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         struct virtio_scsi_inhdr in_hdr;
  41         u8 status;
  42 };
  43
  44 static void blk_done(struct virtqueue *vq)
  45 {
  46         struct virtio_blk *vblk = vq->vdev->priv;
  47         struct virtblk_req *vbr;
  48         unsigned int len;
  49         unsigned long flags;
  50
  51         spin_lock_irqsave(&vblk->lock, flags);
  52         while ((vbr = vblk->vq->vq_ops->get_buf(vblk->vq, &len)) != NULL) {
  53                 int error;
  54
  55                 switch (vbr->status) {
  56                 case VIRTIO_BLK_S_OK:
  57                         error = 0;
  58                         break;
  59                 case VIRTIO_BLK_S_UNSUPP:
  60                         error = -ENOTTY;
  61                         break;
  62                 default:
  63                         error = -EIO;
  64                         break;
  65                 }
  66
  67                 if (blk_pc_request(vbr->req)) {
  68                         vbr->req->resid_len = vbr->in_hdr.residual;
  69                         vbr->req->sense_len = vbr->in_hdr.sense_len;
  70                         vbr->req->errors = vbr->in_hdr.errors;
  71                 }
  72
  73                 __blk_end_request_all(vbr->req, error);
  74                 list_del(&vbr->list);
  75                 mempool_free(vbr, vblk->pool);
  76         }
  77         /* In case queue is stopped waiting for more buffers. */
  78         blk_start_queue(vblk->disk->queue);
  79         spin_unlock_irqrestore(&vblk->lock, flags);
  80 }
  81
  82 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
  83                    struct request *req)
  84 {
  85         unsigned long num, out = 0, in = 0;
  86         struct virtblk_req *vbr;
  87
  88         vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
  89         if (!vbr)
  90                 /* When another request finishes we'll try again. */
  91                 return false;
  92
  93         vbr->req = req;
  94         switch (req->cmd_type) {
  95         case REQ_TYPE_FS:
  96                 vbr->out_hdr.type = 0;
  97                 vbr->out_hdr.sector = blk_rq_pos(vbr->req);
  98                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
  99                 break;
 100         case REQ_TYPE_BLOCK_PC:
 101                 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
 102                 vbr->out_hdr.sector = 0;
 103                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 104                 break;
 105         case REQ_TYPE_LINUX_BLOCK:
 106                 if (req->cmd[0] == REQ_LB_OP_FLUSH) {
 107                         vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
 108                         vbr->out_hdr.sector = 0;
 109                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 110                         break;
 111                 }
 112                 /*FALLTHRU*/
 113         default:
 114                 /* We don't put anything else in the queue. */
 115                 BUG();
 116         }
 117
 118         if (blk_barrier_rq(vbr->req))
 119                 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
 120
 121         sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
 122
 123         /*
 124          * If this is a packet command we need a couple of additional headers.
 125          * Behind the normal outhdr we put a segment with the scsi command
 126          * block, and before the normal inhdr we put the sense data and the
 127          * inhdr with additional status information before the normal inhdr.
 128          */
 129         if (blk_pc_request(vbr->req))
 130                 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
 131
 132         num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
 133
 134         if (blk_pc_request(vbr->req)) {
 135                 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
 136                 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
 137                            sizeof(vbr->in_hdr));
 138         }
 139
 140         sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
 141                    sizeof(vbr->status));
 142
 143         if (num) {
 144                 if (rq_data_dir(vbr->req) == WRITE) {
 145                         vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 146                         out += num;
 147                 } else {
 148                         vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 149                         in += num;
 150                 }
 151         }
 152
 153         if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
 154                 mempool_free(vbr, vblk->pool);
 155                 return false;
 156         }
 157
 158         list_add_tail(&vbr->list, &vblk->reqs);
 159         return true;
 160 }
 161
 162 static void do_virtblk_request(struct request_queue *q)
 163 {
 164         struct virtio_blk *vblk = q->queuedata;
 165         struct request *req;
 166         unsigned int issued = 0;
 167
 168         while ((req = blk_peek_request(q)) != NULL) {
 169                 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 170
 171                 /* If this request fails, stop queue and wait for something to
 172                    finish to restart it. */
 173                 if (!do_req(q, vblk, req)) {
 174                         blk_stop_queue(q);
 175                         break;
 176                 }
 177                 blk_start_request(req);
 178                 issued++;
 179         }
 180
 181         if (issued)
 182                 vblk->vq->vq_ops->kick(vblk->vq);
 183 }
 184
 185 static void virtblk_prepare_flush(struct request_queue *q, struct request *req)
 186 {
 187         req->cmd_type = REQ_TYPE_LINUX_BLOCK;
 188         req->cmd[0] = REQ_LB_OP_FLUSH;
 189 }
 190
 191 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
 192                          unsigned cmd, unsigned long data)
 193 {
 194         struct gendisk *disk = bdev->bd_disk;
 195         struct virtio_blk *vblk = disk->private_data;
 196
 197         /*
 198          * Only allow the generic SCSI ioctls if the host can support it.
