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);
79 __blk_end_request_all(vbr->req, error);
81 mempool_free(vbr, vblk->pool);
83 /* In case queue is stopped waiting for more buffers. */
84 blk_start_queue(vblk->disk->queue);
85 spin_unlock_irqrestore(&vblk->lock, flags);
88 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
91 unsigned long num, out = 0, in = 0;
92 struct virtblk_req *vbr;
94 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
96 /* When another request finishes we'll try again. */
100 switch (req->cmd_type) {
102 vbr->out_hdr.type = 0;
103 vbr->out_hdr.sector = blk_rq_pos(vbr->req);
104 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
106 case REQ_TYPE_BLOCK_PC:
107 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
108 vbr->out_hdr.sector = 0;
109 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
111 case REQ_TYPE_SPECIAL:
112 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
113 vbr->out_hdr.sector = 0;
114 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
116 case REQ_TYPE_LINUX_BLOCK:
117 if (req->cmd[0] == REQ_LB_OP_FLUSH) {
118 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
119 vbr->out_hdr.sector = 0;
120 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
125 /* We don't put anything else in the queue. */
129 if (vbr->req->cmd_flags & REQ_HARDBARRIER)
130 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
132 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
135 * If this is a packet command we need a couple of additional headers.
136 * Behind the normal outhdr we put a segment with the scsi command
137 * block, and before the normal inhdr we put the sense data and the
138 * inhdr with additional status information before the normal inhdr.
140 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
141 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
143 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
145 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
146 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
147 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
148 sizeof(vbr->in_hdr));
151 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
152 sizeof(vbr->status));
155 if (rq_data_dir(vbr->req) == WRITE) {
156 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
159 vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
164 if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
165 mempool_free(vbr, vblk->pool);
169 list_add_tail(&vbr->list, &vblk->reqs);
173 static void do_virtblk_request(struct request_queue *q)
175 struct virtio_blk *vblk = q->queuedata;
177 unsigned int issued = 0;
179 while ((req = blk_peek_request(q)) != NULL) {
180 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
182 /* If this request fails, stop queue and wait for something to
183 finish to restart it. */
184 if (!do_req(q, vblk, req)) {
188 blk_start_request(req);
193 virtqueue_kick(vblk->vq);
196 static void virtblk_prepare_flush(struct request_queue *q, struct request *req)
198 req->cmd_type = REQ_TYPE_LINUX_BLOCK;
199 req->cmd[0] = REQ_LB_OP_FLUSH;
202 /* return id (s/n) string for *disk to *id_str
204 static int virtblk_get_id(struct gendisk *disk, char *id_str)
206 struct virtio_blk *vblk = disk->private_data;
210 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
215 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
221 req->cmd_type = REQ_TYPE_SPECIAL;
222 return blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
225 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
226 unsigned cmd, unsigned long data)
228 struct gendisk *disk = bdev->bd_disk;
229 struct virtio_blk *vblk = disk->private_data;
231 if (cmd == 0x56424944) { /* 'VBID' */
232 void __user *usr_data = (void __user *)data;
233 char id_str[VIRTIO_BLK_ID_BYTES];
236 err = virtblk_get_id(disk, id_str);
237 if (!err && copy_to_user(usr_data, id_str, VIRTIO_BLK_ID_BYTES))
242 * Only allow the generic SCSI ioctls if the host can support it.
244 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
247 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
248 (void __user *)data);
251 /* We provide getgeo only to please some old bootloader/partitioning tools */
252 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
254 struct virtio_blk *vblk = bd->bd_disk->private_data;
255 struct virtio_blk_geometry vgeo;
258 /* see if the host passed in geometry config */
259 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
260 offsetof(struct virtio_blk_config, geometry),
264 geo->heads = vgeo.heads;
265 geo->sectors = vgeo.sectors;
266 geo->cylinders = vgeo.cylinders;
268 /* some standard values, similar to sd */
270 geo->sectors = 1 << 5;
271 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
276 static const struct block_device_operations virtblk_fops = {
277 .locked_ioctl = virtblk_ioctl,
278 .owner = THIS_MODULE,
279 .getgeo = virtblk_getgeo,
282 static int index_to_minor(int index)
284 return index << PART_BITS;
287 static int __devinit virtblk_probe(struct virtio_device *vdev)
289 struct virtio_blk *vblk;
290 struct request_queue *q;
293 u32 v, blk_size, sg_elems, opt_io_size;
295 u8 physical_block_exp, alignment_offset;
297 if (index_to_minor(index) >= 1 << MINORBITS)
300 /* We need to know how many segments before we allocate. */
301 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
302 offsetof(struct virtio_blk_config, seg_max),
305 /* We need at least one SG element, whatever they say. */
306 if (err || !sg_elems)
