drivers/block/virtio_blk.c

   1 //#define DEBUG
   2 #include <linux/spinlock.h>
   3 #include <linux/slab.h>
   4 #include <linux/blkdev.h>
   5 #include <linux/smp_lock.h>
   6 #include <linux/hdreg.h>
   7 #include <linux/virtio.h>
   8 #include <linux/virtio_blk.h>
   9 #include <linux/scatterlist.h>
  10
  11 #define PART_BITS 4
  12
  13 static int major, index;
  14
  15 struct virtio_blk
  16 {
  17         spinlock_t lock;
  18
  19         struct virtio_device *vdev;
  20         struct virtqueue *vq;
  21
  22         /* The disk structure for the kernel. */
  23         struct gendisk *disk;
  24
  25         /* Request tracking. */
  26         struct list_head reqs;
  27
  28         mempool_t *pool;
  29
  30         /* What host tells us, plus 2 for header & tailer. */
  31         unsigned int sg_elems;
  32
  33         /* Scatterlist: can be too big for stack. */
  34         struct scatterlist sg[/*sg_elems*/];
  35 };
  36
  37 struct virtblk_req
  38 {
  39         struct list_head list;
  40         struct request *req;
  41         struct virtio_blk_outhdr out_hdr;
  42         struct virtio_scsi_inhdr in_hdr;
  43         u8 status;
  44 };
  45
  46 static void blk_done(struct virtqueue *vq)
  47 {
  48         struct virtio_blk *vblk = vq->vdev->priv;
  49         struct virtblk_req *vbr;
  50         unsigned int len;
  51         unsigned long flags;
  52
  53         spin_lock_irqsave(&vblk->lock, flags);
  54         while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
  55                 int error;
  56
  57                 switch (vbr->status) {
  58                 case VIRTIO_BLK_S_OK:
  59                         error = 0;
  60                         break;
  61                 case VIRTIO_BLK_S_UNSUPP:
  62                         error = -ENOTTY;
  63                         break;
  64                 default:
  65                         error = -EIO;
  66                         break;
  67                 }
  68
  69                 switch (vbr->req->cmd_type) {
  70                 case REQ_TYPE_BLOCK_PC:
  71                         vbr->req->resid_len = vbr->in_hdr.residual;
  72                         vbr->req->sense_len = vbr->in_hdr.sense_len;
  73                         vbr->req->errors = vbr->in_hdr.errors;
  74                         break;
  75                 case REQ_TYPE_SPECIAL:
  76                         vbr->req->errors = (error != 0);
  77                         break;
  78                 default:
  79                         break;
  80                 }
  81
  82                 __blk_end_request_all(vbr->req, error);
  83                 list_del(&vbr->list);
  84                 mempool_free(vbr, vblk->pool);
  85         }
  86         /* In case queue is stopped waiting for more buffers. */
  87         blk_start_queue(vblk->disk->queue);
  88         spin_unlock_irqrestore(&vblk->lock, flags);
  89 }
  90
  91 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
  92                    struct request *req)
  93 {
  94         unsigned long num, out = 0, in = 0;
  95         struct virtblk_req *vbr;
  96
  97         vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
  98         if (!vbr)
  99                 /* When another request finishes we'll try again. */
 100                 return false;
 101
 102         vbr->req = req;
 103
 104         if (req->cmd_flags & REQ_FLUSH) {
 105                 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
 106                 vbr->out_hdr.sector = 0;
 107                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 108         } else {
 109                 switch (req->cmd_type) {
 110                 case REQ_TYPE_FS:
 111                         vbr->out_hdr.type = 0;
 112                         vbr->out_hdr.sector = blk_rq_pos(vbr->req);
 113                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 114                         break;
 115                 case REQ_TYPE_BLOCK_PC:
 116                         vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
 117                         vbr->out_hdr.sector = 0;
 118                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 119                         break;
 120                 case REQ_TYPE_SPECIAL:
 121                         vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
 122                         vbr->out_hdr.sector = 0;
 123                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 124                         break;
 125                 default:
 126                         /* We don't put anything else in the queue. */
 127                         BUG();
 128                 }
 129         }
 130
 131         if (vbr->req->cmd_flags & REQ_HARDBARRIER)
 132                 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
 133
 134         sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
 135
 136         /*
 137          * If this is a packet command we need a couple of additional headers.
