error
= dmu_read_uio_dnode(zv
->zv_dn
, &uio
, bytes
);
894 /* convert checksum errors into IO errors */
896 error
= SET_ERROR(EIO
);
900 zfs_range_unlock(zvr
->rl
);
902 rw_exit(&zv
->zv_suspend_lock
);
903 blk_generic_end_io_acct(zv
->zv_queue
, READ
, &zv
->zv_disk
->part0
,
905 BIO_END_IO(bio
, -error
);
906 kmem_free(zvr
, sizeof (zv_request_t
));
909 static MAKE_REQUEST_FN_RET
910 zvol_request(struct request_queue
*q
, struct bio
*bio
)
912 zvol_state_t
*zv
= q
->queuedata
;
913 fstrans_cookie_t cookie
= spl_fstrans_mark();
914 uint64_t offset
= BIO_BI_SECTOR(bio
) << 9;
915 uint64_t size
= BIO_BI_SIZE(bio
);
916 int rw
= bio_data_dir(bio
);
919 if (bio_has_data(bio
) && offset
+ size
> zv
->zv_volsize
) {
921 "%s: bad access: offset=%llu, size=%lu\n",
922 zv
->zv_disk
->disk_name
,
923 (long long unsigned)offset
,
924 (long unsigned)size
);
926 BIO_END_IO(bio
, -SET_ERROR(EIO
));
931 boolean_t need_sync
= B_FALSE
;
933 if (unlikely(zv
->zv_flags
& ZVOL_RDONLY
)) {
934 BIO_END_IO(bio
, -SET_ERROR(EROFS
));
939 * To be released in the I/O function. See the comment on
940 * zfs_range_lock below.
942 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
944 /* bio marked as FLUSH need to flush before write */
945 if (bio_is_flush(bio
))
946 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
948 /* Some requests are just for flush and nothing else. */
950 rw_exit(&zv
->zv_suspend_lock
);
955 zvr
= kmem_alloc(sizeof (zv_request_t
), KM_SLEEP
);
960 * To be released in the I/O function. Since the I/O functions
961 * are asynchronous, we take it here synchronously to make
962 * sure overlapped I/Os are properly ordered.
964 zvr
->rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
967 * Sync writes and discards execute zil_commit() which may need
968 * to take a RL_READER lock on the whole block being modified
969 * via its zillog->zl_get_data(): to avoid circular dependency
970 * issues with taskq threads execute these requests
971 * synchronously here in zvol_request().
973 need_sync
= bio_is_fua(bio
) ||
974 zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
975 if (bio_is_discard(bio
) || bio_is_secure_erase(bio
)) {
976 if (zvol_request_sync
|| need_sync
||
977 taskq_dispatch(zvol_taskq
, zvol_discard
, zvr
,
978 TQ_SLEEP
) == TASKQID_INVALID
)
981 if (zvol_request_sync
|| need_sync
||
982 taskq_dispatch(zvol_taskq
, zvol_write
, zvr
,
983 TQ_SLEEP
) == TASKQID_INVALID
)
987 zvr
= kmem_alloc(sizeof (zv_request_t
), KM_SLEEP
);
991 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
993 zvr
->rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
995 if (zvol_request_sync
|| taskq_dispatch(zvol_taskq
,
996 zvol_read
, zvr
, TQ_SLEEP
) == TASKQID_INVALID
)
1001 spl_fstrans_unmark(cookie
);
1002 #ifdef HAVE_MAKE_REQUEST_FN_RET_INT
1004 #elif defined(HAVE_MAKE_REQUEST_FN_RET_QC)
1005 return (BLK_QC_T_NONE
);
1010 zvol_get_done(zgd_t
*zgd
, int error
)
1013 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1015 zfs_range_unlock(zgd
->zgd_rl
);
1017 if (error
== 0 && zgd
->zgd_bp
)
1018 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
1020 kmem_free(zgd
, sizeof (zgd_t
));
1024 * Get data to generate a TX_WRITE intent log record.
1027 zvol_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
1029 zvol_state_t
*zv
= arg
;
1030 uint64_t offset
= lr
->lr_offset
;
1031 uint64_t size
= lr
->lr_length
;
1036 ASSERT(zio
!= NULL
);
1039 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1040 zgd
->zgd_zilog
= zv
->zv_zilog
;
1043 * Write records come in two flavors: immediate and indirect.
1044 * For small writes it's cheaper to store the data with the
1045 * log record (immediate); for large writes it's cheaper to
1046 * sync the data and get a pointer to it (indirect) so that
1047 * we don't have to write the data twice.
1049 if (buf
!= NULL
) { /* immediate write */
1050 zgd
->zgd_rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
1052 error
= dmu_read_by_dnode(zv
->zv_dn
, offset
, size
, buf
,
1053 DMU_READ_NO_PREFETCH
);
1054 } else { /* indirect write */
1056 * Have to lock the whole block to ensure when it's written out
1057 * and its checksum is being calculated that no one can change
1058 * the data. Contrarily to zfs_get_data we need not re-check
1059 * blocksize after we get the lock because it cannot be changed.
1061 size
= zv
->zv_volblocksize
;
1062 offset
= P2ALIGN_TYPED(offset
, size
, uint64_t);
1063 zgd
->zgd_rl
= zfs_range_lock(&zv
->zv_range_lock
, offset
, size
,
1065 error
= dmu_buf_hold_by_dnode(zv
->zv_dn
, offset
, zgd
, &db
,
1066 DMU_READ_NO_PREFETCH
);
1068 blkptr_t
*bp
= &lr
->lr_blkptr
;
1074 ASSERT(db
->db_offset
== offset
);
1075 ASSERT(db
->db_size
== size
);
1077 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1078 zvol_get_done
, zgd
);
1085 zvol_get_done(zgd
, error
);
1087 return (SET_ERROR(error
));
1091 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable.
1094 zvol_insert(zvol_state_t
*zv
)
1096 ASSERT(RW_WRITE_HELD(&zvol_state_lock
));
1097 ASSERT3U(MINOR(zv
->zv_dev
) & ZVOL_MINOR_MASK
, ==, 0);
1098 list_insert_head(&zvol_state_list
, zv
);
1099 hlist_add_head(&zv
->zv_hlink
, ZVOL_HT_HEAD(zv
->zv_hash
));
1103 * Simply remove the zvol from to list of zvols.
