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>
14 #include <linux/blk-mq.h>
15 #include <linux/numa.h>
20 static DEFINE_IDA(vd_index_ida);
22 static struct workqueue_struct *virtblk_wq;
26 struct virtio_device *vdev;
30 /* The disk structure for the kernel. */
33 /* Process context for config space updates */
34 struct work_struct config_work;
36 /* Lock for config space updates */
37 struct mutex config_lock;
39 /* enable config space updates */
42 /* What host tells us, plus 2 for header & tailer. */
43 unsigned int sg_elems;
45 /* Ida index - used to track minor number allocations. */
52 struct virtio_blk_outhdr out_hdr;
53 struct virtio_scsi_inhdr in_hdr;
55 struct scatterlist sg[];
58 static inline int virtblk_result(struct virtblk_req *vbr)
60 switch (vbr->status) {
63 case VIRTIO_BLK_S_UNSUPP:
70 static int __virtblk_add_req(struct virtqueue *vq,
71 struct virtblk_req *vbr,
72 struct scatterlist *data_sg,
75 struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
76 unsigned int num_out = 0, num_in = 0;
77 int type = vbr->out_hdr.type & ~VIRTIO_BLK_T_OUT;
79 sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
80 sgs[num_out++] = &hdr;
83 * If this is a packet command we need a couple of additional headers.
84 * Behind the normal outhdr we put a segment with the scsi command
85 * block, and before the normal inhdr we put the sense data and the
86 * inhdr with additional status information.
88 if (type == VIRTIO_BLK_T_SCSI_CMD) {
89 sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len);
90 sgs[num_out++] = &cmd;
94 if (vbr->out_hdr.type & VIRTIO_BLK_T_OUT)
95 sgs[num_out++] = data_sg;
97 sgs[num_out + num_in++] = data_sg;
100 if (type == VIRTIO_BLK_T_SCSI_CMD) {
101 sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
102 sgs[num_out + num_in++] = &sense;
103 sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
104 sgs[num_out + num_in++] = &inhdr;
107 sg_init_one(&status, &vbr->status, sizeof(vbr->status));
108 sgs[num_out + num_in++] = &status;
110 return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
113 static inline void virtblk_request_done(struct request *req)
115 struct virtblk_req *vbr = req->special;
116 int error = virtblk_result(vbr);
118 if (req->cmd_type == REQ_TYPE_BLOCK_PC) {
119 req->resid_len = vbr->in_hdr.residual;
120 req->sense_len = vbr->in_hdr.sense_len;
121 req->errors = vbr->in_hdr.errors;
122 } else if (req->cmd_type == REQ_TYPE_SPECIAL) {
123 req->errors = (error != 0);
126 blk_mq_end_io(req, error);
129 static void virtblk_done(struct virtqueue *vq)
131 struct virtio_blk *vblk = vq->vdev->priv;
132 bool req_done = false;
133 struct virtblk_req *vbr;
137 spin_lock_irqsave(&vblk->vq_lock, flags);
139 virtqueue_disable_cb(vq);
140 while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
141 blk_mq_complete_request(vbr->req);
144 if (unlikely(virtqueue_is_broken(vq)))
146 } while (!virtqueue_enable_cb(vq));
147 spin_unlock_irqrestore(&vblk->vq_lock, flags);
149 /* In case queue is stopped waiting for more buffers. */
151 blk_mq_start_stopped_hw_queues(vblk->disk->queue);
154 static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req)
156 struct virtio_blk *vblk = hctx->queue->queuedata;
157 struct virtblk_req *vbr = req->special;
160 const bool last = (req->cmd_flags & REQ_END) != 0;
162 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
165 if (req->cmd_flags & REQ_FLUSH) {
166 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
167 vbr->out_hdr.sector = 0;
168 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
170 switch (req->cmd_type) {
172 vbr->out_hdr.type = 0;
173 vbr->out_hdr.sector = blk_rq_pos(vbr->req);
174 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
176 case REQ_TYPE_BLOCK_PC:
177 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
178 vbr->out_hdr.sector = 0;
179 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
181 case REQ_TYPE_SPECIAL:
182 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
183 vbr->out_hdr.sector = 0;
184 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
187 /* We don't put anything else in the queue. */
192 num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg);
194 if (rq_data_dir(vbr->req) == WRITE)
195 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
197 vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
200 spin_lock_irqsave(&vblk->vq_lock, flags);
201 if (__virtblk_add_req(vblk->vq, vbr, vbr->sg, num) < 0) {
202 virtqueue_kick(vblk->vq);
203 spin_unlock_irqrestore(&vblk->vq_lock, flags);
204 blk_mq_stop_hw_queue(hctx);
205 return BLK_MQ_RQ_QUEUE_BUSY;
209 virtqueue_kick(vblk->vq);
211 spin_unlock_irqrestore(&vblk->vq_lock, flags);
212 return BLK_MQ_RQ_QUEUE_OK;
215 /* return id (s/n) string for *disk to *id_str
217 static int virtblk_get_id(struct gendisk *disk, char *id_str)
219 struct virtio_blk *vblk = disk->private_data;
224 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
229 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
235 req->cmd_type = REQ_TYPE_SPECIAL;
236 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
237 blk_put_request(req);
242 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
243 unsigned int cmd, unsigned long data)
245 struct gendisk *disk = bdev->bd_disk;
246 struct virtio_blk *vblk = disk->private_data;
249 * Only allow the generic SCSI ioctls if the host can support it.