 138          * Behind the normal outhdr we put a segment with the scsi command
 139          * block, and before the normal inhdr we put the sense data and the
 140          * inhdr with additional status information before the normal inhdr.
 141          */
 142         if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
 143                 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
 144
 145         num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
 146
 147         if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
 148                 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
 149                 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
 150                            sizeof(vbr->in_hdr));
 151         }
 152
 153         sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
 154                    sizeof(vbr->status));
 155
 156         if (num) {
 157                 if (rq_data_dir(vbr->req) == WRITE) {
 158                         vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 159                         out += num;
 160                 } else {
 161                         vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 162                         in += num;
 163                 }
 164         }
 165
 166         if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
 167                 mempool_free(vbr, vblk->pool);
 168                 return false;
 169         }
 170
 171         list_add_tail(&vbr->list, &vblk->reqs);
 172         return true;
 173 }
 174
 175 static void do_virtblk_request(struct request_queue *q)
 176 {
 177         struct virtio_blk *vblk = q->queuedata;
 178         struct request *req;
 179         unsigned int issued = 0;
 180
 181         while ((req = blk_peek_request(q)) != NULL) {
 182                 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 183
 184                 /* If this request fails, stop queue and wait for something to
 185                    finish to restart it. */
 186                 if (!do_req(q, vblk, req)) {
 187                         blk_stop_queue(q);
 188                         break;
 189                 }
 190                 blk_start_request(req);
 191                 issued++;
 192         }
 193
 194         if (issued)
 195                 virtqueue_kick(vblk->vq);
 196 }
 197
 198 /* return id (s/n) string for *disk to *id_str
 199  */
 200 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 201 {
 202         struct virtio_blk *vblk = disk->private_data;
 203         struct request *req;
 204         struct bio *bio;
 205
 206         bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
 207                            GFP_KERNEL);
 208         if (IS_ERR(bio))
 209                 return PTR_ERR(bio);
 210
 211         req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
 212         if (IS_ERR(req)) {
 213                 bio_put(bio);
 214                 return PTR_ERR(req);
 215         }
 216
 217         req->cmd_type = REQ_TYPE_SPECIAL;
 218         return blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
 219 }
 220
 221 static int virtblk_locked_ioctl(struct block_device *bdev, fmode_t mode,
 222                          unsigned cmd, unsigned long data)
 223 {
 224         struct gendisk *disk = bdev->bd_disk;
 225         struct virtio_blk *vblk = disk->private_data;
 226
 227         if (cmd == 0x56424944) { /* 'VBID' */
 228                 void __user *usr_data = (void __user *)data;
 229                 char id_str[VIRTIO_BLK_ID_BYTES];
 230                 int err;
 231
 232                 err = virtblk_get_id(disk, id_str);
 233                 if (!err && copy_to_user(usr_data, id_str, VIRTIO_BLK_ID_BYTES))
 234                         err = -EFAULT;
 235                 return err;
 236         }
 237         /*
 238          * Only allow the generic SCSI ioctls if the host can support it.
