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 PART_BITS 4
  10
  11 static int major, index;
  12
  13 struct virtio_blk
  14 {
  15         spinlock_t lock;
  16
  17         struct virtio_device *vdev;
  18         struct virtqueue *vq;
  19
  20         /* The disk structure for the kernel. */
  21         struct gendisk *disk;
  22
  23         /* Request tracking. */
  24         struct list_head reqs;
  25
  26         mempool_t *pool;
  27
  28         /* What host tells us, plus 2 for header & tailer. */
  29         unsigned int sg_elems;
  30
  31         /* Scatterlist: can be too big for stack. */
  32         struct scatterlist sg[/*sg_elems*/];
  33 };
  34
  35 struct virtblk_req
  36 {
  37         struct list_head list;
  38         struct request *req;
  39         struct virtio_blk_outhdr out_hdr;
  40         u8 status;
  41 };
  42
  43 static void blk_done(struct virtqueue *vq)
  44 {
  45         struct virtio_blk *vblk = vq->vdev->priv;
  46         struct virtblk_req *vbr;
  47         unsigned int len;
  48         unsigned long flags;
  49
  50         spin_lock_irqsave(&vblk->lock, flags);
  51         while ((vbr = vblk->vq->vq_ops->get_buf(vblk->vq, &len)) != NULL) {
  52                 int error;
  53                 switch (vbr->status) {
  54                 case VIRTIO_BLK_S_OK:
  55                         error = 0;
  56                         break;
  57                 case VIRTIO_BLK_S_UNSUPP:
  58                         error = -ENOTTY;
  59                         break;
  60                 default:
  61                         error = -EIO;
  62                         break;
  63                 }
  64
  65                 __blk_end_request(vbr->req, error, blk_rq_bytes(vbr->req));
  66                 list_del(&vbr->list);
  67                 mempool_free(vbr, vblk->pool);
  68         }
  69         /* In case queue is stopped waiting for more buffers. */
  70         blk_start_queue(vblk->disk->queue);
  71         spin_unlock_irqrestore(&vblk->lock, flags);
  72 }
  73
  74 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
  75                    struct request *req)
  76 {
  77         unsigned long num, out, in;
  78         struct virtblk_req *vbr;
  79
  80         vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
  81         if (!vbr)
  82                 /* When another request finishes we'll try again. */
  83                 return false;
  84
  85         vbr->req = req;
  86         if (blk_fs_request(vbr->req)) {
  87                 vbr->out_hdr.type = 0;
  88                 vbr->out_hdr.sector = vbr->req->sector;
  89                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
  90         } else if (blk_pc_request(vbr->req)) {
  91                 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
  92                 vbr->out_hdr.sector = 0;
  93                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
  94         } else {
  95                 /* We don't put anything else in the queue. */
  96                 BUG();
  97         }
  98
  99         if (blk_barrier_rq(vbr->req))
 100                 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
 101
 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->status, sizeof(vbr->status));
 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 + 2 > vblk->sg_elems);
 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 block_device *bdev, fmode_t mode,
 150                          unsigned cmd, unsigned long data)
 151 {
 152         return scsi_cmd_ioctl(bdev->bd_disk->queue,
 153                               bdev->bd_disk, mode, 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         struct virtio_blk *vblk = bd->bd_disk->private_data;
 161         struct virtio_blk_geometry vgeo;
 162         int err;
 163
 164         /* see if the host passed in geometry config */
 165         err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
 166                                 offsetof(struct virtio_blk_config, geometry),
 167                                 &vgeo);
 168
 169         if (!err) {
 170                 geo->heads = vgeo.heads;
 171                 geo->sectors = vgeo.sectors;
 172                 geo->cylinders = vgeo.cylinders;
 173         } else {
 174                 /* some standard values, similar to sd */
 175                 geo->heads = 1 << 6;
 176                 geo->sectors = 1 << 5;
 177                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 178         }
 179         return 0;
 180 }
 181
 182 static struct block_device_operations virtblk_fops = {
 183         .locked_ioctl = virtblk_ioctl,
 184         .owner  = THIS_MODULE,
 185         .getgeo = virtblk_getgeo,
 186 };
 187
 188 static int index_to_minor(int index)
 189 {
 190         return index << PART_BITS;
 191 }
 192
 193 static int virtblk_probe(struct virtio_device *vdev)
 194 {
 195         struct virtio_blk *vblk;
 196         int err;
 197         u64 cap;
 198         u32 v;
 199         u32 blk_size, sg_elems;
 200
 201         if (index_to_minor(index) >= 1 << MINORBITS)
 202                 return -ENOSPC;
 203
 204         /* We need to know how many segments before we allocate. */
 205         err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
 206                                 offsetof(struct virtio_blk_config, seg_max),
 207                                 &sg_elems);
 208         if (err)
 209                 sg_elems = 1;
 210
 211         /* We need an extra sg elements at head and tail. */
 212         sg_elems += 2;
 213         vdev->priv = vblk = kmalloc(sizeof(*vblk) +
 214                                     sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
 215         if (!vblk) {
 216                 err = -ENOMEM;
 217                 goto out;
 218         }
 219
 220         INIT_LIST_HEAD(&vblk->reqs);
 221         spin_lock_init(&vblk->lock);
 222         vblk->vdev = vdev;
 223         vblk->sg_elems = sg_elems;
 224         sg_init_table(vblk->sg, vblk->sg_elems);
 225
 226         /* We expect one virtqueue, for output. */
 227         vblk->vq = vdev->config->find_vq(vdev, 0, blk_done);
