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, 2018 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.
77 #include <sys/dataset_kstats.h>
79 #include <sys/dmu_traverse.h>
80 #include <sys/dsl_dataset.h>
81 #include <sys/dsl_prop.h>
82 #include <sys/dsl_dir.h>
84 #include <sys/zfeature.h>
85 #include <sys/zil_impl.h>
86 #include <sys/dmu_tx.h>
88 #include <sys/zfs_rlock.h>
89 #include <sys/spa_impl.h>
92 #include <linux/blkdev_compat.h>
93 #include <linux/task_io_accounting_ops.h>
95 unsigned int zvol_inhibit_dev
= 0;
96 unsigned int zvol_major
= ZVOL_MAJOR
;
97 unsigned int zvol_threads
= 32;
98 unsigned int zvol_request_sync
= 0;
99 unsigned int zvol_prefetch_bytes
= (128 * 1024);
100 unsigned long zvol_max_discard_blocks
= 16384;
101 unsigned int zvol_volmode
= ZFS_VOLMODE_GEOM
;
103 static taskq_t
*zvol_taskq
;
104 static krwlock_t zvol_state_lock
;
105 static list_t zvol_state_list
;
107 #define ZVOL_HT_SIZE 1024
108 static struct hlist_head
*zvol_htable
;
109 #define ZVOL_HT_HEAD(hash) (&zvol_htable[(hash) & (ZVOL_HT_SIZE-1)])
111 static struct ida zvol_ida
;
114 * The in-core state of each volume.
117 char zv_name
[MAXNAMELEN
]; /* name */
118 uint64_t zv_volsize
; /* advertised space */
119 uint64_t zv_volblocksize
; /* volume block size */
120 objset_t
*zv_objset
; /* objset handle */
121 uint32_t zv_flags
; /* ZVOL_* flags */
122 uint32_t zv_open_count
; /* open counts */
123 uint32_t zv_changed
; /* disk changed */
124 zilog_t
*zv_zilog
; /* ZIL handle */
125 rangelock_t zv_rangelock
; /* for range locking */
126 dnode_t
*zv_dn
; /* dnode hold */
127 dev_t zv_dev
; /* device id */
128 struct gendisk
*zv_disk
; /* generic disk */
129 struct request_queue
*zv_queue
; /* request queue */
130 dataset_kstats_t zv_kstat
; /* zvol kstats */
131 list_node_t zv_next
; /* next zvol_state_t linkage */
132 uint64_t zv_hash
; /* name hash */
133 struct hlist_node zv_hlink
; /* hash link */
134 kmutex_t zv_state_lock
; /* protects zvol_state_t */
135 atomic_t zv_suspend_ref
; /* refcount for suspend */
136 krwlock_t zv_suspend_lock
; /* suspend lock */
140 ZVOL_ASYNC_CREATE_MINORS
,
141 ZVOL_ASYNC_REMOVE_MINORS
,
142 ZVOL_ASYNC_RENAME_MINORS
,
143 ZVOL_ASYNC_SET_SNAPDEV
,
144 ZVOL_ASYNC_SET_VOLMODE
,
150 char pool
[MAXNAMELEN
];
151 char name1
[MAXNAMELEN
];
152 char name2
[MAXNAMELEN
];
153 zprop_source_t source
;
157 #define ZVOL_RDONLY 0x1
160 zvol_name_hash(const char *name
)
163 uint64_t crc
= -1ULL;
164 uint8_t *p
= (uint8_t *)name
;
165 ASSERT(zfs_crc64_table
[128] == ZFS_CRC64_POLY
);
166 for (i
= 0; i
< MAXNAMELEN
- 1 && *p
; i
++, p
++) {
167 crc
= (crc
>> 8) ^ zfs_crc64_table
[(crc
^ (*p
)) & 0xFF];
173 * Find a zvol_state_t given the full major+minor dev_t. If found,
174 * return with zv_state_lock taken, otherwise, return (NULL) without
175 * taking zv_state_lock.
177 static zvol_state_t
*
178 zvol_find_by_dev(dev_t dev
)
182 rw_enter(&zvol_state_lock
, RW_READER
);
183 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
;
184 zv
= list_next(&zvol_state_list
, zv
)) {
185 mutex_enter(&zv
->zv_state_lock
);
186 if (zv
->zv_dev
== dev
) {
187 rw_exit(&zvol_state_lock
);
190 mutex_exit(&zv
->zv_state_lock
);
192 rw_exit(&zvol_state_lock
);
198 * Find a zvol_state_t given the name and hash generated by zvol_name_hash.
199 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
200 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
201 * before zv_state_lock. The mode argument indicates the mode (including none)
202 * for zv_suspend_lock to be taken.
204 static zvol_state_t
*
205 zvol_find_by_name_hash(const char *name
, uint64_t hash
, int mode
)
208 struct hlist_node
*p
= NULL
;
210 rw_enter(&zvol_state_lock
, RW_READER
);
211 hlist_for_each(p
, ZVOL_HT_HEAD(hash
)) {
212 zv
= hlist_entry(p
, zvol_state_t
, zv_hlink
);
213 mutex_enter(&zv
->zv_state_lock
);
214 if (zv
->zv_hash
== hash
&&
215 strncmp(zv
->zv_name
, name
, MAXNAMELEN
) == 0) {
217 * this is the right zvol, take the locks in the
220 if (mode
!= RW_NONE
&&
221 !rw_tryenter(&zv
->zv_suspend_lock
, mode
)) {
222 mutex_exit(&zv
->zv_state_lock
);
223 rw_enter(&zv
->zv_suspend_lock
, mode
);
224 mutex_enter(&zv
->zv_state_lock
);
226 * zvol cannot be renamed as we continue
227 * to hold zvol_state_lock
229 ASSERT(zv
->zv_hash
== hash
&&
230 strncmp(zv
->zv_name
, name
, MAXNAMELEN
)
233 rw_exit(&zvol_state_lock
);
236 mutex_exit(&zv
->zv_state_lock
);
238 rw_exit(&zvol_state_lock
);
244 * Find a zvol_state_t given the name.
245 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
246 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
247 * before zv_state_lock. The mode argument indicates the mode (including none)
248 * for zv_suspend_lock to be taken.
250 static zvol_state_t
*
251 zvol_find_by_name(const char *name
, int mode
)
253 return (zvol_find_by_name_hash(name
, zvol_name_hash(name
), mode
));
258 * Given a path, return TRUE if path is a ZVOL.
261 zvol_is_zvol(const char *device
)
263 struct block_device
*bdev
;
266 bdev
= vdev_lookup_bdev(device
);
270 major
= MAJOR(bdev
->bd_dev
);
273 if (major
== zvol_major
)
280 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
283 zvol_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
285 zfs_creat_t
*zct
= arg
;
286 nvlist_t
*nvprops
= zct
->zct_props
;
288 uint64_t volblocksize
, volsize
;
290 VERIFY(nvlist_lookup_uint64(nvprops
,
291 zfs_prop_to_name(ZFS_PROP_VOLSIZE
), &volsize
) == 0);
292 if (nvlist_lookup_uint64(nvprops
,
293 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
), &volblocksize
) != 0)
294 volblocksize
= zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE
);
297 * These properties must be removed from the list so the generic
298 * property setting step won't apply to them.
300 VERIFY(nvlist_remove_all(nvprops
,
301 zfs_prop_to_name(ZFS_PROP_VOLSIZE
)) == 0);
302 (void) nvlist_remove_all(nvprops
,
303 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
));
305 error
= dmu_object_claim(os
, ZVOL_OBJ
, DMU_OT_ZVOL
, volblocksize
,
309 error
= zap_create_claim(os
, ZVOL_ZAP_OBJ
, DMU_OT_ZVOL_PROP
,
313 error
= zap_update(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
, tx
);
318 * ZFS_IOC_OBJSET_STATS entry point.
321 zvol_get_stats(objset_t
*os
, nvlist_t
*nv
)
324 dmu_object_info_t
*doi
;
327 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &val
);
329 return (SET_ERROR(error
));
331 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_VOLSIZE
, val
);
332 doi
= kmem_alloc(sizeof (dmu_object_info_t
), KM_SLEEP
);
333 error
= dmu_object_info(os
, ZVOL_OBJ
, doi
);
336 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_VOLBLOCKSIZE
,
337 doi
->doi_data_block_size
);
340 kmem_free(doi
, sizeof (dmu_object_info_t
));
342 return (SET_ERROR(error
));
346 * Sanity check volume size.
349 zvol_check_volsize(uint64_t volsize
, uint64_t blocksize
)
352 return (SET_ERROR(EINVAL
));
354 if (volsize
% blocksize
!= 0)
355 return (SET_ERROR(EINVAL
));
358 if (volsize
- 1 > SPEC_MAXOFFSET_T
)
359 return (SET_ERROR(EOVERFLOW
));
365 * Ensure the zap is flushed then inform the VFS of the capacity change.
368 zvol_update_volsize(uint64_t volsize
, objset_t
*os
)
374 tx
= dmu_tx_create(os
);
375 dmu_tx_hold_zap(tx
, ZVOL_ZAP_OBJ
, TRUE
, NULL
);
376 dmu_tx_mark_netfree(tx
);
377 error
= dmu_tx_assign(tx
, TXG_WAIT
);
380 return (SET_ERROR(error
));
382 txg
= dmu_tx_get_txg(tx
);
384 error
= zap_update(os
, ZVOL_ZAP_OBJ
, "size", 8, 1,
388 txg_wait_synced(dmu_objset_pool(os
), txg
);
391 error
= dmu_free_long_range(os
,
392 ZVOL_OBJ
, volsize
, DMU_OBJECT_END
);
398 * Set ZFS_PROP_VOLSIZE set entry point. Note that modifying the volume
399 * size will result in a udev "change" event being generated.