309 /* We need an extra sg elements at head and tail. */
311 vdev->priv = vblk = kmalloc(sizeof(*vblk) +
312 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
318 INIT_LIST_HEAD(&vblk->reqs);
319 spin_lock_init(&vblk->lock);
321 vblk->sg_elems = sg_elems;
322 sg_init_table(vblk->sg, vblk->sg_elems);
324 /* We expect one virtqueue, for output. */
325 vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
326 if (IS_ERR(vblk->vq)) {
327 err = PTR_ERR(vblk->vq);
331 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
337 /* FIXME: How many partitions? How long is a piece of string? */
338 vblk->disk = alloc_disk(1 << PART_BITS);
344 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
353 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
354 } else if (index < (26 + 1) * 26) {
355 sprintf(vblk->disk->disk_name, "vd%c%c",
356 'a' + index / 26 - 1, 'a' + index % 26);
358 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
359 const unsigned int m2 = (index / 26 - 1) % 26;
360 const unsigned int m3 = index % 26;
361 sprintf(vblk->disk->disk_name, "vd%c%c%c",
362 'a' + m1, 'a' + m2, 'a' + m3);
365 vblk->disk->major = major;
366 vblk->disk->first_minor = index_to_minor(index);
367 vblk->disk->private_data = vblk;
368 vblk->disk->fops = &virtblk_fops;
369 vblk->disk->driverfs_dev = &vdev->dev;
372 /* If barriers are supported, tell block layer that queue is ordered */
373 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
374 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH,
375 virtblk_prepare_flush);
376 else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
377 blk_queue_ordered(q, QUEUE_ORDERED_TAG, NULL);
379 /* If disk is read-only in the host, the guest should obey */
380 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
381 set_disk_ro(vblk->disk, 1);
383 /* Host must always specify the capacity. */
384 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
387 /* If capacity is too big, truncate with warning. */
388 if ((sector_t)cap != cap) {
389 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
390 (unsigned long long)cap);
393 set_capacity(vblk->disk, cap);
395 /* We can handle whatever the host told us to handle. */
396 blk_queue_max_segments(q, vblk->sg_elems-2);
398 /* No need to bounce any requests */
399 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
401 /* No real sector limit. */
402 blk_queue_max_hw_sectors(q, -1U);
404 /* Host can optionally specify maximum segment size and number of
406 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
407 offsetof(struct virtio_blk_config, size_max),
410 blk_queue_max_segment_size(q, v);
412 blk_queue_max_segment_size(q, -1U);
414 /* Host can optionally specify the block size of the device */
415 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
416 offsetof(struct virtio_blk_config, blk_size),
419 blk_queue_logical_block_size(q, blk_size);
421 blk_size = queue_logical_block_size(q);
423 /* Use topology information if available */
424 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
425 offsetof(struct virtio_blk_config, physical_block_exp),
426 &physical_block_exp);
427 if (!err && physical_block_exp)
428 blk_queue_physical_block_size(q,
429 blk_size * (1 << physical_block_exp));
431 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
432 offsetof(struct virtio_blk_config, alignment_offset),
434 if (!err && alignment_offset)
435 blk_queue_alignment_offset(q, blk_size * alignment_offset);
437 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
438 offsetof(struct virtio_blk_config, min_io_size),
440 if (!err && min_io_size)
441 blk_queue_io_min(q, blk_size * min_io_size);
443 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
444 offsetof(struct virtio_blk_config, opt_io_size),
446 if (!err && opt_io_size)
447 blk_queue_io_opt(q, blk_size * opt_io_size);
450 add_disk(vblk->disk);
454 put_disk(vblk->disk);
456 mempool_destroy(vblk->pool);
458 vdev->config->del_vqs(vdev);
465 static void __devexit virtblk_remove(struct virtio_device *vdev)
467 struct virtio_blk *vblk = vdev->priv;
469 /* Nothing should be pending. */
470 BUG_ON(!list_empty(&vblk->reqs));
472 /* Stop all the virtqueues. */
473 vdev->config->reset(vdev);
475 del_gendisk(vblk->disk);
476 blk_cleanup_queue(vblk->disk->queue);
477 put_disk(vblk->disk);
478 mempool_destroy(vblk->pool);
479 vdev->config->del_vqs(vdev);
483 static const struct virtio_device_id id_table[] = {
484 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
488 static unsigned int features[] = {
489 VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
490 VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
491 VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
495 * virtio_blk causes spurious section mismatch warning by
496 * simultaneously referring to a __devinit and a __devexit function.
497 * Use __refdata to avoid this warning.
499 static struct virtio_driver __refdata virtio_blk = {
500 .feature_table = features,
501 .feature_table_size = ARRAY_SIZE(features),
502 .driver.name = KBUILD_MODNAME,
503 .driver.owner = THIS_MODULE,
504 .id_table = id_table,
505 .probe = virtblk_probe,
506 .remove = __devexit_p(virtblk_remove),
509 static int __init init(void)
511 major = register_blkdev(0, "virtblk");
514 return register_virtio_driver(&virtio_blk);
517 static void __exit fini(void)
519 unregister_blkdev(major, "virtblk");
520 unregister_virtio_driver(&virtio_blk);
525 MODULE_DEVICE_TABLE(virtio, id_table);
526 MODULE_DESCRIPTION("Virtio block driver");
527 MODULE_LICENSE("GPL");