2 #include <linux/spinlock.h>
3 #include <linux/slab.h>
4 #include <linux/blkdev.h>
5 #include <linux/hdreg.h>
6 #include <linux/virtio.h>
7 #include <linux/virtio_blk.h>
8 #include <linux/scatterlist.h>
12 static int major, index;
18 struct virtio_device *vdev;
21 /* The disk structure for the kernel. */
24 /* Request tracking. */
25 struct list_head reqs;
29 /* What host tells us, plus 2 for header & tailer. */
30 unsigned int sg_elems;
32 /* Scatterlist: can be too big for stack. */
33 struct scatterlist sg[/*sg_elems*/];
38 struct list_head list;
40 struct virtio_blk_outhdr out_hdr;
41 struct virtio_scsi_inhdr in_hdr;
45 static void blk_done(struct virtqueue *vq)
47 struct virtio_blk *vblk = vq->vdev->priv;
48 struct virtblk_req *vbr;
52 spin_lock_irqsave(&vblk->lock, flags);
53 while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
56 switch (vbr->status) {
60 case VIRTIO_BLK_S_UNSUPP:
68 switch (vbr->req->cmd_type) {
69 case REQ_TYPE_BLOCK_PC:
70 vbr->req->resid_len = vbr->in_hdr.residual;
71 vbr->req->sense_len = vbr->in_hdr.sense_len;
72 vbr->req->errors = vbr->in_hdr.errors;
74 case REQ_TYPE_SPECIAL:
75 vbr->req->errors = (error != 0);
81 __blk_end_request_all(vbr->req, error);
83 mempool_free(vbr, vblk->pool);
85 /* In case queue is stopped waiting for more buffers. */
86 blk_start_queue(vblk->disk->queue);
87 spin_unlock_irqrestore(&vblk->lock, flags);
90 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
93 unsigned long num, out = 0, in = 0;
94 struct virtblk_req *vbr;
96 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
98 /* When another request finishes we'll try again. */
103 if (req->cmd_flags & REQ_FLUSH) {
104 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
105 vbr->out_hdr.sector = 0;
106 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
108 switch (req->cmd_type) {
110 vbr->out_hdr.type = 0;
111 vbr->out_hdr.sector = blk_rq_pos(vbr->req);
112 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
114 case REQ_TYPE_BLOCK_PC:
115 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
116 vbr->out_hdr.sector = 0;
117 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
119 case REQ_TYPE_SPECIAL:
120 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
121 vbr->out_hdr.sector = 0;
122 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
125 /* We don't put anything else in the queue. */
130 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
133 * If this is a packet command we need a couple of additional headers.
134 * Behind the normal outhdr we put a segment with the scsi command
135 * block, and before the normal inhdr we put the sense data and the
136 * inhdr with additional status information before the normal inhdr.
138 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
139 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
141 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
143 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
144 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
145 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
146 sizeof(vbr->in_hdr));
149 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
150 sizeof(vbr->status));
153 if (rq_data_dir(vbr->req) == WRITE) {
154 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
157 vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
162 if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
163 mempool_free(vbr, vblk->pool);
167 list_add_tail(&vbr->list, &vblk->reqs);
171 static void do_virtblk_request(struct request_queue *q)
173 struct virtio_blk *vblk = q->queuedata;
175 unsigned int issued = 0;
177 while ((req = blk_peek_request(q)) != NULL) {
178 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
180 /* If this request fails, stop queue and wait for something to
181 finish to restart it. */
182 if (!do_req(q, vblk, req)) {
186 blk_start_request(req);
191 virtqueue_kick(vblk->vq);
194 /* return id (s/n) string for *disk to *id_str
196 static int virtblk_get_id(struct gendisk *disk, char *id_str)
198 struct virtio_blk *vblk = disk->private_data;
203 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
208 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
214 req->cmd_type = REQ_TYPE_SPECIAL;
215 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
216 blk_put_request(req);
221 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
222 unsigned int cmd, unsigned long data)
224 struct gendisk *disk = bdev->bd_disk;
225 struct virtio_blk *vblk = disk->private_data;
228 * Only allow the generic SCSI ioctls if the host can support it.