1106 zvol_remove(zvol_state_t
*zv
)
1108 ASSERT(RW_WRITE_HELD(&zvol_state_lock
));
1109 list_remove(&zvol_state_list
, zv
);
1110 hlist_del(&zv
->zv_hlink
);
1114 * Setup zv after we just own the zv->objset
1117 zvol_setup_zv(zvol_state_t
*zv
)
1122 objset_t
*os
= zv
->zv_objset
;
1124 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1125 ASSERT(RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1127 error
= dsl_prop_get_integer(zv
->zv_name
, "readonly", &ro
, NULL
);
1129 return (SET_ERROR(error
));
1131 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
1133 return (SET_ERROR(error
));
1135 error
= dnode_hold(os
, ZVOL_OBJ
, FTAG
, &zv
->zv_dn
);
1137 return (SET_ERROR(error
));
1139 set_capacity(zv
->zv_disk
, volsize
>> 9);
1140 zv
->zv_volsize
= volsize
;
1141 zv
->zv_zilog
= zil_open(os
, zvol_get_data
);
1143 if (ro
|| dmu_objset_is_snapshot(os
) ||
1144 !spa_writeable(dmu_objset_spa(os
))) {
1145 set_disk_ro(zv
->zv_disk
, 1);
1146 zv
->zv_flags
|= ZVOL_RDONLY
;
1148 set_disk_ro(zv
->zv_disk
, 0);
1149 zv
->zv_flags
&= ~ZVOL_RDONLY
;
1155 * Shutdown every zv_objset related stuff except zv_objset itself.
1156 * The is the reverse of zvol_setup_zv.
1159 zvol_shutdown_zv(zvol_state_t
*zv
)
1161 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
) &&
1162 RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1164 zil_close(zv
->zv_zilog
);
1165 zv
->zv_zilog
= NULL
;
1167 dnode_rele(zv
->zv_dn
, FTAG
);
1173 if (dsl_dataset_is_dirty(dmu_objset_ds(zv
->zv_objset
)) &&
1174 !(zv
->zv_flags
& ZVOL_RDONLY
))
1175 txg_wait_synced(dmu_objset_pool(zv
->zv_objset
), 0);
1176 (void) dmu_objset_evict_dbufs(zv
->zv_objset
);
1180 * return the proper tag for rollback and recv
1183 zvol_tag(zvol_state_t
*zv
)
1185 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1186 return (zv
->zv_open_count
> 0 ? zv
: NULL
);
1190 * Suspend the zvol for recv and rollback.
1193 zvol_suspend(const char *name
)
1197 zv
= zvol_find_by_name(name
, RW_WRITER
);
1202 /* block all I/O, release in zvol_resume. */
1203 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1204 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1206 atomic_inc(&zv
->zv_suspend_ref
);
1208 if (zv
->zv_open_count
> 0)
1209 zvol_shutdown_zv(zv
);
1212 * do not hold zv_state_lock across suspend/resume to
1213 * avoid locking up zvol lookups
1215 mutex_exit(&zv
->zv_state_lock
);
1217 /* zv_suspend_lock is released in zvol_resume() */
1222 zvol_resume(zvol_state_t
*zv
)
1226 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1228 mutex_enter(&zv
->zv_state_lock
);
1230 if (zv
->zv_open_count
> 0) {
1231 VERIFY0(dmu_objset_hold(zv
->zv_name
, zv
, &zv
->zv_objset
));
1232 VERIFY3P(zv
->zv_objset
->os_dsl_dataset
->ds_owner
, ==, zv
);
1233 VERIFY(dsl_dataset_long_held(zv
->zv_objset
->os_dsl_dataset
));
1234 dmu_objset_rele(zv
->zv_objset
, zv
);
1236 error
= zvol_setup_zv(zv
);
1239 mutex_exit(&zv
->zv_state_lock
);
1241 rw_exit(&zv
->zv_suspend_lock
);
1243 * We need this because we don't hold zvol_state_lock while releasing
1244 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check
1245 * zv_suspend_lock to determine it is safe to free because rwlock is
1246 * not inherent atomic.
1248 atomic_dec(&zv
->zv_suspend_ref
);
1250 return (SET_ERROR(error
));
1254 zvol_first_open(zvol_state_t
*zv
)
1257 int error
, locked
= 0;
1259 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
1260 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1263 * In all other cases the spa_namespace_lock is taken before the
1264 * bdev->bd_mutex lock. But in this case the Linux __blkdev_get()
1265 * function calls fops->open() with the bdev->bd_mutex lock held.
1266 * This deadlock can be easily observed with zvols used as vdevs.
1268 * To avoid a potential lock inversion deadlock we preemptively
1269 * try to take the spa_namespace_lock(). Normally it will not
1270 * be contended and this is safe because spa_open_common() handles
1271 * the case where the caller already holds the spa_namespace_lock.
1273 * When it is contended we risk a lock inversion if we were to
1274 * block waiting for the lock. Luckily, the __blkdev_get()
1275 * function allows us to return -ERESTARTSYS which will result in
1276 * bdev->bd_mutex being dropped, reacquired, and fops->open() being
1277 * called again. This process can be repeated safely until both
1278 * locks are acquired.
1280 if (!mutex_owned(&spa_namespace_lock
)) {
1281 locked
= mutex_tryenter(&spa_namespace_lock
);
1283 return (-SET_ERROR(ERESTARTSYS
));
1286 /* lie and say we're read-only */
1287 error
= dmu_objset_own(zv
->zv_name
, DMU_OST_ZVOL
, 1, zv
, &os
);
1293 error
= zvol_setup_zv(zv
);
1296 dmu_objset_disown(os
, zv
);
1297 zv
->zv_objset
= NULL
;
1302 mutex_exit(&spa_namespace_lock
);
1303 return (SET_ERROR(-error
));
1307 zvol_last_close(zvol_state_t
*zv
)
1309 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
1310 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1312 zvol_shutdown_zv(zv
);
1314 dmu_objset_disown(zv
->zv_objset
, zv
);
1315 zv
->zv_objset
= NULL
;
1319 zvol_open(struct block_device
*bdev
, fmode_t flag
)
1323 boolean_t drop_suspend
= B_TRUE
;
1325 rw_enter(&zvol_state_lock
, RW_READER
);
1327 * Obtain a copy of private_data under the zvol_state_lock to make
1328 * sure that either the result of zvol free code path setting
1329 * bdev->bd_disk->private_data to NULL is observed, or zvol_free()
1330 * is not called on this zv because of the positive zv_open_count.