251 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
254 return scsi_cmd_blk_ioctl(bdev, mode, cmd,
255 (void __user *)data);
258 /* We provide getgeo only to please some old bootloader/partitioning tools */
259 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
261 struct virtio_blk *vblk = bd->bd_disk->private_data;
263 /* see if the host passed in geometry config */
264 if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
265 virtio_cread(vblk->vdev, struct virtio_blk_config,
266 geometry.cylinders, &geo->cylinders);
267 virtio_cread(vblk->vdev, struct virtio_blk_config,
268 geometry.heads, &geo->heads);
269 virtio_cread(vblk->vdev, struct virtio_blk_config,
270 geometry.sectors, &geo->sectors);
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];
324 char *envp[] = { "RESIZE=1", NULL };
327 mutex_lock(&vblk->config_lock);
328 if (!vblk->config_enable)
331 /* Host must always specify the capacity. */
332 virtio_cread(vdev, struct virtio_blk_config, capacity, &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);
352 revalidate_disk(vblk->disk);
353 kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
355 mutex_unlock(&vblk->config_lock);
358 static void virtblk_config_changed(struct virtio_device *vdev)
360 struct virtio_blk *vblk = vdev->priv;
362 queue_work(virtblk_wq, &vblk->config_work);
365 static int init_vq(struct virtio_blk *vblk)
369 /* We expect one virtqueue, for output. */
370 vblk->vq = virtio_find_single_vq(vblk->vdev, virtblk_done, "requests");
371 if (IS_ERR(vblk->vq))
372 err = PTR_ERR(vblk->vq);
378 * Legacy naming scheme used for virtio devices. We are stuck with it for
379 * virtio blk but don't ever use it for any new driver.
381 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
383 const int base = 'z' - 'a' + 1;
384 char *begin = buf + strlen(prefix);
385 char *end = buf + buflen;
395 *--p = 'a' + (index % unit);
396 index = (index / unit) - 1;
397 } while (index >= 0);
399 memmove(begin, p, end - p);
400 memcpy(buf, prefix, strlen(prefix));
405 static int virtblk_get_cache_mode(struct virtio_device *vdev)
410 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
411 struct virtio_blk_config, wce,
414 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE);
419 static void virtblk_update_cache_mode(struct virtio_device *vdev)
421 u8 writeback = virtblk_get_cache_mode(vdev);
422 struct virtio_blk *vblk = vdev->priv;
425 blk_queue_flush(vblk->disk->queue, REQ_FLUSH);
427 blk_queue_flush(vblk->disk->queue, 0);
429 revalidate_disk(vblk->disk);
432 static const char *const virtblk_cache_types[] = {
433 "write through", "write back"
437 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
438 const char *buf, size_t count)
440 struct gendisk *disk = dev_to_disk(dev);
441 struct virtio_blk *vblk = disk->private_data;
442 struct virtio_device *vdev = vblk->vdev;
445 BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
446 for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
447 if (sysfs_streq(buf, virtblk_cache_types[i]))
453 virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
454 virtblk_update_cache_mode(vdev);
459 virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
462 struct gendisk *disk = dev_to_disk(dev);
463 struct virtio_blk *vblk = disk->private_data;
464 u8 writeback = virtblk_get_cache_mode(vblk->vdev);
466 BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
467 return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
470 static const struct device_attribute dev_attr_cache_type_ro =
471 __ATTR(cache_type, S_IRUGO,
472 virtblk_cache_type_show, NULL);
473 static const struct device_attribute dev_attr_cache_type_rw =
474 __ATTR(cache_type, S_IRUGO|S_IWUSR,
475 virtblk_cache_type_show, virtblk_cache_type_store);
