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 bool use_bio;
  18 module_param(use_bio, bool, S_IRUGO);
  19
  20 static int major;
  21 static DEFINE_IDA(vd_index_ida);
  22
  23 struct workqueue_struct *virtblk_wq;
  24
  25 struct virtio_blk
  26 {
  27         struct virtio_device *vdev;
  28         struct virtqueue *vq;
  29         wait_queue_head_t queue_wait;
  30
  31         /* The disk structure for the kernel. */
  32         struct gendisk *disk;
  33
  34         mempool_t *pool;
  35
  36         /* Process context for config space updates */
  37         struct work_struct config_work;
  38
  39         /* Lock for config space updates */
  40         struct mutex config_lock;
  41
  42         /* enable config space updates */
  43         bool config_enable;
  44
  45         /* What host tells us, plus 2 for header & tailer. */
  46         unsigned int sg_elems;
  47
  48         /* Ida index - used to track minor number allocations. */
  49         int index;
  50
  51         /* Scatterlist: can be too big for stack. */
  52         struct scatterlist sg[/*sg_elems*/];
  53 };
  54
  55 struct virtblk_req
  56 {
  57         struct request *req;
  58         struct bio *bio;
  59         struct virtio_blk_outhdr out_hdr;
  60         struct virtio_scsi_inhdr in_hdr;
  61         struct work_struct work;
  62         struct virtio_blk *vblk;
  63         int flags;
  64         u8 status;
  65         struct scatterlist sg[];
  66 };
  67
  68 enum {
  69         VBLK_IS_FLUSH           = 1,
  70         VBLK_REQ_FLUSH          = 2,
  71         VBLK_REQ_DATA           = 4,
  72         VBLK_REQ_FUA            = 8,
  73 };
  74
  75 static inline int virtblk_result(struct virtblk_req *vbr)
  76 {
  77         switch (vbr->status) {
  78         case VIRTIO_BLK_S_OK:
  79                 return 0;
  80         case VIRTIO_BLK_S_UNSUPP:
  81                 return -ENOTTY;
  82         default:
  83                 return -EIO;
  84         }
  85 }
  86
  87 static inline struct virtblk_req *virtblk_alloc_req(struct virtio_blk *vblk,
  88                                                     gfp_t gfp_mask)
  89 {
  90         struct virtblk_req *vbr;
  91
  92         vbr = mempool_alloc(vblk->pool, gfp_mask);
  93         if (!vbr)
  94                 return NULL;
  95
  96         vbr->vblk = vblk;
  97         if (use_bio)
  98                 sg_init_table(vbr->sg, vblk->sg_elems);
  99
 100         return vbr;
 101 }
 102
 103 static void virtblk_add_buf_wait(struct virtio_blk *vblk,
 104                                  struct virtblk_req *vbr,
 105                                  unsigned long out,
 106                                  unsigned long in)
 107 {
 108         DEFINE_WAIT(wait);
 109
 110         for (;;) {
 111                 prepare_to_wait_exclusive(&vblk->queue_wait, &wait,
 112                                           TASK_UNINTERRUPTIBLE);
 113
 114                 spin_lock_irq(vblk->disk->queue->queue_lock);
 115                 if (virtqueue_add_buf(vblk->vq, vbr->sg, out, in, vbr,
 116                                       GFP_ATOMIC) < 0) {
 117                         spin_unlock_irq(vblk->disk->queue->queue_lock);
 118                         io_schedule();
 119                 } else {
 120                         virtqueue_kick(vblk->vq);
 121                         spin_unlock_irq(vblk->disk->queue->queue_lock);
 122                         break;
 123                 }
 124
 125         }
 126
 127         finish_wait(&vblk->queue_wait, &wait);
 128 }
 129
 130 static inline void virtblk_add_req(struct virtblk_req *vbr,
 131                                    unsigned int out, unsigned int in)
 132 {
 133         struct virtio_blk *vblk = vbr->vblk;
 134
 135         spin_lock_irq(vblk->disk->queue->queue_lock);
 136         if (unlikely(virtqueue_add_buf(vblk->vq, vbr->sg, out, in, vbr,
 137                                         GFP_ATOMIC) < 0)) {