1332 zv
= bdev
->bd_disk
->private_data
;
1334 rw_exit(&zvol_state_lock
);
1335 return (SET_ERROR(-ENXIO
));
1338 mutex_enter(&zv
->zv_state_lock
);
1340 * make sure zvol is not suspended during first open
1341 * (hold zv_suspend_lock) and respect proper lock acquisition
1342 * ordering - zv_suspend_lock before zv_state_lock
1344 if (zv
->zv_open_count
== 0) {
1345 if (!rw_tryenter(&zv
->zv_suspend_lock
, RW_READER
)) {
1346 mutex_exit(&zv
->zv_state_lock
);
1347 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1348 mutex_enter(&zv
->zv_state_lock
);
1349 /* check to see if zv_suspend_lock is needed */
1350 if (zv
->zv_open_count
!= 0) {
1351 rw_exit(&zv
->zv_suspend_lock
);
1352 drop_suspend
= B_FALSE
;
1356 drop_suspend
= B_FALSE
;
1358 rw_exit(&zvol_state_lock
);
1360 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1361 ASSERT(zv
->zv_open_count
!= 0 || RW_READ_HELD(&zv
->zv_suspend_lock
));
1363 if (zv
->zv_open_count
== 0) {
1364 error
= zvol_first_open(zv
);
1369 if ((flag
& FMODE_WRITE
) && (zv
->zv_flags
& ZVOL_RDONLY
)) {
1371 goto out_open_count
;
1374 zv
->zv_open_count
++;
1376 mutex_exit(&zv
->zv_state_lock
);
1378 rw_exit(&zv
->zv_suspend_lock
);
1380 check_disk_change(bdev
);
1385 if (zv
->zv_open_count
== 0)
1386 zvol_last_close(zv
);
1389 mutex_exit(&zv
->zv_state_lock
);
1391 rw_exit(&zv
->zv_suspend_lock
);
1392 if (error
== -ERESTARTSYS
)
1395 return (SET_ERROR(error
));
1398 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1403 zvol_release(struct gendisk
*disk
, fmode_t mode
)
1406 boolean_t drop_suspend
= B_TRUE
;
1408 rw_enter(&zvol_state_lock
, RW_READER
);
1409 zv
= disk
->private_data
;
1411 mutex_enter(&zv
->zv_state_lock
);
1412 ASSERT(zv
->zv_open_count
> 0);
1414 * make sure zvol is not suspended during last close
1415 * (hold zv_suspend_lock) and respect proper lock acquisition
1416 * ordering - zv_suspend_lock before zv_state_lock
1418 if (zv
->zv_open_count
== 1) {
1419 if (!rw_tryenter(&zv
->zv_suspend_lock
, RW_READER
)) {
1420 mutex_exit(&zv
->zv_state_lock
);
1421 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1422 mutex_enter(&zv
->zv_state_lock
);
1423 /* check to see if zv_suspend_lock is needed */
1424 if (zv
->zv_open_count
!= 1) {
1425 rw_exit(&zv
->zv_suspend_lock
);
1426 drop_suspend
= B_FALSE
;
1430 drop_suspend
= B_FALSE
;
1432 rw_exit(&zvol_state_lock
);
1434 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1435 ASSERT(zv
->zv_open_count
!= 1 || RW_READ_HELD(&zv
->zv_suspend_lock
));
1437 zv
->zv_open_count
--;
1438 if (zv
->zv_open_count
== 0)
1439 zvol_last_close(zv
);
1441 mutex_exit(&zv
->zv_state_lock
);
1444 rw_exit(&zv
->zv_suspend_lock
);
1446 #ifndef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1452 zvol_ioctl(struct block_device
*bdev
, fmode_t mode
,
1453 unsigned int cmd
, unsigned long arg
)
1455 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
1458 ASSERT3U(zv
->zv_open_count
, >, 0);
1463 invalidate_bdev(bdev
);
1464 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1466 if (dsl_dataset_is_dirty(dmu_objset_ds(zv
->zv_objset
)) &&
1467 !(zv
->zv_flags
& ZVOL_RDONLY
))
1468 txg_wait_synced(dmu_objset_pool(zv
->zv_objset
), 0);
1470 rw_exit(&zv
->zv_suspend_lock
);
1474 mutex_enter(&zv
->zv_state_lock
);
1475 error
= copy_to_user((void *)arg
, zv
->zv_name
, MAXNAMELEN
);
1476 mutex_exit(&zv
->zv_state_lock
);
1484 return (SET_ERROR(error
));
1487 #ifdef CONFIG_COMPAT
1489 zvol_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1490 unsigned cmd
, unsigned long arg
)
1492 return (zvol_ioctl(bdev
, mode
, cmd
, arg
));
1495 #define zvol_compat_ioctl NULL
1499 * Linux 2.6.38 preferred interface.
1501 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS
1503 zvol_check_events(struct gendisk
*disk
, unsigned int clearing
)
1505 unsigned int mask
= 0;
1507 rw_enter(&zvol_state_lock
, RW_READER
);
1509 zvol_state_t
*zv
= disk
->private_data
;
1511 mutex_enter(&zv
->zv_state_lock
);
1512 mask
= zv
->zv_changed
? DISK_EVENT_MEDIA_CHANGE
: 0;
1514 mutex_exit(&zv
->zv_state_lock
);
1517 rw_exit(&zvol_state_lock
);
1522 static int zvol_media_changed(struct gendisk
*disk
)
1526 rw_enter(&zvol_state_lock
, RW_READER
);
1528 zvol_state_t
*zv
= disk
->private_data
;
1530 mutex_enter(&zv
->zv_state_lock
);
1531 changed
= zv
->zv_changed
;
1533 mutex_exit(&zv
->zv_state_lock
);
1536 rw_exit(&zvol_state_lock
);
1542 static int zvol_revalidate_disk(struct gendisk
*disk
)
1544 rw_enter(&zvol_state_lock
, RW_READER
);
1546 zvol_state_t
*zv
= disk
->private_data
;
1548 mutex_enter(&zv
->zv_state_lock
);
1549 set_capacity(zv
->zv_disk
, zv
->zv_volsize
>> SECTOR_BITS
);
1550 mutex_exit(&zv
->zv_state_lock
);
1553 rw_exit(&zvol_state_lock
);
1559 * Provide a simple virtual geometry for legacy compatibility. For devices
1560 * smaller than 1 MiB a small head and sector count is used to allow very
1561 * tiny devices. For devices over 1 Mib a standard head and sector count
1562 * is used to keep the cylinders count reasonable.
1565 zvol_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1567 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
1570 ASSERT3U(zv
->zv_open_count
, >, 0);
1572 sectors
= get_capacity(zv
->zv_disk
);
1574 if (sectors
> 2048) {
1583 geo
->cylinders
= sectors
/ (geo
->heads
* geo
->sectors
);
1588 static struct kobject
*
1589 zvol_probe(dev_t dev
, int *part
, void *arg
)
1592 struct kobject
*kobj
;
1594 zv
= zvol_find_by_dev(dev
);
1595 kobj
= zv
? get_disk_and_module(zv
->zv_disk
) : NULL
;
1596 ASSERT(zv
== NULL
|| MUTEX_HELD(&zv
->zv_state_lock
));
1598 mutex_exit(&zv
->zv_state_lock
);
1603 static struct block_device_operations zvol_ops
= {
1605 .release
= zvol_release
,
1606 .ioctl
= zvol_ioctl
,
1607 .compat_ioctl
= zvol_compat_ioctl
,
1608 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS
1609 .check_events
= zvol_check_events
,
1611 .media_changed
= zvol_media_changed
,
1613 .revalidate_disk
= zvol_revalidate_disk
,
1614 .getgeo
= zvol_getgeo
,
1615 .owner
= THIS_MODULE
,
1619 * Allocate memory for a new zvol_state_t and setup the required
1620 * request queue and generic disk structures for the block device.