477 static struct blk_mq_ops virtio_mq_ops = {
478 .queue_rq = virtio_queue_rq,
479 .map_queue = blk_mq_map_queue,
480 .alloc_hctx = blk_mq_alloc_single_hw_queue,
481 .free_hctx = blk_mq_free_single_hw_queue,
482 .complete = virtblk_request_done,
485 static struct blk_mq_reg virtio_mq_reg = {
486 .ops = &virtio_mq_ops,
489 .numa_node = NUMA_NO_NODE,
490 .flags = BLK_MQ_F_SHOULD_MERGE,
493 static void virtblk_init_vbr(void *data, struct blk_mq_hw_ctx *hctx,
494 struct request *rq, unsigned int nr)
496 struct virtio_blk *vblk = data;
497 struct virtblk_req *vbr = rq->special;
499 sg_init_table(vbr->sg, vblk->sg_elems);
502 static int virtblk_probe(struct virtio_device *vdev)
504 struct virtio_blk *vblk;
505 struct request_queue *q;
509 u32 v, blk_size, sg_elems, opt_io_size;
511 u8 physical_block_exp, alignment_offset;
513 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
519 /* We need to know how many segments before we allocate. */
520 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
521 struct virtio_blk_config, seg_max,
524 /* We need at least one SG element, whatever they say. */
525 if (err || !sg_elems)
528 /* We need an extra sg elements at head and tail. */
530 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
537 vblk->sg_elems = sg_elems;
538 mutex_init(&vblk->config_lock);
540 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
541 vblk->config_enable = true;
546 spin_lock_init(&vblk->vq_lock);
548 /* FIXME: How many partitions? How long is a piece of string? */
549 vblk->disk = alloc_disk(1 << PART_BITS);
555 virtio_mq_reg.cmd_size =
556 sizeof(struct virtblk_req) +
557 sizeof(struct scatterlist) * sg_elems;
559 q = vblk->disk->queue = blk_mq_init_queue(&virtio_mq_reg, vblk);
565 blk_mq_init_commands(q, virtblk_init_vbr, vblk);
569 virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
571 vblk->disk->major = major;
572 vblk->disk->first_minor = index_to_minor(index);
573 vblk->disk->private_data = vblk;
574 vblk->disk->fops = &virtblk_fops;
575 vblk->disk->driverfs_dev = &vdev->dev;
578 /* configure queue flush support */
579 virtblk_update_cache_mode(vdev);
581 /* If disk is read-only in the host, the guest should obey */
582 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
583 set_disk_ro(vblk->disk, 1);
585 /* Host must always specify the capacity. */
586 virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
588 /* If capacity is too big, truncate with warning. */
589 if ((sector_t)cap != cap) {
590 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
591 (unsigned long long)cap);
594 set_capacity(vblk->disk, cap);
596 /* We can handle whatever the host told us to handle. */
597 blk_queue_max_segments(q, vblk->sg_elems-2);
599 /* No need to bounce any requests */
600 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
602 /* No real sector limit. */
603 blk_queue_max_hw_sectors(q, -1U);
605 /* Host can optionally specify maximum segment size and number of
607 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
608 struct virtio_blk_config, size_max, &v);
610 blk_queue_max_segment_size(q, v);
612 blk_queue_max_segment_size(q, -1U);
614 /* Host can optionally specify the block size of the device */
615 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
616 struct virtio_blk_config, blk_size,
619 blk_queue_logical_block_size(q, blk_size);
621 blk_size = queue_logical_block_size(q);
623 /* Use topology information if available */
624 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
625 struct virtio_blk_config, physical_block_exp,
626 &physical_block_exp);
627 if (!err && physical_block_exp)
628 blk_queue_physical_block_size(q,
629 blk_size * (1 << physical_block_exp));
631 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
632 struct virtio_blk_config, alignment_offset,
