2 #include <linux/spinlock.h>
3 #include <linux/slab.h>
4 #include <linux/blkdev.h>
5 #include <linux/hdreg.h>
6 #include <linux/module.h>
7 #include <linux/mutex.h>
8 #include <linux/virtio.h>
9 #include <linux/virtio_blk.h>
10 #include <linux/scatterlist.h>
11 #include <linux/string_helpers.h>
12 #include <scsi/scsi_cmnd.h>
13 #include <linux/idr.h>
18 static DEFINE_IDA(vd_index_ida
);
20 struct workqueue_struct
*virtblk_wq
;
26 struct virtio_device
*vdev
;
29 /* The disk structure for the kernel. */
32 /* Request tracking. */
33 struct list_head reqs
;
37 /* Process context for config space updates */
38 struct work_struct config_work
;
40 /* Lock for config space updates */
41 struct mutex config_lock
;
43 /* enable config space updates */
46 /* What host tells us, plus 2 for header & tailer. */
47 unsigned int sg_elems
;
49 /* Ida index - used to track minor number allocations. */
52 /* Scatterlist: can be too big for stack. */
53 struct scatterlist sg
[/*sg_elems*/];
58 struct list_head list
;
60 struct virtio_blk_outhdr out_hdr
;
61 struct virtio_scsi_inhdr in_hdr
;
65 static void blk_done(struct virtqueue
*vq
)
67 struct virtio_blk
*vblk
= vq
->vdev
->priv
;
68 struct virtblk_req
*vbr
;
72 spin_lock_irqsave(&vblk
->lock
, flags
);
73 while ((vbr
= virtqueue_get_buf(vblk
->vq
, &len
)) != NULL
) {
76 switch (vbr
->status
) {
80 case VIRTIO_BLK_S_UNSUPP
:
88 switch (vbr
->req
->cmd_type
) {
89 case REQ_TYPE_BLOCK_PC
:
90 vbr
->req
->resid_len
= vbr
->in_hdr
.residual
;
91 vbr
->req
->sense_len
= vbr
->in_hdr
.sense_len
;
92 vbr
->req
->errors
= vbr
->in_hdr
.errors
;
94 case REQ_TYPE_SPECIAL
:
95 vbr
->req
->errors
= (error
!= 0);
101 __blk_end_request_all(vbr
->req
, error
);
102 list_del(&vbr
->list
);
103 mempool_free(vbr
, vblk
->pool
);
105 /* In case queue is stopped waiting for more buffers. */
106 blk_start_queue(vblk
->disk
->queue
);
107 spin_unlock_irqrestore(&vblk
->lock
, flags
);
110 static bool do_req(struct request_queue
*q
, struct virtio_blk
*vblk
,
113 unsigned long num
, out
= 0, in
= 0;
114 struct virtblk_req
*vbr
;
116 vbr
= mempool_alloc(vblk
->pool
, GFP_ATOMIC
);
118 /* When another request finishes we'll try again. */
123 if (req
->cmd_flags
& REQ_FLUSH
) {
124 vbr
->out_hdr
.type
= VIRTIO_BLK_T_FLUSH
;
125 vbr
->out_hdr
.sector
= 0;
126 vbr
->out_hdr
.ioprio
= req_get_ioprio(vbr
->req
);
128 switch (req
->cmd_type
) {
130 vbr
->out_hdr
.type
= 0;
131 vbr
->out_hdr
.sector
= blk_rq_pos(vbr
->req
);
132 vbr
->out_hdr
.ioprio
= req_get_ioprio(vbr
->req
);
134 case REQ_TYPE_BLOCK_PC
:
135 vbr
->out_hdr
.type
= VIRTIO_BLK_T_SCSI_CMD
;
136 vbr
->out_hdr
.sector
= 0;
137 vbr
->out_hdr
.ioprio
= req_get_ioprio(vbr
->req
);
139 case REQ_TYPE_SPECIAL
:
140 vbr
->out_hdr
.type
= VIRTIO_BLK_T_GET_ID
;
141 vbr
->out_hdr
.sector
= 0;
142 vbr
->out_hdr
.ioprio
= req_get_ioprio(vbr
->req
);
145 /* We don't put anything else in the queue. */
150 sg_set_buf(&vblk
->sg
[out
++], &vbr
->out_hdr
, sizeof(vbr
->out_hdr
));
153 * If this is a packet command we need a couple of additional headers.
154 * Behind the normal outhdr we put a segment with the scsi command
155 * block, and before the normal inhdr we put the sense data and the
156 * inhdr with additional status information before the normal inhdr.