 199          */
 200         if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
 201                 return -ENOTTY;
 202
 203         return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
 204                               (void __user *)data);
 205 }
 206
 207 /* We provide getgeo only to please some old bootloader/partitioning tools */
 208 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 209 {
 210         struct virtio_blk *vblk = bd->bd_disk->private_data;
 211         struct virtio_blk_geometry vgeo;
 212         int err;
 213
 214         /* see if the host passed in geometry config */
 215         err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
 216                                 offsetof(struct virtio_blk_config, geometry),
 217                                 &vgeo);
 218
 219         if (!err) {
 220                 geo->heads = vgeo.heads;
 221                 geo->sectors = vgeo.sectors;
 222                 geo->cylinders = vgeo.cylinders;
 223         } else {
 224                 /* some standard values, similar to sd */
 225                 geo->heads = 1 << 6;
 226                 geo->sectors = 1 << 5;
 227                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 228         }
 229         return 0;
 230 }
 231
 232 static const struct block_device_operations virtblk_fops = {
 233         .locked_ioctl = virtblk_ioctl,
 234         .owner  = THIS_MODULE,
 235         .getgeo = virtblk_getgeo,
 236 };
 237
 238 static int index_to_minor(int index)
 239 {
 240         return index << PART_BITS;
 241 }
 242
 243 static int __devinit virtblk_probe(struct virtio_device *vdev)
 244 {
 245         struct virtio_blk *vblk;
 246         int err;
 247         u64 cap;
 248         u32 v;
 249         u32 blk_size, sg_elems;
 250
 251         if (index_to_minor(index) >= 1 << MINORBITS)
 252                 return -ENOSPC;
 253
 254         /* We need to know how many segments before we allocate. */
 255         err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
 256                                 offsetof(struct virtio_blk_config, seg_max),
 257                                 &sg_elems);
 258         if (err)
 259                 sg_elems = 1;
 260
 261         /* We need an extra sg elements at head and tail. */
 262         sg_elems += 2;
 263         vdev->priv = vblk = kmalloc(sizeof(*vblk) +
 264                                     sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
 265         if (!vblk) {
 266                 err = -ENOMEM;
 267                 goto out;
 268         }
 269
 270         INIT_LIST_HEAD(&vblk->reqs);
 271         spin_lock_init(&vblk->lock);
 272         vblk->vdev = vdev;
 273         vblk->sg_elems = sg_elems;
 274         sg_init_table(vblk->sg, vblk->sg_elems);
 275
 276         /* We expect one virtqueue, for output. */
 277         vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
 278         if (IS_ERR(vblk->vq)) {
 279                 err = PTR_ERR(vblk->vq);
 280                 goto out_free_vblk;
 281         }
 282
 283         vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
 284         if (!vblk->pool) {
 285                 err = -ENOMEM;
 286                 goto out_free_vq;
 287         }
 288
 289         /* FIXME: How many partitions?  How long is a piece of string? */
 290         vblk->disk = alloc_disk(1 << PART_BITS);
 291         if (!vblk->disk) {
 292                 err = -ENOMEM;
 293                 goto out_mempool;
 294         }
 295
 296         vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
 297         if (!vblk->disk->queue) {
 298                 err = -ENOMEM;
 299                 goto out_put_disk;
 300         }
 301
 302         vblk->disk->queue->queuedata = vblk;
 303
 304         if (index < 26) {
 305                 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
 306         } else if (index < (26 + 1) * 26) {
 307                 sprintf(vblk->disk->disk_name, "vd%c%c",
 308                         'a' + index / 26 - 1, 'a' + index % 26);
 309         } else {
 310                 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
 311                 const unsigned int m2 = (index / 26 - 1) % 26;
 312                 const unsigned int m3 =  index % 26;
 313                 sprintf(vblk->disk->disk_name, "vd%c%c%c",
 314                         'a' + m1, 'a' + m2, 'a' + m3);
 315         }
 316
 317         vblk->disk->major = major;
 318         vblk->disk->first_minor = index_to_minor(index);
 319         vblk->disk->private_data = vblk;
 320         vblk->disk->fops = &virtblk_fops;
 321         vblk->disk->driverfs_dev = &vdev->dev;
 322         index++;
 323
 324         /* If barriers are supported, tell block layer that queue is ordered */
 325         if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