1622 static zvol_state_t
*
1623 zvol_alloc(dev_t dev
, const char *name
)
1628 if (dsl_prop_get_integer(name
, "volmode", &volmode
, NULL
) != 0)
1631 if (volmode
== ZFS_VOLMODE_DEFAULT
)
1632 volmode
= zvol_volmode
;
1634 if (volmode
== ZFS_VOLMODE_NONE
)
1637 zv
= kmem_zalloc(sizeof (zvol_state_t
), KM_SLEEP
);
1639 list_link_init(&zv
->zv_next
);
1641 mutex_init(&zv
->zv_state_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1643 zv
->zv_queue
= blk_alloc_queue(GFP_ATOMIC
);
1644 if (zv
->zv_queue
== NULL
)
1647 blk_queue_make_request(zv
->zv_queue
, zvol_request
);
1648 blk_queue_set_write_cache(zv
->zv_queue
, B_TRUE
, B_TRUE
);
1650 /* Limit read-ahead to a single page to prevent over-prefetching. */
1651 blk_queue_set_read_ahead(zv
->zv_queue
, 1);
1653 /* Disable write merging in favor of the ZIO pipeline. */
1654 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, zv
->zv_queue
);
1656 zv
->zv_disk
= alloc_disk(ZVOL_MINORS
);
1657 if (zv
->zv_disk
== NULL
)
1660 zv
->zv_queue
->queuedata
= zv
;
1662 zv
->zv_open_count
= 0;
1663 strlcpy(zv
->zv_name
, name
, MAXNAMELEN
);
1665 zfs_rlock_init(&zv
->zv_range_lock
);
1666 rw_init(&zv
->zv_suspend_lock
, NULL
, RW_DEFAULT
, NULL
);
1668 zv
->zv_disk
->major
= zvol_major
;
1669 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS
1670 zv
->zv_disk
->events
= DISK_EVENT_MEDIA_CHANGE
;
1673 if (volmode
== ZFS_VOLMODE_DEV
) {
1675 * ZFS_VOLMODE_DEV disable partitioning on ZVOL devices: set
1676 * gendisk->minors = 1 as noted in include/linux/genhd.h.
1677 * Also disable extended partition numbers (GENHD_FL_EXT_DEVT)
1678 * and suppresses partition scanning (GENHD_FL_NO_PART_SCAN)
1679 * setting gendisk->flags accordingly.
1681 zv
->zv_disk
->minors
= 1;
1682 #if defined(GENHD_FL_EXT_DEVT)
1683 zv
->zv_disk
->flags
&= ~GENHD_FL_EXT_DEVT
;
1685 #if defined(GENHD_FL_NO_PART_SCAN)
1686 zv
->zv_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1689 zv
->zv_disk
->first_minor
= (dev
& MINORMASK
);
1690 zv
->zv_disk
->fops
= &zvol_ops
;
1691 zv
->zv_disk
->private_data
= zv
;
1692 zv
->zv_disk
->queue
= zv
->zv_queue
;
1693 snprintf(zv
->zv_disk
->disk_name
, DISK_NAME_LEN
, "%s%d",
1694 ZVOL_DEV_NAME
, (dev
& MINORMASK
));
1699 blk_cleanup_queue(zv
->zv_queue
);
1701 kmem_free(zv
, sizeof (zvol_state_t
));
1707 * Cleanup then free a zvol_state_t which was created by zvol_alloc().
1708 * At this time, the structure is not opened by anyone, is taken off
1709 * the zvol_state_list, and has its private data set to NULL.
1710 * The zvol_state_lock is dropped.
1713 zvol_free(void *arg
)
1715 zvol_state_t
*zv
= arg
;
1717 ASSERT(!RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1718 ASSERT(!MUTEX_HELD(&zv
->zv_state_lock
));
1719 ASSERT(zv
->zv_open_count
== 0);
1720 ASSERT(zv
->zv_disk
->private_data
== NULL
);
1722 rw_destroy(&zv
->zv_suspend_lock
);
1723 zfs_rlock_destroy(&zv
->zv_range_lock
);
1725 del_gendisk(zv
->zv_disk
);
1726 blk_cleanup_queue(zv
->zv_queue
);
1727 put_disk(zv
->zv_disk
);
1729 ida_simple_remove(&zvol_ida
, MINOR(zv
->zv_dev
) >> ZVOL_MINOR_BITS
);
1731 mutex_destroy(&zv
->zv_state_lock
);
1733 kmem_free(zv
, sizeof (zvol_state_t
));
1737 * Create a block device minor node and setup the linkage between it
1738 * and the specified volume. Once this function returns the block
1739 * device is live and ready for use.
1742 zvol_create_minor_impl(const char *name
)
1746 dmu_object_info_t
*doi
;
1752 uint64_t hash
= zvol_name_hash(name
);
1754 if (zvol_inhibit_dev
)
1757 idx
= ida_simple_get(&zvol_ida
, 0, 0, kmem_flags_convert(KM_SLEEP
));
1759 return (SET_ERROR(-idx
));
1760 minor
= idx
<< ZVOL_MINOR_BITS
;
1762 zv
= zvol_find_by_name_hash(name
, hash
, RW_NONE
);
1764 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1765 mutex_exit(&zv
->zv_state_lock
);
1766 ida_simple_remove(&zvol_ida
, idx
);
1767 return (SET_ERROR(EEXIST
));
1770 doi
= kmem_alloc(sizeof (dmu_object_info_t
), KM_SLEEP
);
1772 error
= dmu_objset_own(name
, DMU_OST_ZVOL
, B_TRUE
, FTAG
, &os
);
1776 error
= dmu_object_info(os
, ZVOL_OBJ
, doi
);
1778 goto out_dmu_objset_disown
;
1780 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
1782 goto out_dmu_objset_disown
;
1784 zv
= zvol_alloc(MKDEV(zvol_major
, minor
), name
);
1786 error
= SET_ERROR(EAGAIN
);
1787 goto out_dmu_objset_disown
;
1791 if (dmu_objset_is_snapshot(os
))
1792 zv
->zv_flags
|= ZVOL_RDONLY
;
1794 zv
->zv_volblocksize
= doi
->doi_data_block_size
;
1795 zv
->zv_volsize
= volsize
;
1798 set_capacity(zv
->zv_disk
, zv
->zv_volsize
>> 9);
1800 blk_queue_max_hw_sectors(zv
->zv_queue
, (DMU_MAX_ACCESS
/ 4) >> 9);
1801 blk_queue_max_segments(zv
->zv_queue
, UINT16_MAX
);
1802 blk_queue_max_segment_size(zv
->zv_queue
, UINT_MAX
);
1803 blk_queue_physical_block_size(zv
->zv_queue
, zv
->zv_volblocksize
);
1804 blk_queue_io_opt(zv
->zv_queue
, zv
->zv_volblocksize
);
1805 blk_queue_max_discard_sectors(zv
->zv_queue
,
1806 (zvol_max_discard_blocks
* zv
->zv_volblocksize
) >> 9);
1807 blk_queue_discard_granularity(zv
->zv_queue
, zv
->zv_volblocksize
);
1808 blk_queue_flag_set(QUEUE_FLAG_DISCARD
, zv
->zv_queue
);
1809 #ifdef QUEUE_FLAG_NONROT
1810 blk_queue_flag_set(QUEUE_FLAG_NONROT
, zv
->zv_queue
);
1812 #ifdef QUEUE_FLAG_ADD_RANDOM
1813 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM
, zv
->zv_queue
);
1816 if (spa_writeable(dmu_objset_spa(os
))) {
1817 if (zil_replay_disable
)
1818 zil_destroy(dmu_objset_zil(os
), B_FALSE
);
1820 zil_replay(os
, zv
, zvol_replay_vector
);
1824 * When udev detects the addition of the device it will immediately
1825 * invoke blkid(8) to determine the type of content on the device.