 239          */
 240         if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
 241                 return -ENOTTY;
 242
 243         return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
 244                               (void __user *)data);
 245 }
 246
 247 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
 248                              unsigned int cmd, unsigned long param)
 249 {
 250         int ret;
 251
 252         lock_kernel();
 253         ret = virtblk_locked_ioctl(bdev, mode, cmd, param);
 254         unlock_kernel();
 255
 256         return ret;
 257 }
 258
 259 /* We provide getgeo only to please some old bootloader/partitioning tools */
 260 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 261 {
 262         struct virtio_blk *vblk = bd->bd_disk->private_data;
 263         struct virtio_blk_geometry vgeo;
 264         int err;
 265
 266         /* see if the host passed in geometry config */
 267         err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
 268                                 offsetof(struct virtio_blk_config, geometry),
 269                                 &vgeo);
 270
 271         if (!err) {
 272                 geo->heads = vgeo.heads;
 273                 geo->sectors = vgeo.sectors;
 274                 geo->cylinders = vgeo.cylinders;
 275         } else {
 276                 /* some standard values, similar to sd */
 277                 geo->heads = 1 << 6;
 278                 geo->sectors = 1 << 5;
 279                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 280         }
 281         return 0;
 282 }
 283
 284 static const struct block_device_operations virtblk_fops = {
 285         .ioctl  = virtblk_ioctl,
 286         .owner  = THIS_MODULE,
 287         .getgeo = virtblk_getgeo,
 288 };
 289
 290 static int index_to_minor(int index)
 291 {
 292         return index << PART_BITS;
 293 }
 294
 295 static int __devinit virtblk_probe(struct virtio_device *vdev)
 296 {
 297         struct virtio_blk *vblk;
 298         struct request_queue *q;
 299         int err;
 300         u64 cap;
 301         u32 v, blk_size, sg_elems, opt_io_size;
 302         u16 min_io_size;
 303         u8 physical_block_exp, alignment_offset;
 304
 305         if (index_to_minor(index) >= 1 << MINORBITS)
 306                 return -ENOSPC;
 307
 308         /* We need to know how many segments before we allocate. */
 309         err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
 310                                 offsetof(struct virtio_blk_config, seg_max),
 311                                 &sg_elems);
 312
 313         /* We need at least one SG element, whatever they say. */
 314         if (err || !sg_elems)
 315                 sg_elems = 1;
 316
 317         /* We need an extra sg elements at head and tail. */
 318         sg_elems += 2;
 319         vdev->priv = vblk = kmalloc(sizeof(*vblk) +
 320                                     sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
 321         if (!vblk) {
 322                 err = -ENOMEM;
 323                 goto out;
 324         }
 325
 326         INIT_LIST_HEAD(&vblk->reqs);
 327         spin_lock_init(&vblk->lock);
 328         vblk->vdev = vdev;
 329         vblk->sg_elems = sg_elems;
 330         sg_init_table(vblk->sg, vblk->sg_elems);
 331
 332         /* We expect one virtqueue, for output. */
 333         vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
 334         if (IS_ERR(vblk->vq)) {
 335                 err = PTR_ERR(vblk->vq);
 336                 goto out_free_vblk;
 337         }
 338
 339         vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
 340         if (!vblk->pool) {
 341                 err = -ENOMEM;
 342                 goto out_free_vq;
 343         }
 344
 345         /* FIXME: How many partitions?  How long is a piece of string? */
 346         vblk->disk = alloc_disk(1 << PART_BITS);
 347         if (!vblk->disk) {
 348                 err = -ENOMEM;
 349                 goto out_mempool;
 350         }
 351
 352         q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
 353         if (!q) {
 354                 err = -ENOMEM;
 355                 goto out_put_disk;
 356         }
 357
 358         q->queuedata = vblk;
 359
 360         if (index < 26) {
 361                 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
 362         } else if (index < (26 + 1) * 26) {
 363                 sprintf(vblk->disk->disk_name, "vd%c%c",
 364                         'a' + index / 26 - 1, 'a' + index % 26);
 365         } else {
 366                 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
 367                 const unsigned int m2 = (index / 26 - 1) % 26;
 368                 const unsigned int m3 =  index % 26;
 369                 sprintf(vblk->disk->disk_name, "vd%c%c%c",
 370                         'a' + m1, 'a' + m2, 'a' + m3);
 371         }
 372
 373         vblk->disk->major = major;
 374         vblk->disk->first_minor = index_to_minor(index);
 375         vblk->disk->private_data = vblk;
 376         vblk->disk->fops = &virtblk_fops;
 377         vblk->disk->driverfs_dev = &vdev->dev;
 378         index++;
 379
 380         /* If barriers are supported, tell block layer that queue is ordered */
 381         if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
 382                 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH);
 383         else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
 384                 blk_queue_ordered(q, QUEUE_ORDERED_TAG);
 385
 386         /* If disk is read-only in the host, the guest should obey */
