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 if (blk_pc_request(vbr->req)) {
69 vbr->req->resid_len = vbr->in_hdr.residual;
70 vbr->req->sense_len = vbr->in_hdr.sense_len;
71 vbr->req->errors = vbr->in_hdr.errors;
73 if (blk_special_request(vbr->req))
74 vbr->req->errors = (error != 0);
76 __blk_end_request_all(vbr->req, error);
78 mempool_free(vbr, vblk->pool);
80 /* In case queue is stopped waiting for more buffers. */
81 blk_start_queue(vblk->disk->queue);
82 spin_unlock_irqrestore(&vblk->lock, flags);
85 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
88 unsigned long num, out = 0, in = 0;
89 struct virtblk_req *vbr;
91 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
93 /* When another request finishes we'll try again. */
97 switch (req->cmd_type) {
99 vbr->out_hdr.type = 0;
100 vbr->out_hdr.sector = blk_rq_pos(vbr->req);
101 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
103 case REQ_TYPE_BLOCK_PC:
104 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
105 vbr->out_hdr.sector = 0;
106 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
108 case REQ_TYPE_SPECIAL:
109 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
110 vbr->out_hdr.sector = 0;
111 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
113 case REQ_TYPE_LINUX_BLOCK:
114 if (req->cmd[0] == REQ_LB_OP_FLUSH) {
115 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
116 vbr->out_hdr.sector = 0;
117 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
122 /* We don't put anything else in the queue. */
126 if (blk_barrier_rq(vbr->req))
127 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
129 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
132 * If this is a packet command we need a couple of additional headers.
133 * Behind the normal outhdr we put a segment with the scsi command
134 * block, and before the normal inhdr we put the sense data and the
135 * inhdr with additional status information before the normal inhdr.
137 if (blk_pc_request(vbr->req))
138 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
140 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
142 if (blk_pc_request(vbr->req)) {
143 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
144 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
145 sizeof(vbr->in_hdr));
148 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
149 sizeof(vbr->status));
152 if (rq_data_dir(vbr->req) == WRITE) {
153 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
156 vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
161 if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
162 mempool_free(vbr, vblk->pool);
166 list_add_tail(&vbr->list, &vblk->reqs);
170 static void do_virtblk_request(struct request_queue *q)
172 struct virtio_blk *vblk = q->queuedata;
174 unsigned int issued = 0;
176 while ((req = blk_peek_request(q)) != NULL) {
177 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
179 /* If this request fails, stop queue and wait for something to
180 finish to restart it. */
181 if (!do_req(q, vblk, req)) {
185 blk_start_request(req);
190 virtqueue_kick(vblk->vq);
193 static void virtblk_prepare_flush(struct request_queue *q, struct request *req)
195 req->cmd_type = REQ_TYPE_LINUX_BLOCK;
196 req->cmd[0] = REQ_LB_OP_FLUSH;
199 /* return id (s/n) string for *disk to *id_str
201 static int virtblk_get_id(struct gendisk *disk, char *id_str)
203 struct virtio_blk *vblk = disk->private_data;
207 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
212 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
218 req->cmd_type = REQ_TYPE_SPECIAL;
219 return blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
222 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
223 unsigned cmd, unsigned long data)
225 struct gendisk *disk = bdev->bd_disk;
226 struct virtio_blk *vblk = disk->private_data;
229 * Only allow the generic SCSI ioctls if the host can support it.
231 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
234 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
235 (void __user *)data);
238 /* We provide getgeo only to please some old bootloader/partitioning tools */
239 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
241 struct virtio_blk *vblk = bd->bd_disk->private_data;
242 struct virtio_blk_geometry vgeo;
245 /* see if the host passed in geometry config */
246 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
247 offsetof(struct virtio_blk_config, geometry),
251 geo->heads = vgeo.heads;
252 geo->sectors = vgeo.sectors;
253 geo->cylinders = vgeo.cylinders;
255 /* some standard values, similar to sd */
257 geo->sectors = 1 << 5;
258 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
263 static const struct block_device_operations virtblk_fops = {
264 .locked_ioctl = virtblk_ioctl,
265 .owner = THIS_MODULE,
266 .getgeo = virtblk_getgeo,
269 static int index_to_minor(int index)
271 return index << PART_BITS;
274 static ssize_t virtblk_serial_show(struct device *dev,
275 struct device_attribute *attr, char *buf)
277 struct gendisk *disk = dev_to_disk(dev);
280 /* sysfs gives us a PAGE_SIZE buffer */
281 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
283 buf[VIRTIO_BLK_ID_BYTES] = '\0';
284 err = virtblk_get_id(disk, buf);
288 if (err == -EIO) /* Unsupported? Make it empty. */
293 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
295 static int __devinit virtblk_probe(struct virtio_device *vdev)
297 struct virtio_blk *vblk;
298 struct request_queue *q;
301 u32 v, blk_size, sg_elems, opt_io_size;
303 u8 physical_block_exp, alignment_offset;
305 if (index_to_minor(index) >= 1 << MINORBITS)
308 /* We need to know how many segments before we allocate. */
309 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
310 offsetof(struct virtio_blk_config, seg_max),
313 /* We need at least one SG element, whatever they say. */
314 if (err || !sg_elems)
317 /* We need an extra sg elements at head and tail. */
319 vdev->priv = vblk = kmalloc(sizeof(*vblk) +
320 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
326 INIT_LIST_HEAD(&vblk->reqs);
327 spin_lock_init(&vblk->lock);
329 vblk->sg_elems = sg_elems;
330 sg_init_table(vblk->sg, vblk->sg_elems);
332 /* We expect one virtqueue, for output. */
333 vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
334 if (IS_ERR(vblk->vq)) {
335 err = PTR_ERR(vblk->vq);
339 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
345 /* FIXME: How many partitions? How long is a piece of string? */
346 vblk->disk = alloc_disk(1 << PART_BITS);
352 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
361 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
362 } else if (index < (26 + 1) * 26) {
363 sprintf(vblk->disk->disk_name, "vd%c%c",
364 'a' + index / 26 - 1, 'a' + index % 26);
366 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
367 const unsigned int m2 = (index / 26 - 1) % 26;
368 const unsigned int m3 = index % 26;
369 sprintf(vblk->disk->disk_name, "vd%c%c%c",
370 'a' + m1, 'a' + m2, 'a' + m3);
373 vblk->disk->major = major;
374 vblk->disk->first_minor = index_to_minor(index);
375 vblk->disk->private_data = vblk;
376 vblk->disk->fops = &virtblk_fops;
377 vblk->disk->driverfs_dev = &vdev->dev;
380 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH)) {
382 * If the FLUSH feature is supported we do have support for
383 * flushing a volatile write cache on the host. Use that
384 * to implement write barrier support.
