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.
38 #include <sys/dmu_traverse.h>
39 #include <sys/dsl_dataset.h>
40 #include <sys/dsl_prop.h>
42 #include <sys/zil_impl.h>
44 #include <sys/zfs_rlock.h>
45 #include <sys/zfs_znode.h>
47 #include <linux/blkdev_compat.h>
49 unsigned int zvol_inhibit_dev
= 0;
50 unsigned int zvol_major
= ZVOL_MAJOR
;
51 unsigned int zvol_threads
= 32;
52 unsigned long zvol_max_discard_blocks
= 16384;
54 static taskq_t
*zvol_taskq
;
55 static kmutex_t zvol_state_lock
;
56 static list_t zvol_state_list
;
57 static char *zvol_tag
= "zvol_tag";
60 * The in-core state of each volume.
62 typedef struct zvol_state
{
63 char zv_name
[MAXNAMELEN
]; /* name */
64 uint64_t zv_volsize
; /* advertised space */
65 uint64_t zv_volblocksize
;/* volume block size */
66 objset_t
*zv_objset
; /* objset handle */
67 uint32_t zv_flags
; /* ZVOL_* flags */
68 uint32_t zv_open_count
; /* open counts */
69 uint32_t zv_changed
; /* disk changed */
70 zilog_t
*zv_zilog
; /* ZIL handle */
71 znode_t zv_znode
; /* for range locking */
72 dmu_buf_t
*zv_dbuf
; /* bonus handle */
73 dev_t zv_dev
; /* device id */
74 struct gendisk
*zv_disk
; /* generic disk */
75 struct request_queue
*zv_queue
; /* request queue */
76 spinlock_t zv_lock
; /* request queue lock */
77 list_node_t zv_next
; /* next zvol_state_t linkage */
80 #define ZVOL_RDONLY 0x1
83 * Find the next available range of ZVOL_MINORS minor numbers. The
84 * zvol_state_list is kept in ascending minor order so we simply need
85 * to scan the list for the first gap in the sequence. This allows us
86 * to recycle minor number as devices are created and removed.
89 zvol_find_minor(unsigned *minor
)
94 ASSERT(MUTEX_HELD(&zvol_state_lock
));
95 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
;
96 zv
= list_next(&zvol_state_list
, zv
), *minor
+= ZVOL_MINORS
) {
97 if (MINOR(zv
->zv_dev
) != MINOR(*minor
))
101 /* All minors are in use */
102 if (*minor
>= (1 << MINORBITS
))
109 * Find a zvol_state_t given the full major+minor dev_t.
111 static zvol_state_t
*
112 zvol_find_by_dev(dev_t dev
)
116 ASSERT(MUTEX_HELD(&zvol_state_lock
));
117 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
;
118 zv
= list_next(&zvol_state_list
, zv
)) {
119 if (zv
->zv_dev
== dev
)
127 * Find a zvol_state_t given the name provided at zvol_alloc() time.
129 static zvol_state_t
*
130 zvol_find_by_name(const char *name
)
134 ASSERT(MUTEX_HELD(&zvol_state_lock
));
135 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
;
136 zv
= list_next(&zvol_state_list
, zv
)) {
137 if (!strncmp(zv
->zv_name
, name
, MAXNAMELEN
))
146 * Given a path, return TRUE if path is a ZVOL.
149 zvol_is_zvol(const char *device
)
151 struct block_device
*bdev
;
154 bdev
= lookup_bdev(device
);
158 major
= MAJOR(bdev
->bd_dev
);
161 if (major
== zvol_major
)
168 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
171 zvol_create_cb(objset_t
*os
, void *arg
, cred_t
*cr
, dmu_tx_t
*tx
)
173 zfs_creat_t
*zct
= arg
;
174 nvlist_t
*nvprops
= zct
->zct_props
;
176 uint64_t volblocksize
, volsize
;
178 VERIFY(nvlist_lookup_uint64(nvprops
,
179 zfs_prop_to_name(ZFS_PROP_VOLSIZE
), &volsize
) == 0);
180 if (nvlist_lookup_uint64(nvprops
,
181 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
), &volblocksize
) != 0)
182 volblocksize
= zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE
);
185 * These properties must be removed from the list so the generic
186 * property setting step won't apply to them.
188 VERIFY(nvlist_remove_all(nvprops
,
189 zfs_prop_to_name(ZFS_PROP_VOLSIZE
)) == 0);
190 (void) nvlist_remove_all(nvprops
,
191 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE
));
193 error
= dmu_object_claim(os
, ZVOL_OBJ
, DMU_OT_ZVOL
, volblocksize
,
197 error
= zap_create_claim(os
, ZVOL_ZAP_OBJ
, DMU_OT_ZVOL_PROP
,
201 error
= zap_update(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
, tx
);
206 * ZFS_IOC_OBJSET_STATS entry point.
209 zvol_get_stats(objset_t
*os
, nvlist_t
*nv
)
212 dmu_object_info_t
*doi
;
215 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &val
);
219 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_VOLSIZE
, val
);
220 doi
= kmem_alloc(sizeof(dmu_object_info_t
), KM_SLEEP
);
221 error
= dmu_object_info(os
, ZVOL_OBJ
, doi
);
224 dsl_prop_nvlist_add_uint64(nv
, ZFS_PROP_VOLBLOCKSIZE
,
225 doi
->doi_data_block_size
);
228 kmem_free(doi
, sizeof(dmu_object_info_t
));
234 * Sanity check volume size.
237 zvol_check_volsize(uint64_t volsize
, uint64_t blocksize
)
240 return (SET_ERROR(EINVAL
));
242 if (volsize
% blocksize
!= 0)
243 return (SET_ERROR(EINVAL
));
246 if (volsize
- 1 > MAXOFFSET_T
)
247 return (SET_ERROR(EOVERFLOW
));
253 * Ensure the zap is flushed then inform the VFS of the capacity change.