 326                 blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_DRAIN_FLUSH,
 327                                   virtblk_prepare_flush);
 328         else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
 329                 blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL);
 330
 331         /* If disk is read-only in the host, the guest should obey */
 332         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 333                 set_disk_ro(vblk->disk, 1);
 334
 335         /* Host must always specify the capacity. */
 336         vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
 337                           &cap, sizeof(cap));
 338
 339         /* If capacity is too big, truncate with warning. */
 340         if ((sector_t)cap != cap) {
 341                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 342                          (unsigned long long)cap);
 343                 cap = (sector_t)-1;
 344         }
 345         set_capacity(vblk->disk, cap);
 346
 347         /* We can handle whatever the host told us to handle. */
 348         blk_queue_max_phys_segments(vblk->disk->queue, vblk->sg_elems-2);
 349         blk_queue_max_hw_segments(vblk->disk->queue, vblk->sg_elems-2);
 350
 351         /* No need to bounce any requests */
 352         blk_queue_bounce_limit(vblk->disk->queue, BLK_BOUNCE_ANY);
 353
 354         /* No real sector limit. */
 355         blk_queue_max_sectors(vblk->disk->queue, -1U);
 356
 357         /* Host can optionally specify maximum segment size and number of
 358          * segments. */
 359         err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
 360                                 offsetof(struct virtio_blk_config, size_max),
 361                                 &v);
 362         if (!err)
 363                 blk_queue_max_segment_size(vblk->disk->queue, v);
 364         else
 365                 blk_queue_max_segment_size(vblk->disk->queue, -1U);
 366
 367         /* Host can optionally specify the block size of the device */
 368         err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
 369                                 offsetof(struct virtio_blk_config, blk_size),
 370                                 &blk_size);
 371         if (!err)
 372                 blk_queue_logical_block_size(vblk->disk->queue, blk_size);
 373
 374         add_disk(vblk->disk);
 375         return 0;
 376
 377 out_put_disk:
 378         put_disk(vblk->disk);
 379 out_mempool:
 380         mempool_destroy(vblk->pool);
 381 out_free_vq:
 382         vdev->config->del_vqs(vdev);
 383 out_free_vblk:
 384         kfree(vblk);
 385 out:
 386         return err;
 387 }
 388
 389 static void __devexit virtblk_remove(struct virtio_device *vdev)
 390 {
 391         struct virtio_blk *vblk = vdev->priv;
 392
 393         /* Nothing should be pending. */
 394         BUG_ON(!list_empty(&vblk->reqs));
 395
 396         /* Stop all the virtqueues. */
 397         vdev->config->reset(vdev);
 398
 399         del_gendisk(vblk->disk);
 400         blk_cleanup_queue(vblk->disk->queue);
 401         put_disk(vblk->disk);
 402         mempool_destroy(vblk->pool);
 403         vdev->config->del_vqs(vdev);
 404         kfree(vblk);
 405 }
 406
 407 static struct virtio_device_id id_table[] = {
 408         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 409         { 0 },
 410 };
 411
 412 static unsigned int features[] = {
 413         VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
 414         VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 415         VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH
 416 };
 417
 418 /*
 419  * virtio_blk causes spurious section mismatch warning by
 420  * simultaneously referring to a __devinit and a __devexit function.
 421  * Use __refdata to avoid this warning.
 422  */
 423 static struct virtio_driver __refdata virtio_blk = {
 424         .feature_table = features,
 425         .feature_table_size = ARRAY_SIZE(features),
 426         .driver.name =  KBUILD_MODNAME,
 427         .driver.owner = THIS_MODULE,
 428         .id_table =     id_table,
 429         .probe =        virtblk_probe,
 430         .remove =       __devexit_p(virtblk_remove),
 431 };
 432
 433 static int __init init(void)
 434 {
 435         major = register_blkdev(0, "virtblk");
 436         if (major < 0)
 437                 return major;
 438         return register_virtio_driver(&virtio_blk);
 439 }
 440
 441 static void __exit fini(void)
 442 {
 443         unregister_blkdev(major, "virtblk");
 444         unregister_virtio_driver(&virtio_blk);
 445 }
 446 module_init(init);
 447 module_exit(fini);
 448
 449 MODULE_DEVICE_TABLE(virtio, id_table);
 450 MODULE_DESCRIPTION("Virtio block driver");
 451 MODULE_LICENSE("GPL");