402 zvol_set_volsize(const char *name
, uint64_t volsize
)
405 struct gendisk
*disk
= NULL
;
408 boolean_t owned
= B_FALSE
;
410 error
= dsl_prop_get_integer(name
,
411 zfs_prop_to_name(ZFS_PROP_READONLY
), &readonly
, NULL
);
413 return (SET_ERROR(error
));
415 return (SET_ERROR(EROFS
));
417 zvol_state_t
*zv
= zvol_find_by_name(name
, RW_READER
);
419 ASSERT(zv
== NULL
|| (MUTEX_HELD(&zv
->zv_state_lock
) &&
420 RW_READ_HELD(&zv
->zv_suspend_lock
)));
422 if (zv
== NULL
|| zv
->zv_objset
== NULL
) {
424 rw_exit(&zv
->zv_suspend_lock
);
425 if ((error
= dmu_objset_own(name
, DMU_OST_ZVOL
, B_FALSE
, B_TRUE
,
428 mutex_exit(&zv
->zv_state_lock
);
429 return (SET_ERROR(error
));
438 dmu_object_info_t
*doi
= kmem_alloc(sizeof (*doi
), KM_SLEEP
);
440 if ((error
= dmu_object_info(os
, ZVOL_OBJ
, doi
)) ||
441 (error
= zvol_check_volsize(volsize
, doi
->doi_data_block_size
)))
444 error
= zvol_update_volsize(volsize
, os
);
445 if (error
== 0 && zv
!= NULL
) {
446 zv
->zv_volsize
= volsize
;
451 kmem_free(doi
, sizeof (dmu_object_info_t
));
454 dmu_objset_disown(os
, B_TRUE
, FTAG
);
456 zv
->zv_objset
= NULL
;
458 rw_exit(&zv
->zv_suspend_lock
);
462 mutex_exit(&zv
->zv_state_lock
);
465 revalidate_disk(disk
);
467 return (SET_ERROR(error
));
471 * Sanity check volume block size.
474 zvol_check_volblocksize(const char *name
, uint64_t volblocksize
)
476 /* Record sizes above 128k need the feature to be enabled */
477 if (volblocksize
> SPA_OLD_MAXBLOCKSIZE
) {
481 if ((error
= spa_open(name
, &spa
, FTAG
)) != 0)
484 if (!spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
)) {
485 spa_close(spa
, FTAG
);
486 return (SET_ERROR(ENOTSUP
));
490 * We don't allow setting the property above 1MB,
491 * unless the tunable has been changed.
493 if (volblocksize
> zfs_max_recordsize
)
494 return (SET_ERROR(EDOM
));
496 spa_close(spa
, FTAG
);
499 if (volblocksize
< SPA_MINBLOCKSIZE
||
500 volblocksize
> SPA_MAXBLOCKSIZE
||
502 return (SET_ERROR(EDOM
));
508 * Set ZFS_PROP_VOLBLOCKSIZE set entry point.
511 zvol_set_volblocksize(const char *name
, uint64_t volblocksize
)
517 zv
= zvol_find_by_name(name
, RW_READER
);
520 return (SET_ERROR(ENXIO
));
522 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
523 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
525 if (zv
->zv_flags
& ZVOL_RDONLY
) {
526 mutex_exit(&zv
->zv_state_lock
);
527 rw_exit(&zv
->zv_suspend_lock
);
528 return (SET_ERROR(EROFS
));
531 tx
= dmu_tx_create(zv
->zv_objset
);
532 dmu_tx_hold_bonus(tx
, ZVOL_OBJ
);
533 error
= dmu_tx_assign(tx
, TXG_WAIT
);
537 error
= dmu_object_set_blocksize(zv
->zv_objset
, ZVOL_OBJ
,
538 volblocksize
, 0, tx
);
539 if (error
== ENOTSUP
)
540 error
= SET_ERROR(EBUSY
);
543 zv
->zv_volblocksize
= volblocksize
;
546 mutex_exit(&zv
->zv_state_lock
);
547 rw_exit(&zv
->zv_suspend_lock
);
549 return (SET_ERROR(error
));
553 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
554 * implement DKIOCFREE/free-long-range.
557 zvol_replay_truncate(void *arg1
, void *arg2
, boolean_t byteswap
)
559 zvol_state_t
*zv
= arg1
;
560 lr_truncate_t
*lr
= arg2
;
561 uint64_t offset
, length
;
564 byteswap_uint64_array(lr
, sizeof (*lr
));
566 offset
= lr
->lr_offset
;
567 length
= lr
->lr_length
;
569 return (dmu_free_long_range(zv
->zv_objset
, ZVOL_OBJ
, offset
, length
));
573 * Replay a TX_WRITE ZIL transaction that didn't get committed
574 * after a system failure
577 zvol_replay_write(void *arg1
, void *arg2
, boolean_t byteswap
)
579 zvol_state_t
*zv
= arg1
;
580 lr_write_t
*lr
= arg2
;
581 objset_t
*os
= zv
->zv_objset
;
582 char *data
= (char *)(lr
+ 1); /* data follows lr_write_t */
583 uint64_t offset
, length
;
588 byteswap_uint64_array(lr
, sizeof (*lr
));
590 offset
= lr
->lr_offset
;
591 length
= lr
->lr_length
;
593 /* If it's a dmu_sync() block, write the whole block */
594 if (lr
->lr_common
.lrc_reclen
== sizeof (lr_write_t
)) {
595 uint64_t blocksize
= BP_GET_LSIZE(&lr
->lr_blkptr
);
596 if (length
< blocksize
) {
597 offset
-= offset
% blocksize
;
602 tx
= dmu_tx_create(os
);
603 dmu_tx_hold_write(tx
, ZVOL_OBJ
, offset
, length
);
604 error
= dmu_tx_assign(tx
, TXG_WAIT
);
608 dmu_write(os
, ZVOL_OBJ
, offset
, length
, data
, tx
);
616 zvol_replay_err(void *arg1
, void *arg2
, boolean_t byteswap
)
618 return (SET_ERROR(ENOTSUP
));
622 * Callback vectors for replaying records.
623 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
625 zil_replay_func_t
*zvol_replay_vector
[TX_MAX_TYPE
] = {
626 zvol_replay_err
, /* no such transaction type */
627 zvol_replay_err
, /* TX_CREATE */
628 zvol_replay_err
, /* TX_MKDIR */
629 zvol_replay_err
, /* TX_MKXATTR */
630 zvol_replay_err
, /* TX_SYMLINK */
631 zvol_replay_err
, /* TX_REMOVE */
632 zvol_replay_err
, /* TX_RMDIR */
633 zvol_replay_err
, /* TX_LINK */
634 zvol_replay_err
, /* TX_RENAME */
635 zvol_replay_write
, /* TX_WRITE */
636 zvol_replay_truncate
, /* TX_TRUNCATE */
637 zvol_replay_err
, /* TX_SETATTR */
638 zvol_replay_err
, /* TX_ACL */
639 zvol_replay_err
, /* TX_CREATE_ATTR */
640 zvol_replay_err
, /* TX_CREATE_ACL_ATTR */
641 zvol_replay_err
, /* TX_MKDIR_ACL */
642 zvol_replay_err
, /* TX_MKDIR_ATTR */
643 zvol_replay_err
, /* TX_MKDIR_ACL_ATTR */
644 zvol_replay_err
, /* TX_WRITE2 */
648 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
650 * We store data in the log buffers if it's small enough.
651 * Otherwise we will later flush the data out via dmu_sync().
653 ssize_t zvol_immediate_write_sz
= 32768;
656 zvol_log_write(zvol_state_t
*zv
, dmu_tx_t
*tx
, uint64_t offset
,
657 uint64_t size
, int sync
)
659 uint32_t blocksize
= zv
->zv_volblocksize
;
660 zilog_t
*zilog
= zv
->zv_zilog
;
661 itx_wr_state_t write_state
;
663 if (zil_replaying(zilog
, tx
))
666 if (zilog
->zl_logbias
== ZFS_LOGBIAS_THROUGHPUT
)
667 write_state
= WR_INDIRECT
;
668 else if (!spa_has_slogs(zilog
->zl_spa
) &&
669 size
>= blocksize
&& blocksize
> zvol_immediate_write_sz
)
670 write_state
= WR_INDIRECT
;
672 write_state
= WR_COPIED
;
674 write_state
= WR_NEED_COPY
;
679 itx_wr_state_t wr_state
= write_state
;
682 if (wr_state
== WR_COPIED
&& size
> ZIL_MAX_COPIED_DATA
)
683 wr_state
= WR_NEED_COPY
;
684 else if (wr_state
== WR_INDIRECT
)
685 len
= MIN(blocksize
- P2PHASE(offset
, blocksize
), size
);
687 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
) +
688 (wr_state
== WR_COPIED
? len
: 0));
689 lr
= (lr_write_t
*)&itx
->itx_lr
;
690 if (wr_state
== WR_COPIED
&& dmu_read_by_dnode(zv
->zv_dn
,
691 offset
, len
, lr
+1, DMU_READ_NO_PREFETCH
) != 0) {
692 zil_itx_destroy(itx
);
693 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
694 lr
= (lr_write_t
*)&itx
->itx_lr
;
695 wr_state
= WR_NEED_COPY
;
698 itx
->itx_wr_state
= wr_state
;
699 lr
->lr_foid
= ZVOL_OBJ
;
700 lr
->lr_offset
= offset
;
703 BP_ZERO(&lr
->lr_blkptr
);
705 itx
->itx_private
= zv
;
706 itx
->itx_sync
= sync
;
708 (void) zil_itx_assign(zilog
, itx
, tx
);
715 typedef struct zv_request
{
722 uio_from_bio(uio_t
*uio
, struct bio
*bio
)
724 uio
->uio_bvec
= &bio
->bi_io_vec
[BIO_BI_IDX(bio
)];
725 uio
->uio_skip
= BIO_BI_SKIP(bio
);
726 uio
->uio_resid
= BIO_BI_SIZE(bio
);
727 uio
->uio_iovcnt
= bio
->bi_vcnt
- BIO_BI_IDX(bio
);
728 uio
->uio_loffset
= BIO_BI_SECTOR(bio
) << 9;
729 uio
->uio_limit
= MAXOFFSET_T
;
730 uio
->uio_segflg
= UIO_BVEC
;
734 zvol_write(void *arg
)
738 zv_request_t
*zvr
= arg
;
739 struct bio
*bio
= zvr
->bio
;
741 uio_from_bio(&uio
, bio
);
743 zvol_state_t
*zv
= zvr
->zv
;
744 ASSERT(zv
&& zv
->zv_open_count
> 0);
746 ssize_t start_resid
= uio
.uio_resid
;
747 unsigned long start_jif
= jiffies
;
748 blk_generic_start_io_acct(zv
->zv_queue
, WRITE
, bio_sectors(bio
),
749 &zv