 387         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 388                 set_disk_ro(vblk->disk, 1);
 389
 390         /* Host must always specify the capacity. */
 391         vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
 392                           &cap, sizeof(cap));
 393
 394         /* If capacity is too big, truncate with warning. */
 395         if ((sector_t)cap != cap) {
 396                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 397                          (unsigned long long)cap);
 398                 cap = (sector_t)-1;
 399         }
 400         set_capacity(vblk->disk, cap);
 401
 402         /* We can handle whatever the host told us to handle. */
 403         blk_queue_max_segments(q, vblk->sg_elems-2);
 404
 405         /* No need to bounce any requests */
 406         blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
 407
 408         /* No real sector limit. */
 409         blk_queue_max_hw_sectors(q, -1U);
 410
 411         /* Host can optionally specify maximum segment size and number of
 412          * segments. */
 413         err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
 414                                 offsetof(struct virtio_blk_config, size_max),
 415                                 &v);
 416         if (!err)
 417                 blk_queue_max_segment_size(q, v);
 418         else
 419                 blk_queue_max_segment_size(q, -1U);
 420
 421         /* Host can optionally specify the block size of the device */
 422         err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
 423                                 offsetof(struct virtio_blk_config, blk_size),
 424                                 &blk_size);
 425         if (!err)
 426                 blk_queue_logical_block_size(q, blk_size);
 427         else
 428                 blk_size = queue_logical_block_size(q);
 429
 430         /* Use topology information if available */
 431         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 432                         offsetof(struct virtio_blk_config, physical_block_exp),
 433                         &physical_block_exp);
 434         if (!err && physical_block_exp)
 435                 blk_queue_physical_block_size(q,
 436                                 blk_size * (1 << physical_block_exp));
 437
 438         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 439                         offsetof(struct virtio_blk_config, alignment_offset),
 440                         &alignment_offset);
 441         if (!err && alignment_offset)
 442                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 443
 444         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 445                         offsetof(struct virtio_blk_config, min_io_size),
 446                         &min_io_size);
 447         if (!err && min_io_size)
 448                 blk_queue_io_min(q, blk_size * min_io_size);
 449
 450         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 451                         offsetof(struct virtio_blk_config, opt_io_size),
 452                         &opt_io_size);
 453         if (!err && opt_io_size)
 454                 blk_queue_io_opt(q, blk_size * opt_io_size);
 455
 456
 457         add_disk(vblk->disk);
 458         return 0;
 459
 460 out_put_disk:
 461         put_disk(vblk->disk);
 462 out_mempool:
 463         mempool_destroy(vblk->pool);
 464 out_free_vq:
 465         vdev->config->del_vqs(vdev);
 466 out_free_vblk:
 467         kfree(vblk);
 468 out:
 469         return err;
 470 }
 471
 472 static void __devexit virtblk_remove(struct virtio_device *vdev)
 473 {
 474         struct virtio_blk *vblk = vdev->priv;
 475
 476         /* Nothing should be pending. */
 477         BUG_ON(!list_empty(&vblk->reqs));
 478
 479         /* Stop all the virtqueues. */
 480         vdev->config->reset(vdev);
 481
 482         del_gendisk(vblk->disk);
 483         blk_cleanup_queue(vblk->disk->queue);
 484         put_disk(vblk->disk);
 485         mempool_destroy(vblk->pool);
 486         vdev->config->del_vqs(vdev);
 487         kfree(vblk);
 488 }
 489
 490 static const struct virtio_device_id id_table[] = {
 491         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 492         { 0 },
 493 };
 494
 495 static unsigned int features[] = {
 496         VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
 497         VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 498         VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
 499 };
 500
 501 /*
 502  * virtio_blk causes spurious section mismatch warning by
 503  * simultaneously referring to a __devinit and a __devexit function.
 504  * Use __refdata to avoid this warning.
 505  */
 506 static struct virtio_driver __refdata virtio_blk = {
 507         .feature_table = features,
 508         .feature_table_size = ARRAY_SIZE(features),
 509         .driver.name =  KBUILD_MODNAME,
 510         .driver.owner = THIS_MODULE,
 511         .id_table =     id_table,
 512         .probe =        virtblk_probe,
 513         .remove =       __devexit_p(virtblk_remove),
 514 };
 515
 516 static int __init init(void)
 517 {
 518         major = register_blkdev(0, "virtblk");
 519         if (major < 0)
 520                 return major;
 521         return register_virtio_driver(&virtio_blk);
 522 }
 523
 524 static void __exit fini(void)
 525 {
 526         unregister_blkdev(major, "virtblk");
 527         unregister_virtio_driver(&virtio_blk);
 528 }
 529 module_init(init);
 530 module_exit(fini);
 531
 532 MODULE_DEVICE_TABLE(virtio, id_table);
 533 MODULE_DESCRIPTION("Virtio block driver");
 534 MODULE_LICENSE("GPL");