 138                 spin_unlock_irq(vblk->disk->queue->queue_lock);
 139                 virtblk_add_buf_wait(vblk, vbr, out, in);
 140                 return;
 141         }
 142         virtqueue_kick(vblk->vq);
 143         spin_unlock_irq(vblk->disk->queue->queue_lock);
 144 }
 145
 146 static int virtblk_bio_send_flush(struct virtblk_req *vbr)
 147 {
 148         unsigned int out = 0, in = 0;
 149
 150         vbr->flags |= VBLK_IS_FLUSH;
 151         vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
 152         vbr->out_hdr.sector = 0;
 153         vbr->out_hdr.ioprio = 0;
 154         sg_set_buf(&vbr->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
 155         sg_set_buf(&vbr->sg[out + in++], &vbr->status, sizeof(vbr->status));
 156
 157         virtblk_add_req(vbr, out, in);
 158
 159         return 0;
 160 }
 161
 162 static int virtblk_bio_send_data(struct virtblk_req *vbr)
 163 {
 164         struct virtio_blk *vblk = vbr->vblk;
 165         unsigned int num, out = 0, in = 0;
 166         struct bio *bio = vbr->bio;
 167
 168         vbr->flags &= ~VBLK_IS_FLUSH;
 169         vbr->out_hdr.type = 0;
 170         vbr->out_hdr.sector = bio->bi_sector;
 171         vbr->out_hdr.ioprio = bio_prio(bio);
 172
 173         sg_set_buf(&vbr->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
 174
 175         num = blk_bio_map_sg(vblk->disk->queue, bio, vbr->sg + out);
 176
 177         sg_set_buf(&vbr->sg[num + out + in++], &vbr->status,
 178                    sizeof(vbr->status));
 179
 180         if (num) {
 181                 if (bio->bi_rw & REQ_WRITE) {
 182                         vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 183                         out += num;
 184                 } else {
 185                         vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 186                         in += num;
 187                 }
 188         }
 189
 190         virtblk_add_req(vbr, out, in);
 191
 192         return 0;
 193 }
 194
 195 static void virtblk_bio_send_data_work(struct work_struct *work)
 196 {
 197         struct virtblk_req *vbr;
 198
 199         vbr = container_of(work, struct virtblk_req, work);
 200
 201         virtblk_bio_send_data(vbr);
 202 }
 203
 204 static void virtblk_bio_send_flush_work(struct work_struct *work)
 205 {
 206         struct virtblk_req *vbr;
 207
 208         vbr = container_of(work, struct virtblk_req, work);
 209
 210         virtblk_bio_send_flush(vbr);
 211 }
 212
 213 static inline void virtblk_request_done(struct virtblk_req *vbr)
 214 {
 215         struct virtio_blk *vblk = vbr->vblk;
 216         struct request *req = vbr->req;
 217         int error = virtblk_result(vbr);
 218
 219         if (req->cmd_type == REQ_TYPE_BLOCK_PC) {
 220                 req->resid_len = vbr->in_hdr.residual;
 221                 req->sense_len = vbr->in_hdr.sense_len;
 222                 req->errors = vbr->in_hdr.errors;
 223         } else if (req->cmd_type == REQ_TYPE_SPECIAL) {
 224                 req->errors = (error != 0);
 225         }
 226
 227         __blk_end_request_all(req, error);
 228         mempool_free(vbr, vblk->pool);
 229 }
 230
 231 static inline void virtblk_bio_flush_done(struct virtblk_req *vbr)
 232 {
 233         struct virtio_blk *vblk = vbr->vblk;
 234
 235         if (vbr->flags & VBLK_REQ_DATA) {
 236                 /* Send out the actual write data */
 237                 INIT_WORK(&vbr->work, virtblk_bio_send_data_work);
 238                 queue_work(virtblk_wq, &vbr->work);
 239         } else {
 240                 bio_endio(vbr->bio, virtblk_result(vbr));
 241                 mempool_free(vbr, vblk->pool);
 242         }
 243 }
 244
 245 static inline void virtblk_bio_data_done(struct virtblk_req *vbr)
 246 {
 247         struct virtio_blk *vblk = vbr->vblk;
 248
 249         if (unlikely(vbr->flags & VBLK_REQ_FUA)) {
 250                 /* Send out a flush before end the bio */
 251                 vbr->flags &= ~VBLK_REQ_DATA;
