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 #define PART_BITS 4
  11
  12 static int major, index;
  13
  14 struct virtio_blk
  15 {
  16         spinlock_t lock;
  17
  18         struct virtio_device *vdev;
  19         struct virtqueue *vq;
  20
  21         /* The disk structure for the kernel. */
  22         struct gendisk *disk;
  23
  24         /* Request tracking. */
  25         struct list_head reqs;
  26
  27         mempool_t *pool;
  28
  29         /* Scatterlist: can be too big for stack. */
  30         struct scatterlist sg[VIRTIO_MAX_SG];
  31 };
  32
  33 struct virtblk_req
  34 {
  35         struct list_head list;
  36         struct request *req;
  37         struct virtio_blk_outhdr out_hdr;
  38         struct virtio_blk_inhdr in_hdr;
  39 };
  40
  41 static void blk_done(struct virtqueue *vq)
  42 {
  43         struct virtio_blk *vblk = vq->vdev->priv;
  44         struct virtblk_req *vbr;
  45         unsigned int len;
  46         unsigned long flags;
  47
  48         spin_lock_irqsave(&vblk->lock, flags);
  49         while ((vbr = vblk->vq->vq_ops->get_buf(vblk->vq, &len)) != NULL) {
  50                 int uptodate;
  51                 switch (vbr->in_hdr.status) {
  52                 case VIRTIO_BLK_S_OK:
  53                         uptodate = 1;
  54                         break;
  55                 case VIRTIO_BLK_S_UNSUPP:
  56                         uptodate = -ENOTTY;
  57                         break;
  58                 default:
  59                         uptodate = 0;
  60                         break;
  61                 }
  62
  63                 end_dequeued_request(vbr->req, uptodate);
  64                 list_del(&vbr->list);
  65                 mempool_free(vbr, vblk->pool);
  66         }
  67         /* In case queue is stopped waiting for more buffers. */
  68         blk_start_queue(vblk->disk->queue);
  69         spin_unlock_irqrestore(&vblk->lock, flags);
  70 }
  71
  72 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
  73                    struct request *req)
  74 {
  75         unsigned long num, out, in;
  76         struct virtblk_req *vbr;
  77
  78         vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
  79         if (!vbr)
  80                 /* When another request finishes we'll try again. */
  81                 return false;
  82
  83         vbr->req = req;
  84         if (blk_fs_request(vbr->req)) {
  85                 vbr->out_hdr.type = 0;
  86                 vbr->out_hdr.sector = vbr->req->sector;
  87                 vbr->out_hdr.ioprio = vbr->req->ioprio;
  88         } else if (blk_pc_request(vbr->req)) {
  89                 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
  90                 vbr->out_hdr.sector = 0;
  91                 vbr->out_hdr.ioprio = vbr->req->ioprio;
  92         } else {
  93                 /* We don't put anything else in the queue. */
  94                 BUG();
  95         }
  96
  97         if (blk_barrier_rq(vbr->req))
  98                 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
  99
 100         /* This init could be done at vblk creation time */
 101         sg_init_table(vblk->sg, VIRTIO_MAX_SG);
 102         sg_set_buf(&vblk->sg[0], &vbr->out_hdr, sizeof(vbr->out_hdr));
 103         num = blk_rq_map_sg(q, vbr->req, vblk->sg+1);
 104         sg_set_buf(&vblk->sg[num+1], &vbr->in_hdr, sizeof(vbr->in_hdr));
 105
 106         if (rq_data_dir(vbr->req) == WRITE) {
 107                 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 108                 out = 1 + num;
 109                 in = 1;
 110         } else {
 111                 vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 112                 out = 1;
 113                 in = 1 + num;
 114         }
 115
 116         if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr)) {
 117                 mempool_free(vbr, vblk->pool);
 118                 return false;
 119         }
 120
 121         list_add_tail(&vbr->list, &vblk->reqs);
 122         return true;
 123 }
 124
 125 static void do_virtblk_request(struct request_queue *q)
 126 {
 127         struct virtio_blk *vblk = NULL;
 128         struct request *req;
 129         unsigned int issued = 0;
 130
 131         while ((req = elv_next_request(q)) != NULL) {
 132                 vblk = req->rq_disk->private_data;
 133                 BUG_ON(req->nr_phys_segments > ARRAY_SIZE(vblk->sg));
 134
 135                 /* If this request fails, stop queue and wait for something to
 136                    finish to restart it. */
 137                 if (!do_req(q, vblk, req)) {
 138                         blk_stop_queue(q);
 139                         break;
 140                 }
 141                 blkdev_dequeue_request(req);
 142                 issued++;
 143         }
 144
 145         if (issued)
 146                 vblk->vq->vq_ops->kick(vblk->vq);
 147 }
 148
 149 static int virtblk_ioctl(struct inode *inode, struct file *filp,
 150                          unsigned cmd, unsigned long data)
 151 {
 152         return scsi_cmd_ioctl(filp, inode->i_bdev->bd_disk->queue,
 153                               inode->i_bdev->bd_disk, cmd,
 154                               (void __user *)data);
 155 }
 156
 157 /* We provide getgeo only to please some old bootloader/partitioning tools */
 158 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 159 {
 160         /* some standard values, similar to sd */
 161         geo->heads = 1 << 6;
 162         geo->sectors = 1 << 5;
 163         geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 164         return 0;
 165 }
 166
 167 static struct block_device_operations virtblk_fops = {
 168         .ioctl  = virtblk_ioctl,
 169         .owner  = THIS_MODULE,
 170         .getgeo = virtblk_getgeo,
 171 };
