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         if (blk_fs_request(vbr->req)) {
  95                 vbr->out_hdr.type = 0;
  96                 vbr->out_hdr.sector = blk_rq_pos(vbr->req);
  97                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
  98         } else if (blk_pc_request(vbr->req)) {
  99                 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
 100                 vbr->out_hdr.sector = 0;
 101                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 102         } else {
 103                 /* We don't put anything else in the queue. */
 104                 BUG();
 105         }
 106
 107         if (blk_barrier_rq(vbr->req))
 108                 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
 109
 110         sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
 111
 112         /*
 113          * If this is a packet command we need a couple of additional headers.
 114          * Behind the normal outhdr we put a segment with the scsi command
 115          * block, and before the normal inhdr we put the sense data and the
 116          * inhdr with additional status information before the normal inhdr.
 117          */
 118         if (blk_pc_request(vbr->req))
 119                 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
 120
 121         num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
 122
 123         if (blk_pc_request(vbr->req)) {
 124                 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
 125                 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
 126                            sizeof(vbr->in_hdr));
 127         }
 128
 129         sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
 130                    sizeof(vbr->status));
 131
 132         if (num) {
 133                 if (rq_data_dir(vbr->req) == WRITE) {
 134                         vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 135                         out += num;
 136                 } else {
 137                         vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 138                         in += num;
 139                 }
 140         }
 141
 142         if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr)) {
 143                 mempool_free(vbr, vblk->pool);
 144                 return false;
 145         }
 146
 147         list_add_tail(&vbr->list, &vblk->reqs);
 148         return true;
 149 }
 150
 151 static void do_virtblk_request(struct request_queue *q)
 152 {
 153         struct virtio_blk *vblk = q->queuedata;
 154         struct request *req;
 155         unsigned int issued = 0;
 156
 157         while ((req = blk_peek_request(q)) != NULL) {
 158                 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 159
 160                 /* If this request fails, stop queue and wait for something to
 161                    finish to restart it. */
 162                 if (!do_req(q, vblk, req)) {
 163                         blk_stop_queue(q);
 164                         break;
 165                 }
 166                 blk_start_request(req);
 167                 issued++;
 168         }
 169
 170         if (issued)
 171                 vblk->vq->vq_ops->kick(vblk->vq);
 172 }
 173
 174 /* return ATA identify data
 175  */
 176 static int virtblk_identify(struct gendisk *disk, void *argp)
 177 {
 178         struct virtio_blk *vblk = disk->private_data;
 179         void *opaque;
 180         int err = -ENOMEM;
 181
 182         opaque = kmalloc(VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
 183         if (!opaque)
 184                 goto out;
 185
 186         err = virtio_config_buf(vblk->vdev, VIRTIO_BLK_F_IDENTIFY,
 187                 offsetof(struct virtio_blk_config, identify), opaque,
 188                 VIRTIO_BLK_ID_BYTES);
 189
 190         if (err)
 191                 goto out_kfree;
 192
 193         if (copy_to_user(argp, opaque, VIRTIO_BLK_ID_BYTES))
 194                 err = -EFAULT;
 195
 196 out_kfree:
 197         kfree(opaque);
 198 out:
 199         return err;
 200 }
 201
 202 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
 203                          unsigned cmd, unsigned long data)
 204 {
 205         struct gendisk *disk = bdev->bd_disk;
 206         struct virtio_blk *vblk = disk->private_data;
 207         void __user *argp = (void __user *)data;
 208
 209         if (cmd == HDIO_GET_IDENTITY)
 210                 return virtblk_identify(disk, argp);
 211
 212         /*
 213          * Only allow the generic SCSI ioctls if the host can support it.
