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/hdreg.h>
   6 #include <linux/module.h>
   7 #include <linux/mutex.h>
   8 #include <linux/virtio.h>
   9 #include <linux/virtio_blk.h>
  10 #include <linux/scatterlist.h>
  11 #include <linux/string_helpers.h>
  12 #include <scsi/scsi_cmnd.h>
  13 #include <linux/idr.h>
  14
  15 #define PART_BITS 4
  16
  17 static int major;
  18 static DEFINE_IDA(vd_index_ida);
  19
  20 struct workqueue_struct *virtblk_wq;
  21
  22 struct virtio_blk
  23 {
  24         struct virtio_device *vdev;
  25         struct virtqueue *vq;
  26
  27         /* The disk structure for the kernel. */
  28         struct gendisk *disk;
  29
  30         mempool_t *pool;
  31
  32         /* Process context for config space updates */
  33         struct work_struct config_work;
  34
  35         /* Lock for config space updates */
  36         struct mutex config_lock;
  37
  38         /* enable config space updates */
  39         bool config_enable;
  40
  41         /* What host tells us, plus 2 for header & tailer. */
  42         unsigned int sg_elems;
  43
  44         /* Ida index - used to track minor number allocations. */
  45         int index;
  46
  47         /* Scatterlist: can be too big for stack. */
  48         struct scatterlist sg[/*sg_elems*/];
  49 };
  50
  51 struct virtblk_req
  52 {
  53         struct request *req;
  54         struct virtio_blk_outhdr out_hdr;
  55         struct virtio_scsi_inhdr in_hdr;
  56         u8 status;
  57 };
  58
  59 static void blk_done(struct virtqueue *vq)
  60 {
  61         struct virtio_blk *vblk = vq->vdev->priv;
  62         struct virtblk_req *vbr;
  63         unsigned int len;
  64         unsigned long flags;
  65
  66         spin_lock_irqsave(vblk->disk->queue->queue_lock, flags);
  67         while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
  68                 int error;
  69
  70                 switch (vbr->status) {
  71                 case VIRTIO_BLK_S_OK:
  72                         error = 0;
  73                         break;
  74                 case VIRTIO_BLK_S_UNSUPP:
  75                         error = -ENOTTY;
  76                         break;
  77                 default:
  78                         error = -EIO;
  79                         break;
  80                 }
  81
  82                 switch (vbr->req->cmd_type) {
  83                 case REQ_TYPE_BLOCK_PC:
  84                         vbr->req->resid_len = vbr->in_hdr.residual;
  85                         vbr->req->sense_len = vbr->in_hdr.sense_len;
  86                         vbr->req->errors = vbr->in_hdr.errors;
  87                         break;
  88                 case REQ_TYPE_SPECIAL:
  89                         vbr->req->errors = (error != 0);
  90                         break;
  91                 default:
  92                         break;
  93                 }
  94
  95                 __blk_end_request_all(vbr->req, error);
  96                 mempool_free(vbr, vblk->pool);
  97         }
  98         /* In case queue is stopped waiting for more buffers. */
  99         blk_start_queue(vblk->disk->queue);
 100         spin_unlock_irqrestore(vblk->disk->queue->queue_lock, flags);
 101 }
 102
 103 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
 104                    struct request *req)
 105 {
 106         unsigned long num, out = 0, in = 0;
 107         struct virtblk_req *vbr;
 108
 109         vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
 110         if (!vbr)
 111                 /* When another request finishes we'll try again. */
 112                 return false;
 113
 114         vbr->req = req;
 115
 116         if (req->cmd_flags & REQ_FLUSH) {
 117                 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
 118                 vbr->out_hdr.sector = 0;
 119                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 120         } else {
 121                 switch (req->cmd_type) {
 122                 case REQ_TYPE_FS:
 123                         vbr->out_hdr.type = 0;
 124                         vbr->out_hdr.sector = blk_rq_pos(vbr->req);
 125                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 126                         break;
 127                 case REQ_TYPE_BLOCK_PC:
 128                         vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
 129                         vbr->out_hdr.sector = 0;
 130                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 131                         break;
 132                 case REQ_TYPE_SPECIAL:
 133                         vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
 134                         vbr->out_hdr.sector = 0;
 135                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 136                         break;
 137                 default:
 138                         /* We don't put anything else in the queue. */
 139                         BUG();
 140                 }
 141         }
 142
 143         sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
 144
 145         /*
 146          * If this is a packet command we need a couple of additional headers.