 228         if (IS_ERR(vblk->vq)) {
 229                 err = PTR_ERR(vblk->vq);
 230                 goto out_free_vblk;
 231         }
 232
 233         vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
 234         if (!vblk->pool) {
 235                 err = -ENOMEM;
 236                 goto out_free_vq;
 237         }
 238
 239         /* FIXME: How many partitions?  How long is a piece of string? */
 240         vblk->disk = alloc_disk(1 << PART_BITS);
 241         if (!vblk->disk) {
 242                 err = -ENOMEM;
 243                 goto out_mempool;
 244         }
 245
 246         vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
 247         if (!vblk->disk->queue) {
 248                 err = -ENOMEM;
 249                 goto out_put_disk;
 250         }
 251
 252         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, vblk->disk->queue);
 253
 254         if (index < 26) {
 255                 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
 256         } else if (index < (26 + 1) * 26) {
 257                 sprintf(vblk->disk->disk_name, "vd%c%c",
 258                         'a' + index / 26 - 1, 'a' + index % 26);
 259         } else {
 260                 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
 261                 const unsigned int m2 = (index / 26 - 1) % 26;
 262                 const unsigned int m3 =  index % 26;
 263                 sprintf(vblk->disk->disk_name, "vd%c%c%c",
 264                         'a' + m1, 'a' + m2, 'a' + m3);
 265         }
 266
 267         vblk->disk->major = major;
 268         vblk->disk->first_minor = index_to_minor(index);
 269         vblk->disk->private_data = vblk;
 270         vblk->disk->fops = &virtblk_fops;
 271         vblk->disk->driverfs_dev = &vdev->dev;
 272         index++;
 273
 274         /* If barriers are supported, tell block layer that queue is ordered */
 275         if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
 276                 blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL);
 277
 278         /* If disk is read-only in the host, the guest should obey */
 279         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 280                 set_disk_ro(vblk->disk, 1);
 281
 282         /* Host must always specify the capacity. */
 283         vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
 284                           &cap, sizeof(cap));
 285
 286         /* If capacity is too big, truncate with warning. */
 287         if ((sector_t)cap != cap) {
 288                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 289                          (unsigned long long)cap);
 290                 cap = (sector_t)-1;
 291         }
 292         set_capacity(vblk->disk, cap);
 293
 294         /* We can handle whatever the host told us to handle. */
 295         blk_queue_max_phys_segments(vblk->disk->queue, vblk->sg_elems-2);
 296         blk_queue_max_hw_segments(vblk->disk->queue, vblk->sg_elems-2);
 297
 298         /* No real sector limit. */
 299         blk_queue_max_sectors(vblk->disk->queue, -1U);
 300
 301         /* Host can optionally specify maximum segment size and number of
 302          * segments. */
 303         err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
 304                                 offsetof(struct virtio_blk_config, size_max),
 305                                 &v);
 306         if (!err)
 307                 blk_queue_max_segment_size(vblk->disk->queue, v);
 308         else
 309                 blk_queue_max_segment_size(vblk->disk->queue, -1U);
 310
 311         /* Host can optionally specify the block size of the device */
 312         err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
 313                                 offsetof(struct virtio_blk_config, blk_size),
 314                                 &blk_size);
 315         if (!err)
 316                 blk_queue_hardsect_size(vblk->disk->queue, blk_size);
 317
 318         add_disk(vblk->disk);
 319         return 0;
 320
 321 out_put_disk:
 322         put_disk(vblk->disk);
 323 out_mempool:
 324         mempool_destroy(vblk->pool);
 325 out_free_vq:
 326         vdev->config->del_vq(vblk->vq);
 327 out_free_vblk:
 328         kfree(vblk);
 329 out:
 330         return err;
 331 }
 332
 333 static void virtblk_remove(struct virtio_device *vdev)
 334 {
 335         struct virtio_blk *vblk = vdev->priv;
 336
 337         /* Nothing should be pending. */
 338         BUG_ON(!list_empty(&vblk->reqs));
 339
 340         /* Stop all the virtqueues. */
 341         vdev->config->reset(vdev);
 342
 343         del_gendisk(vblk->disk);
 344         blk_cleanup_queue(vblk->disk->queue);
 345         put_disk(vblk->disk);
 346         mempool_destroy(vblk->pool);
 347         vdev->config->del_vq(vblk->vq);
 348         kfree(vblk);
 349 }
 350
 351 static struct virtio_device_id id_table[] = {
 352         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 353         { 0 },
 354 };
 355
 356 static unsigned int features[] = {
 357         VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
 358         VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 359 };
 360
 361 static struct virtio_driver virtio_blk = {
 362         .feature_table = features,
 363         .feature_table_size = ARRAY_SIZE(features),
 364         .driver.name =  KBUILD_MODNAME,
 365         .driver.owner = THIS_MODULE,
 366         .id_table =     id_table,
 367         .probe =        virtblk_probe,
 368         .remove =       __devexit_p(virtblk_remove),
 369 };
 370
 371 static int __init init(void)
 372 {
 373         major = register_blkdev(0, "virtblk");
 374         if (major < 0)
 375                 return major;
 376         return register_virtio_driver(&virtio_blk);
 377 }
 378
 379 static void __exit fini(void)
 380 {
 381         unregister_blkdev(major, "virtblk");
 382         unregister_virtio_driver(&virtio_blk);
 383 }
 384 module_init(init);
 385 module_exit(fini);
 386
 387 MODULE_DEVICE_TABLE(virtio, id_table);
 388 MODULE_DESCRIPTION("Virtio block driver");
 389 MODULE_LICENSE("GPL");