158 if (vbr
->req
->cmd_type
== REQ_TYPE_BLOCK_PC
)
159 sg_set_buf(&vblk
->sg
[out
++], vbr
->req
->cmd
, vbr
->req
->cmd_len
);
161 num
= blk_rq_map_sg(q
, vbr
->req
, vblk
->sg
+ out
);
163 if (vbr
->req
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
164 sg_set_buf(&vblk
->sg
[num
+ out
+ in
++], vbr
->req
->sense
, SCSI_SENSE_BUFFERSIZE
);
165 sg_set_buf(&vblk
->sg
[num
+ out
+ in
++], &vbr
->in_hdr
,
166 sizeof(vbr
->in_hdr
));
169 sg_set_buf(&vblk
->sg
[num
+ out
+ in
++], &vbr
->status
,
170 sizeof(vbr
->status
));
173 if (rq_data_dir(vbr
->req
) == WRITE
) {
174 vbr
->out_hdr
.type
|= VIRTIO_BLK_T_OUT
;
177 vbr
->out_hdr
.type
|= VIRTIO_BLK_T_IN
;
182 if (virtqueue_add_buf(vblk
->vq
, vblk
->sg
, out
, in
, vbr
, GFP_ATOMIC
)<0) {
183 mempool_free(vbr
, vblk
->pool
);
187 list_add_tail(&vbr
->list
, &vblk
->reqs
);
191 static void do_virtblk_request(struct request_queue
*q
)
193 struct virtio_blk
*vblk
= q
->queuedata
;
195 unsigned int issued
= 0;
197 while ((req
= blk_peek_request(q
)) != NULL
) {
198 BUG_ON(req
->nr_phys_segments
+ 2 > vblk
->sg_elems
);
200 /* If this request fails, stop queue and wait for something to
201 finish to restart it. */
202 if (!do_req(q
, vblk
, req
)) {
206 blk_start_request(req
);
211 virtqueue_kick(vblk
->vq
);
214 /* return id (s/n) string for *disk to *id_str
216 static int virtblk_get_id(struct gendisk
*disk
, char *id_str
)
218 struct virtio_blk
*vblk
= disk
->private_data
;
223 bio
= bio_map_kern(vblk
->disk
->queue
, id_str
, VIRTIO_BLK_ID_BYTES
,
228 req
= blk_make_request(vblk
->disk
->queue
, bio
, GFP_KERNEL
);
234 req
->cmd_type
= REQ_TYPE_SPECIAL
;
235 err
= blk_execute_rq(vblk
->disk
->queue
, vblk
->disk
, req
, false);
236 blk_put_request(req
);
241 static int virtblk_ioctl(struct block_device
*bdev
, fmode_t mode
,
242 unsigned int cmd
, unsigned long data
)
244 struct gendisk
*disk
= bdev
->bd_disk
;
245 struct virtio_blk
*vblk
= disk
->private_data
;
248 * Only allow the generic SCSI ioctls if the host can support it.
250 if (!virtio_has_feature(vblk
->vdev
, VIRTIO_BLK_F_SCSI
))
253 return scsi_cmd_blk_ioctl(bdev
, mode
, cmd
,
254 (void __user
*)data
);
257 /* We provide getgeo only to please some old bootloader/partitioning tools */
258 static int virtblk_getgeo(struct block_device
*bd
, struct hd_geometry
*geo
)
260 struct virtio_blk
*vblk
= bd
->bd_disk
->private_data
;
261 struct virtio_blk_geometry vgeo
;
264 /* see if the host passed in geometry config */
265 err
= virtio_config_val(vblk
->vdev
, VIRTIO_BLK_F_GEOMETRY
,
266 offsetof(struct virtio_blk_config
, geometry
),
270 geo
->heads
= vgeo
.heads
;
271 geo
->sectors
= vgeo
.sectors
;
272 geo
->cylinders
= vgeo
.cylinders
;
274 /* some standard values, similar to sd */
276 geo
->sectors
= 1 << 5;
277 geo
->cylinders
= get_capacity(bd
->bd_disk
) >> 11;
282 static const struct block_device_operations virtblk_fops
= {
283 .ioctl
= virtblk_ioctl
,
284 .owner
= THIS_MODULE
,
285 .getgeo
= virtblk_getgeo
,
288 static int index_to_minor(int index
)
290 return index
<< PART_BITS
;
293 static int minor_to_index(int minor
)
295 return minor
>> PART_BITS
;
298 static ssize_t
virtblk_serial_show(struct device
*dev
,
299 struct device_attribute
*attr
, char *buf
)
301 struct gendisk
*disk
= dev_to_disk(dev
);