 147          * Behind the normal outhdr we put a segment with the scsi command
 148          * block, and before the normal inhdr we put the sense data and the
 149          * inhdr with additional status information before the normal inhdr.
 150          */
 151         if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
 152                 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
 153
 154         num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
 155
 156         if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
 157                 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
 158                 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
 159                            sizeof(vbr->in_hdr));
 160         }
 161
 162         sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
 163                    sizeof(vbr->status));
 164
 165         if (num) {
 166                 if (rq_data_dir(vbr->req) == WRITE) {
 167                         vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 168                         out += num;
 169                 } else {
 170                         vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 171                         in += num;
 172                 }
 173         }
 174
 175         if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr, GFP_ATOMIC)<0) {
 176                 mempool_free(vbr, vblk->pool);
 177                 return false;
 178         }
 179
 180         return true;
 181 }
 182
 183 static void do_virtblk_request(struct request_queue *q)
 184 {
 185         struct virtio_blk *vblk = q->queuedata;
 186         struct request *req;
 187         unsigned int issued = 0;
 188
 189         while ((req = blk_peek_request(q)) != NULL) {
 190                 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 191
 192                 /* If this request fails, stop queue and wait for something to
 193                    finish to restart it. */
 194                 if (!do_req(q, vblk, req)) {
 195                         blk_stop_queue(q);
 196                         break;
 197                 }
 198                 blk_start_request(req);
 199                 issued++;
 200         }
 201
 202         if (issued)
 203                 virtqueue_kick(vblk->vq);
 204 }
 205
 206 /* return id (s/n) string for *disk to *id_str
 207  */
 208 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 209 {
 210         struct virtio_blk *vblk = disk->private_data;
 211         struct request *req;
 212         struct bio *bio;
 213         int err;
 214
 215         bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
 216                            GFP_KERNEL);
 217         if (IS_ERR(bio))
 218                 return PTR_ERR(bio);
 219
 220         req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
 221         if (IS_ERR(req)) {
 222                 bio_put(bio);
 223                 return PTR_ERR(req);
 224         }
 225
 226         req->cmd_type = REQ_TYPE_SPECIAL;
 227         err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
 228         blk_put_request(req);
 229
 230         return err;
 231 }
 232
 233 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
 234                              unsigned int cmd, unsigned long data)
 235 {
 236         struct gendisk *disk = bdev->bd_disk;
 237         struct virtio_blk *vblk = disk->private_data;
 238
 239         /*
 240          * Only allow the generic SCSI ioctls if the host can support it.
 241          */
 242         if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
 243                 return -ENOTTY;
 244
 245         return scsi_cmd_blk_ioctl(bdev, mode, cmd,
 246                                   (void __user *)data);
 247 }
 248
 249 /* We provide getgeo only to please some old bootloader/partitioning tools */
 250 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 251 {
 252         struct virtio_blk *vblk = bd->bd_disk->private_data;
 253         struct virtio_blk_geometry vgeo;
 254         int err;
 255
 256         /* see if the host passed in geometry config */
 257         err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
 258                                 offsetof(struct virtio_blk_config, geometry),
 259                                 &vgeo);
 260
 261         if (!err) {
 262                 geo->heads = vgeo.heads;
 263                 geo->sectors = vgeo.sectors;
 264                 geo->cylinders = vgeo.cylinders;
 265         } else {
 266                 /* some standard values, similar to sd */
 267                 geo->heads = 1 << 6;
 268                 geo->sectors = 1 << 5;
