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;
24 struct virtio_device *vdev;
27 /* The disk structure for the kernel. */
30 /* Request tracking. */
31 struct list_head reqs;
35 /* Process context for config space updates */
36 struct work_struct config_work;
38 /* Lock for config space updates */
39 struct mutex config_lock;
41 /* enable config space updates */
44 /* What host tells us, plus 2 for header & tailer. */
45 unsigned int sg_elems;
47 /* Ida index - used to track minor number allocations. */
50 /* Scatterlist: can be too big for stack. */
51 struct scatterlist sg[/*sg_elems*/];
56 struct list_head list;
58 struct virtio_blk_outhdr out_hdr;
59 struct virtio_scsi_inhdr in_hdr;
63 static void blk_done(struct virtqueue *vq)
65 struct virtio_blk *vblk = vq->vdev->priv;
66 struct virtblk_req *vbr;
70 spin_lock_irqsave(vblk->disk->queue->queue_lock, flags);
71 while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
74 switch (vbr->status) {
78 case VIRTIO_BLK_S_UNSUPP:
86 switch (vbr->req->cmd_type) {
87 case REQ_TYPE_BLOCK_PC:
88 vbr->req->resid_len = vbr->in_hdr.residual;
89 vbr->req->sense_len = vbr->in_hdr.sense_len;
90 vbr->req->errors = vbr->in_hdr.errors;
92 case REQ_TYPE_SPECIAL:
93 vbr->req->errors = (error != 0);
99 __blk_end_request_all(vbr->req, error);
100 list_del(&vbr->list);
101 mempool_free(vbr, vblk->pool);
103 /* In case queue is stopped waiting for more buffers. */
104 blk_start_queue(vblk->disk->queue);
105 spin_unlock_irqrestore(vblk->disk->queue->queue_lock, flags);
108 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
111 unsigned long num, out = 0, in = 0;
112 struct virtblk_req *vbr;
114 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
116 /* When another request finishes we'll try again. */
121 if (req->cmd_flags & REQ_FLUSH) {
122 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
123 vbr->out_hdr.sector = 0;
124 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
126 switch (req->cmd_type) {
128 vbr->out_hdr.type = 0;
129 vbr->out_hdr.sector = blk_rq_pos(vbr->req);
130 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
132 case REQ_TYPE_BLOCK_PC:
133 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
134 vbr->out_hdr.sector = 0;
135 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
137 case REQ_TYPE_SPECIAL:
138 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
139 vbr->out_hdr.sector = 0;
140 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
143 /* We don't put anything else in the queue. */
148 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
151 * If this is a packet command we need a couple of additional headers.
152 * Behind the normal outhdr we put a segment with the scsi command
153 * block, and before the normal inhdr we put the sense data and the
154 * inhdr with additional status information before the normal inhdr.
156 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
157 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
159 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
161 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
162 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
163 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
164 sizeof(vbr->in_hdr));
167 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
168 sizeof(vbr->status));
171 if (rq_data_dir(vbr->req) == WRITE) {
172 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
175 vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
180 if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
181 mempool_free(vbr, vblk->pool);
185 list_add_tail(&vbr->list, &vblk->reqs);
189 static void do_virtblk_request(struct request_queue *q)
191 struct virtio_blk *vblk = q->queuedata;
193 unsigned int issued = 0;
195 while ((req = blk_peek_request(q)) != NULL) {
196 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
198 /* If this request fails, stop queue and wait for something to
199 finish to restart it. */
200 if (!do_req(q, vblk, req)) {
204 blk_start_request(req);
209 virtqueue_kick(vblk->vq);
212 /* return id (s/n) string for *disk to *id_str
214 static int virtblk_get_id(struct gendisk *disk, char *id_str)
216 struct virtio_blk *vblk = disk->private_data;
221 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
226 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
232 req->cmd_type = REQ_TYPE_SPECIAL;
233 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
234 blk_put_request(req);
239 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
240 unsigned int cmd, unsigned long data)
242 struct gendisk *disk = bdev->bd_disk;
243 struct virtio_blk *vblk = disk->private_data;
246 * Only allow the generic SCSI ioctls if the host can support it.