304 /* sysfs gives us a PAGE_SIZE buffer */
305 BUILD_BUG_ON(PAGE_SIZE
< VIRTIO_BLK_ID_BYTES
);
307 buf
[VIRTIO_BLK_ID_BYTES
] = '\0';
308 err
= virtblk_get_id(disk
, buf
);
312 if (err
== -EIO
) /* Unsupported? Make it empty. */
317 DEVICE_ATTR(serial
, S_IRUGO
, virtblk_serial_show
, NULL
);
319 static void virtblk_config_changed_work(struct work_struct
*work
)
321 struct virtio_blk
*vblk
=
322 container_of(work
, struct virtio_blk
, config_work
);
323 struct virtio_device
*vdev
= vblk
->vdev
;
324 struct request_queue
*q
= vblk
->disk
->queue
;
325 char cap_str_2
[10], cap_str_10
[10];
328 mutex_lock(&vblk
->config_lock
);
329 if (!vblk
->config_enable
)
332 /* Host must always specify the capacity. */
333 vdev
->config
->get(vdev
, offsetof(struct virtio_blk_config
, capacity
),
334 &capacity
, sizeof(capacity
));
336 /* If capacity is too big, truncate with warning. */
337 if ((sector_t
)capacity
!= capacity
) {
338 dev_warn(&vdev
->dev
, "Capacity %llu too large: truncating\n",
339 (unsigned long long)capacity
);
340 capacity
= (sector_t
)-1;
343 size
= capacity
* queue_logical_block_size(q
);
344 string_get_size(size
, STRING_UNITS_2
, cap_str_2
, sizeof(cap_str_2
));
345 string_get_size(size
, STRING_UNITS_10
, cap_str_10
, sizeof(cap_str_10
));
347 dev_notice(&vdev
->dev
,
348 "new size: %llu %d-byte logical blocks (%s/%s)\n",
349 (unsigned long long)capacity
,
350 queue_logical_block_size(q
),
351 cap_str_10
, cap_str_2
);
353 set_capacity(vblk
->disk
, capacity
);
354 revalidate_disk(vblk
->disk
);
356 mutex_unlock(&vblk
->config_lock
);
359 static void virtblk_config_changed(struct virtio_device
*vdev
)
361 struct virtio_blk
*vblk
= vdev
->priv
;
363 queue_work(virtblk_wq
, &vblk
->config_work
);
366 static int init_vq(struct virtio_blk
*vblk
)
370 /* We expect one virtqueue, for output. */
371 vblk
->vq
= virtio_find_single_vq(vblk
->vdev
, blk_done
, "requests");
372 if (IS_ERR(vblk
->vq
))
373 err
= PTR_ERR(vblk
->vq
);
378 static int __devinit
virtblk_probe(struct virtio_device
*vdev
)
380 struct virtio_blk
*vblk
;
381 struct request_queue
*q
;
384 u32 v
, blk_size
, sg_elems
, opt_io_size
;
386 u8 physical_block_exp
, alignment_offset
;
388 err
= ida_simple_get(&vd_index_ida
, 0, minor_to_index(1 << MINORBITS
),
394 /* We need to know how many segments before we allocate. */
395 err
= virtio_config_val(vdev
, VIRTIO_BLK_F_SEG_MAX
,
396 offsetof(struct virtio_blk_config
, seg_max
),
399 /* We need at least one SG element, whatever they say. */
400 if (err
|| !sg_elems
)
403 /* We need an extra sg elements at head and tail. */
405 vdev
->priv
= vblk
= kmalloc(sizeof(*vblk
) +
406 sizeof(vblk
->sg
[0]) * sg_elems
, GFP_KERNEL
);
412 INIT_LIST_HEAD(&vblk
->reqs
);
413 spin_lock_init(&vblk
->lock
);
415 vblk
->sg_elems
= sg_elems
;
416 sg_init_table(vblk
->sg
, vblk
->sg_elems
);
417 mutex_init(&vblk
->config_lock
);
418 INIT_WORK(&vblk
->config_work
, virtblk_config_changed_work
);
419 vblk
->config_enable
= true;
425 vblk
->pool
= mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req
));
431 /* FIXME: How many partitions? How long is a piece of string? */
432 vblk
->disk
= alloc_disk(1 << PART_BITS
);
438 q
= vblk
->disk
->queue
= blk_init_queue(do_virtblk_request
, &vblk
->lock
);
447 sprintf(vblk
->disk
->disk_name