256 zvol_update_volsize(zvol_state_t
*zv
, uint64_t volsize
, objset_t
*os
)
258 struct block_device
*bdev
;
262 ASSERT(MUTEX_HELD(&zvol_state_lock
));
264 tx
= dmu_tx_create(os
);
265 dmu_tx_hold_zap(tx
, ZVOL_ZAP_OBJ
, TRUE
, NULL
);
266 error
= dmu_tx_assign(tx
, TXG_WAIT
);
272 error
= zap_update(os
, ZVOL_ZAP_OBJ
, "size", 8, 1,
279 error
= dmu_free_long_range(os
,
280 ZVOL_OBJ
, volsize
, DMU_OBJECT_END
);
284 bdev
= bdget_disk(zv
->zv_disk
, 0);
286 return (SET_ERROR(EIO
));
289 * Added check_disk_size_change() helper function.
291 #ifdef HAVE_CHECK_DISK_SIZE_CHANGE
292 set_capacity(zv
->zv_disk
, volsize
>> 9);
293 zv
->zv_volsize
= volsize
;
294 check_disk_size_change(zv
->zv_disk
, bdev
);
296 zv
->zv_volsize
= volsize
;
298 (void) check_disk_change(bdev
);
299 #endif /* HAVE_CHECK_DISK_SIZE_CHANGE */
307 * Set ZFS_PROP_VOLSIZE set entry point.
310 zvol_set_volsize(const char *name
, uint64_t volsize
)
313 dmu_object_info_t
*doi
;
318 error
= dsl_prop_get_integer(name
,
319 zfs_prop_to_name(ZFS_PROP_READONLY
), &readonly
, NULL
);
323 return (SET_ERROR(EROFS
));
325 mutex_enter(&zvol_state_lock
);
327 zv
= zvol_find_by_name(name
);
329 error
= SET_ERROR(ENXIO
);
333 doi
= kmem_alloc(sizeof(dmu_object_info_t
), KM_SLEEP
);
335 error
= dmu_objset_hold(name
, FTAG
, &os
);
339 if ((error
= dmu_object_info(os
, ZVOL_OBJ
, doi
)) != 0 ||
340 (error
= zvol_check_volsize(volsize
,doi
->doi_data_block_size
)) != 0)
343 VERIFY(dsl_prop_get_integer(name
, "readonly", &readonly
, NULL
) == 0);
345 error
= SET_ERROR(EROFS
);
349 if (get_disk_ro(zv
->zv_disk
) || (zv
->zv_flags
& ZVOL_RDONLY
)) {
350 error
= SET_ERROR(EROFS
);
354 error
= zvol_update_volsize(zv
, volsize
, os
);
356 kmem_free(doi
, sizeof(dmu_object_info_t
));
359 dmu_objset_rele(os
, FTAG
);
361 mutex_exit(&zvol_state_lock
);
367 * Sanity check volume block size.
370 zvol_check_volblocksize(uint64_t volblocksize
)
372 if (volblocksize
< SPA_MINBLOCKSIZE
||
373 volblocksize
> SPA_MAXBLOCKSIZE
||
375 return (SET_ERROR(EDOM
));
381 * Set ZFS_PROP_VOLBLOCKSIZE set entry point.
384 zvol_set_volblocksize(const char *name
, uint64_t volblocksize
)
390 mutex_enter(&zvol_state_lock
);
392 zv
= zvol_find_by_name(name
);
394 error
= SET_ERROR(ENXIO
);
398 if (get_disk_ro(zv
->zv_disk
) || (zv
->zv_flags
& ZVOL_RDONLY
)) {
399 error
= SET_ERROR(EROFS
);
403 tx
= dmu_tx_create(zv
->zv_objset
);
404 dmu_tx_hold_bonus(tx
, ZVOL_OBJ
);
405 error
= dmu_tx_assign(tx
, TXG_WAIT
);
409 error
= dmu_object_set_blocksize(zv
->zv_objset
, ZVOL_OBJ
,
410 volblocksize
, 0, tx
);
411 if (error
== ENOTSUP
)
412 error
= SET_ERROR(EBUSY
);
415 zv
->zv_volblocksize
= volblocksize
;
418 mutex_exit(&zvol_state_lock
);
424 * Replay a TX_WRITE ZIL transaction that didn't get committed
425 * after a system failure
428 zvol_replay_write(zvol_state_t
*zv
, lr_write_t
*lr
, boolean_t byteswap
)
430 objset_t
*os
= zv
->zv_objset
;
431 char *data
= (char *)(lr
+ 1); /* data follows lr_write_t */
432 uint64_t off
= lr
->lr_offset
;
433 uint64_t len
= lr
->lr_length
;
438 byteswap_uint64_array(lr
, sizeof (*lr
));
440 tx
= dmu_tx_create(os
);
441 dmu_tx_hold_write(tx
, ZVOL_OBJ
, off
, len
);
442 error
= dmu_tx_assign(tx
, TXG_WAIT
);
446 dmu_write(os
, ZVOL_OBJ
, off
, len
, data
, tx
);
454 zvol_replay_err(zvol_state_t
*zv
, lr_t
*lr
, boolean_t byteswap
)
456 return (SET_ERROR(ENOTSUP
));
460 * Callback vectors for replaying records.
461 * Only TX_WRITE is needed for zvol.
