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/virtio.h>
8 #include <linux/virtio_blk.h>
9 #include <linux/scatterlist.h>
10 #include <linux/string_helpers.h>
11 #include <scsi/scsi_cmnd.h>
12 #include <linux/idr.h>
17 static DEFINE_IDA(vd_index_ida);
19 struct workqueue_struct *virtblk_wq;
25 struct virtio_device *vdev;
28 /* The disk structure for the kernel. */
31 /* Request tracking. */
32 struct list_head reqs;
36 /* Process context for config space updates */
37 struct work_struct config_work;
39 /* What host tells us, plus 2 for header & tailer. */
40 unsigned int sg_elems;
42 /* Ida index - used to track minor number allocations. */
45 /* Scatterlist: can be too big for stack. */
46 struct scatterlist sg[/*sg_elems*/];
51 struct list_head list;
53 struct virtio_blk_outhdr out_hdr;
54 struct virtio_scsi_inhdr in_hdr;
58 static void blk_done(struct virtqueue *vq)
60 struct virtio_blk *vblk = vq->vdev->priv;
61 struct virtblk_req *vbr;
65 spin_lock_irqsave(&vblk->lock, flags);
66 while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
69 switch (vbr->status) {
73 case VIRTIO_BLK_S_UNSUPP:
81 switch (vbr->req->cmd_type) {
82 case REQ_TYPE_BLOCK_PC:
83 vbr->req->resid_len = vbr->in_hdr.residual;
84 vbr->req->sense_len = vbr->in_hdr.sense_len;
85 vbr->req->errors = vbr->in_hdr.errors;
87 case REQ_TYPE_SPECIAL:
88 vbr->req->errors = (error != 0);
94 __blk_end_request_all(vbr->req, error);
96 mempool_free(vbr, vblk->pool);
98 /* In case queue is stopped waiting for more buffers. */
99 blk_start_queue(vblk->disk->queue);
100 spin_unlock_irqrestore(&vblk->lock, flags);
103 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
106 unsigned long num, out = 0, in = 0;
107 struct virtblk_req *vbr;
109 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
111 /* When another request finishes we'll try again. */
116 if (req->cmd_flags & REQ_FLUSH) {
117 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
118 vbr->out_hdr.sector = 0;
119 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
121 switch (req->cmd_type) {
123 vbr->out_hdr.type = 0;
124 vbr->out_hdr.sector = blk_rq_pos(vbr->req);
125 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
127 case REQ_TYPE_BLOCK_PC:
128 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
129 vbr->out_hdr.sector = 0;
130 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
132 case REQ_TYPE_SPECIAL:
133 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
134 vbr->out_hdr.sector = 0;
135 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
138 /* We don't put anything else in the queue. */
143 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
146 * If this is a packet command we need a couple of additional headers.
147 * Behind the normal outhdr we put a segment with the scsi command
148 * block, and before the normal inhdr we put the sense data and the
149 * inhdr with additional status information before the normal inhdr.
151 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
152 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
154 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
156 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
157 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
158 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
159 sizeof(vbr->in_hdr));
162 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
163 sizeof(vbr->status));
166 if (rq_data_dir(vbr->req) == WRITE) {
167 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
170 vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
175 if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr, GFP_ATOMIC)<0) {
176 mempool_free(vbr, vblk->pool);
180 list_add_tail(&vbr->list, &vblk->reqs);
184 static void do_virtblk_request(struct request_queue *q)
186 struct virtio_blk *vblk = q->queuedata;
188 unsigned int issued = 0;
190 while ((req = blk_peek_request(q)) != NULL) {
191 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
193 /* If this request fails, stop queue and wait for something to
194 finish to restart it. */
195 if (!do_req(q, vblk, req)) {
199 blk_start_request(req);
204 virtqueue_kick(vblk->vq);
207 /* return id (s/n) string for *disk to *id_str
209 static int virtblk_get_id(struct gendisk *disk, char *id_str)
211 struct virtio_blk *vblk = disk->private_data;
216 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
221 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
227 req->cmd_type = REQ_TYPE_SPECIAL;
228 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
229 blk_put_request(req);
234 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
235 unsigned int cmd, unsigned long data)
237 struct gendisk *disk = bdev->bd_disk;
238 struct virtio_blk *vblk = disk->private_data;
241 * Only allow the generic SCSI ioctls if the host can support it.