248 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
251 return scsi_cmd_blk_ioctl(bdev, mode, cmd,
252 (void __user *)data);
255 /* We provide getgeo only to please some old bootloader/partitioning tools */
256 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
258 struct virtio_blk *vblk = bd->bd_disk->private_data;
259 struct virtio_blk_geometry vgeo;
262 /* see if the host passed in geometry config */
263 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
264 offsetof(struct virtio_blk_config, geometry),
268 geo->heads = vgeo.heads;
269 geo->sectors = vgeo.sectors;
270 geo->cylinders = vgeo.cylinders;
272 /* some standard values, similar to sd */
274 geo->sectors = 1 << 5;
275 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
280 static const struct block_device_operations virtblk_fops = {
281 .ioctl = virtblk_ioctl,
282 .owner = THIS_MODULE,
283 .getgeo = virtblk_getgeo,
286 static int index_to_minor(int index)
288 return index << PART_BITS;
291 static int minor_to_index(int minor)
293 return minor >> PART_BITS;
296 static ssize_t virtblk_serial_show(struct device *dev,
297 struct device_attribute *attr, char *buf)
299 struct gendisk *disk = dev_to_disk(dev);
302 /* sysfs gives us a PAGE_SIZE buffer */
303 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
305 buf[VIRTIO_BLK_ID_BYTES] = '\0';
306 err = virtblk_get_id(disk, buf);
310 if (err == -EIO) /* Unsupported? Make it empty. */
315 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
317 static void virtblk_config_changed_work(struct work_struct *work)
319 struct virtio_blk *vblk =
320 container_of(work, struct virtio_blk, config_work);
321 struct virtio_device *vdev = vblk->vdev;
322 struct request_queue *q = vblk->disk->queue;
323 char cap_str_2[10], cap_str_10[10];
326 mutex_lock(&vblk->config_lock);
327 if (!vblk->config_enable)
330 /* Host must always specify the capacity. */
331 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
332 &capacity, sizeof(capacity));
334 /* If capacity is too big, truncate with warning. */
335 if ((sector_t)capacity != capacity) {
336 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
337 (unsigned long long)capacity);
338 capacity = (sector_t)-1;
341 size = capacity * queue_logical_block_size(q);
342 string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
343 string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
345 dev_notice(&vdev->dev,
346 "new size: %llu %d-byte logical blocks (%s/%s)\n",
347 (unsigned long long)capacity,
348 queue_logical_block_size(q),
349 cap_str_10, cap_str_2);
351 set_capacity(vblk->disk, capacity);
353 mutex_unlock(&vblk->config_lock);
356 static void virtblk_config_changed(struct virtio_device *vdev)
358 struct virtio_blk *vblk = vdev->priv;
360 queue_work(virtblk_wq, &vblk->config_work);
363 static int __devinit virtblk_probe(struct virtio_device *vdev)
365 struct virtio_blk *vblk;
366 struct request_queue *q;
369 u32 v, blk_size, sg_elems, opt_io_size;
371 u8 physical_block_exp, alignment_offset;
373 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
379 /* We need to know how many segments before we allocate. */
380 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
381 offsetof(struct virtio_blk_config, seg_max),
384 /* We need at least one SG element, whatever they say. */
385 if (err || !sg_elems)
388 /* We need an extra sg elements at head and tail. */
390 vdev->priv = vblk = kmalloc(sizeof(*vblk) +
391 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
397 INIT_LIST_HEAD(&vblk->reqs);
399 vblk->sg_elems = sg_elems;
400 sg_init_table(vblk->sg, vblk->sg_elems);
401 mutex_init(&vblk->config_lock);
402 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
403 vblk->config_enable = true;
405 /* We expect one virtqueue, for output. */
406 vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
407 if (IS_ERR(vblk->vq)) {
408 err = PTR_ERR(vblk->vq);
412 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
418 /* FIXME: How many partitions? How long is a piece of string? */
419 vblk->disk = alloc_disk(1 << PART_BITS);
425 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, NULL);
434 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
435 } else if (index < (26 + 1) * 26) {
436 sprintf(vblk->disk->disk_name, "vd%c%c",
437 'a' + index / 26 - 1, 'a' + index % 26);
439 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
440 const unsigned int m2 = (index / 26 - 1) % 26;
441 const unsigned int m3 = index % 26;
442 sprintf(vblk->disk->disk_name, "vd%c%c%c",
443 'a' + m1, 'a' + m2, 'a' + m3);
446 vblk->disk->major = major;
447 vblk->disk->first_minor = index_to_minor(index);
448 vblk->disk->private_data = vblk;
449 vblk->disk->fops = &virtblk_fops;
450 vblk->disk->driverfs_dev = &vdev->dev;
453 /* configure queue flush support */