463 zil_replay_func_t zvol_replay_vector
[TX_MAX_TYPE
] = {
464 (zil_replay_func_t
)zvol_replay_err
, /* no such transaction type */
465 (zil_replay_func_t
)zvol_replay_err
, /* TX_CREATE */
466 (zil_replay_func_t
)zvol_replay_err
, /* TX_MKDIR */
467 (zil_replay_func_t
)zvol_replay_err
, /* TX_MKXATTR */
468 (zil_replay_func_t
)zvol_replay_err
, /* TX_SYMLINK */
469 (zil_replay_func_t
)zvol_replay_err
, /* TX_REMOVE */
470 (zil_replay_func_t
)zvol_replay_err
, /* TX_RMDIR */
471 (zil_replay_func_t
)zvol_replay_err
, /* TX_LINK */
472 (zil_replay_func_t
)zvol_replay_err
, /* TX_RENAME */
473 (zil_replay_func_t
)zvol_replay_write
, /* TX_WRITE */
474 (zil_replay_func_t
)zvol_replay_err
, /* TX_TRUNCATE */
475 (zil_replay_func_t
)zvol_replay_err
, /* TX_SETATTR */
476 (zil_replay_func_t
)zvol_replay_err
, /* TX_ACL */
480 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
482 * We store data in the log buffers if it's small enough.
483 * Otherwise we will later flush the data out via dmu_sync().
485 ssize_t zvol_immediate_write_sz
= 32768;
488 zvol_log_write(zvol_state_t
*zv
, dmu_tx_t
*tx
,
489 uint64_t offset
, uint64_t size
, int sync
)
491 uint32_t blocksize
= zv
->zv_volblocksize
;
492 zilog_t
*zilog
= zv
->zv_zilog
;
494 ssize_t immediate_write_sz
;
496 if (zil_replaying(zilog
, tx
))
499 immediate_write_sz
= (zilog
->zl_logbias
== ZFS_LOGBIAS_THROUGHPUT
)
500 ? 0 : zvol_immediate_write_sz
;
501 slogging
= spa_has_slogs(zilog
->zl_spa
) &&
502 (zilog
->zl_logbias
== ZFS_LOGBIAS_LATENCY
);
508 itx_wr_state_t write_state
;
511 * Unlike zfs_log_write() we can be called with
512 * up to DMU_MAX_ACCESS/2 (5MB) writes.
514 if (blocksize
> immediate_write_sz
&& !slogging
&&
515 size
>= blocksize
&& offset
% blocksize
== 0) {
516 write_state
= WR_INDIRECT
; /* uses dmu_sync */
519 write_state
= WR_COPIED
;
520 len
= MIN(ZIL_MAX_LOG_DATA
, size
);
522 write_state
= WR_NEED_COPY
;
523 len
= MIN(ZIL_MAX_LOG_DATA
, size
);
526 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
) +
527 (write_state
== WR_COPIED
? len
: 0));
528 lr
= (lr_write_t
*)&itx
->itx_lr
;
529 if (write_state
== WR_COPIED
&& dmu_read(zv
->zv_objset
,
530 ZVOL_OBJ
, offset
, len
, lr
+1, DMU_READ_NO_PREFETCH
) != 0) {
531 zil_itx_destroy(itx
);
532 itx
= zil_itx_create(TX_WRITE
, sizeof (*lr
));
533 lr
= (lr_write_t
*)&itx
->itx_lr
;
534 write_state
= WR_NEED_COPY
;
537 itx
->itx_wr_state
= write_state
;
538 if (write_state
== WR_NEED_COPY
)
540 lr
->lr_foid
= ZVOL_OBJ
;
541 lr
->lr_offset
= offset
;
544 BP_ZERO(&lr
->lr_blkptr
);
546 itx
->itx_private
= zv
;
547 itx
->itx_sync
= sync
;
549 (void) zil_itx_assign(zilog
, itx
, tx
);
557 * Common write path running under the zvol taskq context. This function
558 * is responsible for copying the request structure data in to the DMU and
559 * signaling the request queue with the result of the copy.
562 zvol_write(void *arg
)
564 struct request
*req
= (struct request
*)arg
;
565 struct request_queue
*q
= req
->q
;
566 zvol_state_t
*zv
= q
->queuedata
;
567 uint64_t offset
= blk_rq_pos(req
) << 9;
568 uint64_t size
= blk_rq_bytes(req
);
574 * Annotate this call path with a flag that indicates that it is
575 * unsafe to use KM_SLEEP during memory allocations due to the
576 * potential for a deadlock. KM_PUSHPAGE should be used instead.
578 ASSERT(!(current
->flags
& PF_NOFS
));
579 current
->flags
|= PF_NOFS
;
581 if (req
->cmd_flags
& VDEV_REQ_FLUSH
)
582 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
585 * Some requests are just for flush and nothing else.
588 blk_end_request(req
, 0, size
);
592 rl
= zfs_range_lock(&zv
->zv_znode
, offset
, size
, RL_WRITER
);
594 tx
= dmu_tx_create(zv
->zv_objset
);
595 dmu_tx_hold_write(tx
, ZVOL_OBJ
, offset
, size
);
597 /* This will only fail for ENOSPC */
598 error
= dmu_tx_assign(tx
, TXG_WAIT
);
601 zfs_range_unlock(rl
);
602 blk_end_request(req
, -error
, size
);
606 error
= dmu_write_req(zv
->zv_objset
, ZVOL_OBJ
, req
, tx
);
608 zvol_log_write(zv
, tx
, offset
, size
,
609 req
->cmd_flags
& VDEV_REQ_FUA
);
612 zfs_range_unlock(rl
);
614 if ((req
->cmd_flags
& VDEV_REQ_FUA
) ||
615 zv
->zv_objset
->os_sync
== ZFS_SYNC_ALWAYS
)
616 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
618 blk_end_request(req
, -error
, size
);
620 current
->flags
&= ~PF_NOFS
;
623 #ifdef HAVE_BLK_QUEUE_DISCARD
625 zvol_discard(void *arg
)
627 struct request
*req
= (struct request
*)arg
;
628 struct request_queue
*q
= req
->q
;
629 zvol_state_t
*zv
= q
->queuedata
;
630 uint64_t start
= blk_rq_pos(req
) << 9;
631 uint64_t end
= start
+ blk_rq_bytes(req
);
636 * Annotate this call path with a flag that indicates that it is
637 * unsafe to use KM_SLEEP during memory allocations due to the
638 * potential for a deadlock. KM_PUSHPAGE should be used instead.
