drivers/block/virtio_blk.c

   1 //#define DEBUG
   2 #include <linux/spinlock.h>
   3 #include <linux/blkdev.h>
   4 #include <linux/hdreg.h>
   5 #include <linux/virtio.h>
   6 #include <linux/virtio_blk.h>
   7 #include <linux/scatterlist.h>
   8
   9 #define VIRTIO_MAX_SG   (3+MAX_PHYS_SEGMENTS)
  10
  11 static unsigned char virtblk_index = 'a';
  12 struct virtio_blk
  13 {
  14         spinlock_t lock;
  15
  16         struct virtio_device *vdev;
  17         struct virtqueue *vq;
  18
  19         /* The disk structure for the kernel. */
  20         struct gendisk *disk;
  21
  22         /* Request tracking. */
  23         struct list_head reqs;
  24
  25         mempool_t *pool;
  26
  27         /* Scatterlist: can be too big for stack. */
  28         struct scatterlist sg[VIRTIO_MAX_SG];
  29 };
  30
  31 struct virtblk_req
  32 {
  33         struct list_head list;
  34         struct request *req;
  35         struct virtio_blk_outhdr out_hdr;
  36         struct virtio_blk_inhdr in_hdr;
  37 };
  38
  39 static bool blk_done(struct virtqueue *vq)
  40 {
  41         struct virtio_blk *vblk = vq->vdev->priv;
  42         struct virtblk_req *vbr;
  43         unsigned int len;
  44         unsigned long flags;
  45
  46         spin_lock_irqsave(&vblk->lock, flags);
  47         while ((vbr = vblk->vq->vq_ops->get_buf(vblk->vq, &len)) != NULL) {
  48                 int uptodate;
  49                 switch (vbr->in_hdr.status) {
  50                 case VIRTIO_BLK_S_OK:
  51                         uptodate = 1;
  52                         break;
  53                 case VIRTIO_BLK_S_UNSUPP:
  54                         uptodate = -ENOTTY;
  55                         break;
  56                 default:
  57                         uptodate = 0;
  58                         break;
  59                 }
  60
  61                 end_dequeued_request(vbr->req, uptodate);
  62                 list_del(&vbr->list);
  63                 mempool_free(vbr, vblk->pool);
  64         }
  65         /* In case queue is stopped waiting for more buffers. */
  66         blk_start_queue(vblk->disk->queue);
  67         spin_unlock_irqrestore(&vblk->lock, flags);
  68         return true;
  69 }
  70
  71 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
  72                    struct request *req)
  73 {
  74         unsigned long num, out, in;
  75         struct virtblk_req *vbr;
  76
  77         vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
  78         if (!vbr)
  79                 /* When another request finishes we'll try again. */
  80                 return false;
  81
  82         vbr->req = req;
  83         if (blk_fs_request(vbr->req)) {
  84                 vbr->out_hdr.type = 0;
  85                 vbr->out_hdr.sector = vbr->req->sector;
  86                 vbr->out_hdr.ioprio = vbr->req->ioprio;
  87         } else if (blk_pc_request(vbr->req)) {
  88                 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
  89                 vbr->out_hdr.sector = 0;
  90                 vbr->out_hdr.ioprio = vbr->req->ioprio;
  91         } else {
  92                 /* We don't put anything else in the queue. */
  93                 BUG();
  94         }
  95
  96         if (blk_barrier_rq(vbr->req))
  97                 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
  98
  99         /* This init could be done at vblk creation time */
 100         sg_init_table(vblk->sg, VIRTIO_MAX_SG);
 101         sg_set_buf(&vblk->sg[0], &vbr->out_hdr, sizeof(vbr->out_hdr));
 102         num = blk_rq_map_sg(q, vbr->req, vblk->sg+1);
 103         sg_set_buf(&vblk->sg[num+1], &vbr->in_hdr, sizeof(vbr->in_hdr));
 104
 105         if (rq_data_dir(vbr->req) == WRITE) {
 106                 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 107                 out = 1 + num;
 108                 in = 1;
 109         } else {
 110                 vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 111                 out = 1;
 112                 in = 1 + num;
 113         }
 114
 115         if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr)) {
 116                 mempool_free(vbr, vblk->pool);
 117                 return false;
 118         }
 119
 120         list_add_tail(&vbr->list, &vblk->reqs);
 121         return true;
 122 }
 123
 124 static void do_virtblk_request(struct request_queue *q)
 125 {
 126         struct virtio_blk *vblk = NULL;
 127         struct request *req;
 128         unsigned int issued = 0;
 129
 130         while ((req = elv_next_request(q)) != NULL) {
 131                 vblk = req->rq_disk->private_data;
 132                 BUG_ON(req->nr_phys_segments > ARRAY_SIZE(vblk->sg));
 133
 134                 /* If this request fails, stop queue and wait for something to
 135                    finish to restart it. */
 136                 if (!do_req(q, vblk, req)) {
 137                         blk_stop_queue(q);
 138                         break;
 139                 }
 140                 blkdev_dequeue_request(req);
 141                 issued++;
 142         }
 143
 144         if (issued)
 145                 vblk->vq->vq_ops->kick(vblk->vq);
 146 }
 147
 148 static int virtblk_ioctl(struct inode *inode, struct file *filp,
 149                          unsigned cmd, unsigned long data)
 150 {
 151         return scsi_cmd_ioctl(filp, inode->i_bdev->bd_disk->queue,
 152                               inode->i_bdev->bd_disk, cmd,