 252                 INIT_WORK(&vbr->work, virtblk_bio_send_flush_work);
 253                 queue_work(virtblk_wq, &vbr->work);
 254         } else {
 255                 bio_endio(vbr->bio, virtblk_result(vbr));
 256                 mempool_free(vbr, vblk->pool);
 257         }
 258 }
 259
 260 static inline void virtblk_bio_done(struct virtblk_req *vbr)
 261 {
 262         if (unlikely(vbr->flags & VBLK_IS_FLUSH))
 263                 virtblk_bio_flush_done(vbr);
 264         else
 265                 virtblk_bio_data_done(vbr);
 266 }
 267
 268 static void virtblk_done(struct virtqueue *vq)
 269 {
 270         struct virtio_blk *vblk = vq->vdev->priv;
 271         bool bio_done = false, req_done = false;
 272         struct virtblk_req *vbr;
 273         unsigned long flags;
 274         unsigned int len;
 275
 276         spin_lock_irqsave(vblk->disk->queue->queue_lock, flags);
 277         do {
 278                 virtqueue_disable_cb(vq);
 279                 while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
 280                         if (vbr->bio) {
 281                                 virtblk_bio_done(vbr);
 282                                 bio_done = true;
 283                         } else {
 284                                 virtblk_request_done(vbr);
 285                                 req_done = true;
 286                         }
 287                 }
 288         } while (!virtqueue_enable_cb(vq));
 289         /* In case queue is stopped waiting for more buffers. */
 290         if (req_done)
 291                 blk_start_queue(vblk->disk->queue);
 292         spin_unlock_irqrestore(vblk->disk->queue->queue_lock, flags);
 293
 294         if (bio_done)
 295                 wake_up(&vblk->queue_wait);
 296 }
 297
 298 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
 299                    struct request *req)
 300 {
 301         unsigned long num, out = 0, in = 0;
 302         struct virtblk_req *vbr;
 303
 304         vbr = virtblk_alloc_req(vblk, GFP_ATOMIC);
 305         if (!vbr)
 306                 /* When another request finishes we'll try again. */
 307                 return false;
 308
 309         vbr->req = req;
 310         vbr->bio = NULL;
 311         if (req->cmd_flags & REQ_FLUSH) {
 312                 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
 313                 vbr->out_hdr.sector = 0;
 314                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 315         } else {
 316                 switch (req->cmd_type) {
 317                 case REQ_TYPE_FS:
 318                         vbr->out_hdr.type = 0;
 319                         vbr->out_hdr.sector = blk_rq_pos(vbr->req);
 320                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 321                         break;
 322                 case REQ_TYPE_BLOCK_PC:
 323                         vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
 324                         vbr->out_hdr.sector = 0;
 325                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 326                         break;
 327                 case REQ_TYPE_SPECIAL:
 328                         vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
 329                         vbr->out_hdr.sector = 0;
 330                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 331                         break;
 332                 default:
 333                         /* We don't put anything else in the queue. */
 334                         BUG();
 335                 }
 336         }
 337
 338         sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
 339
 340         /*
 341          * If this is a packet command we need a couple of additional headers.
 342          * Behind the normal outhdr we put a segment with the scsi command
 343          * block, and before the normal inhdr we put the sense data and the
 344          * inhdr with additional status information before the normal inhdr.