640 ASSERT(!(current
->flags
& PF_NOFS
));
641 current
->flags
|= PF_NOFS
;
643 if (end
> zv
->zv_volsize
) {
644 blk_end_request(req
, -EIO
, blk_rq_bytes(req
));
649 * Align the request to volume block boundaries. If we don't,
650 * then this will force dnode_free_range() to zero out the
651 * unaligned parts, which is slow (read-modify-write) and
652 * useless since we are not freeing any space by doing so.
654 start
= P2ROUNDUP(start
, zv
->zv_volblocksize
);
655 end
= P2ALIGN(end
, zv
->zv_volblocksize
);
658 blk_end_request(req
, 0, blk_rq_bytes(req
));
662 rl
= zfs_range_lock(&zv
->zv_znode
, start
, end
- start
, RL_WRITER
);
664 error
= dmu_free_long_range(zv
->zv_objset
, ZVOL_OBJ
, start
, end
- start
);
667 * TODO: maybe we should add the operation to the log.
670 zfs_range_unlock(rl
);
672 blk_end_request(req
, -error
, blk_rq_bytes(req
));
674 current
->flags
&= ~PF_NOFS
;
676 #endif /* HAVE_BLK_QUEUE_DISCARD */
679 * Common read path running under the zvol taskq context. This function
680 * is responsible for copying the requested data out of the DMU and in to
681 * a linux request structure. It then must signal the request queue with
682 * an error code describing the result of the copy.
687 struct request
*req
= (struct request
*)arg
;
688 struct request_queue
*q
= req
->q
;
689 zvol_state_t
*zv
= q
->queuedata
;
690 uint64_t offset
= blk_rq_pos(req
) << 9;
691 uint64_t size
= blk_rq_bytes(req
);
696 blk_end_request(req
, 0, size
);
700 rl
= zfs_range_lock(&zv
->zv_znode
, offset
, size
, RL_READER
);
702 error
= dmu_read_req(zv
->zv_objset
, ZVOL_OBJ
, req
);
704 zfs_range_unlock(rl
);
706 /* convert checksum errors into IO errors */
708 error
= SET_ERROR(EIO
);
710 blk_end_request(req
, -error
, size
);
714 * Request will be added back to the request queue and retried if
715 * it cannot be immediately dispatched to the taskq for handling
718 zvol_dispatch(task_func_t func
, struct request
*req
)
720 if (!taskq_dispatch(zvol_taskq
, func
, (void *)req
, TQ_NOSLEEP
))
721 blk_requeue_request(req
->q
, req
);
725 * Common request path. Rather than registering a custom make_request()
726 * function we use the generic Linux version. This is done because it allows
727 * us to easily merge read requests which would otherwise we performed
728 * synchronously by the DMU. This is less critical in write case where the
729 * DMU will perform the correct merging within a transaction group. Using
730 * the generic make_request() also let's use leverage the fact that the
731 * elevator with ensure correct ordering in regards to barrior IOs. On
732 * the downside it means that in the write case we end up doing request
733 * merging twice once in the elevator and once in the DMU.
735 * The request handler is called under a spin lock so all the real work
736 * is handed off to be done in the context of the zvol taskq. This function
737 * simply performs basic request sanity checking and hands off the request.
740 zvol_request(struct request_queue
*q
)
742 zvol_state_t
*zv
= q
->queuedata
;
746 while ((req
= blk_fetch_request(q
)) != NULL
) {
747 size
= blk_rq_bytes(req
);
749 if (size
!= 0 && blk_rq_pos(req
) + blk_rq_sectors(req
) >
750 get_capacity(zv
->zv_disk
)) {
752 "%s: bad access: block=%llu, count=%lu\n",
753 req
->rq_disk
->disk_name
,
754 (long long unsigned)blk_rq_pos(req
),
755 (long unsigned)blk_rq_sectors(req
));
756 __blk_end_request(req
, -EIO
, size
);
760 if (!blk_fs_request(req
)) {
761 printk(KERN_INFO
"%s: non-fs cmd\n",
762 req
->rq_disk
->disk_name
);
763 __blk_end_request(req
, -EIO
, size
);
767 switch (rq_data_dir(req
)) {
769 zvol_dispatch(zvol_read
, req
);
772 if (unlikely(get_disk_ro(zv
->zv_disk
)) ||
773 unlikely(zv
->zv_flags
& ZVOL_RDONLY
)) {
774 __blk_end_request(req
, -EROFS
, size
);
778 #ifdef HAVE_BLK_QUEUE_DISCARD
779 if (req
->cmd_flags
& VDEV_REQ_DISCARD
) {
780 zvol_dispatch(zvol_discard
, req
);
783 #endif /* HAVE_BLK_QUEUE_DISCARD */
785 zvol_dispatch(zvol_write
, req
);
788 printk(KERN_INFO
"%s: unknown cmd: %d\n",
789 req
->rq_disk
->disk_name
, (int)rq_data_dir(req
));
790 __blk_end_request(req
, -EIO
, size
);
797 zvol_get_done(zgd_t
*zgd
, int error
)
800 dmu_buf_rele(zgd
->zgd_db
, zgd
);
802 zfs_range_unlock(zgd
->zgd_rl
);
804 if (error
== 0 && zgd
->zgd_bp
)
805 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
807 kmem_free(zgd
, sizeof (zgd_t
));
811 * Get data to generate a TX_WRITE intent log record.
814 zvol_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
816 zvol_state_t
*zv
= arg
;
817 objset_t
*os
= zv
->zv_objset
;
818 uint64_t offset
= lr
->lr_offset
;
819 uint64_t size
= lr
->lr_length
;
827 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_PUSHPAGE
);
828 zgd
->zgd_zilog
= zv
->zv_zilog
;
829 zgd
->zgd_rl
= zfs_range_lock(&zv
->zv_znode
, offset
, size
, RL_READER
);
832 * Write records come in two flavors: immediate and indirect.