 153                               (void __user *)data);
 154 }
 155
 156 static struct block_device_operations virtblk_fops = {
 157         .ioctl = virtblk_ioctl,
 158         .owner = THIS_MODULE,
 159 };
 160
 161 static int virtblk_probe(struct virtio_device *vdev)
 162 {
 163         struct virtio_blk *vblk;
 164         int err, major;
 165         void *token;
 166         unsigned int len;
 167         u64 cap;
 168         u32 v;
 169
 170         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
 171         if (!vblk) {
 172                 err = -ENOMEM;
 173                 goto out;
 174         }
 175
 176         INIT_LIST_HEAD(&vblk->reqs);
 177         spin_lock_init(&vblk->lock);
 178         vblk->vdev = vdev;
 179
 180         /* We expect one virtqueue, for output. */
 181         vblk->vq = vdev->config->find_vq(vdev, blk_done);
 182         if (IS_ERR(vblk->vq)) {
 183                 err = PTR_ERR(vblk->vq);
 184                 goto out_free_vblk;
 185         }
 186
 187         vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
 188         if (!vblk->pool) {
 189                 err = -ENOMEM;
 190                 goto out_free_vq;
 191         }
 192
 193         major = register_blkdev(0, "virtblk");
 194         if (major < 0) {
 195                 err = major;
 196                 goto out_mempool;
 197         }
 198
 199         /* FIXME: How many partitions?  How long is a piece of string? */
 200         vblk->disk = alloc_disk(1 << 4);
 201         if (!vblk->disk) {
 202                 err = -ENOMEM;
 203                 goto out_unregister_blkdev;
 204         }
 205
 206         vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
 207         if (!vblk->disk->queue) {
 208                 err = -ENOMEM;
 209                 goto out_put_disk;
 210         }
 211
 212         sprintf(vblk->disk->disk_name, "vd%c", virtblk_index++);
 213         vblk->disk->major = major;
 214         vblk->disk->first_minor = 0;
 215         vblk->disk->private_data = vblk;
 216         vblk->disk->fops = &virtblk_fops;
 217
 218         /* If barriers are supported, tell block layer that queue is ordered */
 219         token = vdev->config->find(vdev, VIRTIO_CONFIG_BLK_F, &len);
 220         if (virtio_use_bit(vdev, token, len, VIRTIO_BLK_F_BARRIER))
 221                 blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL);
 222
 223         err = virtio_config_val(vdev, VIRTIO_CONFIG_BLK_F_CAPACITY, &cap);
 224         if (err) {
 225                 dev_err(&vdev->dev, "Bad/missing capacity in config\n");
 226                 goto out_cleanup_queue;
 227         }
 228
 229         /* If capacity is too big, truncate with warning. */
 230         if ((sector_t)cap != cap) {
 231                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 232                          (unsigned long long)cap);
 233                 cap = (sector_t)-1;
 234         }
 235         set_capacity(vblk->disk, cap);
 236
 237         err = virtio_config_val(vdev, VIRTIO_CONFIG_BLK_F_SIZE_MAX, &v);
 238         if (!err)
 239                 blk_queue_max_segment_size(vblk->disk->queue, v);
 240         else if (err != -ENOENT) {
 241                 dev_err(&vdev->dev, "Bad SIZE_MAX in config\n");
 242                 goto out_cleanup_queue;
 243         }
 244
 245         err = virtio_config_val(vdev, VIRTIO_CONFIG_BLK_F_SEG_MAX, &v);
 246         if (!err)
 247                 blk_queue_max_hw_segments(vblk->disk->queue, v);
 248         else if (err != -ENOENT) {
 249                 dev_err(&vdev->dev, "Bad SEG_MAX in config\n");
 250                 goto out_cleanup_queue;
 251         }
 252
 253         add_disk(vblk->disk);
 254         return 0;
 255
 256 out_cleanup_queue:
 257         blk_cleanup_queue(vblk->disk->queue);
 258 out_put_disk:
 259         put_disk(vblk->disk);
 260 out_unregister_blkdev:
 261         unregister_blkdev(major, "virtblk");
 262 out_mempool:
 263         mempool_destroy(vblk->pool);
 264 out_free_vq:
 265         vdev->config->del_vq(vblk->vq);
 266 out_free_vblk:
 267         kfree(vblk);
 268 out:
 269         return err;
 270 }
 271
 272 static void virtblk_remove(struct virtio_device *vdev)
 273 {
 274         struct virtio_blk *vblk = vdev->priv;
 275         int major = vblk->disk->major;
 276
 277         BUG_ON(!list_empty(&vblk->reqs));
 278         blk_cleanup_queue(vblk->disk->queue);
 279         put_disk(vblk->disk);
 280         unregister_blkdev(major, "virtblk");
 281         mempool_destroy(vblk->pool);
 282         /* There should be nothing in the queue now, so no need to shutdown */
 283         vdev->config->del_vq(vblk->vq);
 284         kfree(vblk);
 285 }
 286
 287 static struct virtio_device_id id_table[] = {
 288         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 289         { 0 },
 290 };
 291
 292 static struct virtio_driver virtio_blk = {
 293         .driver.name =  KBUILD_MODNAME,
 294         .driver.owner = THIS_MODULE,
 295         .id_table =     id_table,
 296         .probe =        virtblk_probe,
 297         .remove =       __devexit_p(virtblk_remove),
 298 };
 299
 300 static int __init init(void)
 301 {
 302         return register_virtio_driver(&virtio_blk);
 303 }
 304
 305 static void __exit fini(void)
 306 {
 307         unregister_virtio_driver(&virtio_blk);
 308 }
 309 module_init(init);
 310 module_exit(fini);
 311
 312 MODULE_DEVICE_TABLE(virtio, id_table);
 313 MODULE_DESCRIPTION("Virtio block driver");
 314 MODULE_LICENSE("GPL");