 214          */
 215         if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
 216                 return -ENOTTY;
 217
 218         return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
 219 }
 220
 221 /* We provide getgeo only to please some old bootloader/partitioning tools */
 222 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 223 {
 224         struct virtio_blk *vblk = bd->bd_disk->private_data;
 225         struct virtio_blk_geometry vgeo;
 226         int err;
 227
 228         /* see if the host passed in geometry config */
 229         err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
 230                                 offsetof(struct virtio_blk_config, geometry),
 231                                 &vgeo);
 232
 233         if (!err) {
 234                 geo->heads = vgeo.heads;
 235                 geo->sectors = vgeo.sectors;
 236                 geo->cylinders = vgeo.cylinders;
 237         } else {
 238                 /* some standard values, similar to sd */
 239                 geo->heads = 1 << 6;
 240                 geo->sectors = 1 << 5;
 241                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 242         }
 243         return 0;
 244 }
 245
 246 static struct block_device_operations virtblk_fops = {
 247         .locked_ioctl = virtblk_ioctl,
 248         .owner  = THIS_MODULE,
 249         .getgeo = virtblk_getgeo,
 250 };
 251
 252 static int index_to_minor(int index)
 253 {
 254         return index << PART_BITS;
 255 }
 256
 257 static int __devinit virtblk_probe(struct virtio_device *vdev)
 258 {
 259         struct virtio_blk *vblk;
 260         int err;
 261         u64 cap;
 262         u32 v;
 263         u32 blk_size, sg_elems;
 264
 265         if (index_to_minor(index) >= 1 << MINORBITS)
 266                 return -ENOSPC;
 267
 268         /* We need to know how many segments before we allocate. */
 269         err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
 270                                 offsetof(struct virtio_blk_config, seg_max),
 271                                 &sg_elems);
 272         if (err)
 273                 sg_elems = 1;
 274
 275         /* We need an extra sg elements at head and tail. */
 276         sg_elems += 2;
 277         vdev->priv = vblk = kmalloc(sizeof(*vblk) +
 278                                     sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
 279         if (!vblk) {
 280                 err = -ENOMEM;
 281                 goto out;
 282         }
 283
 284         INIT_LIST_HEAD(&vblk->reqs);
 285         spin_lock_init(&vblk->lock);
 286         vblk->vdev = vdev;
 287         vblk->sg_elems = sg_elems;
 288         sg_init_table(vblk->sg, vblk->sg_elems);
 289
 290         /* We expect one virtqueue, for output. */
 291         vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
 292         if (IS_ERR(vblk->vq)) {
 293                 err = PTR_ERR(vblk->vq);
 294                 goto out_free_vblk;
 295         }
 296
 297         vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
 298         if (!vblk->pool) {
 299                 err = -ENOMEM;
 300                 goto out_free_vq;
 301         }
 302
 303         /* FIXME: How many partitions?  How long is a piece of string? */
 304         vblk->disk = alloc_disk(1 << PART_BITS);
 305         if (!vblk->disk) {
 306                 err = -ENOMEM;
 307                 goto out_mempool;
 308         }
 309
 310         vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
 311         if (!vblk->disk->queue) {
 312                 err = -ENOMEM;
 313                 goto out_put_disk;
 314         }
 315
 316         vblk->disk->queue->queuedata = vblk;
 317         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, vblk->disk->queue);
 318
 319         if (index < 26) {
 320                 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
 321         } else if (index < (26 + 1) * 26) {
 322                 sprintf(vblk->disk->disk_name, "vd%c%c",
 323                         'a' + index / 26 - 1, 'a' + index % 26);
 324         } else {
 325                 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
 326                 const unsigned int m2 = (index / 26 - 1) % 26;
 327                 const unsigned int m3 =  index % 26;
 328                 sprintf(vblk->disk->disk_name, "vd%c%c%c",
 329                         'a' + m1, 'a' + m2, 'a' + m3);
 330         }
 331
 332         vblk->disk->major = major;
 333         vblk->disk->first_minor = index_to_minor(index);
 334         vblk->disk->private_data = vblk;
 335         vblk->disk->fops = &virtblk_fops;
 336         vblk->disk->driverfs_dev = &vdev->dev;
 337         index++;
 338