833 * For small writes it's cheaper to store the data with the
834 * log record (immediate); for large writes it's cheaper to
835 * sync the data and get a pointer to it (indirect) so that
836 * we don't have to write the data twice.
838 if (buf
!= NULL
) { /* immediate write */
839 error
= dmu_read(os
, ZVOL_OBJ
, offset
, size
, buf
,
840 DMU_READ_NO_PREFETCH
);
842 size
= zv
->zv_volblocksize
;
843 offset
= P2ALIGN_TYPED(offset
, size
, uint64_t);
844 error
= dmu_buf_hold(os
, ZVOL_OBJ
, offset
, zgd
, &db
,
845 DMU_READ_NO_PREFETCH
);
848 zgd
->zgd_bp
= &lr
->lr_blkptr
;
851 ASSERT(db
->db_offset
== offset
);
852 ASSERT(db
->db_size
== size
);
854 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
862 zvol_get_done(zgd
, error
);
868 * The zvol_state_t's are inserted in increasing MINOR(dev_t) order.
871 zvol_insert(zvol_state_t
*zv_insert
)
873 zvol_state_t
*zv
= NULL
;
875 ASSERT(MUTEX_HELD(&zvol_state_lock
));
876 ASSERT3U(MINOR(zv_insert
->zv_dev
) & ZVOL_MINOR_MASK
, ==, 0);
877 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
;
878 zv
= list_next(&zvol_state_list
, zv
)) {
879 if (MINOR(zv
->zv_dev
) > MINOR(zv_insert
->zv_dev
))
883 list_insert_before(&zvol_state_list
, zv
, zv_insert
);
887 * Simply remove the zvol from to list of zvols.
890 zvol_remove(zvol_state_t
*zv_remove
)
892 ASSERT(MUTEX_HELD(&zvol_state_lock
));
893 list_remove(&zvol_state_list
, zv_remove
);
897 zvol_first_open(zvol_state_t
*zv
)
906 * In all other cases the spa_namespace_lock is taken before the
907 * bdev->bd_mutex lock. But in this case the Linux __blkdev_get()
908 * function calls fops->open() with the bdev->bd_mutex lock held.
910 * To avoid a potential lock inversion deadlock we preemptively
911 * try to take the spa_namespace_lock(). Normally it will not
912 * be contended and this is safe because spa_open_common() handles
913 * the case where the caller already holds the spa_namespace_lock.
915 * When it is contended we risk a lock inversion if we were to
916 * block waiting for the lock. Luckily, the __blkdev_get()
917 * function allows us to return -ERESTARTSYS which will result in
918 * bdev->bd_mutex being dropped, reacquired, and fops->open() being
919 * called again. This process can be repeated safely until both
920 * locks are acquired.
922 if (!mutex_owned(&spa_namespace_lock
)) {
923 locked
= mutex_tryenter(&spa_namespace_lock
);
925 return (-SET_ERROR(ERESTARTSYS
));
928 /* lie and say we're read-only */
929 error
= dmu_objset_own(zv
->zv_name
, DMU_OST_ZVOL
, 1, zvol_tag
, &os
);
933 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
935 dmu_objset_disown(os
, zvol_tag
);
940 error
= dmu_bonus_hold(os
, ZVOL_OBJ
, zvol_tag
, &zv
->zv_dbuf
);
942 dmu_objset_disown(os
, zvol_tag
);
946 set_capacity(zv
->zv_disk
, volsize
>> 9);
947 zv
->zv_volsize
= volsize
;
948 zv
->zv_zilog
= zil_open(os
, zvol_get_data
);
950 VERIFY(dsl_prop_get_integer(zv
->zv_name
, "readonly", &ro
, NULL
) == 0);
951 if (ro
|| dmu_objset_is_snapshot(os
) ||
952 !spa_writeable(dmu_objset_spa(os
))) {
953 set_disk_ro(zv
->zv_disk
, 1);
954 zv
->zv_flags
|= ZVOL_RDONLY
;
956 set_disk_ro(zv
->zv_disk
, 0);
957 zv
->zv_flags
&= ~ZVOL_RDONLY
;
962 mutex_exit(&spa_namespace_lock
);
968 zvol_last_close(zvol_state_t
*zv
)
970 zil_close(zv
->zv_zilog
);
973 dmu_buf_rele(zv
->zv_dbuf
, zvol_tag
);
979 if (dsl_dataset_is_dirty(dmu_objset_ds(zv
->zv_objset
)) &&
980 !(zv
->zv_flags
& ZVOL_RDONLY
))
981 txg_wait_synced(dmu_objset_pool(zv
->zv_objset
), 0);
982 (void) dmu_objset_evict_dbufs(zv
->zv_objset
);
984 dmu_objset_disown(zv
->zv_objset
, zvol_tag
);
985 zv
->zv_objset
= NULL
;
989 zvol_open(struct block_device
*bdev
, fmode_t flag
)
991 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
992 int error
= 0, drop_mutex
= 0;
995 * If the caller is already holding the mutex do not take it
996 * again, this will happen as part of zvol_create_minor().
997 * Once add_disk() is called the device is live and the kernel
998 * will attempt to open it to read the partition information.