->zv_disk
->part0
);
752 bio_is_fua(bio
) || zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
754 uint64_t volsize
= zv
->zv_volsize
;
755 while (uio
.uio_resid
> 0 && uio
.uio_loffset
< volsize
) {
756 uint64_t bytes
= MIN(uio
.uio_resid
, DMU_MAX_ACCESS
>> 1);
757 uint64_t off
= uio
.uio_loffset
;
758 dmu_tx_t
*tx
= dmu_tx_create(zv
->zv_objset
);
760 if (bytes
> volsize
- off
) /* don't write past the end */
761 bytes
= volsize
- off
;
763 dmu_tx_hold_write(tx
, ZVOL_OBJ
, off
, bytes
);
765 /* This will only fail for ENOSPC */
766 error
= dmu_tx_assign(tx
, TXG_WAIT
);
771 error
= dmu_write_uio_dnode(zv
->zv_dn
, &uio
, bytes
, tx
);
773 zvol_log_write(zv
, tx
, off
, bytes
, sync
);
780 rangelock_exit(zvr
->lr
);
782 int64_t nwritten
= start_resid
- uio
.uio_resid
;
783 dataset_kstats_update_write_kstats(&zv
->zv_kstat
, nwritten
);
784 task_io_account_write(nwritten
);
787 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
789 rw_exit(&zv
->zv_suspend_lock
);
790 blk_generic_end_io_acct(zv
->zv_queue
, WRITE
, &zv
->zv_disk
->part0
,
792 BIO_END_IO(bio
, -error
);
793 kmem_free(zvr
, sizeof (zv_request_t
));
797 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
800 zvol_log_truncate(zvol_state_t
*zv
, dmu_tx_t
*tx
, uint64_t off
, uint64_t len
,
805 zilog_t
*zilog
= zv
->zv_zilog
;
807 if (zil_replaying(zilog
, tx
))
810 itx
= zil_itx_create(TX_TRUNCATE
, sizeof (*lr
));
811 lr
= (lr_truncate_t
*)&itx
->itx_lr
;
812 lr
->lr_foid
= ZVOL_OBJ
;
816 itx
->itx_sync
= sync
;
817 zil_itx_assign(zilog
, itx
, tx
);
821 zvol_discard(void *arg
)
823 zv_request_t
*zvr
= arg
;
824 struct bio
*bio
= zvr
->bio
;
825 zvol_state_t
*zv
= zvr
->zv
;
826 uint64_t start
= BIO_BI_SECTOR(bio
) << 9;
827 uint64_t size
= BIO_BI_SIZE(bio
);
828 uint64_t end
= start
+ size
;
832 unsigned long start_jif
;
834 ASSERT(zv
&& zv
->zv_open_count
> 0);
837 blk_generic_start_io_acct(zv
->zv_queue
, WRITE
, bio_sectors(bio
),
838 &zv
->zv_disk
->part0
);
840 sync
= bio_is_fua(bio
) || zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
842 if (end
> zv
->zv_volsize
) {
843 error
= SET_ERROR(EIO
);
848 * Align the request to volume block boundaries when a secure erase is
849 * not required. This will prevent dnode_free_range() from zeroing out
850 * the unaligned parts which is slow (read-modify-write) and useless
851 * since we are not freeing any space by doing so.
853 if (!bio_is_secure_erase(bio
)) {
854 start
= P2ROUNDUP(start
, zv
->zv_volblocksize
);
855 end
= P2ALIGN(end
, zv
->zv_volblocksize
);
862 tx
= dmu_tx_create(zv
->zv_objset
);
863 dmu_tx_mark_netfree(tx
);
864 error
= dmu_tx_assign(tx
, TXG_WAIT
);
868 zvol_log_truncate(zv
, tx
, start
, size
, B_TRUE
);
870 error
= dmu_free_long_range(zv
->zv_objset
,
871 ZVOL_OBJ
, start
, size
);
874 rangelock_exit(zvr
->lr
);
876 if (error
== 0 && sync
)
877 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
879 rw_exit(&zv
->zv_suspend_lock
);
880 blk_generic_end_io_acct(zv
->zv_queue
, WRITE
, &zv
->zv_disk
->part0
,
882 BIO_END_IO(bio
, -error
);
883 kmem_free(zvr
, sizeof (zv_request_t
));
891 zv_request_t
*zvr
= arg
;
892 struct bio
*bio
= zvr
->bio
;
894 uio_from_bio(&uio
, bio
);
896 zvol_state_t
*zv
= zvr
->zv
;
897 ASSERT(zv
&& zv
->zv_open_count
> 0);
899 ssize_t start_resid
= uio
.uio_resid
;
900 unsigned long start_jif
= jiffies
;
901 blk_generic_start_io_acct(zv
->zv_queue
, READ
, bio_sectors(bio
),
902 &zv
->zv_disk
->part0
);
904 uint64_t volsize
= zv
->zv_volsize
;
905 while (uio
.uio_resid
> 0 && uio
.uio_loffset
< volsize
) {
906 uint64_t bytes
= MIN(uio
.uio_resid
, DMU_MAX_ACCESS
>> 1);
908 /* don't read past the end */
909 if (bytes
> volsize
- uio
.uio_loffset
)
910 bytes
= volsize
- uio
.uio_loffset
;
912 error
= dmu_read_uio_dnode(zv
->zv_dn
, &uio
, bytes
);
914 /* convert checksum errors into IO errors */
916 error
= SET_ERROR(EIO
);
920 rangelock_exit(zvr
->lr
);
922 int64_t nread
= start_resid
- uio
.uio_resid
;
923 dataset_kstats_update_read_kstats(&zv
->zv_kstat
, nread
);
924 task_io_account_read(nread
);
926 rw_exit(&zv
->zv_suspend_lock
);
927 blk_generic_end_io_acct(zv
->zv_queue
, READ
, &zv
->zv_disk
->part0
,
929 BIO_END_IO(bio
, -error
);
930 kmem_free(zvr
, sizeof (zv_request_t
));
933 static MAKE_REQUEST_FN_RET
934 zvol_request(struct request_queue
*q
, struct bio
*bio
)
936 zvol_state_t
*zv
= q
->queuedata
;
937 fstrans_cookie_t cookie
= spl_fstrans_mark();
938 uint64_t offset
= BIO_BI_SECTOR(bio
) << 9;
939 uint64_t size
= BIO_BI_SIZE(bio
);
940 int rw
= bio_data_dir(bio
);
943 if (bio_has_data(bio
) && offset
+ size
> zv
->zv_volsize
) {
945 "%s: bad access: offset=%llu, size=%lu\n",
946 zv
->zv_disk
->disk_name
,
947 (long long unsigned)offset
,
948 (long unsigned)size
);
950 BIO_END_IO(bio
, -SET_ERROR(EIO
));
955 boolean_t need_sync
= B_FALSE
;
957 if (unlikely(zv
->zv_flags
& ZVOL_RDONLY
)) {
958 BIO_END_IO(bio
, -SET_ERROR(EROFS
));
963 * To be released in the I/O function. See the comment on
964 * zfs_range_lock below.
966 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
968 /* bio marked as FLUSH need to flush before write */
969 if (bio_is_flush(bio
))
970 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
972 /* Some requests are just for flush and nothing else. */
974 rw_exit(&zv
->zv_suspend_lock
);
979 zvr
= kmem_alloc(sizeof (zv_request_t
), KM_SLEEP
);
984 * To be released in the I/O function. Since the I/O functions
985 * are asynchronous, we take it here synchronously to make
986 * sure overlapped I/Os are properly ordered.
988 zvr
->lr
= rangelock_enter(&zv
->zv_rangelock
, offset
, size
,
991 * Sync writes and discards execute zil_commit() which may need
992 * to take a RL_READER lock on the whole block being modified
993 * via its zillog->zl_get_data(): to avoid circular dependency
994 * issues with taskq threads execute these requests
995 * synchronously here in zvol_request().
997 need_sync
= bio_is_fua(bio
) ||
998 zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
;
999 if (bio_is_discard(bio
) || bio_is_secure_erase(bio
)) {
1000 if (zvol_request_sync
|| need_sync
||
1001 taskq_dispatch(zvol_taskq
, zvol_discard
, zvr
,
1002 TQ_SLEEP
) == TASKQID_INVALID
)
1005 if (zvol_request_sync
|| need_sync
||
1006 taskq_dispatch(zvol_taskq
, zvol_write
, zvr
,
1007 TQ_SLEEP
) == TASKQID_INVALID
)
1011 zvr
= kmem_alloc(sizeof (zv_request_t
), KM_SLEEP
);
1015 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1017 zvr
->lr
= rangelock_enter(&zv
->zv_rangelock
, offset
, size
,
1019 if (zvol_request_sync
|| taskq_dispatch(zvol_taskq
,
1020 zvol_read
, zvr
, TQ_SLEEP
) == TASKQID_INVALID
)
1025 spl_fstrans_unmark(cookie
);
1026 #ifdef HAVE_MAKE_REQUEST_FN_RET_INT
1028 #elif defined(HAVE_MAKE_REQUEST_FN_RET_QC)
1029 return (BLK_QC_T_NONE
);
1035 zvol_get_done(zgd_t
*zgd
, int error
)
1038 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1040 rangelock_exit(zgd
->zgd_lr
);
1042 kmem_free(zgd
, sizeof (zgd_t
));
1046 * Get data to generate a TX_WRITE intent log record.
1049 zvol_get_data(void *arg
, lr_write_t
*lr
, char *buf
, struct lwb
*lwb
, zio_t
*zio
)
1051 zvol_state_t
*zv
= arg
;
1052 uint64_t offset
= lr
->lr_offset
;
1053 uint64_t size
= lr
->lr_length
;
1058 ASSERT3P(lwb
, !=, NULL
);
1059 ASSERT3P(zio
, !=, NULL
);
1060 ASSERT3U(size
, !=, 0);
1062 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1066 * Write records come in two flavors: immediate and indirect.
1067 * For small writes it's cheaper to store the data with the
1068 * log record (immediate); for large writes it's cheaper to
1069 * sync the data and get a pointer to it (indirect) so that
1070 * we don't have to write the data twice.