634 if (!err && alignment_offset)
635 blk_queue_alignment_offset(q, blk_size * alignment_offset);
637 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
638 struct virtio_blk_config, min_io_size,
640 if (!err && min_io_size)
641 blk_queue_io_min(q, blk_size * min_io_size);
643 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
644 struct virtio_blk_config, opt_io_size,
646 if (!err && opt_io_size)
647 blk_queue_io_opt(q, blk_size * opt_io_size);
649 add_disk(vblk->disk);
650 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
654 if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
655 err = device_create_file(disk_to_dev(vblk->disk),
656 &dev_attr_cache_type_rw);
658 err = device_create_file(disk_to_dev(vblk->disk),
659 &dev_attr_cache_type_ro);
665 del_gendisk(vblk->disk);
666 blk_cleanup_queue(vblk->disk->queue);
668 put_disk(vblk->disk);
670 vdev->config->del_vqs(vdev);
674 ida_simple_remove(&vd_index_ida, index);
679 static void virtblk_remove(struct virtio_device *vdev)
681 struct virtio_blk *vblk = vdev->priv;
682 int index = vblk->index;
685 /* Prevent config work handler from accessing the device. */
686 mutex_lock(&vblk->config_lock);
687 vblk->config_enable = false;
688 mutex_unlock(&vblk->config_lock);
690 del_gendisk(vblk->disk);
691 blk_cleanup_queue(vblk->disk->queue);
693 /* Stop all the virtqueues. */
694 vdev->config->reset(vdev);
696 flush_work(&vblk->config_work);
698 refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount);
699 put_disk(vblk->disk);
700 vdev->config->del_vqs(vdev);
703 /* Only free device id if we don't have any users */
705 ida_simple_remove(&vd_index_ida, index);
708 #ifdef CONFIG_PM_SLEEP
709 static int virtblk_freeze(struct virtio_device *vdev)
711 struct virtio_blk *vblk = vdev->priv;
713 /* Ensure we don't receive any more interrupts */
714 vdev->config->reset(vdev);
716 /* Prevent config work handler from accessing the device. */
717 mutex_lock(&vblk->config_lock);
718 vblk->config_enable = false;
719 mutex_unlock(&vblk->config_lock);
721 flush_work(&vblk->config_work);
723 blk_mq_stop_hw_queues(vblk->disk->queue);
725 vdev->config->del_vqs(vdev);
729 static int virtblk_restore(struct virtio_device *vdev)
731 struct virtio_blk *vblk = vdev->priv;
734 vblk->config_enable = true;
735 ret = init_vq(vdev->priv);
737 blk_mq_start_stopped_hw_queues(vblk->disk->queue);
743 static const struct virtio_device_id id_table[] = {
744 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
748 static unsigned int features[] = {
749 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
750 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
751 VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE
754 static struct virtio_driver virtio_blk = {
755 .feature_table = features,
756 .feature_table_size = ARRAY_SIZE(features),
757 .driver.name = KBUILD_MODNAME,
758 .driver.owner = THIS_MODULE,
759 .id_table = id_table,
760 .probe = virtblk_probe,
761 .remove = virtblk_remove,
762 .config_changed = virtblk_config_changed,
763 #ifdef CONFIG_PM_SLEEP
764 .freeze = virtblk_freeze,
765 .restore = virtblk_restore,
769 static int __init init(void)
773 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
777 major = register_blkdev(0, "virtblk");
780 goto out_destroy_workqueue;
783 error = register_virtio_driver(&virtio_blk);
785 goto out_unregister_blkdev;
788 out_unregister_blkdev:
789 unregister_blkdev(major, "virtblk");
790 out_destroy_workqueue:
791 destroy_workqueue(virtblk_wq);
795 static void __exit fini(void)
797 unregister_blkdev(major, "virtblk");
798 unregister_virtio_driver(&virtio_blk);
799 destroy_workqueue(virtblk_wq);
804 MODULE_DEVICE_TABLE(virtio, id_table);
805 MODULE_DESCRIPTION("Virtio block driver");
806 MODULE_LICENSE("GPL");