1000 if (!mutex_owned(&zvol_state_lock
)) {
1001 mutex_enter(&zvol_state_lock
);
1005 ASSERT3P(zv
, !=, NULL
);
1007 if (zv
->zv_open_count
== 0) {
1008 error
= zvol_first_open(zv
);
1013 if ((flag
& FMODE_WRITE
) &&
1014 (get_disk_ro(zv
->zv_disk
) || (zv
->zv_flags
& ZVOL_RDONLY
))) {
1016 goto out_open_count
;
1019 zv
->zv_open_count
++;
1022 if (zv
->zv_open_count
== 0)
1023 zvol_last_close(zv
);
1027 mutex_exit(&zvol_state_lock
);
1029 check_disk_change(bdev
);
1034 #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1039 zvol_release(struct gendisk
*disk
, fmode_t mode
)
1041 zvol_state_t
*zv
= disk
->private_data
;
1044 if (!mutex_owned(&zvol_state_lock
)) {
1045 mutex_enter(&zvol_state_lock
);
1049 ASSERT3P(zv
, !=, NULL
);
1050 ASSERT3U(zv
->zv_open_count
, >, 0);
1051 zv
->zv_open_count
--;
1052 if (zv
->zv_open_count
== 0)
1053 zvol_last_close(zv
);
1056 mutex_exit(&zvol_state_lock
);
1058 #ifndef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
1064 zvol_ioctl(struct block_device
*bdev
, fmode_t mode
,
1065 unsigned int cmd
, unsigned long arg
)
1067 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
1071 return (-SET_ERROR(ENXIO
));
1075 zil_commit(zv
->zv_zilog
, ZVOL_OBJ
);
1078 error
= copy_to_user((void *)arg
, zv
->zv_name
, MAXNAMELEN
);
1090 #ifdef CONFIG_COMPAT
1092 zvol_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1093 unsigned cmd
, unsigned long arg
)
1095 return zvol_ioctl(bdev
, mode
, cmd
, arg
);
1098 #define zvol_compat_ioctl NULL
1101 static int zvol_media_changed(struct gendisk
*disk
)
1103 zvol_state_t
*zv
= disk
->private_data
;
1105 return zv
->zv_changed
;
1108 static int zvol_revalidate_disk(struct gendisk
*disk
)
1110 zvol_state_t
*zv
= disk
->private_data
;
1113 set_capacity(zv
->zv_disk
, zv
->zv_volsize
>> 9);
1119 * Provide a simple virtual geometry for legacy compatibility. For devices
1120 * smaller than 1 MiB a small head and sector count is used to allow very
1121 * tiny devices. For devices over 1 Mib a standard head and sector count
1122 * is used to keep the cylinders count reasonable.
1125 zvol_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1127 zvol_state_t
*zv
= bdev
->bd_disk
->private_data
;
1128 sector_t sectors
= get_capacity(zv
->zv_disk
);
1130 if (sectors
> 2048) {
1139 geo
->cylinders
= sectors
/ (geo
->heads
* geo
->sectors
);
1144 static struct kobject
*
1145 zvol_probe(dev_t dev
, int *part
, void *arg
)
1148 struct kobject
*kobj
;
1150 mutex_enter(&zvol_state_lock
);
1151 zv
= zvol_find_by_dev(dev
);
1152 kobj
= zv
? get_disk(zv
->zv_disk
) : NULL
;
1153 mutex_exit(&zvol_state_lock
);
1158 #ifdef HAVE_BDEV_BLOCK_DEVICE_OPERATIONS
1159 static struct block_device_operations zvol_ops
= {
1161 .release
= zvol_release
,
1162 .ioctl
= zvol_ioctl
,
1163 .compat_ioctl
= zvol_compat_ioctl
,
1164 .media_changed
= zvol_media_changed
,
1165 .revalidate_disk
= zvol_revalidate_disk
,
1166 .getgeo
= zvol_getgeo
,
1167 .owner
= THIS_MODULE
,
1170 #else /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */
1173 zvol_open_by_inode(struct inode
*inode
, struct file
*file
)
1175 return zvol_open(inode
->i_bdev
, file
->f_mode
);
1179 zvol_release_by_inode(struct inode
*inode
, struct file
*file
)
1181 return zvol_release(inode
->i_bdev
->bd_disk
, file
->f_mode
);
1185 zvol_ioctl_by_inode(struct inode
*inode
, struct file
*file
,
1186 unsigned int cmd
, unsigned long arg
)
1188 if (file
== NULL
|| inode
== NULL
)
1190 return zvol_ioctl(inode
->i_bdev
, file
->f_mode
, cmd
, arg
);
1193 # ifdef CONFIG_COMPAT
1195 zvol_compat_ioctl_by_inode(struct file
*file
,
1196 unsigned int cmd
, unsigned long arg
)
1200 return zvol_compat_ioctl(file
->f_dentry
->d_inode
->i_bdev
,
1201 file
->f_mode
, cmd
, arg
);
1204 # define zvol_compat_ioctl_by_inode NULL
1207 static struct block_device_operations zvol_ops
= {
1208 .open
= zvol_open_by_inode
,
1209 .release
= zvol_release_by_inode
,
1210 .ioctl
= zvol_ioctl_by_inode
,
1211 .compat_ioctl
= zvol_compat_ioctl_by_inode
,
1212 .media_changed
= zvol_media_changed
,
1213 .revalidate_disk
= zvol_revalidate_disk
,
1214 .getgeo
= zvol_getgeo
,
1215 .owner
= THIS_MODULE
,
1217 #endif /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */
1220 * Allocate memory for a new zvol_state_t and setup the required
1221 * request queue and generic disk structures for the block device.