 345          */
 346         if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
 347                 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
 348
 349         num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
 350
 351         if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
 352                 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
 353                 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
 354                            sizeof(vbr->in_hdr));
 355         }
 356
 357         sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
 358                    sizeof(vbr->status));
 359
 360         if (num) {
 361                 if (rq_data_dir(vbr->req) == WRITE) {
 362                         vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 363                         out += num;
 364                 } else {
 365                         vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 366                         in += num;
 367                 }
 368         }
 369
 370         if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr,
 371                               GFP_ATOMIC) < 0) {
 372                 mempool_free(vbr, vblk->pool);
 373                 return false;
 374         }
 375
 376         return true;
 377 }
 378
 379 static void virtblk_request(struct request_queue *q)
 380 {
 381         struct virtio_blk *vblk = q->queuedata;
 382         struct request *req;
 383         unsigned int issued = 0;
 384
 385         while ((req = blk_peek_request(q)) != NULL) {
 386                 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 387
 388                 /* If this request fails, stop queue and wait for something to
 389                    finish to restart it. */
 390                 if (!do_req(q, vblk, req)) {
 391                         blk_stop_queue(q);
 392                         break;
 393                 }
 394                 blk_start_request(req);
 395                 issued++;
 396         }
 397
 398         if (issued)
 399                 virtqueue_kick(vblk->vq);
 400 }
 401
 402 static void virtblk_make_request(struct request_queue *q, struct bio *bio)
 403 {
 404         struct virtio_blk *vblk = q->queuedata;
 405         struct virtblk_req *vbr;
 406
 407         BUG_ON(bio->bi_phys_segments + 2 > vblk->sg_elems);
 408
 409         vbr = virtblk_alloc_req(vblk, GFP_NOIO);
 410         if (!vbr) {
 411                 bio_endio(bio, -ENOMEM);
 412                 return;
 413         }
 414
 415         vbr->bio = bio;
 416         vbr->flags = 0;
 417         if (bio->bi_rw & REQ_FLUSH)
 418                 vbr->flags |= VBLK_REQ_FLUSH;
 419         if (bio->bi_rw & REQ_FUA)
 420                 vbr->flags |= VBLK_REQ_FUA;
 421         if (bio->bi_size)
 422                 vbr->flags |= VBLK_REQ_DATA;
 423
 424         if (unlikely(vbr->flags & VBLK_REQ_FLUSH))
 425                 virtblk_bio_send_flush(vbr);
 426         else
 427                 virtblk_bio_send_data(vbr);
 428 }
 429
 430 /* return id (s/n) string for *disk to *id_str
 431  */
 432 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 433 {
 434         struct virtio_blk *vblk = disk->private_data;
 435         struct request *req;
 436         struct bio *bio;
 437         int err;
 438
 439         bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
 440                            GFP_KERNEL);
 441         if (IS_ERR(bio))
 442                 return PTR_ERR(bio);
 443
 444         req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
 445         if (IS_ERR(req)) {
 446                 bio_put(bio);
 447                 return PTR_ERR(req);
 448         }
 449
 450         req->cmd_type = REQ_TYPE_SPECIAL;
 451         err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
 452         blk_put_request(req);
 453
 454         return err;
 455 }
 456
 457 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
 458                              unsigned int cmd, unsigned long data)
 459 {
 460         struct gendisk *disk = bdev->bd_disk;
 461         struct virtio_blk *vblk = disk->private_data;
 462
 463         /*
 464          * Only allow the generic SCSI ioctls if the host can support it.
 465          */
 466         if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
 467                 return -ENOTTY;
 468
 469         return scsi_cmd_blk_ioctl(bdev, mode, cmd,
 470                                   (void __user *)data);
 471 }
 472
 473 /* We provide getgeo only to please some old bootloader/partitioning tools */
 474 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 475 {
 476         struct virtio_blk *vblk = bd->bd_disk->private_data;
 477         struct virtio_blk_geometry vgeo;
 478         int err;
 479
 480         /* see if the host passed in geometry config */
 481         err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
 482                                 offsetof(struct virtio_blk_config, geometry),
 483                                 &vgeo);