 172
 173 static int index_to_minor(int index)
 174 {
 175         return index << PART_BITS;
 176 }
 177
 178 static int virtblk_probe(struct virtio_device *vdev)
 179 {
 180         struct virtio_blk *vblk;
 181         int err;
 182         u64 cap;
 183         u32 v;
 184
 185         if (index_to_minor(index) >= 1 << MINORBITS)
 186                 return -ENOSPC;
 187
 188         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
 189         if (!vblk) {
 190                 err = -ENOMEM;
 191                 goto out;
 192         }
 193
 194         INIT_LIST_HEAD(&vblk->reqs);
 195         spin_lock_init(&vblk->lock);
 196         vblk->vdev = vdev;
 197
 198         /* We expect one virtqueue, for output. */
 199         vblk->vq = vdev->config->find_vq(vdev, 0, blk_done);
 200         if (IS_ERR(vblk->vq)) {
 201                 err = PTR_ERR(vblk->vq);
 202                 goto out_free_vblk;
 203         }
 204
 205         vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
 206         if (!vblk->pool) {
 207                 err = -ENOMEM;
 208                 goto out_free_vq;
 209         }
 210
 211         /* FIXME: How many partitions?  How long is a piece of string? */
 212         vblk->disk = alloc_disk(1 << PART_BITS);
 213         if (!vblk->disk) {
 214                 err = -ENOMEM;
 215                 goto out_mempool;
 216         }
 217
 218         vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
 219         if (!vblk->disk->queue) {
 220                 err = -ENOMEM;
 221                 goto out_put_disk;
 222         }
 223
 224         if (index < 26) {
 225                 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
 226         } else if (index < (26 + 1) * 26) {
 227                 sprintf(vblk->disk->disk_name, "vd%c%c",
 228                         'a' + index / 26 - 1, 'a' + index % 26);
 229         } else {
 230                 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
 231                 const unsigned int m2 = (index / 26 - 1) % 26;
 232                 const unsigned int m3 =  index % 26;
 233                 sprintf(vblk->disk->disk_name, "vd%c%c%c",
 234                         'a' + m1, 'a' + m2, 'a' + m3);
 235         }
 236
 237         vblk->disk->major = major;
 238         vblk->disk->first_minor = index_to_minor(index);
 239         vblk->disk->private_data = vblk;
 240         vblk->disk->fops = &virtblk_fops;
 241         index++;
 242
 243         /* If barriers are supported, tell block layer that queue is ordered */
 244         if (vdev->config->feature(vdev, VIRTIO_BLK_F_BARRIER))
 245                 blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL);
 246
 247         /* Host must always specify the capacity. */
 248         __virtio_config_val(vdev, offsetof(struct virtio_blk_config, capacity),
 249                             &cap);
 250
 251         /* If capacity is too big, truncate with warning. */
 252         if ((sector_t)cap != cap) {
 253                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 254                          (unsigned long long)cap);
 255                 cap = (sector_t)-1;
 256         }
 257         set_capacity(vblk->disk, cap);
 258
 259         /* Host can optionally specify maximum segment size and number of
 260          * segments. */
 261         err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
 262                                 offsetof(struct virtio_blk_config, size_max),
 263                                 &v);
 264         if (!err)
 265                 blk_queue_max_segment_size(vblk->disk->queue, v);
 266
 267         err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
 268                                 offsetof(struct virtio_blk_config, seg_max),
 269                                 &v);
 270         if (!err)
 271                 blk_queue_max_hw_segments(vblk->disk->queue, v);
 272
 273         add_disk(vblk->disk);
 274         return 0;
 275
 276 out_put_disk:
 277         put_disk(vblk->disk);
 278 out_mempool:
 279         mempool_destroy(vblk->pool);
 280 out_free_vq:
 281         vdev->config->del_vq(vblk->vq);
 282 out_free_vblk:
 283         kfree(vblk);
 284 out:
 285         return err;
 286 }
 287
 288 static void virtblk_remove(struct virtio_device *vdev)
 289 {
 290         struct virtio_blk *vblk = vdev->priv;
 291         int major = vblk->disk->major;
 292
 293         /* Nothing should be pending. */
 294         BUG_ON(!list_empty(&vblk->reqs));
 295
 296         /* Stop all the virtqueues. */
 297         vdev->config->reset(vdev);
 298
 299         blk_cleanup_queue(vblk->disk->queue);
 300         put_disk(vblk->disk);
 301         unregister_blkdev(major, "virtblk");
 302         mempool_destroy(vblk->pool);
 303         vdev->config->del_vq(vblk->vq);
 304         kfree(vblk);
 305 }
 306
 307 static struct virtio_device_id id_table[] = {
 308         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 309         { 0 },
 310 };
 311
 312 static struct virtio_driver virtio_blk = {
 313         .driver.name =  KBUILD_MODNAME,
 314         .driver.owner = THIS_MODULE,
 315         .id_table =     id_table,
 316         .probe =        virtblk_probe,
 317         .remove =       __devexit_p(virtblk_remove),
 318 };
 319
 320 static int __init init(void)
 321 {
 322         major = register_blkdev(0, "virtblk");
 323         if (major < 0)
 324                 return major;
 325         return register_virtio_driver(&virtio_blk);
 326 }
 327
 328 static void __exit fini(void)
 329 {
 330         unregister_blkdev(major, "virtblk");
 331         unregister_virtio_driver(&virtio_blk);
 332 }
 333 module_init(init);
 334 module_exit(fini);
 335
 336 MODULE_DEVICE_TABLE(virtio, id_table);
 337 MODULE_DESCRIPTION("Virtio block driver");
 338 MODULE_LICENSE("GPL");