386 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH,
387 virtblk_prepare_flush);
388 } else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER)) {
390 * If the BARRIER feature is supported the host expects us
391 * to order request by tags. This implies there is not
392 * volatile write cache on the host, and that the host
393 * never re-orders outstanding I/O. This feature is not
394 * useful for real life scenarious and deprecated.
396 blk_queue_ordered(q, QUEUE_ORDERED_TAG, NULL);
399 * If the FLUSH feature is not supported we must assume that
400 * the host does not perform any kind of volatile write
401 * caching. We still need to drain the queue to provider
402 * proper barrier semantics.
404 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN, NULL);
407 /* If disk is read-only in the host, the guest should obey */
408 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
409 set_disk_ro(vblk->disk, 1);
411 /* Host must always specify the capacity. */
412 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
415 /* If capacity is too big, truncate with warning. */
416 if ((sector_t)cap != cap) {
417 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
418 (unsigned long long)cap);
421 set_capacity(vblk->disk, cap);
423 /* We can handle whatever the host told us to handle. */
424 blk_queue_max_segments(q, vblk->sg_elems-2);
426 /* No need to bounce any requests */
427 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
429 /* No real sector limit. */
430 blk_queue_max_hw_sectors(q, -1U);
432 /* Host can optionally specify maximum segment size and number of
434 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
435 offsetof(struct virtio_blk_config, size_max),
438 blk_queue_max_segment_size(q, v);
440 blk_queue_max_segment_size(q, -1U);
442 /* Host can optionally specify the block size of the device */
443 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
444 offsetof(struct virtio_blk_config, blk_size),
447 blk_queue_logical_block_size(q, blk_size);
449 blk_size = queue_logical_block_size(q);
451 /* Use topology information if available */
452 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
453 offsetof(struct virtio_blk_config, physical_block_exp),
454 &physical_block_exp);
455 if (!err && physical_block_exp)
456 blk_queue_physical_block_size(q,
457 blk_size * (1 << physical_block_exp));
459 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
460 offsetof(struct virtio_blk_config, alignment_offset),
462 if (!err && alignment_offset)
463 blk_queue_alignment_offset(q, blk_size * alignment_offset);
465 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
466 offsetof(struct virtio_blk_config, min_io_size),
468 if (!err && min_io_size)
469 blk_queue_io_min(q, blk_size * min_io_size);
471 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
472 offsetof(struct virtio_blk_config, opt_io_size),
474 if (!err && opt_io_size)
475 blk_queue_io_opt(q, blk_size * opt_io_size);
478 add_disk(vblk->disk);
479 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
486 del_gendisk(vblk->disk);
487 blk_cleanup_queue(vblk->disk->queue);
489 put_disk(vblk->disk);
491 mempool_destroy(vblk->pool);
493 vdev->config->del_vqs(vdev);
500 static void __devexit virtblk_remove(struct virtio_device *vdev)
502 struct virtio_blk *vblk = vdev->priv;
504 /* Nothing should be pending. */
505 BUG_ON(!list_empty(&vblk->reqs));
507 /* Stop all the virtqueues. */
508 vdev->config->reset(vdev);
510 del_gendisk(vblk->disk);
511 blk_cleanup_queue(vblk->disk->queue);
512 put_disk(vblk->disk);
513 mempool_destroy(vblk->pool);
514 vdev->config->del_vqs(vdev);
518 static const struct virtio_device_id id_table[] = {
519 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
523 static unsigned int features[] = {
524 VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
525 VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
526 VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
530 * virtio_blk causes spurious section mismatch warning by
531 * simultaneously referring to a __devinit and a __devexit function.
532 * Use __refdata to avoid this warning.
534 static struct virtio_driver __refdata virtio_blk = {
535 .feature_table = features,
536 .feature_table_size = ARRAY_SIZE(features),
537 .driver.name = KBUILD_MODNAME,
538 .driver.owner = THIS_MODULE,
539 .id_table = id_table,
540 .probe = virtblk_probe,
541 .remove = __devexit_p(virtblk_remove),
544 static int __init init(void)
546 major = register_blkdev(0, "virtblk");
549 return register_virtio_driver(&virtio_blk);
552 static void __exit fini(void)
554 unregister_blkdev(major, "virtblk");
555 unregister_virtio_driver(&virtio_blk);
560 MODULE_DEVICE_TABLE(virtio, id_table);
561 MODULE_DESCRIPTION("Virtio block driver");
562 MODULE_LICENSE("GPL");