1072 if (buf
!= NULL
) { /* immediate write */
1073 zgd
->zgd_lr
= rangelock_enter(&zv
->zv_rangelock
, offset
, size
,
1075 error
= dmu_read_by_dnode(zv
->zv_dn
, offset
, size
, buf
,
1076 DMU_READ_NO_PREFETCH
);
1077 } else { /* indirect write */
1079 * Have to lock the whole block to ensure when it's written out
1080 * and its checksum is being calculated that no one can change
1081 * the data. Contrarily to zfs_get_data we need not re-check
1082 * blocksize after we get the lock because it cannot be changed.
1084 size
= zv
->zv_volblocksize
;
1085 offset
= P2ALIGN_TYPED(offset
, size
, uint64_t);
1086 zgd
->zgd_lr
= rangelock_enter(&zv
->zv_rangelock
, offset
, size
,
1088 error
= dmu_buf_hold_by_dnode(zv
->zv_dn
, offset
, zgd
, &db
,
1089 DMU_READ_NO_PREFETCH
);
1091 blkptr_t
*bp
= &lr
->lr_blkptr
;
1097 ASSERT(db
->db_offset
== offset
);
1098 ASSERT(db
->db_size
== size
);
1100 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1101 zvol_get_done
, zgd
);
1108 zvol_get_done(zgd
, error
);
1110 return (SET_ERROR(error
));
1114 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable.
1117 zvol_insert(zvol_state_t
*zv
)
1119 ASSERT(RW_WRITE_HELD(&zvol_state_lock
));
1120 ASSERT3U(MINOR(zv
->zv_dev
) & ZVOL_MINOR_MASK
, ==, 0);
1121 list_insert_head(&zvol_state_list
, zv
);
1122 hlist_add_head(&zv
->zv_hlink
, ZVOL_HT_HEAD(zv
->zv_hash
));
1126 * Simply remove the zvol from to list of zvols.
1129 zvol_remove(zvol_state_t
*zv
)
1131 ASSERT(RW_WRITE_HELD(&zvol_state_lock
));
1132 list_remove(&zvol_state_list
, zv
);
1133 hlist_del(&zv
->zv_hlink
);
1137 * Setup zv after we just own the zv->objset
1140 zvol_setup_zv(zvol_state_t
*zv
)
1145 objset_t
*os
= zv
->zv_objset
;
1147 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1148 ASSERT(RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1150 error
= dsl_prop_get_integer(zv
->zv_name
, "readonly", &ro
, NULL
);
1152 return (SET_ERROR(error
));
1154 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
1156 return (SET_ERROR(error
));
1158 error
= dnode_hold(os
, ZVOL_OBJ
, FTAG
, &zv
->zv_dn
);
1160 return (SET_ERROR(error
));
1162 set_capacity(zv
->zv_disk
, volsize
>> 9);
1163 zv
->zv_volsize
= volsize
;
1164 zv
->zv_zilog
= zil_open(os
, zvol_get_data
);
1166 if (ro
|| dmu_objset_is_snapshot(os
) ||
1167 !spa_writeable(dmu_objset_spa(os
))) {
1168 set_disk_ro(zv
->zv_disk
, 1);
1169 zv
->zv_flags
|= ZVOL_RDONLY
;
1171 set_disk_ro(zv
->zv_disk
, 0);
1172 zv
->zv_flags
&= ~ZVOL_RDONLY
;
1178 * Shutdown every zv_objset related stuff except zv_objset itself.
1179 * The is the reverse of zvol_setup_zv.
1182 zvol_shutdown_zv(zvol_state_t
*zv
)
1184 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
) &&
1185 RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1187 zil_close(zv
->zv_zilog
);
1188 zv
->zv_zilog
= NULL
;
1190 dnode_rele(zv
->zv_dn
, FTAG
);
1194 * Evict cached data. We must write out any dirty data before
1195 * disowning the dataset.
1197 if (!(zv
->zv_flags
& ZVOL_RDONLY
))
1198 txg_wait_synced(dmu_objset_pool(zv
->zv_objset
), 0);
1199 (void) dmu_objset_evict_dbufs(zv
->zv_objset
);
1203 * return the proper tag for rollback and recv
1206 zvol_tag(zvol_state_t
*zv
)
1208 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1209 return (zv
->zv_open_count
> 0 ? zv
: NULL
);
1213 * Suspend the zvol for recv and rollback.
1216 zvol_suspend(const char *name
)
1220 zv
= zvol_find_by_name(name
, RW_WRITER
);
1225 /* block all I/O, release in zvol_resume. */
1226 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1227 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1229 atomic_inc(&zv
->zv_suspend_ref
);
1231 if (zv
->zv_open_count
> 0)
1232 zvol_shutdown_zv(zv
);
1235 * do not hold zv_state_lock across suspend/resume to
1236 * avoid locking up zvol lookups
1238 mutex_exit(&zv
->zv_state_lock
);
1240 /* zv_suspend_lock is released in zvol_resume() */
1245 zvol_resume(zvol_state_t
*zv
)
1249 ASSERT(RW_WRITE_HELD(&zv
->zv_suspend_lock
));
1251 mutex_enter(&zv
->zv_state_lock
);
1253 if (zv
->zv_open_count
> 0) {
1254 VERIFY0(dmu_objset_hold(zv
->zv_name
, zv
, &zv
->zv_objset
));
1255 VERIFY3P(zv
->zv_objset
->os_dsl_dataset
->ds_owner
, ==, zv
);
1256 VERIFY(dsl_dataset_long_held(zv
->zv_objset
->os_dsl_dataset
));
1257 dmu_objset_rele(zv
->zv_objset
, zv
);
1259 error
= zvol_setup_zv(zv
);
1262 mutex_exit(&zv
->zv_state_lock
);
1264 rw_exit(&zv
->zv_suspend_lock
);
1266 * We need this because we don't hold zvol_state_lock while releasing
1267 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check
1268 * zv_suspend_lock to determine it is safe to free because rwlock is
1269 * not inherent atomic.
1271 atomic_dec(&zv
->zv_suspend_ref
);
1273 return (SET_ERROR(error
));
1277 zvol_first_open(zvol_state_t
*zv
, boolean_t readonly
)
1280 int error
, locked
= 0;
1283 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
1284 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1287 * In all other cases the spa_namespace_lock is taken before the
1288 * bdev->bd_mutex lock. But in this case the Linux __blkdev_get()
1289 * function calls fops->open() with the bdev->bd_mutex lock held.
1290 * This deadlock can be easily observed with zvols used as vdevs.
1292 * To avoid a potential lock inversion deadlock we preemptively
1293 * try to take the spa_namespace_lock(). Normally it will not
1294 * be contended and this is safe because spa_open_common() handles
1295 * the case where the caller already holds the spa_namespace_lock.
1297 * When it is contended we risk a lock inversion if we were to
1298 * block waiting for the lock. Luckily, the __blkdev_get()
1299 * function allows us to return -ERESTARTSYS which will result in
1300 * bdev->bd_mutex being dropped, reacquired, and fops->open() being
1301 * called again. This process can be repeated safely until both
1302 * locks are acquired.
1304 if (!mutex_owned(&spa_namespace_lock
)) {
1305 locked
= mutex_tryenter(&spa_namespace_lock
);
1307 return (-SET_ERROR(ERESTARTSYS
));
1310 ro
= (readonly
|| (strchr(zv
->zv_name
, '@') != NULL
));
1311 error
= dmu_objset_own(zv
->zv_name
, DMU_OST_ZVOL
, ro
, B_TRUE
, zv
, &os
);
1317 error
= zvol_setup_zv(zv
);
1320 dmu_objset_disown(os
, 1, zv
);
1321 zv
->zv_objset
= NULL
;
1326 mutex_exit(&spa_namespace_lock
);
1327 return (SET_ERROR(-error
));
1331 zvol_last_close(zvol_state_t
*zv
)
1333 ASSERT(RW_READ_HELD(&zv
->zv_suspend_lock
));
1334 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1336 zvol_shutdown_zv(zv
);
1338 dmu_objset_disown(zv
->zv_objset
, 1, zv
);
1339 zv
->zv_objset
= NULL
;
1343 zvol_open(struct block_device
*bdev
, fmode_t flag
)
1347 boolean_t drop_suspend
= B_TRUE
;
1349 rw_enter(&zvol_state_lock
, RW_READER
);
1351 * Obtain a copy of private_data under the zvol_state_lock to make
1352 * sure that either the result of zvol free code path setting
1353 * bdev->bd_disk->private_data to NULL is observed, or zvol_free()
1354 * is not called on this zv because of the positive zv_open_count.