 269                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 270         }
 271         return 0;
 272 }
 273
 274 static const struct block_device_operations virtblk_fops = {
 275         .ioctl  = virtblk_ioctl,
 276         .owner  = THIS_MODULE,
 277         .getgeo = virtblk_getgeo,
 278 };
 279
 280 static int index_to_minor(int index)
 281 {
 282         return index << PART_BITS;
 283 }
 284
 285 static int minor_to_index(int minor)
 286 {
 287         return minor >> PART_BITS;
 288 }
 289
 290 static ssize_t virtblk_serial_show(struct device *dev,
 291                                 struct device_attribute *attr, char *buf)
 292 {
 293         struct gendisk *disk = dev_to_disk(dev);
 294         int err;
 295
 296         /* sysfs gives us a PAGE_SIZE buffer */
 297         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 298
 299         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 300         err = virtblk_get_id(disk, buf);
 301         if (!err)
 302                 return strlen(buf);
 303
 304         if (err == -EIO) /* Unsupported? Make it empty. */
 305                 return 0;
 306
 307         return err;
 308 }
 309 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
 310
 311 static void virtblk_config_changed_work(struct work_struct *work)
 312 {
 313         struct virtio_blk *vblk =
 314                 container_of(work, struct virtio_blk, config_work);
 315         struct virtio_device *vdev = vblk->vdev;
 316         struct request_queue *q = vblk->disk->queue;
 317         char cap_str_2[10], cap_str_10[10];
 318         u64 capacity, size;
 319
 320         mutex_lock(&vblk->config_lock);
 321         if (!vblk->config_enable)
 322                 goto done;
 323
 324         /* Host must always specify the capacity. */
 325         vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
 326                           &capacity, sizeof(capacity));
 327
 328         /* If capacity is too big, truncate with warning. */
 329         if ((sector_t)capacity != capacity) {
 330                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 331                          (unsigned long long)capacity);
 332                 capacity = (sector_t)-1;
 333         }
 334
 335         size = capacity * queue_logical_block_size(q);
 336         string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 337         string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 338
 339         dev_notice(&vdev->dev,
 340                   "new size: %llu %d-byte logical blocks (%s/%s)\n",
 341                   (unsigned long long)capacity,
 342                   queue_logical_block_size(q),
 343                   cap_str_10, cap_str_2);
 344
 345         set_capacity(vblk->disk, capacity);
 346         revalidate_disk(vblk->disk);
 347 done:
 348         mutex_unlock(&vblk->config_lock);
 349 }
 350
 351 static void virtblk_config_changed(struct virtio_device *vdev)
 352 {
 353         struct virtio_blk *vblk = vdev->priv;
 354
 355         queue_work(virtblk_wq, &vblk->config_work);
 356 }
 357
 358 static int init_vq(struct virtio_blk *vblk)
 359 {
 360         int err = 0;
 361
 362         /* We expect one virtqueue, for output. */
 363         vblk->vq = virtio_find_single_vq(vblk->vdev, blk_done, "requests");
 364         if (IS_ERR(vblk->vq))
 365                 err = PTR_ERR(vblk->vq);
 366
 367         return err;
 368 }
 369
 370 /*
 371  * Legacy naming scheme used for virtio devices.  We are stuck with it for
 372  * virtio blk but don't ever use it for any new driver.
 373  */
 374 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
 375 {
 376         const int base = 'z' - 'a' + 1;
 377         char *begin = buf + strlen(prefix);
 378         char *end = buf + buflen;
 379         char *p;
 380         int unit;
 381
 382         p = end - 1;
 383         *p = '\0';
 384         unit = base;
 385         do {
 386                 if (p == begin)
 387                         return -EINVAL;
 388                 *--p = 'a' + (index % unit);
 389                 index = (index / unit) - 1;
 390         } while (index >= 0);
 391
 392         memmove(begin, p, end - p);
 393         memcpy(buf, prefix, strlen(prefix));
 394
 395         return 0;
 396 }
 397
 398 static int virtblk_get_cache_mode(struct virtio_device *vdev)
 399 {
 400         u8 writeback;
 401         int err;
 402
 403         err = virtio_config_val(vdev, VIRTIO_BLK_F_CONFIG_WCE,
 404                                 offsetof(struct virtio_blk_config, wce),
 405                                 &writeback);
 406         if (err)
 407                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE);
 408
 409         return writeback;