454 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
455 blk_queue_flush(q, REQ_FLUSH);
457 /* If disk is read-only in the host, the guest should obey */
458 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
459 set_disk_ro(vblk->disk, 1);
461 /* Host must always specify the capacity. */
462 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
465 /* If capacity is too big, truncate with warning. */
466 if ((sector_t)cap != cap) {
467 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
468 (unsigned long long)cap);
471 set_capacity(vblk->disk, cap);
473 /* We can handle whatever the host told us to handle. */
474 blk_queue_max_segments(q, vblk->sg_elems-2);
476 /* No need to bounce any requests */
477 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
479 /* No real sector limit. */
480 blk_queue_max_hw_sectors(q, -1U);
482 /* Host can optionally specify maximum segment size and number of
484 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
485 offsetof(struct virtio_blk_config, size_max),
488 blk_queue_max_segment_size(q, v);
490 blk_queue_max_segment_size(q, -1U);
492 /* Host can optionally specify the block size of the device */
493 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
494 offsetof(struct virtio_blk_config, blk_size),
497 blk_queue_logical_block_size(q, blk_size);
499 blk_size = queue_logical_block_size(q);
501 /* Use topology information if available */
502 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
503 offsetof(struct virtio_blk_config, physical_block_exp),
504 &physical_block_exp);
505 if (!err && physical_block_exp)
506 blk_queue_physical_block_size(q,
507 blk_size * (1 << physical_block_exp));
509 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
510 offsetof(struct virtio_blk_config, alignment_offset),
512 if (!err && alignment_offset)
513 blk_queue_alignment_offset(q, blk_size * alignment_offset);
515 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
516 offsetof(struct virtio_blk_config, min_io_size),
518 if (!err && min_io_size)
519 blk_queue_io_min(q, blk_size * min_io_size);
521 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
522 offsetof(struct virtio_blk_config, opt_io_size),
524 if (!err && opt_io_size)
525 blk_queue_io_opt(q, blk_size * opt_io_size);
528 add_disk(vblk->disk);
529 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
536 del_gendisk(vblk->disk);
537 blk_cleanup_queue(vblk->disk->queue);
539 put_disk(vblk->disk);
541 mempool_destroy(vblk->pool);
543 vdev->config->del_vqs(vdev);
547 ida_simple_remove(&vd_index_ida, index);
552 static void __devexit virtblk_remove(struct virtio_device *vdev)
554 struct virtio_blk *vblk = vdev->priv;
555 int index = vblk->index;
557 /* Prevent config work handler from accessing the device. */
558 mutex_lock(&vblk->config_lock);
559 vblk->config_enable = false;
560 mutex_unlock(&vblk->config_lock);
562 /* Nothing should be pending. */
563 BUG_ON(!list_empty(&vblk->reqs));
565 /* Stop all the virtqueues. */
566 vdev->config->reset(vdev);
568 flush_work(&vblk->config_work);
570 del_gendisk(vblk->disk);
571 blk_cleanup_queue(vblk->disk->queue);
572 put_disk(vblk->disk);
573 mempool_destroy(vblk->pool);
574 vdev->config->del_vqs(vdev);
576 ida_simple_remove(&vd_index_ida, index);
579 static const struct virtio_device_id id_table[] = {
580 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
584 static unsigned int features[] = {
585 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
586 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
587 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
591 * virtio_blk causes spurious section mismatch warning by
592 * simultaneously referring to a __devinit and a __devexit function.
593 * Use __refdata to avoid this warning.
595 static struct virtio_driver __refdata virtio_blk = {
596 .feature_table = features,
597 .feature_table_size = ARRAY_SIZE(features),
598 .driver.name = KBUILD_MODNAME,
599 .driver.owner = THIS_MODULE,
600 .id_table = id_table,
601 .probe = virtblk_probe,
602 .remove = __devexit_p(virtblk_remove),
603 .config_changed = virtblk_config_changed,
606 static int __init init(void)
610 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
614 major = register_blkdev(0, "virtblk");
617 goto out_destroy_workqueue;
620 error = register_virtio_driver(&virtio_blk);
622 goto out_unregister_blkdev;
625 out_unregister_blkdev:
626 unregister_blkdev(major, "virtblk");
627 out_destroy_workqueue:
628 destroy_workqueue(virtblk_wq);
632 static void __exit fini(void)
634 unregister_blkdev(major, "virtblk");
635 unregister_virtio_driver(&virtio_blk);
636 destroy_workqueue(virtblk_wq);
641 MODULE_DEVICE_TABLE(virtio, id_table);
642 MODULE_DESCRIPTION("Virtio block driver");
643 MODULE_LICENSE("GPL");