1356 zv
= bdev
->bd_disk
->private_data
;
1358 rw_exit(&zvol_state_lock
);
1359 return (SET_ERROR(-ENXIO
));
1362 mutex_enter(&zv
->zv_state_lock
);
1364 * make sure zvol is not suspended during first open
1365 * (hold zv_suspend_lock) and respect proper lock acquisition
1366 * ordering - zv_suspend_lock before zv_state_lock
1368 if (zv
->zv_open_count
== 0) {
1369 if (!rw_tryenter(&zv
->zv_suspend_lock
, RW_READER
)) {
1370 mutex_exit(&zv
->zv_state_lock
);
1371 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1372 mutex_enter(&zv
->zv_state_lock
);
1373 /* check to see if zv_suspend_lock is needed */
1374 if (zv
->zv_open_count
!= 0) {
1375 rw_exit(&zv
->zv_suspend_lock
);
1376 drop_suspend
= B_FALSE
;
1380 drop_suspend
= B_FALSE
;
1382 rw_exit(&zvol_state_lock
);
1384 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1385 ASSERT(zv
->zv_open_count
!= 0 || RW_READ_HELD(&zv
->zv_suspend_lock
));
1387 if (zv
->zv_open_count
== 0) {
1388 error
= zvol_first_open(zv
, !(flag
& FMODE_WRITE
));
1393 if ((flag
& FMODE_WRITE
) && (zv
->zv_flags
& ZVOL_RDONLY
)) {
1395 goto out_open_count
;
1398 zv
->zv_open_count
++;
1400 mutex_exit(&zv
->zv_state_lock
);
1402 rw_exit(&zv
->zv_suspend_lock
);
1404 check_disk_change(bdev
);
1409 if (zv
->zv_open_count
== 0)
1410 zvol_last_close(zv
);
1413 mutex_exit(&zv
->zv_state_lock
);
1415 rw_exit(&zv
->zv_suspend_lock
);
1416 if (error
== -ERESTARTSYS
)
1419 return (SET_ERROR(error
));
1422 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1427 zvol_release(struct gendisk
*disk
, fmode_t mode
)
1430 boolean_t drop_suspend
= B_TRUE
;
1432 rw_enter(&zvol_state_lock
, RW_READER
);
1433 zv
= disk
->private_data
;
1435 mutex_enter(&zv
->zv_state_lock
);
1436 ASSERT(zv
->zv_open_count
> 0);
1438 * make sure zvol is not suspended during last close
1439 * (hold zv_suspend_lock) and respect proper lock acquisition
1440 * ordering - zv_suspend_lock before zv_state_lock
1442 if (zv
->zv_open_count
== 1) {
1443 if (!rw_tryenter(&zv
->zv_suspend_lock
, RW_READER
)) {
1444 mutex_exit(&zv
->zv_state_lock
);
1445 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1446 mutex_enter(&zv
->zv_state_lock
);
1447 /* check to see if zv_suspend_lock is needed */
1448 if (zv
->zv_open_count
!= 1) {
1449 rw_exit(&zv
->zv_suspend_lock
);
1450 drop_suspend
= B_FALSE
;
1454 drop_suspend
= B_FALSE
;
1456 rw_exit(&zvol_state_lock
);
1458 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1459 ASSERT(zv
->zv_open_count
!= 1 || RW_READ_HELD(&zv
->zv_suspend_lock
));
1461 zv
->zv_open_count
--;
1462 if (zv
->zv_open_count
== 0)
1463 zvol_last_close(zv
);
1465 mutex_exit(&zv
->zv_state_lock
);
1468 rw_exit(&zv
->zv_suspend_lock
);
1470 #ifndef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1476 zvol_ioctl(struct block_device
*bdev
, fmode_t mode
,
1477 unsigned int cmd
, unsigned long arg
)
1479 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
1482 ASSERT3U(zv
->zv_open_count
, >, 0);
1487 invalidate_bdev(bdev
);
1488 rw_enter(&zv
->zv_suspend_lock
, RW_READER
);
1490 if (!(zv
->zv_flags
& ZVOL_RDONLY
))
1491 txg_wait_synced(dmu_objset_pool(zv
->zv_objset
), 0);
1493 rw_exit(&zv
->zv_suspend_lock
);
1497 mutex_enter(&zv
->zv_state_lock
);
1498 error
= copy_to_user((void *)arg
, zv
->zv_name
, MAXNAMELEN
);
1499 mutex_exit(&zv
->zv_state_lock
);
1507 return (SET_ERROR(error
));
1510 #ifdef CONFIG_COMPAT
1512 zvol_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1513 unsigned cmd
, unsigned long arg
)
1515 return (zvol_ioctl(bdev
, mode
, cmd
, arg
));
1518 #define zvol_compat_ioctl NULL
1522 * Linux 2.6.38 preferred interface.
1524 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS
1526 zvol_check_events(struct gendisk
*disk
, unsigned int clearing
)
1528 unsigned int mask
= 0;
1530 rw_enter(&zvol_state_lock
, RW_READER
);
1532 zvol_state_t
*zv
= disk
->private_data
;
1534 mutex_enter(&zv
->zv_state_lock
);
1535 mask
= zv
->zv_changed
? DISK_EVENT_MEDIA_CHANGE
: 0;
1537 mutex_exit(&zv
->zv_state_lock
);
1540 rw_exit(&zvol_state_lock
);
1545 static int zvol_media_changed(struct gendisk
*disk
)
1549 rw_enter(&zvol_state_lock
, RW_READER
);
1551 zvol_state_t
*zv
= disk
->private_data
;
1553 mutex_enter(&zv
->zv_state_lock
);
1554 changed
= zv
->zv_changed
;
1556 mutex_exit(&zv
->zv_state_lock
);
1559 rw_exit(&zvol_state_lock
);
1565 static int zvol_revalidate_disk(struct gendisk
*disk
)
1567 rw_enter(&zvol_state_lock
, RW_READER
);
1569 zvol_state_t
*zv
= disk
->private_data
;
1571 mutex_enter(&zv
->zv_state_lock
);
1572 set_capacity(zv
->zv_disk
, zv
->zv_volsize
>> SECTOR_BITS
);
1573 mutex_exit(&zv
->zv_state_lock
);
1576 rw_exit(&zvol_state_lock
);
1582 * Provide a simple virtual geometry for legacy compatibility. For devices
1583 * smaller than 1 MiB a small head and sector count is used to allow very
1584 * tiny devices. For devices over 1 Mib a standard head and sector count
1585 * is used to keep the cylinders count reasonable.
1588 zvol_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1590 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
1593 ASSERT3U(zv
->zv_open_count
, >, 0);
1595 sectors
= get_capacity(zv
->zv_disk
);
1597 if (sectors
> 2048) {
1606 geo
->cylinders
= sectors
/ (geo
->heads
* geo
->sectors
);
1611 static struct kobject
*
1612 zvol_probe(dev_t dev
, int *part
, void *arg
)
1615 struct kobject
*kobj
;
1617 zv
= zvol_find_by_dev(dev
);
1618 kobj
= zv
? get_disk_and_module(zv
->zv_disk
) : NULL
;
1619 ASSERT(zv
== NULL
|| MUTEX_HELD(&zv
->zv_state_lock
));
1621 mutex_exit(&zv
->zv_state_lock
);
1626 static struct block_device_operations zvol_ops
= {
1628 .release
= zvol_release
,
1629 .ioctl
= zvol_ioctl
,
1630 .compat_ioctl
= zvol_compat_ioctl
,
1631 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS
1632 .check_events
= zvol_check_events
,
1634 .media_changed
= zvol_media_changed
,
1636 .revalidate_disk
= zvol_revalidate_disk
,
1637 .getgeo
= zvol_getgeo
,
1638 .owner
= THIS_MODULE
,
1642 * Allocate memory for a new zvol_state_t and setup the required
1643 * request queue and generic disk structures for the block device.
1645 static zvol_state_t
*
1646 zvol_alloc(dev_t dev
, const char *name
)
1651 if (dsl_prop_get_integer(name
, "volmode", &volmode
, NULL
) != 0)
1654 if (volmode
== ZFS_VOLMODE_DEFAULT
)
1655 volmode
= zvol_volmode
;
1657 if (volmode
== ZFS_VOLMODE_NONE
)
1660 zv
= kmem_zalloc(sizeof (zvol_state_t
), KM_SLEEP
);
1662 list_link_init(&zv
->zv_next
);
1664 mutex_init(&zv
->zv_state_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1666 zv
->zv_queue
= blk_alloc_queue(GFP_ATOMIC
);
1667 if (zv
->zv_queue
== NULL
)
1670 blk_queue_make_request(zv
->zv_queue
, zvol_request
);
1671 blk_queue_set_write_cache(zv
->zv_queue
, B_TRUE
, B_TRUE
);
1673 /* Limit read-ahead to a single page to prevent over-prefetching. */
1674 blk_queue_set_read_ahead(zv
->zv_queue
, 1);
1676 /* Disable write merging in favor of the ZIO pipeline. */
1677 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, zv
->zv_queue
);
1679 zv
->zv_disk
= alloc_disk(ZVOL_MINORS
);
1680 if (zv
->zv_disk
== NULL
)
1683 zv
->zv_queue
->queuedata
= zv
;
1685 zv
->zv_open_count
= 0;
1686 strlcpy(zv
->zv_name
, name
, MAXNAMELEN
);
1688 rangelock_init(&zv
->zv_rangelock
, NULL
, NULL
);
1689 rw_init(&zv
->zv_suspend_lock
, NULL
, RW_DEFAULT
, NULL
);
1691 zv
->zv_disk
->major
= zvol_major
;
1692 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS
1693 zv
->zv_disk
->events
= DISK_EVENT_MEDIA_CHANGE
;
1696 if (volmode
== ZFS_VOLMODE_DEV
) {
1698 * ZFS_VOLMODE_DEV disable partitioning on ZVOL devices: set
1699 * gendisk->minors = 1 as noted in include/linux/genhd.h.
1700 * Also disable extended partition numbers (GENHD_FL_EXT_DEVT)
1701 * and suppresses partition scanning (GENHD_FL_NO_PART_SCAN)
1702 * setting gendisk->flags accordingly.
1704 zv
->zv_disk
->minors
= 1;
1705 #if defined(GENHD_FL_EXT_DEVT)
1706 zv
->zv_disk
->flags
&= ~GENHD_FL_EXT_DEVT
;
1708 #if defined(GENHD_FL_NO_PART_SCAN)
1709 zv
->zv_disk
->flags
|= GENHD_FL_NO_PART_SCAN
;
1712 zv
->zv_disk
->first_minor
= (dev
& MINORMASK
);
1713 zv
->zv_disk
->fops
= &zvol_ops
;
1714 zv
->zv_disk
->private_data
= zv
;
1715 zv
->zv_disk
->queue
= zv
->zv_queue
;
1716 snprintf(zv
->zv_disk
->disk_name
, DISK_NAME_LEN
, "%s%d",
1717 ZVOL_DEV_NAME
, (dev
& MINORMASK
));
1722 blk_cleanup_queue(zv
->zv_queue
);
1724 kmem_free(zv
, sizeof (zvol_state_t
));
1730 * Cleanup then free a zvol_state_t which was created by zvol_alloc().
1731 * At this time, the structure is not opened by anyone, is taken off
1732 * the zvol_state_list, and has its private data set to NULL.
1733 * The zvol_state_lock is dropped.
1736 zvol_free(void *arg
)
1738 zvol_state_t
*zv
= arg
;
1740 ASSERT(!RW_LOCK_HELD(&zv
->zv_suspend_lock
));
1741 ASSERT(!MUTEX_HELD(&zv
->zv_state_lock
));
1742 ASSERT(zv
->zv_open_count
== 0);
1743 ASSERT(zv
->zv_disk
->private_data
== NULL
);
1745 rw_destroy(&zv
->zv_suspend_lock
);
1746 rangelock_fini(&zv
->zv_rangelock
);
1748 del_gendisk(zv
->zv_disk
);
1749 blk_cleanup_queue(zv
->zv_queue
);
1750 put_disk(zv
->zv_disk
);
1752 ida_simple_remove(&zvol_ida
, MINOR(zv
->zv_dev
) >> ZVOL_MINOR_BITS
);
1754 mutex_destroy(&zv
->zv_state_lock
);
1755 dataset_kstats_destroy(&zv
->zv_kstat
);
1757 kmem_free(zv
, sizeof (zvol_state_t
));
1761 * Create a block device minor node and setup the linkage between it
1762 * and the specified volume. Once this function returns the block
1763 * device is live and ready for use.