, "vd%c", 'a' + index
% 26);
448 } else if (index
< (26 + 1) * 26) {
449 sprintf(vblk
->disk
->disk_name
, "vd%c%c",
450 'a' + index
/ 26 - 1, 'a' + index
% 26);
452 const unsigned int m1
= (index
/ 26 - 1) / 26 - 1;
453 const unsigned int m2
= (index
/ 26 - 1) % 26;
454 const unsigned int m3
= index
% 26;
455 sprintf(vblk
->disk
->disk_name
, "vd%c%c%c",
456 'a' + m1
, 'a' + m2
, 'a' + m3
);
459 vblk
->disk
->major
= major
;
460 vblk
->disk
->first_minor
= index_to_minor(index
);
461 vblk
->disk
->private_data
= vblk
;
462 vblk
->disk
->fops
= &virtblk_fops
;
463 vblk
->disk
->driverfs_dev
= &vdev
->dev
;
466 /* configure queue flush support */
467 if (virtio_has_feature(vdev
, VIRTIO_BLK_F_FLUSH
))
468 blk_queue_flush(q
, REQ_FLUSH
);
470 /* If disk is read-only in the host, the guest should obey */
471 if (virtio_has_feature(vdev
, VIRTIO_BLK_F_RO
))
472 set_disk_ro(vblk
->disk
, 1);
474 /* Host must always specify the capacity. */
475 vdev
->config
->get(vdev
, offsetof(struct virtio_blk_config
, capacity
),
478 /* If capacity is too big, truncate with warning. */
479 if ((sector_t
)cap
!= cap
) {
480 dev_warn(&vdev
->dev
, "Capacity %llu too large: truncating\n",
481 (unsigned long long)cap
);
484 set_capacity(vblk
->disk
, cap
);
486 /* We can handle whatever the host told us to handle. */
487 blk_queue_max_segments(q
, vblk
->sg_elems
-2);
489 /* No need to bounce any requests */
490 blk_queue_bounce_limit(q
, BLK_BOUNCE_ANY
);
492 /* No real sector limit. */
493 blk_queue_max_hw_sectors(q
, -1U);
495 /* Host can optionally specify maximum segment size and number of
497 err
= virtio_config_val(vdev
, VIRTIO_BLK_F_SIZE_MAX
,
498 offsetof(struct virtio_blk_config
, size_max
),
501 blk_queue_max_segment_size(q
, v
);
503 blk_queue_max_segment_size(q
, -1U);
505 /* Host can optionally specify the block size of the device */
506 err
= virtio_config_val(vdev
, VIRTIO_BLK_F_BLK_SIZE
,
507 offsetof(struct virtio_blk_config
, blk_size
),
510 blk_queue_logical_block_size(q
, blk_size
);
512 blk_size
= queue_logical_block_size(q
);
514 /* Use topology information if available */
515 err
= virtio_config_val(vdev
, VIRTIO_BLK_F_TOPOLOGY
,
516 offsetof(struct virtio_blk_config
, physical_block_exp
),
517 &physical_block_exp
);
518 if (!err
&& physical_block_exp
)
519 blk_queue_physical_block_size(q
,
520 blk_size
* (1 << physical_block_exp
));
522 err
= virtio_config_val(vdev
, VIRTIO_BLK_F_TOPOLOGY
,
523 offsetof(struct virtio_blk_config
, alignment_offset
),
525 if (!err
&& alignment_offset
)
526 blk_queue_alignment_offset(q
, blk_size
* alignment_offset
);
528 err
= virtio_config_val(vdev
, VIRTIO_BLK_F_TOPOLOGY
,
529 offsetof(struct virtio_blk_config
, min_io_size
),
531 if (!err
&& min_io_size
)
532 blk_queue_io_min(q
, blk_size
* min_io_size
);
534 err
= virtio_config_val(vdev
, VIRTIO_BLK_F_TOPOLOGY
,
535 offsetof(struct virtio_blk_config
, opt_io_size
),
537 if (!err
&& opt_io_size
)
538 blk_queue_io_opt(q
, blk_size
* opt_io_size
);
541 add_disk(vblk
->disk
);
542 err
= device_create_file(disk_to_dev(vblk
->disk
), &dev_attr_serial
);
549 del_gendisk(vblk
->disk
);
550 blk_cleanup_queue(vblk
->disk
->queue
);
552 put_disk(vblk
->disk
);
554 mempool_destroy(vblk
->pool
);
556 vdev
->config
->del_vqs(vdev
);
560 ida_simple_remove(&vd_index_ida