1223 static zvol_state_t
*
1224 zvol_alloc(dev_t dev
, const char *name
)
1229 zv
= kmem_zalloc(sizeof (zvol_state_t
), KM_SLEEP
);
1231 spin_lock_init(&zv
->zv_lock
);
1232 list_link_init(&zv
->zv_next
);
1234 zv
->zv_queue
= blk_init_queue(zvol_request
, &zv
->zv_lock
);
1235 if (zv
->zv_queue
== NULL
)
1238 #ifdef HAVE_ELEVATOR_CHANGE
1239 error
= elevator_change(zv
->zv_queue
, "noop");
1240 #endif /* HAVE_ELEVATOR_CHANGE */
1242 printk("ZFS: Unable to set \"%s\" scheduler for zvol %s: %d\n",
1243 "noop", name
, error
);
1247 #ifdef HAVE_BLK_QUEUE_FLUSH
1248 blk_queue_flush(zv
->zv_queue
, VDEV_REQ_FLUSH
| VDEV_REQ_FUA
);
1250 blk_queue_ordered(zv
->zv_queue
, QUEUE_ORDERED_DRAIN
, NULL
);
1251 #endif /* HAVE_BLK_QUEUE_FLUSH */
1253 zv
->zv_disk
= alloc_disk(ZVOL_MINORS
);
1254 if (zv
->zv_disk
== NULL
)
1257 zv
->zv_queue
->queuedata
= zv
;
1259 zv
->zv_open_count
= 0;
1260 strlcpy(zv
->zv_name
, name
, MAXNAMELEN
);
1262 mutex_init(&zv
->zv_znode
.z_range_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1263 avl_create(&zv
->zv_znode
.z_range_avl
, zfs_range_compare
,
1264 sizeof (rl_t
), offsetof(rl_t
, r_node
));
1265 zv
->zv_znode
.z_is_zvol
= TRUE
;
1267 zv
->zv_disk
->major
= zvol_major
;
1268 zv
->zv_disk
->first_minor
= (dev
& MINORMASK
);
1269 zv
->zv_disk
->fops
= &zvol_ops
;
1270 zv
->zv_disk
->private_data
= zv
;
1271 zv
->zv_disk
->queue
= zv
->zv_queue
;
1272 snprintf(zv
->zv_disk
->disk_name
, DISK_NAME_LEN
, "%s%d",
1273 ZVOL_DEV_NAME
, (dev
& MINORMASK
));
1278 blk_cleanup_queue(zv
->zv_queue
);
1280 kmem_free(zv
, sizeof (zvol_state_t
));
1286 * Cleanup then free a zvol_state_t which was created by zvol_alloc().
1289 zvol_free(zvol_state_t
*zv
)
1291 avl_destroy(&zv
->zv_znode
.z_range_avl
);
1292 mutex_destroy(&zv
->zv_znode
.z_range_lock
);
1294 del_gendisk(zv
->zv_disk
);
1295 blk_cleanup_queue(zv
->zv_queue
);
1296 put_disk(zv
->zv_disk
);
1298 kmem_free(zv
, sizeof (zvol_state_t
));
1302 __zvol_snapdev_hidden(const char *name
)
1309 parent
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1310 (void) strlcpy(parent
, name
, MAXPATHLEN
);
1312 if ((atp
= strrchr(parent
, '@')) != NULL
) {
1314 error
= dsl_prop_get_integer(parent
, "snapdev", &snapdev
, NULL
);
1315 if ((error
== 0) && (snapdev
== ZFS_SNAPDEV_HIDDEN
))
1316 error
= SET_ERROR(ENODEV
);
1318 kmem_free(parent
, MAXPATHLEN
);
1323 __zvol_create_minor(const char *name
, boolean_t ignore_snapdev
)
1327 dmu_object_info_t
*doi
;
1332 ASSERT(MUTEX_HELD(&zvol_state_lock
));
1334 zv
= zvol_find_by_name(name
);
1336 error
= SET_ERROR(EEXIST
);
1340 if (ignore_snapdev
== B_FALSE
) {
1341 error
= __zvol_snapdev_hidden(name
);
1346 doi
= kmem_alloc(sizeof(dmu_object_info_t
), KM_SLEEP
);
1348 error
= dmu_objset_own(name
, DMU_OST_ZVOL
, B_TRUE
, zvol_tag
, &os
);
1352 error
= dmu_object_info(os
, ZVOL_OBJ
, doi
);
1354 goto out_dmu_objset_disown
;
1356 error
= zap_lookup(os
, ZVOL_ZAP_OBJ
, "size", 8, 1, &volsize
);
1358 goto out_dmu_objset_disown
;
1360 error
= zvol_find_minor(&minor
);
1362 goto out_dmu_objset_disown
;
1364 zv
= zvol_alloc(MKDEV(zvol_major
, minor
), name
);
1366 error
= SET_ERROR(EAGAIN
);
1367 goto out_dmu_objset_disown
;
1370 if (dmu_objset_is_snapshot(os
))
1371 zv
->zv_flags
|= ZVOL_RDONLY
;
1373 zv
->zv_volblocksize
= doi
->doi_data_block_size
;
1374 zv
->zv_volsize
= volsize
;
1377 set_capacity(zv
->zv_disk
, zv
->zv_volsize
>> 9);
1379 blk_queue_max_hw_sectors(zv
->zv_queue
, UINT_MAX
);
1380 blk_queue_max_segments(zv
->zv_queue
, UINT16_MAX
);
1381 blk_queue_max_segment_size(zv
->zv_queue
, UINT_MAX
);
1382 blk_queue_physical_block_size(zv
->zv_queue
, zv
->zv_volblocksize
);
1383 blk_queue_io_opt(zv
->zv_queue
, zv
->zv_volblocksize
);
1384 #ifdef HAVE_BLK_QUEUE_DISCARD
1385 blk_queue_max_discard_sectors(zv
->zv_queue
,
1386 (zvol_max_discard_blocks
* zv
->zv_volblocksize
) >> 9);
1387 blk_queue_discard_granularity(zv
->zv_queue
, zv
->zv_volblocksize
);
1388 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, zv
->zv_queue
);
1390 #ifdef HAVE_BLK_QUEUE_NONROT
1391 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, zv
->zv_queue
);
1394 if (spa_writeable(dmu_objset_spa(os
))) {
1395 if (zil_replay_disable
)
1396 zil_destroy(dmu_objset_zil(os
), B_FALSE
);
1398 zil_replay(os
, zv
, zvol_replay_vector
);
1401 zv
->zv_objset
= NULL
;
1402 out_dmu_objset_disown
:
1403 dmu_objset_disown(os
, zvol_tag
);
1405 kmem_free(doi
, sizeof(dmu_object_info_t
));
1410 add_disk(zv
->zv_disk
);
1417 * Create a block device minor node and setup the linkage between it
1418 * and the specified volume. Once this function returns the block
1419 * device is live and ready for use.