1766 zvol_create_minor_impl(const char *name
)
1770 dmu_object_info_t
*doi
;
1776 uint64_t hash
= zvol_name_hash(name
);
1778 if (zvol_inhibit_dev
)
1781 idx
= ida_simple_get(&zvol_ida
, 0, 0, kmem_flags_convert(KM_SLEEP
));
1783 return (SET_ERROR(-idx
));
1784 minor
= idx
<< ZVOL_MINOR_BITS
;
1786 zv
= zvol_find_by_name_hash(name
, hash
, RW_NONE
);
1788 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1789 mutex_exit(&zv
->zv_state_lock
);
1790 ida_simple_remove(&zvol_ida
, idx
);
1791 return (SET_ERROR(EEXIST
));
1794 doi
= kmem_alloc(sizeof (dmu_object_info_t
), KM_SLEEP
);
1796 error
= dmu_objset_own(name
, DMU_OST_ZVOL
, B_TRUE
, B_TRUE
, FTAG
, &os
);
1800 error
= dmu_object_info(os
, ZVOL_OBJ
, doi
);
1802 goto out_dmu_objset_disown
;
1804 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
1806 goto out_dmu_objset_disown
;
1808 zv
= zvol_alloc(MKDEV(zvol_major
, minor
), name
);
1810 error
= SET_ERROR(EAGAIN
);
1811 goto out_dmu_objset_disown
;
1815 if (dmu_objset_is_snapshot(os
))
1816 zv
->zv_flags
|= ZVOL_RDONLY
;
1818 zv
->zv_volblocksize
= doi
->doi_data_block_size
;
1819 zv
->zv_volsize
= volsize
;
1822 set_capacity(zv
->zv_disk
, zv
->zv_volsize
>> 9);
1824 blk_queue_max_hw_sectors(zv
->zv_queue
, (DMU_MAX_ACCESS
/ 4) >> 9);
1825 blk_queue_max_segments(zv
->zv_queue
, UINT16_MAX
);
1826 blk_queue_max_segment_size(zv
->zv_queue
, UINT_MAX
);
1827 blk_queue_physical_block_size(zv
->zv_queue
, zv
->zv_volblocksize
);
1828 blk_queue_io_opt(zv
->zv_queue
, zv
->zv_volblocksize
);
1829 blk_queue_max_discard_sectors(zv
->zv_queue
,
1830 (zvol_max_discard_blocks
* zv
->zv_volblocksize
) >> 9);
1831 blk_queue_discard_granularity(zv
->zv_queue
, zv
->zv_volblocksize
);
1832 blk_queue_flag_set(QUEUE_FLAG_DISCARD
, zv
->zv_queue
);
1833 #ifdef QUEUE_FLAG_NONROT
1834 blk_queue_flag_set(QUEUE_FLAG_NONROT
, zv
->zv_queue
);
1836 #ifdef QUEUE_FLAG_ADD_RANDOM
1837 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM
, zv
->zv_queue
);
1840 if (spa_writeable(dmu_objset_spa(os
))) {
1841 if (zil_replay_disable
)
1842 zil_destroy(dmu_objset_zil(os
), B_FALSE
);
1844 zil_replay(os
, zv
, zvol_replay_vector
);
1846 ASSERT3P(zv
->zv_kstat
.dk_kstats
, ==, NULL
);
1847 dataset_kstats_create(&zv
->zv_kstat
, zv
->zv_objset
);
1850 * When udev detects the addition of the device it will immediately
1851 * invoke blkid(8) to determine the type of content on the device.
1852 * Prefetching the blocks commonly scanned by blkid(8) will speed
1855 len
= MIN(MAX(zvol_prefetch_bytes
, 0), SPA_MAXBLOCKSIZE
);
1857 dmu_prefetch(os
, ZVOL_OBJ
, 0, 0, len
, ZIO_PRIORITY_SYNC_READ
);
1858 dmu_prefetch(os
, ZVOL_OBJ
, 0, volsize
- len
, len
,
1859 ZIO_PRIORITY_SYNC_READ
);
1862 zv
->zv_objset
= NULL
;
1863 out_dmu_objset_disown
:
1864 dmu_objset_disown(os
, B_TRUE
, FTAG
);
1866 kmem_free(doi
, sizeof (dmu_object_info_t
));
1869 rw_enter(&zvol_state_lock
, RW_WRITER
);
1871 rw_exit(&zvol_state_lock
);
1872 add_disk(zv
->zv_disk
);
1874 ida_simple_remove(&zvol_ida
, idx
);
1877 return (SET_ERROR(error
));
1881 * Rename a block device minor mode for the specified volume.
1884 zvol_rename_minor(zvol_state_t
*zv
, const char *newname
)
1886 int readonly
= get_disk_ro(zv
->zv_disk
);
1888 ASSERT(RW_LOCK_HELD(&zvol_state_lock
));
1889 ASSERT(MUTEX_HELD(&zv
->zv_state_lock
));
1891 strlcpy(zv
->zv_name
, newname
, sizeof (zv
->zv_name
));
1893 /* move to new hashtable entry */
1894 zv
->zv_hash
= zvol_name_hash(zv
->zv_name
);
1895 hlist_del(&zv
->zv_hlink
);
1896 hlist_add_head(&zv
->zv_hlink
, ZVOL_HT_HEAD(zv
->zv_hash
));
1899 * The block device's read-only state is briefly changed causing
1900 * a KOBJ_CHANGE uevent to be issued. This ensures udev detects
1901 * the name change and fixes the symlinks. This does not change
1902 * ZVOL_RDONLY in zv->zv_flags so the actual read-only state never
1903 * changes. This would normally be done using kobject_uevent() but
1904 * that is a GPL-only symbol which is why we need this workaround.
1906 set_disk_ro(zv
->zv_disk
, !readonly
);
1907 set_disk_ro(zv
->zv_disk
, readonly
);
1910 typedef struct minors_job
{
1920 * Prefetch zvol dnodes for the minors_job
1923 zvol_prefetch_minors_impl(void *arg
)
1925 minors_job_t
*job
= arg
;
1926 char *dsname
= job
->name
;
1927 objset_t
*os
= NULL
;
1929 job
->error
= dmu_objset_own(dsname
, DMU_OST_ZVOL
, B_TRUE
, B_TRUE
,
1931 if (job
->error
== 0) {
1932 dmu_prefetch(os
, ZVOL_OBJ
, 0, 0, 0, ZIO_PRIORITY_SYNC_READ
);
1933 dmu_objset_disown(os
, B_TRUE
, FTAG
);
1938 * Mask errors to continue dmu_objset_find() traversal
1941 zvol_create_snap_minor_cb(const char *dsname
, void *arg
)
1943 minors_job_t
*j
= arg
;
1944 list_t
*minors_list
= j
->list
;
1945 const char *name
= j
->name
;
1947 ASSERT0(MUTEX_HELD(&spa_namespace_lock
));
1949 /* skip the designated dataset */
1950 if (name
&& strcmp(dsname
, name
) == 0)
1953 /* at this point, the dsname should name a snapshot */
1954 if (strchr(dsname
, '@') == 0) {
1955 dprintf("zvol_create_snap_minor_cb(): "
1956 "%s is not a shapshot name\n", dsname
);
1959 char *n
= strdup(dsname
);
1963 job
= kmem_alloc(sizeof (minors_job_t
), KM_SLEEP
);
1965 job
->list
= minors_list
;
1967 list_insert_tail(minors_list
, job
);
1968 /* don't care if dispatch fails, because job->error is 0 */
1969 taskq_dispatch(system_taskq
, zvol_prefetch_minors_impl
, job
,
1977 * Mask errors to continue dmu_objset_find() traversal
1980 zvol_create_minors_cb(const char *dsname
, void *arg
)
1984 list_t
*minors_list
= arg
;
1986 ASSERT0(MUTEX_HELD(&spa_namespace_lock
));
1988 error
= dsl_prop_get_integer(dsname
, "snapdev", &snapdev
, NULL
);
1993 * Given the name and the 'snapdev' property, create device minor nodes
1994 * with the linkages to zvols/snapshots as needed.
1995 * If the name represents a zvol, create a minor node for the zvol, then
1996 * check if its snapshots are 'visible', and if so, iterate over the
1997 * snapshots and create device minor nodes for those.
1999 if (strchr(dsname
, '@') == 0) {
2001 char *n
= strdup(dsname
);
2005 job
= kmem_alloc(sizeof (minors_job_t
), KM_SLEEP
);
2007 job
->list
= minors_list
;
2009 list_insert_tail(minors_list
, job
);
2010 /* don't care if dispatch fails, because job->error is 0 */
2011 taskq_dispatch(system_taskq
, zvol_prefetch_minors_impl
, job
,
2014 if (snapdev
== ZFS_SNAPDEV_VISIBLE
) {
2016 * traverse snapshots only, do not traverse children,
2017 * and skip the 'dsname'
2019 error
= dmu_objset_find((char *)dsname
,
2020 zvol_create_snap_minor_cb
, (void *)job
,
2024 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n",
2032 * Create minors for the specified dataset, including children and snapshots.
2033 * Pay attention to the 'snapdev' property and iterate over the snapshots
2034 * only if they are 'visible'. This approach allows one to assure that the
2035 * snapshot metadata is read from disk only if it is needed.
2037 * The name can represent a dataset to be recursively scanned for zvols and
2038 * their snapshots, or a single zvol snapshot. If the name represents a
2039 * dataset, the scan is performed in two nested stages:
2040 * - scan the dataset for zvols, and
2041 * - for each zvol, create a minor node, then check if the zvol's snapshots
2042 * are 'visible', and only then iterate over the snapshots if needed
2044 * If the name represents a snapshot, a check is performed if the snapshot is
2045 * 'visible' (which also verifies that the parent is a zvol), and if so,
2046 * a minor node for that snapshot is created.