1826 * Prefetching the blocks commonly scanned by blkid(8) will speed
1829 len
= MIN(MAX(zvol_prefetch_bytes
, 0), SPA_MAXBLOCKSIZE
);
1831 dmu_prefetch(os
, ZVOL_OBJ
, 0, 0, len
, ZIO_PRIORITY_SYNC_READ
);
1832 dmu_prefetch(os
, ZVOL_OBJ
, 0, volsize
- len
, len
,
1833 ZIO_PRIORITY_SYNC_READ
);
1836 zv
->zv_objset
= NULL
;
1837 out_dmu_objset_disown
:
1838 dmu_objset_disown(os
, FTAG
);
1840 kmem_free(doi
, sizeof (dmu_object_info_t
));
1843 rw_enter(&zvol_state_lock
, RW_WRITER
);
1845 rw_exit(&zvol_state_lock
);
1846 add_disk(zv
->zv_disk
);
1848 ida_simple_remove(&zvol_ida
, idx
);
1851 return (SET_ERROR(error
));
1855 * Rename a block device minor mode for the specified volume.
1858 zvol_rename_minor(zvol_state_t
*zv
, const char *newname
)
1860 int readonly
= get_disk_ro(zv
->zv_disk
);
1862 ASSERT(RW_LOCK_HELD(&zvol_state_lock
));
1863 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1865 strlcpy(zv
->zv_name
, newname
, sizeof (zv
->zv_name
));
1867 /* move to new hashtable entry */
1868 zv
->zv_hash
= zvol_name_hash(zv
->zv_name
);
1869 hlist_del(&zv
->zv_hlink
);
1870 hlist_add_head(&zv
->zv_hlink
, ZVOL_HT_HEAD(zv
->zv_hash
));
1873 * The block device's read-only state is briefly changed causing
1874 * a KOBJ_CHANGE uevent to be issued. This ensures udev detects
1875 * the name change and fixes the symlinks. This does not change
1876 * ZVOL_RDONLY in zv->zv_flags so the actual read-only state never
1877 * changes. This would normally be done using kobject_uevent() but
1878 * that is a GPL-only symbol which is why we need this workaround.
1880 set_disk_ro(zv
->zv_disk
, !readonly
);
1881 set_disk_ro(zv
->zv_disk
, readonly
);
1884 typedef struct minors_job
{
1894 * Prefetch zvol dnodes for the minors_job
1897 zvol_prefetch_minors_impl(void *arg
)
1899 minors_job_t
*job
= arg
;
1900 char *dsname
= job
->name
;
1901 objset_t
*os
= NULL
;
1903 job
->error
= dmu_objset_own(dsname
, DMU_OST_ZVOL
, B_TRUE
, FTAG
,
1905 if (job
->error
== 0) {
1906 dmu_prefetch(os
, ZVOL_OBJ
, 0, 0, 0, ZIO_PRIORITY_SYNC_READ
);
1907 dmu_objset_disown(os
, FTAG
);
1912 * Mask errors to continue dmu_objset_find() traversal
1915 zvol_create_snap_minor_cb(const char *dsname
, void *arg
)
1917 minors_job_t
*j
= arg
;
1918 list_t
*minors_list
= j
->list
;
1919 const char *name
= j
->name
;
1921 ASSERT0(MUTEX_HELD(&spa_namespace_lock
));
1923 /* skip the designated dataset */
1924 if (name
&& strcmp(dsname
, name
) == 0)
1927 /* at this point, the dsname should name a snapshot */
1928 if (strchr(dsname
, '@') == 0) {
1929 dprintf("zvol_create_snap_minor_cb(): "
1930 "%s is not a shapshot name\n", dsname
);
1933 char *n
= strdup(dsname
);
1937 job
= kmem_alloc(sizeof (minors_job_t
), KM_SLEEP
);
1939 job
->list
= minors_list
;
1941 list_insert_tail(minors_list
, job
);
1942 /* don't care if dispatch fails, because job->error is 0 */
1943 taskq_dispatch(system_taskq
, zvol_prefetch_minors_impl
, job
,
1951 * Mask errors to continue dmu_objset_find() traversal
1954 zvol_create_minors_cb(const char *dsname
, void *arg
)
1958 list_t
*minors_list
= arg
;
1960 ASSERT0(MUTEX_HELD(&spa_namespace_lock
));
1962 error
= dsl_prop_get_integer(dsname
, "snapdev", &snapdev
, NULL
);
1967 * Given the name and the 'snapdev' property, create device minor nodes
1968 * with the linkages to zvols/snapshots as needed.
1969 * If the name represents a zvol, create a minor node for the zvol, then
1970 * check if its snapshots are 'visible', and if so, iterate over the
1971 * snapshots and create device minor nodes for those.
1973 if (strchr(dsname
, '@') == 0) {
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
,
1988 if (snapdev
== ZFS_SNAPDEV_VISIBLE
) {
1990 * traverse snapshots only, do not traverse children,
1991 * and skip the 'dsname'
1993 error
= dmu_objset_find((char *)dsname
,
1994 zvol_create_snap_minor_cb
, (void *)job
,
1998 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n",
2006 * Create minors for the specified dataset, including children and snapshots.
2007 * Pay attention to the 'snapdev' property and iterate over the snapshots
2008 * only if they are 'visible'. This approach allows one to assure that the
2009 * snapshot metadata is read from disk only if it is needed.
2011 * The name can represent a dataset to be recursively scanned for zvols and
2012 * their snapshots, or a single zvol snapshot. If the name represents a
2013 * dataset, the scan is performed in two nested stages:
2014 * - scan the dataset for zvols, and
2015 * - for each zvol, create a minor node, then check if the zvol's snapshots
2016 * are 'visible', and only then iterate over the snapshots if needed
2018 * If the name represents a snapshot, a check is performed if the snapshot is
2019 * 'visible' (which also verifies that the parent is a zvol), and if so,
2020 * a minor node for that snapshot is created.