 484
 485         if (!err) {
 486                 geo->heads = vgeo.heads;
 487                 geo->sectors = vgeo.sectors;
 488                 geo->cylinders = vgeo.cylinders;
 489         } else {
 490                 /* some standard values, similar to sd */
 491                 geo->heads = 1 << 6;
 492                 geo->sectors = 1 << 5;
 493                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 494         }
 495         return 0;
 496 }
 497
 498 static const struct block_device_operations virtblk_fops = {
 499         .ioctl  = virtblk_ioctl,
 500         .owner  = THIS_MODULE,
 501         .getgeo = virtblk_getgeo,
 502 };
 503
 504 static int index_to_minor(int index)
 505 {
 506         return index << PART_BITS;
 507 }
 508
 509 static int minor_to_index(int minor)
 510 {
 511         return minor >> PART_BITS;
 512 }
 513
 514 static ssize_t virtblk_serial_show(struct device *dev,
 515                                 struct device_attribute *attr, char *buf)
 516 {
 517         struct gendisk *disk = dev_to_disk(dev);
 518         int err;
 519
 520         /* sysfs gives us a PAGE_SIZE buffer */
 521         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 522
 523         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 524         err = virtblk_get_id(disk, buf);
 525         if (!err)
 526                 return strlen(buf);
 527
 528         if (err == -EIO) /* Unsupported? Make it empty. */
 529                 return 0;
 530
 531         return err;
 532 }
 533 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
 534
 535 static void virtblk_config_changed_work(struct work_struct *work)
 536 {
 537         struct virtio_blk *vblk =
 538                 container_of(work, struct virtio_blk, config_work);
 539         struct virtio_device *vdev = vblk->vdev;
 540         struct request_queue *q = vblk->disk->queue;
 541         char cap_str_2[10], cap_str_10[10];
 542         u64 capacity, size;
 543
 544         mutex_lock(&vblk->config_lock);
 545         if (!vblk->config_enable)
 546                 goto done;
 547
 548         /* Host must always specify the capacity. */
 549         vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
 550                           &capacity, sizeof(capacity));
 551
 552         /* If capacity is too big, truncate with warning. */
 553         if ((sector_t)capacity != capacity) {
 554                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 555                          (unsigned long long)capacity);
 556                 capacity = (sector_t)-1;
 557         }
 558
 559         size = capacity * queue_logical_block_size(q);
 560         string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 561         string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 562
 563         dev_notice(&vdev->dev,
 564                   "new size: %llu %d-byte logical blocks (%s/%s)\n",
 565                   (unsigned long long)capacity,
 566                   queue_logical_block_size(q),
 567                   cap_str_10, cap_str_2);
 568
 569         set_capacity(vblk->disk, capacity);
 570         revalidate_disk(vblk->disk);
 571 done:
 572         mutex_unlock(&vblk->config_lock);
 573 }
 574
 575 static void virtblk_config_changed(struct virtio_device *vdev)
 576 {
 577         struct virtio_blk *vblk = vdev->priv;
 578
 579         queue_work(virtblk_wq, &vblk->config_work);
 580 }
 581
 582 static int init_vq(struct virtio_blk *vblk)
 583 {
 584         int err = 0;
 585
 586         /* We expect one virtqueue, for output. */
 587         vblk->vq = virtio_find_single_vq(vblk->vdev, virtblk_done, "requests");
 588         if (IS_ERR(vblk->vq))
 589                 err = PTR_ERR(vblk->vq);
 590
 591         return err;
 592 }
 593
 594 /*
 595  * Legacy naming scheme used for virtio devices.  We are stuck with it for
 596  * virtio blk but don't ever use it for any new driver.
 597  */
 598 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
 599 {
 600         const int base = 'z' - 'a' + 1;
 601         char *begin = buf + strlen(prefix);
 602         char *end = buf + buflen;
 603         char *p;
 604         int unit;
 605
 606         p = end - 1;
 607         *p = '\0';
 608         unit = base;
 609         do {
 610                 if (p == begin)
 611                         return -EINVAL;
 612                 *--p = 'a' + (index % unit);
 613                 index = (index / unit) - 1;
 614         } while (index >= 0);
 615
 616         memmove(begin, p, end - p);
 617         memcpy(buf, prefix, strlen(prefix));
 618
 619         return 0;
 620 }
 621
 622 static int virtblk_get_cache_mode(struct virtio_device *vdev)
 623 {
 624         u8 writeback;
 625         int err;
 626
 627         err = virtio_config_val(vdev, VIRTIO_BLK_F_CONFIG_WCE,
 628                                 offsetof(struct virtio_blk_config, wce),
 629                                 &writeback);
 630         if (err)