2049 zvol_create_minors_impl(const char *name
)
2052 fstrans_cookie_t cookie
;
2057 if (zvol_inhibit_dev
)
2061 * This is the list for prefetch jobs. Whenever we found a match
2062 * during dmu_objset_find, we insert a minors_job to the list and do
2063 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need
2064 * any lock because all list operation is done on the current thread.
2066 * We will use this list to do zvol_create_minor_impl after prefetch
2067 * so we don't have to traverse using dmu_objset_find again.
2069 list_create(&minors_list
, sizeof (minors_job_t
),
2070 offsetof(minors_job_t
, link
));
2072 parent
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
2073 (void) strlcpy(parent
, name
, MAXPATHLEN
);
2075 if ((atp
= strrchr(parent
, '@')) != NULL
) {
2079 error
= dsl_prop_get_integer(parent
, "snapdev",
2082 if (error
== 0 && snapdev
== ZFS_SNAPDEV_VISIBLE
)
2083 error
= zvol_create_minor_impl(name
);
2085 cookie
= spl_fstrans_mark();
2086 error
= dmu_objset_find(parent
, zvol_create_minors_cb
,
2087 &minors_list
, DS_FIND_CHILDREN
);
2088 spl_fstrans_unmark(cookie
);
2091 kmem_free(parent
, MAXPATHLEN
);
2092 taskq_wait_outstanding(system_taskq
, 0);
2095 * Prefetch is completed, we can do zvol_create_minor_impl
2098 while ((job
= list_head(&minors_list
)) != NULL
) {
2099 list_remove(&minors_list
, job
);
2101 zvol_create_minor_impl(job
->name
);
2103 kmem_free(job
, sizeof (minors_job_t
));
2106 list_destroy(&minors_list
);
2108 return (SET_ERROR(error
));
2112 * Remove minors for specified dataset including children and snapshots.
2115 zvol_remove_minors_impl(const char *name
)
2117 zvol_state_t
*zv
, *zv_next
;
2118 int namelen
= ((name
) ? strlen(name
) : 0);
2119 taskqid_t t
, tid
= TASKQID_INVALID
;
2122 if (zvol_inhibit_dev
)
2125 list_create(&free_list
, sizeof (zvol_state_t
),
2126 offsetof(zvol_state_t
, zv_next
));
2128 rw_enter(&zvol_state_lock
, RW_WRITER
);
2130 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2131 zv_next
= list_next(&zvol_state_list
, zv
);
2133 mutex_enter(&zv
->zv_state_lock
);
2134 if (name
== NULL
|| strcmp(zv
->zv_name
, name
) == 0 ||
2135 (strncmp(zv
->zv_name
, name
, namelen
) == 0 &&
2136 (zv
->zv_name
[namelen
] == '/' ||
2137 zv
->zv_name
[namelen
] == '@'))) {
2139 * By holding zv_state_lock here, we guarantee that no
2140 * one is currently using this zv
2143 /* If in use, leave alone */
2144 if (zv
->zv_open_count
> 0 ||
2145 atomic_read(&zv
->zv_suspend_ref
)) {
2146 mutex_exit(&zv
->zv_state_lock
);
2153 * Cleared while holding zvol_state_lock as a writer
2154 * which will prevent zvol_open() from opening it.
2156 zv
->zv_disk
->private_data
= NULL
;
2158 /* Drop zv_state_lock before zvol_free() */
2159 mutex_exit(&zv
->zv_state_lock
);
2161 /* Try parallel zv_free, if failed do it in place */
2162 t
= taskq_dispatch(system_taskq
, zvol_free
, zv
,
2164 if (t
== TASKQID_INVALID
)
2165 list_insert_head(&free_list
, zv
);
2169 mutex_exit(&zv
->zv_state_lock
);
2172 rw_exit(&zvol_state_lock
);
2174 /* Drop zvol_state_lock before calling zvol_free() */
2175 while ((zv
= list_head(&free_list
)) != NULL
) {
2176 list_remove(&free_list
, zv
);
2180 if (tid
!= TASKQID_INVALID
)
2181 taskq_wait_outstanding(system_taskq
, tid
);
2184 /* Remove minor for this specific volume only */
2186 zvol_remove_minor_impl(const char *name
)
2188 zvol_state_t
*zv
= NULL
, *zv_next
;
2190 if (zvol_inhibit_dev
)
2193 rw_enter(&zvol_state_lock
, RW_WRITER
);
2195 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2196 zv_next
= list_next(&zvol_state_list
, zv
);
2198 mutex_enter(&zv
->zv_state_lock
);
2199 if (strcmp(zv
->zv_name
, name
) == 0) {
2201 * By holding zv_state_lock here, we guarantee that no
2202 * one is currently using this zv
2205 /* If in use, leave alone */
2206 if (zv
->zv_open_count
> 0 ||
2207 atomic_read(&zv
->zv_suspend_ref
)) {
2208 mutex_exit(&zv
->zv_state_lock
);
2214 * Cleared while holding zvol_state_lock as a writer
2215 * which will prevent zvol_open() from opening it.
2217 zv
->zv_disk
->private_data
= NULL
;
2219 mutex_exit(&zv
->zv_state_lock
);
2222 mutex_exit(&zv
->zv_state_lock
);
2226 /* Drop zvol_state_lock before calling zvol_free() */
2227 rw_exit(&zvol_state_lock
);
2234 * Rename minors for specified dataset including children and snapshots.
2237 zvol_rename_minors_impl(const char *oldname
, const char *newname
)
2239 zvol_state_t
*zv
, *zv_next
;
2240 int oldnamelen
, newnamelen
;
2242 if (zvol_inhibit_dev
)
2245 oldnamelen
= strlen(oldname
);
2246 newnamelen
= strlen(newname
);
2248 rw_enter(&zvol_state_lock
, RW_READER
);
2250 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
2251 zv_next
= list_next(&zvol_state_list
, zv
);
2253 mutex_enter(&zv
->zv_state_lock
);
2255 /* If in use, leave alone */
2256 if (zv
->zv_open_count
> 0) {
2257 mutex_exit(&zv
->zv_state_lock
);
2261 if (strcmp(zv
->zv_name
, oldname
) == 0) {
2262 zvol_rename_minor(zv
, newname
);
2263 } else if (strncmp(zv
->zv_name
, oldname
, oldnamelen
) == 0 &&
2264 (zv
->zv_name
[oldnamelen
] == '/' ||
2265 zv
->zv_name
[oldnamelen
] == '@')) {
2266 char *name
= kmem_asprintf("%s%c%s", newname
,
2267 zv
->zv_name
[oldnamelen
],
2268 zv
->zv_name
+ oldnamelen
+ 1);
2269 zvol_rename_minor(zv
, name
);
2273 mutex_exit(&zv
->zv_state_lock
);
2276 rw_exit(&zvol_state_lock
);
2279 typedef struct zvol_snapdev_cb_arg
{
2281 } zvol_snapdev_cb_arg_t
;
2284 zvol_set_snapdev_cb(const char *dsname
, void *param
)
2286 zvol_snapdev_cb_arg_t
*arg
= param
;
2288 if (strchr(dsname
, '@') == NULL
)
2291 switch (arg
->snapdev
) {
2292 case ZFS_SNAPDEV_VISIBLE
:
2293 (void) zvol_create_minor_impl(dsname
);
2295 case ZFS_SNAPDEV_HIDDEN
:
2296 (void) zvol_remove_minor_impl(dsname
);
2304 zvol_set_snapdev_impl(char *name
, uint64_t snapdev
)
2306 zvol_snapdev_cb_arg_t arg
= {snapdev
};
2307 fstrans_cookie_t cookie
= spl_fstrans_mark();
2309 * The zvol_set_snapdev_sync() sets snapdev appropriately
2310 * in the dataset hierarchy. Here, we only scan snapshots.
2312 dmu_objset_find(name
, zvol_set_snapdev_cb
, &arg
, DS_FIND_SNAPSHOTS
);
2313 spl_fstrans_unmark(cookie
);
2316 typedef struct zvol_volmode_cb_arg
{
2318 } zvol_volmode_cb_arg_t
;
2321 zvol_set_volmode_impl(char *name
, uint64_t volmode
)
2323 fstrans_cookie_t cookie
= spl_fstrans_mark();
2325 if (strchr(name
, '@') != NULL
)
2329 * It's unfortunate we need to remove minors before we create new ones:
2330 * this is necessary because our backing gendisk (zvol_state->zv_disk)
2331 * coule be different when we set, for instance, volmode from "geom"
2332 * to "dev" (or vice versa).
2333 * A possible optimization is to modify our consumers so we don't get
2334 * called when "volmode" does not change.
2337 case ZFS_VOLMODE_NONE
:
2338 (void) zvol_remove_minor_impl(name
);
2340 case ZFS_VOLMODE_GEOM
:
2341 case ZFS_VOLMODE_DEV
:
2342 (void) zvol_remove_minor_impl(name
);
2343 (void) zvol_create_minor_impl(name
);
2345 case ZFS_VOLMODE_DEFAULT
:
2346 (void) zvol_remove_minor_impl(name
);
2347 if (zvol_volmode
== ZFS_VOLMODE_NONE
)
2349 else /* if zvol_volmode is invalid defaults to "geom" */
2350 (void) zvol_create_minor_impl(name
);
2354 spl_fstrans_unmark(cookie
);
2357 static zvol_task_t
*
2358 zvol_task_alloc(zvol_async_op_t op
, const char *name1
, const char *name2
,
2364 /* Never allow tasks on hidden names. */
2365 if (name1
[0] == '$')
2368 task
= kmem_zalloc(sizeof (zvol_task_t
), KM_SLEEP
);
2370 task
->value
= value
;
2371 delim
= strchr(name1
, '/');
2372 strlcpy(task
->pool
, name1
, delim
? (delim
- name1
+ 1) : MAXNAMELEN
);
2374 strlcpy(task
->name1
, name1
, MAXNAMELEN
);
2376 strlcpy(task
->name2
, name2
, MAXNAMELEN
);
2382 zvol_task_free(zvol_task_t
*task
)
2384 kmem_free(task
, sizeof (zvol_task_t
));
2388 * The worker thread function performed asynchronously.