2023 zvol_create_minors_impl(const char *name
)
2026 fstrans_cookie_t cookie
;
2031 if (zvol_inhibit_dev
)
2035 * This is the list for prefetch jobs. Whenever we found a match
2036 * during dmu_objset_find, we insert a minors_job to the list and do
2037 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need
2038 * any lock because all list operation is done on the current thread.
2040 * We will use this list to do zvol_create_minor_impl after prefetch
2041 * so we don't have to traverse using dmu_objset_find again.
2043 list_create(&minors_list
, sizeof (minors_job_t
),
2044 offsetof(minors_job_t
, link
));
2046 parent
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
2047 (void) strlcpy(parent
, name
, MAXPATHLEN
);
2049 if ((atp
= strrchr(parent
, '@')) != NULL
) {
2053 error
= dsl_prop_get_integer(parent
, "snapdev",
2056 if (error
== 0 && snapdev
== ZFS_SNAPDEV_VISIBLE
)
2057 error
= zvol_create_minor_impl(name
);
2059 cookie
= spl_fstrans_mark();
2060 error
= dmu_objset_find(parent
, zvol_create_minors_cb
,
2061 &minors_list
, DS_FIND_CHILDREN
);
2062 spl_fstrans_unmark(cookie
);
2065 kmem_free(parent
, MAXPATHLEN
);
2066 taskq_wait_outstanding(system_taskq
, 0);
2069 * Prefetch is completed, we can do zvol_create_minor_impl
2072 while ((job
= list_head(&minors_list
)) != NULL
) {
2073 list_remove(&minors_list
, job
);
2075 zvol_create_minor_impl(job
->name
);
2077 kmem_free(job
, sizeof (minors_job_t
));
2080 list_destroy(&minors_list
);
2082 return (SET_ERROR(error
));
2086 * Remove minors for specified dataset including children and snapshots.
2089 zvol_remove_minors_impl(const char *name
)
2091 zvol_state_t
*zv
, *zv_next
;
2092 int namelen
= ((name
) ? strlen(name
) : 0);
2093 taskqid_t t
, tid
= TASKQID_INVALID
;
2096 if (zvol_inhibit_dev
)
2099 list_create(&free_list
, sizeof (zvol_state_t
),
2100 offsetof(zvol_state_t
, zv_next
));
2102 rw_enter(&zvol_state_lock
, RW_WRITER
);
2104 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2105 zv_next
= list_next(&zvol_state_list
, zv
);
2107 mutex_enter(&zv
->zv_state_lock
);
2108 if (name
== NULL
|| strcmp(zv
->zv_name
, name
) == 0 ||
2109 (strncmp(zv
->zv_name
, name
, namelen
) == 0 &&
2110 (zv
->zv_name
[namelen
] == '/' ||
2111 zv
->zv_name
[namelen
] == '@'))) {
2113 * By holding zv_state_lock here, we guarantee that no
2114 * one is currently using this zv
2117 /* If in use, leave alone */
2118 if (zv
->zv_open_count
> 0 ||
2119 atomic_read(&zv
->zv_suspend_ref
)) {
2120 mutex_exit(&zv
->zv_state_lock
);
2127 * Cleared while holding zvol_state_lock as a writer
2128 * which will prevent zvol_open() from opening it.
2130 zv
->zv_disk
->private_data
= NULL
;
2132 /* Drop zv_state_lock before zvol_free() */
2133 mutex_exit(&zv
->zv_state_lock
);
2135 /* Try parallel zv_free, if failed do it in place */
2136 t
= taskq_dispatch(system_taskq
, zvol_free
, zv
,
2138 if (t
== TASKQID_INVALID
)
2139 list_insert_head(&free_list
, zv
);
2143 mutex_exit(&zv
->zv_state_lock
);
2146 rw_exit(&zvol_state_lock
);
2148 /* Drop zvol_state_lock before calling zvol_free() */
2149 while ((zv
= list_head(&free_list
)) != NULL
) {
2150 list_remove(&free_list
, zv
);
2154 if (tid
!= TASKQID_INVALID
)
2155 taskq_wait_outstanding(system_taskq
, tid
);
2158 /* Remove minor for this specific volume only */
2160 zvol_remove_minor_impl(const char *name
)
2162 zvol_state_t
*zv
= NULL
, *zv_next
;
2164 if (zvol_inhibit_dev
)
2167 rw_enter(&zvol_state_lock
, RW_WRITER
);
2169 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2170 zv_next
= list_next(&zvol_state_list
, zv
);
2172 mutex_enter(&zv
->zv_state_lock
);
2173 if (strcmp(zv
->zv_name
, name
) == 0) {
2175 * By holding zv_state_lock here, we guarantee that no
2176 * one is currently using this zv
2179 /* If in use, leave alone */
2180 if (zv
->zv_open_count
> 0 ||
2181 atomic_read(&zv
->zv_suspend_ref
)) {
2182 mutex_exit(&zv
->zv_state_lock
);
2188 * Cleared while holding zvol_state_lock as a writer
2189 * which will prevent zvol_open() from opening it.
2191 zv
->zv_disk
->private_data
= NULL
;
2193 mutex_exit(&zv
->zv_state_lock
);
2196 mutex_exit(&zv
->zv_state_lock
);
2200 /* Drop zvol_state_lock before calling zvol_free() */
2201 rw_exit(&zvol_state_lock
);
2208 * Rename minors for specified dataset including children and snapshots.