243 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
246 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
247 (void __user *)data);
250 /* We provide getgeo only to please some old bootloader/partitioning tools */
251 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
253 struct virtio_blk *vblk = bd->bd_disk->private_data;
254 struct virtio_blk_geometry vgeo;
257 /* see if the host passed in geometry config */
258 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
259 offsetof(struct virtio_blk_config, geometry),
263 geo->heads = vgeo.heads;
264 geo->sectors = vgeo.sectors;
265 geo->cylinders = vgeo.cylinders;
267 /* some standard values, similar to sd */
269 geo->sectors = 1 << 5;
270 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
275 static const struct block_device_operations virtblk_fops = {
276 .ioctl = virtblk_ioctl,
277 .owner = THIS_MODULE,
278 .getgeo = virtblk_getgeo,
281 static int index_to_minor(int index)
283 return index << PART_BITS;
286 static int minor_to_index(int minor)
288 return minor >> PART_BITS;
291 static ssize_t virtblk_serial_show(struct device *dev,
292 struct device_attribute *attr, char *buf)
294 struct gendisk *disk = dev_to_disk(dev);
297 /* sysfs gives us a PAGE_SIZE buffer */
298 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
300 buf[VIRTIO_BLK_ID_BYTES] = '\0';
301 err = virtblk_get_id(disk, buf);
305 if (err == -EIO) /* Unsupported? Make it empty. */
310 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
312 static void virtblk_config_changed_work(struct work_struct *work)
314 struct virtio_blk *vblk =
315 container_of(work, struct virtio_blk, config_work);
316 struct virtio_device *vdev = vblk->vdev;
317 struct request_queue *q = vblk->disk->queue;
318 char cap_str_2[10], cap_str_10[10];
321 /* Host must always specify the capacity. */
322 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
323 &capacity, sizeof(capacity));
325 /* If capacity is too big, truncate with warning. */
326 if ((sector_t)capacity != capacity) {
327 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
328 (unsigned long long)capacity);
329 capacity = (sector_t)-1;
332 size = capacity * queue_logical_block_size(q);
333 string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
334 string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
336 dev_notice(&vdev->dev,
337 "new size: %llu %d-byte logical blocks (%s/%s)\n",
338 (unsigned long long)capacity,
339 queue_logical_block_size(q),
340 cap_str_10, cap_str_2);
342 set_capacity(vblk->disk, capacity);
345 static void virtblk_config_changed(struct virtio_device *vdev)
347 struct virtio_blk *vblk = vdev->priv;
349 queue_work(virtblk_wq, &vblk->config_work);
352 static int __devinit virtblk_probe(struct virtio_device *vdev)
354 struct virtio_blk *vblk;
355 struct request_queue *q;
358 u32 v, blk_size, sg_elems, opt_io_size;
360 u8 physical_block_exp, alignment_offset;
362 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
368 /* We need to know how many segments before we allocate. */
369 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
370 offsetof(struct virtio_blk_config, seg_max),
373 /* We need at least one SG element, whatever they say. */
374 if (err || !sg_elems)
377 /* We need an extra sg elements at head and tail. */
379 vdev->priv = vblk = kmalloc(sizeof(*vblk) +
380 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
386 INIT_LIST_HEAD(&vblk->reqs);
387 spin_lock_init(&vblk->lock);
389 vblk->sg_elems = sg_elems;
390 sg_init_table(vblk->sg, vblk->sg_elems);
391 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
393 /* We expect one virtqueue, for output. */
394 vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
395 if (IS_ERR(vblk->vq)) {
396 err = PTR_ERR(vblk->vq);
400 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
406 /* FIXME: How many partitions? How long is a piece of string? */
407 vblk->disk = alloc_disk(1 << PART_BITS);
413 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
422 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
423 } else if (index < (26 + 1) * 26) {
424 sprintf(vblk->disk->disk_name, "vd%c%c",
425 'a' + index / 26 - 1, 'a' + index % 26);
427 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
428 const unsigned int m2 = (index / 26 - 1) % 26;
429 const unsigned int m3 = index % 26;
430 sprintf(vblk->disk->disk_name, "vd%c%c%c",
431 'a' + m1, 'a' + m2, 'a' + m3);
434 vblk->disk->major = major;
435 vblk->disk->first_minor = index_to_minor(index);
436 vblk->disk->private_data = vblk;
437 vblk->disk->fops = &virtblk_fops;
438 vblk->disk->driverfs_dev = &vdev->dev;