 410 }
 411
 412 static void virtblk_update_cache_mode(struct virtio_device *vdev)
 413 {
 414         u8 writeback = virtblk_get_cache_mode(vdev);
 415         struct virtio_blk *vblk = vdev->priv;
 416
 417         if (writeback)
 418                 blk_queue_flush(vblk->disk->queue, REQ_FLUSH);
 419         else
 420                 blk_queue_flush(vblk->disk->queue, 0);
 421
 422         revalidate_disk(vblk->disk);
 423 }
 424
 425 static const char *const virtblk_cache_types[] = {
 426         "write through", "write back"
 427 };
 428
 429 static ssize_t
 430 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
 431                          const char *buf, size_t count)
 432 {
 433         struct gendisk *disk = dev_to_disk(dev);
 434         struct virtio_blk *vblk = disk->private_data;
 435         struct virtio_device *vdev = vblk->vdev;
 436         int i;
 437         u8 writeback;
 438
 439         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
 440         for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
 441                 if (sysfs_streq(buf, virtblk_cache_types[i]))
 442                         break;
 443
 444         if (i < 0)
 445                 return -EINVAL;
 446
 447         writeback = i;
 448         vdev->config->set(vdev,
 449                           offsetof(struct virtio_blk_config, wce),
 450                           &writeback, sizeof(writeback));
 451
 452         virtblk_update_cache_mode(vdev);
 453         return count;
 454 }
 455
 456 static ssize_t
 457 virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
 458                          char *buf)
 459 {
 460         struct gendisk *disk = dev_to_disk(dev);
 461         struct virtio_blk *vblk = disk->private_data;
 462         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
 463
 464         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
 465         return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
 466 }
 467
 468 static const struct device_attribute dev_attr_cache_type_ro =
 469         __ATTR(cache_type, S_IRUGO,
 470                virtblk_cache_type_show, NULL);
 471 static const struct device_attribute dev_attr_cache_type_rw =
 472         __ATTR(cache_type, S_IRUGO|S_IWUSR,
 473                virtblk_cache_type_show, virtblk_cache_type_store);
 474
 475 static int __devinit virtblk_probe(struct virtio_device *vdev)
 476 {
 477         struct virtio_blk *vblk;
 478         struct request_queue *q;
 479         int err, index;
 480         u64 cap;
 481         u32 v, blk_size, sg_elems, opt_io_size;
 482         u16 min_io_size;
 483         u8 physical_block_exp, alignment_offset;
 484
 485         err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
 486                              GFP_KERNEL);
 487         if (err < 0)
 488                 goto out;
 489         index = err;
 490
 491         /* We need to know how many segments before we allocate. */
 492         err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
 493                                 offsetof(struct virtio_blk_config, seg_max),
 494                                 &sg_elems);
 495
 496         /* We need at least one SG element, whatever they say. */
 497         if (err || !sg_elems)
 498                 sg_elems = 1;
 499
 500         /* We need an extra sg elements at head and tail. */
 501         sg_elems += 2;
 502         vdev->priv = vblk = kmalloc(sizeof(*vblk) +
 503                                     sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
 504         if (!vblk) {
 505                 err = -ENOMEM;
 506                 goto out_free_index;
 507         }
 508
 509         vblk->vdev = vdev;
 510         vblk->sg_elems = sg_elems;
 511         sg_init_table(vblk->sg, vblk->sg_elems);
 512         mutex_init(&vblk->config_lock);
 513         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 514         vblk->config_enable = true;
 515
 516         err = init_vq(vblk);
 517         if (err)
 518                 goto out_free_vblk;
 519
 520         vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
 521         if (!vblk->pool) {
 522                 err = -ENOMEM;
 523                 goto out_free_vq;
 524         }
 525
 526         /* FIXME: How many partitions?  How long is a piece of string? */
 527         vblk->disk = alloc_disk(1 << PART_BITS);
 528         if (!vblk->disk) {
 529                 err = -ENOMEM;
 530                 goto out_mempool;
 531         }
 532
 533         q = vblk->disk->queue = blk_init_queue(do_virtblk_request, NULL);
 534         if (!q) {
 535                 err = -ENOMEM;
 536                 goto out_put_disk;
 537         }
 538