2211 zvol_rename_minors_impl(const char *oldname
, const char *newname
)
2213 zvol_state_t
*zv
, *zv_next
;
2214 int oldnamelen
, newnamelen
;
2216 if (zvol_inhibit_dev
)
2219 oldnamelen
= strlen(oldname
);
2220 newnamelen
= strlen(newname
);
2222 rw_enter(&zvol_state_lock
, RW_READER
);
2224 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2225 zv_next
= list_next(&zvol_state_list
, zv
);
2227 mutex_enter(&zv
->zv_state_lock
);
2229 /* If in use, leave alone */
2230 if (zv
->zv_open_count
> 0) {
2231 mutex_exit(&zv
->zv_state_lock
);
2235 if (strcmp(zv
->zv_name
, oldname
) == 0) {
2236 zvol_rename_minor(zv
, newname
);
2237 } else if (strncmp(zv
->zv_name
, oldname
, oldnamelen
) == 0 &&
2238 (zv
->zv_name
[oldnamelen
] == '/' ||
2239 zv
->zv_name
[oldnamelen
] == '@')) {
2240 char *name
= kmem_asprintf("%s%c%s", newname
,
2241 zv
->zv_name
[oldnamelen
],
2242 zv
->zv_name
+ oldnamelen
+ 1);
2243 zvol_rename_minor(zv
, name
);
2244 kmem_free(name
, strlen(name
+ 1));
2247 mutex_exit(&zv
->zv_state_lock
);
2250 rw_exit(&zvol_state_lock
);
2253 typedef struct zvol_snapdev_cb_arg
{
2255 } zvol_snapdev_cb_arg_t
;
2258 zvol_set_snapdev_cb(const char *dsname
, void *param
)
2260 zvol_snapdev_cb_arg_t
*arg
= param
;
2262 if (strchr(dsname
, '@') == NULL
)
2265 switch (arg
->snapdev
) {
2266 case ZFS_SNAPDEV_VISIBLE
:
2267 (void) zvol_create_minor_impl(dsname
);
2269 case ZFS_SNAPDEV_HIDDEN
:
2270 (void) zvol_remove_minor_impl(dsname
);
2278 zvol_set_snapdev_impl(char *name
, uint64_t snapdev
)
2280 zvol_snapdev_cb_arg_t arg
= {snapdev
};
2281 fstrans_cookie_t cookie
= spl_fstrans_mark();
2283 * The zvol_set_snapdev_sync() sets snapdev appropriately
2284 * in the dataset hierarchy. Here, we only scan snapshots.
2286 dmu_objset_find(name
, zvol_set_snapdev_cb
, &arg
, DS_FIND_SNAPSHOTS
);
2287 spl_fstrans_unmark(cookie
);
2290 typedef struct zvol_volmode_cb_arg
{
2292 } zvol_volmode_cb_arg_t
;
2295 zvol_set_volmode_impl(char *name
, uint64_t volmode
)
2297 fstrans_cookie_t cookie
= spl_fstrans_mark();
2299 if (strchr(name
, '@') != NULL
)
2303 * It's unfortunate we need to remove minors before we create new ones:
2304 * this is necessary because our backing gendisk (zvol_state->zv_disk)
2305 * coule be different when we set, for instance, volmode from "geom"
2306 * to "dev" (or vice versa).
2307 * A possible optimization is to modify our consumers so we don't get
2308 * called when "volmode" does not change.
2311 case ZFS_VOLMODE_NONE
:
2312 (void) zvol_remove_minor_impl(name
);
2314 case ZFS_VOLMODE_GEOM
:
2315 case ZFS_VOLMODE_DEV
:
2316 (void) zvol_remove_minor_impl(name
);
2317 (void) zvol_create_minor_impl(name
);
2319 case ZFS_VOLMODE_DEFAULT
:
2320 (void) zvol_remove_minor_impl(name
);
2321 if (zvol_volmode
== ZFS_VOLMODE_NONE
)
2323 else /* if zvol_volmode is invalid defaults to "geom" */
2324 (void) zvol_create_minor_impl(name
);
2328 spl_fstrans_unmark(cookie
);
2331 static zvol_task_t
*
2332 zvol_task_alloc(zvol_async_op_t op
, const char *name1
, const char *name2
,
2338 /* Never allow tasks on hidden names. */
2339 if (name1
[0] == '$')
2342 task
= kmem_zalloc(sizeof (zvol_task_t
), KM_SLEEP
);
2344 task
->value
= value
;
2345 delim
= strchr(name1
, '/');
2346 strlcpy(task
->pool
, name1
, delim
? (delim
- name1
+ 1) : MAXNAMELEN
);
2348 strlcpy(task
->name1
, name1
, MAXNAMELEN
);
2350 strlcpy(task
->name2
, name2
, MAXNAMELEN
);
2356 zvol_task_free(zvol_task_t
*task
)
2358 kmem_free(task
, sizeof (zvol_task_t
));
2362 * The worker thread function performed asynchronously.
2365 zvol_task_cb(void *param
)
2367 zvol_task_t
*task
= (zvol_task_t
*)param
;
2370 case ZVOL_ASYNC_CREATE_MINORS
:
2371 (void) zvol_create_minors_impl(task
->name1
);
2373 case ZVOL_ASYNC_REMOVE_MINORS
:
2374 zvol_remove_minors_impl(task
->name1
);
2376 case ZVOL_ASYNC_RENAME_MINORS
:
2377 zvol_rename_minors_impl(task
->name1
, task
->name2
);
2379 case ZVOL_ASYNC_SET_SNAPDEV
:
2380 zvol_set_snapdev_impl(task
->name1
, task
->value
);
2382 case ZVOL_ASYNC_SET_VOLMODE
:
2383 zvol_set_volmode_impl(task
->name1
, task
->value
);
2390 zvol_task_free(task
);
2393 typedef struct zvol_set_prop_int_arg
{
2394 const char *zsda_name
;
2395 uint64_t zsda_value
;
2396 zprop_source_t zsda_source
;
2398 } zvol_set_prop_int_arg_t
;
2401 * Sanity check the dataset for safe use by the sync task. No additional
2402 * conditions are imposed.
2405 zvol_set_snapdev_check(void *arg
, dmu_tx_t
*tx
)
2407 zvol_set_prop_int_arg_t
*zsda
= arg
;
2408 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2412 error
= dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
);
2416 dsl_dir_rele(dd
, FTAG
);
2423 zvol_set_snapdev_sync_cb(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2425 char dsname
[MAXNAMELEN
];
2429 dsl_dataset_name(ds
, dsname
);
2430 if (dsl_prop_get_int_ds(ds
, "snapdev", &snapdev
) != 0)
2432 task
= zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV
, dsname
, NULL
, snapdev
);
2436 (void) taskq_dispatch(dp
->dp_spa
->spa_zvol_taskq
, zvol_task_cb
,
2442 * Traverse all child datasets and apply snapdev appropriately.
2443 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
2444 * dataset and read the effective "snapdev" on every child in the callback
2445 * function: this is because the value is not guaranteed to be the same in the
2446 * whole dataset hierarchy.
2449 zvol_set_snapdev_sync(void *arg
, dmu_tx_t
*tx
)
2451 zvol_set_prop_int_arg_t
*zsda
= arg
;
2452 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2457 VERIFY0(dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
));
2460 error
= dsl_dataset_hold(dp
, zsda
->zsda_name
, FTAG
, &ds
);
2462 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_SNAPDEV
),
2463 zsda
->zsda_source
, sizeof (zsda
->zsda_value
), 1,
2464 &zsda
->zsda_value
, zsda
->zsda_tx
);
2465 dsl_dataset_rele(ds
, FTAG
);
2467 dmu_objset_find_dp(dp
, dd
->dd_object
, zvol_set_snapdev_sync_cb
,
2468 zsda
, DS_FIND_CHILDREN
);
2470 dsl_dir_rele(dd
, FTAG
);
2474 zvol_set_snapdev(const char *ddname
, zprop_source_t source
, uint64_t snapdev
)
2476 zvol_set_prop_int_arg_t zsda
;
2478 zsda
.zsda_name
= ddname
;
2479 zsda
.zsda_source
= source
;
2480 zsda
.zsda_value
= snapdev
;
2482 return (dsl_sync_task(ddname
, zvol_set_snapdev_check
,
2483 zvol_set_snapdev_sync
, &zsda
, 0, ZFS_SPACE_CHECK_NONE
));
2487 * Sanity check the dataset for safe use by the sync task. No additional
2488 * conditions are imposed.