441 /* configure queue flush support */
442 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
443 blk_queue_flush(q, REQ_FLUSH);
445 /* If disk is read-only in the host, the guest should obey */
446 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
447 set_disk_ro(vblk->disk, 1);
449 /* Host must always specify the capacity. */
450 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
453 /* If capacity is too big, truncate with warning. */
454 if ((sector_t)cap != cap) {
455 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
456 (unsigned long long)cap);
459 set_capacity(vblk->disk, cap);
461 /* We can handle whatever the host told us to handle. */
462 blk_queue_max_segments(q, vblk->sg_elems-2);
464 /* No need to bounce any requests */
465 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
467 /* No real sector limit. */
468 blk_queue_max_hw_sectors(q, -1U);
470 /* Host can optionally specify maximum segment size and number of
472 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
473 offsetof(struct virtio_blk_config, size_max),
476 blk_queue_max_segment_size(q, v);
478 blk_queue_max_segment_size(q, -1U);
480 /* Host can optionally specify the block size of the device */
481 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
482 offsetof(struct virtio_blk_config, blk_size),
485 blk_queue_logical_block_size(q, blk_size);
487 blk_size = queue_logical_block_size(q);
489 /* Use topology information if available */
490 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
491 offsetof(struct virtio_blk_config, physical_block_exp),
492 &physical_block_exp);
493 if (!err && physical_block_exp)
494 blk_queue_physical_block_size(q,
495 blk_size * (1 << physical_block_exp));
497 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
498 offsetof(struct virtio_blk_config, alignment_offset),
500 if (!err && alignment_offset)
501 blk_queue_alignment_offset(q, blk_size * alignment_offset);
503 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
504 offsetof(struct virtio_blk_config, min_io_size),
506 if (!err && min_io_size)
507 blk_queue_io_min(q, blk_size * min_io_size);
509 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
510 offsetof(struct virtio_blk_config, opt_io_size),
512 if (!err && opt_io_size)
513 blk_queue_io_opt(q, blk_size * opt_io_size);
516 add_disk(vblk->disk);
517 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
524 del_gendisk(vblk->disk);
525 blk_cleanup_queue(vblk->disk->queue);
527 put_disk(vblk->disk);
529 mempool_destroy(vblk->pool);
531 vdev->config->del_vqs(vdev);
535 ida_simple_remove(&vd_index_ida, index);
540 static void __devexit virtblk_remove(struct virtio_device *vdev)
542 struct virtio_blk *vblk = vdev->priv;
543 int index = vblk->index;
545 flush_work(&vblk->config_work);
547 /* Nothing should be pending. */
548 BUG_ON(!list_empty(&vblk->reqs));
550 /* Stop all the virtqueues. */
551 vdev->config->reset(vdev);
553 del_gendisk(vblk->disk);
554 blk_cleanup_queue(vblk->disk->queue);
555 put_disk(vblk->disk);
556 mempool_destroy(vblk->pool);
557 vdev->config->del_vqs(vdev);
559 ida_simple_remove(&vd_index_ida, index);
562 static const struct virtio_device_id id_table[] = {
563 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
567 static unsigned int features[] = {
568 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
569 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
570 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
574 * virtio_blk causes spurious section mismatch warning by
575 * simultaneously referring to a __devinit and a __devexit function.
576 * Use __refdata to avoid this warning.
578 static struct virtio_driver __refdata virtio_blk = {
579 .feature_table = features,
580 .feature_table_size = ARRAY_SIZE(features),
581 .driver.name = KBUILD_MODNAME,
582 .driver.owner = THIS_MODULE,
583 .id_table = id_table,
584 .probe = virtblk_probe,
585 .remove = __devexit_p(virtblk_remove),
586 .config_changed = virtblk_config_changed,
589 static int __init init(void)
593 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
597 major = register_blkdev(0, "virtblk");
600 goto out_destroy_workqueue;
603 error = register_virtio_driver(&virtio_blk);
605 goto out_unregister_blkdev;
608 out_unregister_blkdev:
609 unregister_blkdev(major, "virtblk");
610 out_destroy_workqueue:
611 destroy_workqueue(virtblk_wq);
615 static void __exit fini(void)
617 unregister_blkdev(major, "virtblk");
618 unregister_virtio_driver(&virtio_blk);
619 destroy_workqueue(virtblk_wq);
624 MODULE_DEVICE_TABLE(virtio, id_table);
625 MODULE_DESCRIPTION("Virtio block driver");
626 MODULE_LICENSE("GPL");