2391 zvol_task_cb(void *param
)
2393 zvol_task_t
*task
= (zvol_task_t
*)param
;
2396 case ZVOL_ASYNC_CREATE_MINORS
:
2397 (void) zvol_create_minors_impl(task
->name1
);
2399 case ZVOL_ASYNC_REMOVE_MINORS
:
2400 zvol_remove_minors_impl(task
->name1
);
2402 case ZVOL_ASYNC_RENAME_MINORS
:
2403 zvol_rename_minors_impl(task
->name1
, task
->name2
);
2405 case ZVOL_ASYNC_SET_SNAPDEV
:
2406 zvol_set_snapdev_impl(task
->name1
, task
->value
);
2408 case ZVOL_ASYNC_SET_VOLMODE
:
2409 zvol_set_volmode_impl(task
->name1
, task
->value
);
2416 zvol_task_free(task
);
2419 typedef struct zvol_set_prop_int_arg
{
2420 const char *zsda_name
;
2421 uint64_t zsda_value
;
2422 zprop_source_t zsda_source
;
2424 } zvol_set_prop_int_arg_t
;
2427 * Sanity check the dataset for safe use by the sync task. No additional
2428 * conditions are imposed.
2431 zvol_set_snapdev_check(void *arg
, dmu_tx_t
*tx
)
2433 zvol_set_prop_int_arg_t
*zsda
= arg
;
2434 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2438 error
= dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
);
2442 dsl_dir_rele(dd
, FTAG
);
2449 zvol_set_snapdev_sync_cb(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2451 char dsname
[MAXNAMELEN
];
2455 dsl_dataset_name(ds
, dsname
);
2456 if (dsl_prop_get_int_ds(ds
, "snapdev", &snapdev
) != 0)
2458 task
= zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV
, dsname
, NULL
, snapdev
);
2462 (void) taskq_dispatch(dp
->dp_spa
->spa_zvol_taskq
, zvol_task_cb
,
2468 * Traverse all child datasets and apply snapdev appropriately.
2469 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
2470 * dataset and read the effective "snapdev" on every child in the callback
2471 * function: this is because the value is not guaranteed to be the same in the
2472 * whole dataset hierarchy.
2475 zvol_set_snapdev_sync(void *arg
, dmu_tx_t
*tx
)
2477 zvol_set_prop_int_arg_t
*zsda
= arg
;
2478 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2483 VERIFY0(dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
));
2486 error
= dsl_dataset_hold(dp
, zsda
->zsda_name
, FTAG
, &ds
);
2488 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_SNAPDEV
),
2489 zsda
->zsda_source
, sizeof (zsda
->zsda_value
), 1,
2490 &zsda
->zsda_value
, zsda
->zsda_tx
);
2491 dsl_dataset_rele(ds
, FTAG
);
2493 dmu_objset_find_dp(dp
, dd
->dd_object
, zvol_set_snapdev_sync_cb
,
2494 zsda
, DS_FIND_CHILDREN
);
2496 dsl_dir_rele(dd
, FTAG
);
2500 zvol_set_snapdev(const char *ddname
, zprop_source_t source
, uint64_t snapdev
)
2502 zvol_set_prop_int_arg_t zsda
;
2504 zsda
.zsda_name
= ddname
;
2505 zsda
.zsda_source
= source
;
2506 zsda
.zsda_value
= snapdev
;
2508 return (dsl_sync_task(ddname
, zvol_set_snapdev_check
,
2509 zvol_set_snapdev_sync
, &zsda
, 0, ZFS_SPACE_CHECK_NONE
));
2513 * Sanity check the dataset for safe use by the sync task. No additional
2514 * conditions are imposed.
2517 zvol_set_volmode_check(void *arg
, dmu_tx_t
*tx
)
2519 zvol_set_prop_int_arg_t
*zsda
= arg
;
2520 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2524 error
= dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
);
2528 dsl_dir_rele(dd
, FTAG
);
2535 zvol_set_volmode_sync_cb(dsl_pool_t
*dp
, dsl_dataset_t
*ds
, void *arg
)
2537 char dsname
[MAXNAMELEN
];
2541 dsl_dataset_name(ds
, dsname
);
2542 if (dsl_prop_get_int_ds(ds
, "volmode", &volmode
) != 0)
2544 task
= zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE
, dsname
, NULL
, volmode
);
2548 (void) taskq_dispatch(dp
->dp_spa
->spa_zvol_taskq
, zvol_task_cb
,
2554 * Traverse all child datasets and apply volmode appropriately.
2555 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
2556 * dataset and read the effective "volmode" on every child in the callback
2557 * function: this is because the value is not guaranteed to be the same in the
2558 * whole dataset hierarchy.
2561 zvol_set_volmode_sync(void *arg
, dmu_tx_t
*tx
)
2563 zvol_set_prop_int_arg_t
*zsda
= arg
;
2564 dsl_pool_t
*dp
= dmu_tx_pool(tx
);
2569 VERIFY0(dsl_dir_hold(dp
, zsda
->zsda_name
, FTAG
, &dd
, NULL
));
2572 error
= dsl_dataset_hold(dp
, zsda
->zsda_name
, FTAG
, &ds
);
2574 dsl_prop_set_sync_impl(ds
, zfs_prop_to_name(ZFS_PROP_VOLMODE
),
2575 zsda
->zsda_source
, sizeof (zsda
->zsda_value
), 1,
2576 &zsda
->zsda_value
, zsda
->zsda_tx
);
2577 dsl_dataset_rele(ds
, FTAG
);
2580 dmu_objset_find_dp(dp
, dd
->dd_object
, zvol_set_volmode_sync_cb
,
2581 zsda
, DS_FIND_CHILDREN
);
2583 dsl_dir_rele(dd
, FTAG
);
2587 zvol_set_volmode(const char *ddname
, zprop_source_t source
, uint64_t volmode
)
2589 zvol_set_prop_int_arg_t zsda
;
2591 zsda
.zsda_name
= ddname
;
2592 zsda
.zsda_source
= source
;
2593 zsda
.zsda_value
= volmode
;
2595 return (dsl_sync_task(ddname
, zvol_set_volmode_check
,
2596 zvol_set_volmode_sync
, &zsda
, 0, ZFS_SPACE_CHECK_NONE
));
2600 zvol_create_minors(spa_t
*spa
, const char *name
, boolean_t async
)
2605 task
= zvol_task_alloc(ZVOL_ASYNC_CREATE_MINORS
, name
, NULL
, ~0ULL);
2609 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2610 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2611 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2615 zvol_remove_minors(spa_t
*spa
, const char *name
, boolean_t async
)
2620 task
= zvol_task_alloc(ZVOL_ASYNC_REMOVE_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_rename_minors(spa_t
*spa
, const char *name1
, const char *name2
,
2636 task
= zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS
, name1
, name2
, ~0ULL);
2640 id
= taskq_dispatch(spa
->spa_zvol_taskq
, zvol_task_cb
, task
, TQ_SLEEP
);
2641 if ((async
== B_FALSE
) && (id
!= TASKQID_INVALID
))
2642 taskq_wait_id(spa
->spa_zvol_taskq
, id
);
2648 int threads
= MIN(MAX(zvol_threads
, 1), 1024);
2651 list_create(&zvol_state_list
, sizeof (zvol_state_t
),
2652 offsetof(zvol_state_t
, zv_next
));
2653 rw_init(&zvol_state_lock
, NULL
, RW_DEFAULT
, NULL
);
2654 ida_init(&zvol_ida
);
2656 zvol_taskq
= taskq_create(ZVOL_DRIVER
, threads
, maxclsyspri
,
2657 threads
* 2, INT_MAX
, TASKQ_PREPOPULATE
| TASKQ_DYNAMIC
);
2658 if (zvol_taskq
== NULL
) {
2659 printk(KERN_INFO
"ZFS: taskq_create() failed\n");
2664 zvol_htable
= kmem_alloc(ZVOL_HT_SIZE
* sizeof (struct hlist_head
),
2670 for (i
= 0; i
< ZVOL_HT_SIZE
; i
++)
2671 INIT_HLIST_HEAD(&zvol_htable
[i
]);
2673 error
= register_blkdev(zvol_major
, ZVOL_DRIVER
);
2675 printk(KERN_INFO
"ZFS: register_blkdev() failed %d\n", error
);
2679 blk_register_region(MKDEV(zvol_major
, 0), 1UL << MINORBITS
,
2680 THIS_MODULE
, zvol_probe
, NULL
, NULL
);
2685 kmem_free(zvol_htable
, ZVOL_HT_SIZE
* sizeof (struct hlist_head
));
2687 taskq_destroy(zvol_taskq
);
2689 ida_destroy(&zvol_ida
);
2690 rw_destroy(&zvol_state_lock
);
2691 list_destroy(&zvol_state_list
);
2693 return (SET_ERROR(error
));
2699 zvol_remove_minors_impl(NULL
);
2701 blk_unregister_region(MKDEV(zvol_major
, 0), 1UL << MINORBITS
);
2702 unregister_blkdev(zvol_major
, ZVOL_DRIVER
);
2703 kmem_free(zvol_htable
, ZVOL_HT_SIZE
* sizeof (struct hlist_head
));
2705 taskq_destroy(zvol_taskq
);
2706 list_destroy(&zvol_state_list
);
2707 rw_destroy(&zvol_state_lock
);
2709 ida_destroy(&zvol_ida
);
2713 module_param(zvol_inhibit_dev
, uint
, 0644);
2714 MODULE_PARM_DESC(zvol_inhibit_dev
, "Do not create zvol device nodes");
2716 module_param(zvol_major
, uint
, 0444);
2717 MODULE_PARM_DESC(zvol_major
, "Major number for zvol device");
2719 module_param(zvol_threads
, uint
, 0444);
2720 MODULE_PARM_DESC(zvol_threads
, "Max number of threads to handle I/O requests");
2722 module_param(zvol_request_sync
, uint
, 0644);
2723 MODULE_PARM_DESC(zvol_request_sync
, "Synchronously handle bio requests");
2725 module_param(zvol_max_discard_blocks
, ulong
, 0444);
2726 MODULE_PARM_DESC(zvol_max_discard_blocks
, "Max number of blocks to discard");
2728 module_param(zvol_prefetch_bytes
, uint
, 0644);
2729 MODULE_PARM_DESC(zvol_prefetch_bytes
, "Prefetch N bytes at zvol start+end");
2731 module_param(zvol_volmode
, uint
, 0644);
2732 MODULE_PARM_DESC(zvol_volmode
, "Default volmode property value");