, index
);
565 static void __devexit
virtblk_remove(struct virtio_device
*vdev
)
567 struct virtio_blk
*vblk
= vdev
->priv
;
568 int index
= vblk
->index
;
570 /* Prevent config work handler from accessing the device. */
571 mutex_lock(&vblk
->config_lock
);
572 vblk
->config_enable
= false;
573 mutex_unlock(&vblk
->config_lock
);
575 /* Nothing should be pending. */
576 BUG_ON(!list_empty(&vblk
->reqs
));
578 /* Stop all the virtqueues. */
579 vdev
->config
->reset(vdev
);
581 flush_work(&vblk
->config_work
);
583 del_gendisk(vblk
->disk
);
584 blk_cleanup_queue(vblk
->disk
->queue
);
585 put_disk(vblk
->disk
);
586 mempool_destroy(vblk
->pool
);
587 vdev
->config
->del_vqs(vdev
);
589 ida_simple_remove(&vd_index_ida
, index
);
593 static int virtblk_freeze(struct virtio_device
*vdev
)
595 struct virtio_blk
*vblk
= vdev
->priv
;
597 /* Ensure we don't receive any more interrupts */
598 vdev
->config
->reset(vdev
);
600 /* Prevent config work handler from accessing the device. */
601 mutex_lock(&vblk
->config_lock
);
602 vblk
->config_enable
= false;
603 mutex_unlock(&vblk
->config_lock
);
605 flush_work(&vblk
->config_work
);
607 spin_lock_irq(vblk
->disk
->queue
->queue_lock
);
608 blk_stop_queue(vblk
->disk
->queue
);
609 spin_unlock_irq(vblk
->disk
->queue
->queue_lock
);
610 blk_sync_queue(vblk
->disk
->queue
);
612 vdev
->config
->del_vqs(vdev
);
616 static int virtblk_restore(struct virtio_device
*vdev
)
618 struct virtio_blk
*vblk
= vdev
->priv
;
621 vblk
->config_enable
= true;
622 ret
= init_vq(vdev
->priv
);
624 spin_lock_irq(vblk
->disk
->queue
->queue_lock
);
625 blk_start_queue(vblk
->disk
->queue
);
626 spin_unlock_irq(vblk
->disk
->queue
->queue_lock
);
632 static const struct virtio_device_id id_table
[] = {
633 { VIRTIO_ID_BLOCK
, VIRTIO_DEV_ANY_ID
},
637 static unsigned int features
[] = {
638 VIRTIO_BLK_F_SEG_MAX
, VIRTIO_BLK_F_SIZE_MAX
, VIRTIO_BLK_F_GEOMETRY
,
639 VIRTIO_BLK_F_RO
, VIRTIO_BLK_F_BLK_SIZE
, VIRTIO_BLK_F_SCSI
,
640 VIRTIO_BLK_F_FLUSH
, VIRTIO_BLK_F_TOPOLOGY
644 * virtio_blk causes spurious section mismatch warning by
645 * simultaneously referring to a __devinit and a __devexit function.
646 * Use __refdata to avoid this warning.
648 static struct virtio_driver __refdata virtio_blk
= {
649 .feature_table
= features
,
650 .feature_table_size
= ARRAY_SIZE(features
),
651 .driver
.name
= KBUILD_MODNAME
,
652 .driver
.owner
= THIS_MODULE
,
653 .id_table
= id_table
,
654 .probe
= virtblk_probe
,
655 .remove
= __devexit_p(virtblk_remove
),
656 .config_changed
= virtblk_config_changed
,
658 .freeze
= virtblk_freeze
,
659 .restore
= virtblk_restore
,
663 static int __init
init(void)
667 virtblk_wq
= alloc_workqueue("virtio-blk", 0, 0);
671 major
= register_blkdev(0, "virtblk");
674 goto out_destroy_workqueue
;
677 error
= register_virtio_driver(&virtio_blk
);
679 goto out_unregister_blkdev
;
682 out_unregister_blkdev
:
683 unregister_blkdev(major
, "virtblk");
684 out_destroy_workqueue
:
685 destroy_workqueue(virtblk_wq
);
689 static void __exit
fini(void)
691 unregister_blkdev(major
, "virtblk");
692 unregister_virtio_driver(&virtio_blk
);
693 destroy_workqueue(virtblk_wq
);
698 MODULE_DEVICE_TABLE(virtio
, id_table
);
699 MODULE_DESCRIPTION("Virtio block driver");
700 MODULE_LICENSE("GPL");