 631                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE);
 632
 633         return writeback;
 634 }
 635
 636 static void virtblk_update_cache_mode(struct virtio_device *vdev)
 637 {
 638         u8 writeback = virtblk_get_cache_mode(vdev);
 639         struct virtio_blk *vblk = vdev->priv;
 640
 641         if (writeback)
 642                 blk_queue_flush(vblk->disk->queue, REQ_FLUSH);
 643         else
 644                 blk_queue_flush(vblk->disk->queue, 0);
 645
 646         revalidate_disk(vblk->disk);
 647 }
 648
 649 static const char *const virtblk_cache_types[] = {
 650         "write through", "write back"
 651 };
 652
 653 static ssize_t
 654 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
 655                          const char *buf, size_t count)
 656 {
 657         struct gendisk *disk = dev_to_disk(dev);
 658         struct virtio_blk *vblk = disk->private_data;
 659         struct virtio_device *vdev = vblk->vdev;
 660         int i;
 661         u8 writeback;
 662
 663         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
 664         for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
 665                 if (sysfs_streq(buf, virtblk_cache_types[i]))
 666                         break;
 667
 668         if (i < 0)
 669                 return -EINVAL;
 670
 671         writeback = i;
 672         vdev->config->set(vdev,
 673                           offsetof(struct virtio_blk_config, wce),
 674                           &writeback, sizeof(writeback));
 675
 676         virtblk_update_cache_mode(vdev);
 677         return count;
 678 }
 679
 680 static ssize_t
 681 virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
 682                          char *buf)
 683 {
 684         struct gendisk *disk = dev_to_disk(dev);
 685         struct virtio_blk *vblk = disk->private_data;
 686         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
 687
 688         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
 689         return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
 690 }
 691
 692 static const struct device_attribute dev_attr_cache_type_ro =
 693         __ATTR(cache_type, S_IRUGO,
 694                virtblk_cache_type_show, NULL);
 695 static const struct device_attribute dev_attr_cache_type_rw =
 696         __ATTR(cache_type, S_IRUGO|S_IWUSR,
 697                virtblk_cache_type_show, virtblk_cache_type_store);
 698
 699 static int __devinit virtblk_probe(struct virtio_device *vdev)
 700 {
 701         struct virtio_blk *vblk;
 702         struct request_queue *q;
 703         int err, index;
 704         int pool_size;
 705
 706         u64 cap;
 707         u32 v, blk_size, sg_elems, opt_io_size;
 708         u16 min_io_size;
 709         u8 physical_block_exp, alignment_offset;
 710
 711         err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
 712                              GFP_KERNEL);
 713         if (err < 0)
 714                 goto out;
 715         index = err;
 716
 717         /* We need to know how many segments before we allocate. */
 718         err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
 719                                 offsetof(struct virtio_blk_config, seg_max),
 720                                 &sg_elems);
 721
 722         /* We need at least one SG element, whatever they say. */
 723         if (err || !sg_elems)
 724                 sg_elems = 1;
 725
 726         /* We need an extra sg elements at head and tail. */
 727         sg_elems += 2;
 728         vdev->priv = vblk = kmalloc(sizeof(*vblk) +
 729                                     sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
 730         if (!vblk) {
 731                 err = -ENOMEM;
 732                 goto out_free_index;
 733         }
 734
 735         init_waitqueue_head(&vblk->queue_wait);
 736         vblk->vdev = vdev;
 737         vblk->sg_elems = sg_elems;
 738         sg_init_table(vblk->sg, vblk->sg_elems);
 739         mutex_init(&vblk->config_lock);
 740
 741         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 742         vblk->config_enable = true;
 743
 744         err = init_vq(vblk);
 745         if (err)
 746                 goto out_free_vblk;
 747
 748         pool_size = sizeof(struct virtblk_req);
 749         if (use_bio)
 750                 pool_size += sizeof(struct scatterlist) * sg_elems;
 751         vblk->pool = mempool_create_kmalloc_pool(1, pool_size);
 752         if (!vblk->pool) {
 753                 err = -ENOMEM;
 754                 goto out_free_vq;
 755         }
 756
 757         /* FIXME: How many partitions?  How long is a piece of string? */
 758         vblk->disk = alloc_disk(1 << PART_BITS);
 759         if (!vblk->disk) {
 760                 err = -ENOMEM;
 761                 goto out_mempool;
 762         }
 763
 764         q = vblk->disk->queue = blk_init_queue(virtblk_request, NULL);
 765         if (!q) {
 766                 err = -ENOMEM;
 767                 goto out_put_disk;
 768         }
 769
 770         if (use_bio)