1422 zvol_create_minor(const char *name
)
1426 mutex_enter(&zvol_state_lock
);
1427 error
= __zvol_create_minor(name
, B_FALSE
);
1428 mutex_exit(&zvol_state_lock
);
1434 __zvol_remove_minor(const char *name
)
1438 ASSERT(MUTEX_HELD(&zvol_state_lock
));
1440 zv
= zvol_find_by_name(name
);
1442 return (SET_ERROR(ENXIO
));
1444 if (zv
->zv_open_count
> 0)
1445 return (SET_ERROR(EBUSY
));
1454 * Remove a block device minor node for the specified volume.
1457 zvol_remove_minor(const char *name
)
1461 mutex_enter(&zvol_state_lock
);
1462 error
= __zvol_remove_minor(name
);
1463 mutex_exit(&zvol_state_lock
);
1469 zvol_create_minors_cb(const char *dsname
, void *arg
)
1471 if (strchr(dsname
, '/') == NULL
)
1474 (void) __zvol_create_minor(dsname
, B_FALSE
);
1479 * Create minors for specified pool, if pool is NULL create minors
1480 * for all available pools.
1483 zvol_create_minors(char *pool
)
1488 if (zvol_inhibit_dev
)
1491 mutex_enter(&zvol_state_lock
);
1493 error
= dmu_objset_find(pool
, zvol_create_minors_cb
,
1494 NULL
, DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
1496 mutex_enter(&spa_namespace_lock
);
1497 while ((spa
= spa_next(spa
)) != NULL
) {
1498 error
= dmu_objset_find(spa_name(spa
), zvol_create_minors_cb
, NULL
,
1499 DS_FIND_CHILDREN
| DS_FIND_SNAPSHOTS
);
1503 mutex_exit(&spa_namespace_lock
);
1505 mutex_exit(&zvol_state_lock
);
1511 * Remove minors for specified pool, if pool is NULL remove all minors.
1514 zvol_remove_minors(const char *pool
)
1516 zvol_state_t
*zv
, *zv_next
;
1519 if (zvol_inhibit_dev
)
1522 str
= kmem_zalloc(MAXNAMELEN
, KM_SLEEP
);
1524 (void) strncpy(str
, pool
, strlen(pool
));
1525 (void) strcat(str
, "/");
1528 mutex_enter(&zvol_state_lock
);
1529 for (zv
= list_head(&zvol_state_list
); zv
!= NULL
; zv
= zv_next
) {
1530 zv_next
= list_next(&zvol_state_list
, zv
);
1532 if (pool
== NULL
|| !strncmp(str
, zv
->zv_name
, strlen(str
))) {
1537 mutex_exit(&zvol_state_lock
);
1538 kmem_free(str
, MAXNAMELEN
);
1542 snapdev_snapshot_changed_cb(const char *dsname
, void *arg
) {
1543 uint64_t snapdev
= *(uint64_t *) arg
;
1545 if (strchr(dsname
, '@') == NULL
)
1549 case ZFS_SNAPDEV_VISIBLE
:
1550 mutex_enter(&zvol_state_lock
);
1551 (void) __zvol_create_minor(dsname
, B_TRUE
);
1552 mutex_exit(&zvol_state_lock
);
1554 case ZFS_SNAPDEV_HIDDEN
:
1555 (void) zvol_remove_minor(dsname
);
1562 zvol_set_snapdev(const char *dsname
, uint64_t snapdev
) {
1563 (void) dmu_objset_find((char *) dsname
, snapdev_snapshot_changed_cb
,
1564 &snapdev
, DS_FIND_SNAPSHOTS
| DS_FIND_CHILDREN
);
1565 /* caller should continue to modify snapdev property */
1575 list_create(&zvol_state_list
, sizeof (zvol_state_t
),
1576 offsetof(zvol_state_t
, zv_next
));
1577 mutex_init(&zvol_state_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1579 zvol_taskq
= taskq_create(ZVOL_DRIVER
, zvol_threads
, maxclsyspri
,
1580 zvol_threads
, INT_MAX
, TASKQ_PREPOPULATE
);
1581 if (zvol_taskq
== NULL
) {
1582 printk(KERN_INFO
"ZFS: taskq_create() failed\n");
1587 error
= register_blkdev(zvol_major
, ZVOL_DRIVER
);
1589 printk(KERN_INFO
"ZFS: register_blkdev() failed %d\n", error
);
1593 blk_register_region(MKDEV(zvol_major
, 0), 1UL << MINORBITS
,
1594 THIS_MODULE
, zvol_probe
, NULL
, NULL
);
1599 taskq_destroy(zvol_taskq
);
1601 mutex_destroy(&zvol_state_lock
);
1602 list_destroy(&zvol_state_list
);
1610 zvol_remove_minors(NULL
);
1611 blk_unregister_region(MKDEV(zvol_major
, 0), 1UL << MINORBITS
);
1612 unregister_blkdev(zvol_major
, ZVOL_DRIVER
);
1613 taskq_destroy(zvol_taskq
);
1614 mutex_destroy(&zvol_state_lock
);
1615 list_destroy(&zvol_state_list
);
1618 module_param(zvol_inhibit_dev
, uint
, 0644);
1619 MODULE_PARM_DESC(zvol_inhibit_dev
, "Do not create zvol device nodes");
1621 module_param(zvol_major
, uint
, 0444);
1622 MODULE_PARM_DESC(zvol_major
, "Major number for zvol device");
1624 module_param(zvol_threads
, uint
, 0444);
1625 MODULE_PARM_DESC(zvol_threads
, "Number of threads for zvol device");
1627 module_param(zvol_max_discard_blocks
, ulong
, 0444);
1628 MODULE_PARM_DESC(zvol_max_discard_blocks
, "Max number of blocks to discard at once");