2491 zvol_set_volmode_check(void *arg
, dmu_tx_t
*tx
)
2493 zvol_set_prop_int_arg_t
*zsda
= arg
;
2494 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2498 error
= dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
);
2502 dsl_dir_rele(dd
, FTAG
);
2509 zvol_set_volmode_sync_cb(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2511 char dsname
[MAXNAMELEN
];
2515 dsl_dataset_name(ds
, dsname
);
2516 if (dsl_prop_get_int_ds(ds
, "volmode", &volmode
) != 0)
2518 task
= zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE
, dsname
, NULL
, volmode
);
2522 (void) taskq_dispatch(dp
->dp_spa
->spa_zvol_taskq
, zvol_task_cb
,
2528 * Traverse all child datasets and apply volmode appropriately.
2529 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
2530 * dataset and read the effective "volmode" on every child in the callback
2531 * function: this is because the value is not guaranteed to be the same in the
2532 * whole dataset hierarchy.
2535 zvol_set_volmode_sync(void *arg
, dmu_tx_t
*tx
)
2537 zvol_set_prop_int_arg_t
*zsda
= arg
;
2538 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2543 VERIFY0(dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
));
2546 error
= dsl_dataset_hold(dp
, zsda
->zsda_name
, FTAG
, &ds
);
2548 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_VOLMODE
),
2549 zsda
->zsda_source
, sizeof (zsda
->zsda_value
), 1,
2550 &zsda
->zsda_value
, zsda
->zsda_tx
);
2551 dsl_dataset_rele(ds
, FTAG
);
2554 dmu_objset_find_dp(dp
, dd
->dd_object
, zvol_set_volmode_sync_cb
,
2555 zsda
, DS_FIND_CHILDREN
);
2557 dsl_dir_rele(dd
, FTAG
);
2561 zvol_set_volmode(const char *ddname
, zprop_source_t source
, uint64_t volmode
)
2563 zvol_set_prop_int_arg_t zsda
;
2565 zsda
.zsda_name
= ddname
;
2566 zsda
.zsda_source
= source
;
2567 zsda
.zsda_value
= volmode
;
2569 return (dsl_sync_task(ddname
, zvol_set_volmode_check
,
2570 zvol_set_volmode_sync
, &zsda
, 0, ZFS_SPACE_CHECK_NONE
));
2574 zvol_create_minors(spa_t
*spa
, const char *name
, boolean_t async
)
2579 task
= zvol_task_alloc(ZVOL_ASYNC_CREATE_MINORS
, name
, NULL
, ~0ULL);
2583 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2584 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2585 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2589 zvol_remove_minors(spa_t
*spa
, const char *name
, boolean_t async
)
2594 task
= zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS
, name
, NULL
, ~0ULL);
2598 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2599 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2600 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2604 zvol_rename_minors(spa_t
*spa
, const char *name1
, const char *name2
,
2610 task
= zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS
, name1
, name2
, ~0ULL);
2614 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2615 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2616 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2622 int threads
= MIN(MAX(zvol_threads
, 1), 1024);
2625 list_create(&zvol_state_list
, sizeof (zvol_state_t
),
2626 offsetof(zvol_state_t
, zv_next
));
2627 rw_init(&zvol_state_lock
, NULL
, RW_DEFAULT
, NULL
);
2628 ida_init(&zvol_ida
);
2630 zvol_taskq
= taskq_create(ZVOL_DRIVER
, threads
, maxclsyspri
,
2631 threads
* 2, INT_MAX
, TASKQ_PREPOPULATE
| TASKQ_DYNAMIC
);
2632 if (zvol_taskq
== NULL
) {
2633 printk(KERN_INFO
"ZFS: taskq_create() failed\n");
2638 zvol_htable
= kmem_alloc(ZVOL_HT_SIZE
* sizeof (struct hlist_head
),
2644 for (i
= 0; i
< ZVOL_HT_SIZE
; i
++)
2645 INIT_HLIST_HEAD(&zvol_htable
[i
]);
2647 error
= register_blkdev(zvol_major
, ZVOL_DRIVER
);
2649 printk(KERN_INFO
"ZFS: register_blkdev() failed %d\n", error
);
2653 blk_register_region(MKDEV(zvol_major
, 0), 1UL << MINORBITS
,
2654 THIS_MODULE
, zvol_probe
, NULL
, NULL
);
2659 kmem_free(zvol_htable
, ZVOL_HT_SIZE
* sizeof (struct hlist_head
));
2661 taskq_destroy(zvol_taskq
);
2663 ida_destroy(&zvol_ida
);
2664 rw_destroy(&zvol_state_lock
);
2665 list_destroy(&zvol_state_list
);
2667 return (SET_ERROR(error
));
2673 zvol_remove_minors_impl(NULL
);
2675 blk_unregister_region(MKDEV(zvol_major
, 0), 1UL << MINORBITS
);
2676 unregister_blkdev(zvol_major
, ZVOL_DRIVER
);
2677 kmem_free(zvol_htable
, ZVOL_HT_SIZE
* sizeof (struct hlist_head
));
2679 taskq_destroy(zvol_taskq
);
2680 list_destroy(&zvol_state_list
);
2681 rw_destroy(&zvol_state_lock
);
2683 ida_destroy(&zvol_ida
);
2687 module_param(zvol_inhibit_dev
, uint
, 0644);
2688 MODULE_PARM_DESC(zvol_inhibit_dev
, "Do not create zvol device nodes");
2690 module_param(zvol_major
, uint
, 0444);
2691 MODULE_PARM_DESC(zvol_major
, "Major number for zvol device");
2693 module_param(zvol_threads
, uint
, 0444);
2694 MODULE_PARM_DESC(zvol_threads
, "Max number of threads to handle I/O requests");
2696 module_param(zvol_request_sync
, uint
, 0644);
2697 MODULE_PARM_DESC(zvol_request_sync
, "Synchronously handle bio requests");
2699 module_param(zvol_max_discard_blocks
, ulong
, 0444);
2700 MODULE_PARM_DESC(zvol_max_discard_blocks
, "Max number of blocks to discard");
2702 module_param(zvol_prefetch_bytes
, uint
, 0644);
2703 MODULE_PARM_DESC(zvol_prefetch_bytes
, "Prefetch N bytes at zvol start+end");
2705 module_param(zvol_volmode
, uint
, 0644);
2706 MODULE_PARM_DESC(zvol_volmode
, "Default volmode property value");