 539         q->queuedata = vblk;
 540
 541         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
 542
 543         vblk->disk->major = major;
 544         vblk->disk->first_minor = index_to_minor(index);
 545         vblk->disk->private_data = vblk;
 546         vblk->disk->fops = &virtblk_fops;
 547         vblk->disk->driverfs_dev = &vdev->dev;
 548         vblk->index = index;
 549
 550         /* configure queue flush support */
 551         virtblk_update_cache_mode(vdev);
 552
 553         /* If disk is read-only in the host, the guest should obey */
 554         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 555                 set_disk_ro(vblk->disk, 1);
 556
 557         /* Host must always specify the capacity. */
 558         vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
 559                           &cap, sizeof(cap));
 560
 561         /* If capacity is too big, truncate with warning. */
 562         if ((sector_t)cap != cap) {
 563                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 564                          (unsigned long long)cap);
 565                 cap = (sector_t)-1;
 566         }
 567         set_capacity(vblk->disk, cap);
 568
 569         /* We can handle whatever the host told us to handle. */
 570         blk_queue_max_segments(q, vblk->sg_elems-2);
 571
 572         /* No need to bounce any requests */
 573         blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
 574
 575         /* No real sector limit. */
 576         blk_queue_max_hw_sectors(q, -1U);
 577
 578         /* Host can optionally specify maximum segment size and number of
 579          * segments. */
 580         err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
 581                                 offsetof(struct virtio_blk_config, size_max),
 582                                 &v);
 583         if (!err)
 584                 blk_queue_max_segment_size(q, v);
 585         else
 586                 blk_queue_max_segment_size(q, -1U);
 587
 588         /* Host can optionally specify the block size of the device */
 589         err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
 590                                 offsetof(struct virtio_blk_config, blk_size),
 591                                 &blk_size);
 592         if (!err)
 593                 blk_queue_logical_block_size(q, blk_size);
 594         else
 595                 blk_size = queue_logical_block_size(q);
 596
 597         /* Use topology information if available */
 598         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 599                         offsetof(struct virtio_blk_config, physical_block_exp),
 600                         &physical_block_exp);
 601         if (!err && physical_block_exp)
 602                 blk_queue_physical_block_size(q,
 603                                 blk_size * (1 << physical_block_exp));
 604
 605         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 606                         offsetof(struct virtio_blk_config, alignment_offset),
 607                         &alignment_offset);
 608         if (!err && alignment_offset)
 609                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 610
 611         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 612                         offsetof(struct virtio_blk_config, min_io_size),
 613                         &min_io_size);
 614         if (!err && min_io_size)
 615                 blk_queue_io_min(q, blk_size * min_io_size);
 616
 617         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 618                         offsetof(struct virtio_blk_config, opt_io_size),
 619                         &opt_io_size);
 620         if (!err && opt_io_size)
 621                 blk_queue_io_opt(q, blk_size * opt_io_size);
 622
 623
 624         add_disk(vblk->disk);
 625         err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
 626         if (err)
 627                 goto out_del_disk;
 628
 629         if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
 630                 err = device_create_file(disk_to_dev(vblk->disk),
 631                                          &dev_attr_cache_type_rw);
 632         else
 633                 err = device_create_file(disk_to_dev(vblk->disk),
 634                                          &dev_attr_cache_type_ro);
 635         if (err)
 636                 goto out_del_disk;
 637         return 0;
 638
 639 out_del_disk:
 640         del_gendisk(vblk->disk);
 641         blk_cleanup_queue(vblk->disk->queue);
 642 out_put_disk:
 643         put_disk(vblk->disk);
 644 out_mempool:
 645         mempool_destroy(vblk->pool);
 646 out_free_vq:
 647         vdev->config->del_vqs(vdev);
 648 out_free_vblk:
 649         kfree(vblk);
 650 out_free_index:
 651         ida_simple_remove(&vd_index_ida, index);
 652 out:
 653         return err;
 654 }
 655
 656 static void __devexit virtblk_remove(struct virtio_device *vdev)
 657 {
 658         struct virtio_blk *vblk = vdev->priv;
 659         int index = vblk->index;
 660
 661         /* Prevent config work handler from accessing the device. */
 662         mutex_lock(&vblk->config_lock);
 663         vblk->config_enable = false;
 664         mutex_unlock(&vblk->config_lock);
 665
 666         del_gendisk(vblk->disk);
 667         blk_cleanup_queue(vblk->disk->queue);
 668
 669         /* Stop all the virtqueues. */
 670         vdev->config->reset(vdev);
 671
 672         flush_work(&vblk->config_work);
 673
 674         put_disk(vblk->disk);
 675         mempool_destroy(vblk->pool);
 676         vdev->config->del_vqs(vdev);
 677         kfree(vblk);
 678         ida_simple_remove(&vd_index_ida, index);
 679 }
 680
 681 #ifdef CONFIG_PM
 682 static int virtblk_freeze(struct virtio_device *vdev)
 683 {
 684         struct virtio_blk *vblk = vdev->priv;
 685
 686         /* Ensure we don't receive any more interrupts */
 687         vdev->config->reset(vdev);
 688
 689         /* Prevent config work handler from accessing the device. */
 690         mutex_lock(&vblk->config_lock);
 691         vblk->config_enable = false;
 692         mutex_unlock(&vblk->config_lock);
 693
 694         flush_work(&vblk->config_work);
 695
 696         spin_lock_irq(vblk->disk->queue->queue_lock);
 697         blk_stop_queue(vblk->disk->queue);
 698         spin_unlock_irq(vblk->disk->queue->queue_lock);
 699         blk_sync_queue(vblk->disk->queue);
 700
 701         vdev->config->del_vqs(vdev);
 702         return 0;
 703 }
 704
 705 static int virtblk_restore(struct virtio_device *vdev)
 706 {
 707         struct virtio_blk *vblk = vdev->priv;
 708         int ret;
 709
 710         vblk->config_enable = true;
 711         ret = init_vq(vdev->priv);
 712         if (!ret) {
 713                 spin_lock_irq(vblk->disk->queue->queue_lock);
 714                 blk_start_queue(vblk->disk->queue);
 715                 spin_unlock_irq(vblk->disk->queue->queue_lock);
 716         }
 717         return ret;
 718 }
 719 #endif
 720
 721 static const struct virtio_device_id id_table[] = {
 722         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 723         { 0 },
 724 };
 725
 726 static unsigned int features[] = {
 727         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 728         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
 729         VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE
 730 };
 731
 732 /*
 733  * virtio_blk causes spurious section mismatch warning by
 734  * simultaneously referring to a __devinit and a __devexit function.
 735  * Use __refdata to avoid this warning.
 736  */
 737 static struct virtio_driver __refdata virtio_blk = {
 738         .feature_table          = features,
 739         .feature_table_size     = ARRAY_SIZE(features),
 740         .driver.name            = KBUILD_MODNAME,
 741         .driver.owner           = THIS_MODULE,
 742         .id_table               = id_table,
 743         .probe                  = virtblk_probe,
 744         .remove                 = __devexit_p(virtblk_remove),
 745         .config_changed         = virtblk_config_changed,
 746 #ifdef CONFIG_PM
 747         .freeze                 = virtblk_freeze,
 748         .restore                = virtblk_restore,
 749 #endif
 750 };
 751
 752 static int __init init(void)
 753 {
 754         int error;
 755
 756         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
 757         if (!virtblk_wq)
 758                 return -ENOMEM;
 759
 760         major = register_blkdev(0, "virtblk");
 761         if (major < 0) {
 762                 error = major;
 763                 goto out_destroy_workqueue;
 764         }
 765
 766         error = register_virtio_driver(&virtio_blk);
 767         if (error)
 768                 goto out_unregister_blkdev;
 769         return 0;
 770
 771 out_unregister_blkdev:
 772         unregister_blkdev(major, "virtblk");
 773 out_destroy_workqueue:
 774         destroy_workqueue(virtblk_wq);
 775         return error;
 776 }
 777
 778 static void __exit fini(void)
 779 {
 780         unregister_blkdev(major, "virtblk");
 781         unregister_virtio_driver(&virtio_blk);
 782         destroy_workqueue(virtblk_wq);
 783 }
 784 module_init(init);
 785 module_exit(fini);
 786
 787 MODULE_DEVICE_TABLE(virtio, id_table);
 788 MODULE_DESCRIPTION("Virtio block driver");
 789 MODULE_LICENSE("GPL");