 339         /* If barriers are supported, tell block layer that queue is ordered */
 340         if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
 341                 blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL);
 342
 343         /* If disk is read-only in the host, the guest should obey */
 344         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 345                 set_disk_ro(vblk->disk, 1);
 346
 347         /* Host must always specify the capacity. */
 348         vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
 349                           &cap, sizeof(cap));
 350
 351         /* If capacity is too big, truncate with warning. */
 352         if ((sector_t)cap != cap) {
 353                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 354                          (unsigned long long)cap);
 355                 cap = (sector_t)-1;
 356         }
 357         set_capacity(vblk->disk, cap);
 358
 359         /* We can handle whatever the host told us to handle. */
 360         blk_queue_max_phys_segments(vblk->disk->queue, vblk->sg_elems-2);
 361         blk_queue_max_hw_segments(vblk->disk->queue, vblk->sg_elems-2);
 362
 363         /* No need to bounce any requests */
 364         blk_queue_bounce_limit(vblk->disk->queue, BLK_BOUNCE_ANY);
 365
 366         /* No real sector limit. */
 367         blk_queue_max_sectors(vblk->disk->queue, -1U);
 368
 369         /* Host can optionally specify maximum segment size and number of
 370          * segments. */
 371         err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
 372                                 offsetof(struct virtio_blk_config, size_max),
 373                                 &v);
 374         if (!err)
 375                 blk_queue_max_segment_size(vblk->disk->queue, v);
 376         else
 377                 blk_queue_max_segment_size(vblk->disk->queue, -1U);
 378
 379         /* Host can optionally specify the block size of the device */
 380         err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
 381                                 offsetof(struct virtio_blk_config, blk_size),
 382                                 &blk_size);
 383         if (!err)
 384                 blk_queue_logical_block_size(vblk->disk->queue, blk_size);
 385
 386         add_disk(vblk->disk);
 387         return 0;
 388
 389 out_put_disk:
 390         put_disk(vblk->disk);
 391 out_mempool:
 392         mempool_destroy(vblk->pool);
 393 out_free_vq:
 394         vdev->config->del_vqs(vdev);
 395 out_free_vblk:
 396         kfree(vblk);
 397 out:
 398         return err;
 399 }
 400
 401 static void __devexit virtblk_remove(struct virtio_device *vdev)
 402 {
 403         struct virtio_blk *vblk = vdev->priv;
 404
 405         /* Nothing should be pending. */
 406         BUG_ON(!list_empty(&vblk->reqs));
 407
 408         /* Stop all the virtqueues. */
 409         vdev->config->reset(vdev);
 410
 411         del_gendisk(vblk->disk);
 412         blk_cleanup_queue(vblk->disk->queue);
 413         put_disk(vblk->disk);
 414         mempool_destroy(vblk->pool);
 415         vdev->config->del_vqs(vdev);
 416         kfree(vblk);
 417 }
 418
 419 static struct virtio_device_id id_table[] = {
 420         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 421         { 0 },
 422 };
 423
 424 static unsigned int features[] = {
 425         VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
 426         VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 427         VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_IDENTIFY
 428 };
 429
 430 /*
 431  * virtio_blk causes spurious section mismatch warning by
 432  * simultaneously referring to a __devinit and a __devexit function.
 433  * Use __refdata to avoid this warning.
 434  */
 435 static struct virtio_driver __refdata virtio_blk = {
 436         .feature_table = features,
 437         .feature_table_size = ARRAY_SIZE(features),
 438         .driver.name =  KBUILD_MODNAME,
 439         .driver.owner = THIS_MODULE,
 440         .id_table =     id_table,
 441         .probe =        virtblk_probe,
 442         .remove =       __devexit_p(virtblk_remove),
 443 };
 444
 445 static int __init init(void)
 446 {
 447         major = register_blkdev(0, "virtblk");
 448         if (major < 0)
 449                 return major;
 450         return register_virtio_driver(&virtio_blk);
 451 }
 452
 453 static void __exit fini(void)
 454 {
 455         unregister_blkdev(major, "virtblk");
 456         unregister_virtio_driver(&virtio_blk);
 457 }
 458 module_init(init);
 459 module_exit(fini);
 460
 461 MODULE_DEVICE_TABLE(virtio, id_table);
 462 MODULE_DESCRIPTION("Virtio block driver");
 463 MODULE_LICENSE("GPL");