230 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
233 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
234 (void __user *)data);
237 /* We provide getgeo only to please some old bootloader/partitioning tools */
238 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
240 struct virtio_blk *vblk = bd->bd_disk->private_data;
241 struct virtio_blk_geometry vgeo;
244 /* see if the host passed in geometry config */
245 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
246 offsetof(struct virtio_blk_config, geometry),
250 geo->heads = vgeo.heads;
251 geo->sectors = vgeo.sectors;
252 geo->cylinders = vgeo.cylinders;
254 /* some standard values, similar to sd */
256 geo->sectors = 1 << 5;
257 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
262 static const struct block_device_operations virtblk_fops = {
263 .ioctl = virtblk_ioctl,
264 .owner = THIS_MODULE,
265 .getgeo = virtblk_getgeo,
268 static int index_to_minor(int index)
270 return index << PART_BITS;
273 static ssize_t virtblk_serial_show(struct device *dev,
274 struct device_attribute *attr, char *buf)
276 struct gendisk *disk = dev_to_disk(dev);
279 /* sysfs gives us a PAGE_SIZE buffer */
280 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
282 buf[VIRTIO_BLK_ID_BYTES] = '\0';
283 err = virtblk_get_id(disk, buf);
287 if (err == -EIO) /* Unsupported? Make it empty. */
292 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
294 static int __devinit virtblk_probe(struct virtio_device *vdev)
296 struct virtio_blk *vblk;
297 struct request_queue *q;
300 u32 v, blk_size, sg_elems, opt_io_size;
302 u8 physical_block_exp, alignment_offset;
304 if (index_to_minor(index) >= 1 << MINORBITS)
307 /* We need to know how many segments before we allocate. */
308 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
309 offsetof(struct virtio_blk_config, seg_max),
312 /* We need at least one SG element, whatever they say. */
313 if (err || !sg_elems)
316 /* We need an extra sg elements at head and tail. */
318 vdev->priv = vblk = kmalloc(sizeof(*vblk) +
319 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
325 INIT_LIST_HEAD(&vblk->reqs);
326 spin_lock_init(&vblk->lock);
328 vblk->sg_elems = sg_elems;
329 sg_init_table(vblk->sg, vblk->sg_elems);
331 /* We expect one virtqueue, for output. */
332 vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
333 if (IS_ERR(vblk->vq)) {
334 err = PTR_ERR(vblk->vq);
338 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
344 /* FIXME: How many partitions? How long is a piece of string? */
345 vblk->disk = alloc_disk(1 << PART_BITS);
351 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
360 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
361 } else if (index < (26 + 1) * 26) {
362 sprintf(vblk->disk->disk_name, "vd%c%c",
363 'a' + index / 26 - 1, 'a' + index % 26);
365 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
366 const unsigned int m2 = (index / 26 - 1) % 26;
367 const unsigned int m3 = index % 26;
368 sprintf(vblk->disk->disk_name, "vd%c%c%c",
369 'a' + m1, 'a' + m2, 'a' + m3);
372 vblk->disk->major = major;
373 vblk->disk->first_minor = index_to_minor(index);
374 vblk->disk->private_data = vblk;
375 vblk->disk->fops = &virtblk_fops;
376 vblk->disk->driverfs_dev = &vdev->dev;
379 /* configure queue flush support */
380 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
381 blk_queue_flush(q, REQ_FLUSH);
383 /* If disk is read-only in the host, the guest should obey */
384 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
385 set_disk_ro(vblk->disk, 1);
387 /* Host must always specify the capacity. */
388 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
391 /* If capacity is too big, truncate with warning. */
392 if ((sector_t)cap != cap) {
393 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
394 (unsigned long long)cap);
397 set_capacity(vblk->disk, cap);
399 /* We can handle whatever the host told us to handle. */
400 blk_queue_max_segments(q, vblk->sg_elems-2);
402 /* No need to bounce any requests */
403 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
405 /* No real sector limit. */
406 blk_queue_max_hw_sectors(q, -1U);
408 /* Host can optionally specify maximum segment size and number of
410 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
411 offsetof(struct virtio_blk_config, size_max),
414 blk_queue_max_segment_size(q, v);
416 blk_queue_max_segment_size(q, -1U);
418 /* Host can optionally specify the block size of the device */
419 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
420 offsetof(struct virtio_blk_config, blk_size),
423 blk_queue_logical_block_size(q, blk_size);
425 blk_size = queue_logical_block_size(q);
427 /* Use topology information if available */
428 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
429 offsetof(struct virtio_blk_config, physical_block_exp),
430 &physical_block_exp);
431 if (!err && physical_block_exp)
432 blk_queue_physical_block_size(q,
433 blk_size * (1 << physical_block_exp));
435 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
436 offsetof(struct virtio_blk_config, alignment_offset),
438 if (!err && alignment_offset)
439 blk_queue_alignment_offset(q, blk_size * alignment_offset);
441 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
442 offsetof(struct virtio_blk_config, min_io_size),
444 if (!err && min_io_size)
445 blk_queue_io_min(q, blk_size * min_io_size);
447 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
448 offsetof(struct virtio_blk_config, opt_io_size),
450 if (!err && opt_io_size)
451 blk_queue_io_opt(q, blk_size * opt_io_size);
454 add_disk(vblk->disk);
455 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
462 del_gendisk(vblk->disk);
463 blk_cleanup_queue(vblk->disk->queue);
465 put_disk(vblk->disk);
467 mempool_destroy(vblk->pool);
469 vdev->config->del_vqs(vdev);
476 static void __devexit virtblk_remove(struct virtio_device *vdev)
478 struct virtio_blk *vblk = vdev->priv;
480 /* Nothing should be pending. */
481 BUG_ON(!list_empty(&vblk->reqs));
483 /* Stop all the virtqueues. */
484 vdev->config->reset(vdev);
486 del_gendisk(vblk->disk);
487 blk_cleanup_queue(vblk->disk->queue);
488 put_disk(vblk->disk);
489 mempool_destroy(vblk->pool);
490 vdev->config->del_vqs(vdev);
494 static const struct virtio_device_id id_table[] = {
495 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
499 static unsigned int features[] = {
500 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
501 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
502 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
506 * virtio_blk causes spurious section mismatch warning by
507 * simultaneously referring to a __devinit and a __devexit function.
508 * Use __refdata to avoid this warning.
510 static struct virtio_driver __refdata virtio_blk = {
511 .feature_table = features,
512 .feature_table_size = ARRAY_SIZE(features),
513 .driver.name = KBUILD_MODNAME,
514 .driver.owner = THIS_MODULE,
515 .id_table = id_table,
516 .probe = virtblk_probe,
517 .remove = __devexit_p(virtblk_remove),
520 static int __init init(void)
522 major = register_blkdev(0, "virtblk");
525 return register_virtio_driver(&virtio_blk);
528 static void __exit fini(void)
530 unregister_blkdev(major, "virtblk");
531 unregister_virtio_driver(&virtio_blk);
536 MODULE_DEVICE_TABLE(virtio, id_table);
537 MODULE_DESCRIPTION("Virtio block driver");
538 MODULE_LICENSE("GPL");