 771                 blk_queue_make_request(q, virtblk_make_request);
 772         q->queuedata = vblk;
 773
 774         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
 775
 776         vblk->disk->major = major;
 777         vblk->disk->first_minor = index_to_minor(index);
 778         vblk->disk->private_data = vblk;
 779         vblk->disk->fops = &virtblk_fops;
 780         vblk->disk->driverfs_dev = &vdev->dev;
 781         vblk->index = index;
 782
 783         /* configure queue flush support */
 784         virtblk_update_cache_mode(vdev);
 785
 786         /* If disk is read-only in the host, the guest should obey */
 787         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 788                 set_disk_ro(vblk->disk, 1);
 789
 790         /* Host must always specify the capacity. */
 791         vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
 792                           &cap, sizeof(cap));
 793
 794         /* If capacity is too big, truncate with warning. */
 795         if ((sector_t)cap != cap) {
 796                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 797                          (unsigned long long)cap);
 798                 cap = (sector_t)-1;
 799         }
 800         set_capacity(vblk->disk, cap);
 801
 802         /* We can handle whatever the host told us to handle. */
 803         blk_queue_max_segments(q, vblk->sg_elems-2);
 804
 805         /* No need to bounce any requests */
 806         blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
 807
 808         /* No real sector limit. */
 809         blk_queue_max_hw_sectors(q, -1U);
 810
 811         /* Host can optionally specify maximum segment size and number of
 812          * segments. */
 813         err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
 814                                 offsetof(struct virtio_blk_config, size_max),
 815                                 &v);
 816         if (!err)
 817                 blk_queue_max_segment_size(q, v);
 818         else
 819                 blk_queue_max_segment_size(q, -1U);
 820
 821         /* Host can optionally specify the block size of the device */
 822         err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
 823                                 offsetof(struct virtio_blk_config, blk_size),
 824                                 &blk_size);
 825         if (!err)
 826                 blk_queue_logical_block_size(q, blk_size);
 827         else
 828                 blk_size = queue_logical_block_size(q);
 829
 830         /* Use topology information if available */
 831         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 832                         offsetof(struct virtio_blk_config, physical_block_exp),
 833                         &physical_block_exp);
 834         if (!err && physical_block_exp)
 835                 blk_queue_physical_block_size(q,
 836                                 blk_size * (1 << physical_block_exp));
 837
 838         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 839                         offsetof(struct virtio_blk_config, alignment_offset),
 840                         &alignment_offset);
 841         if (!err && alignment_offset)
 842                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 843
 844         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 845                         offsetof(struct virtio_blk_config, min_io_size),
 846                         &min_io_size);
 847         if (!err && min_io_size)
 848                 blk_queue_io_min(q, blk_size * min_io_size);
 849
 850         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 851                         offsetof(struct virtio_blk_config, opt_io_size),
 852                         &opt_io_size);
 853         if (!err && opt_io_size)
 854                 blk_queue_io_opt(q, blk_size * opt_io_size);
 855
 856         add_disk(vblk->disk);
 857         err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
 858         if (err)
 859                 goto out_del_disk;
 860
 861         if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
 862                 err = device_create_file(disk_to_dev(vblk->disk),
 863                                          &dev_attr_cache_type_rw);
 864         else
 865                 err = device_create_file(disk_to_dev(vblk->disk),
 866                                          &dev_attr_cache_type_ro);
 867         if (err)
 868                 goto out_del_disk;
 869         return 0;
 870
 871 out_del_disk:
 872         del_gendisk(vblk->disk);
 873         blk_cleanup_queue(vblk->disk->queue);
 874 out_put_disk:
 875         put_disk(vblk->disk);
 876 out_mempool:
 877         mempool_destroy(vblk->pool);
 878 out_free_vq:
 879         vdev->config->del_vqs(vdev);
 880 out_free_vblk:
 881         kfree(vblk);
 882 out_free_index:
 883         ida_simple_remove(&vd_index_ida, index);
 884 out:
 885         return err;
 886 }
 887
 888 static void __devexit virtblk_remove(struct virtio_device *vdev)
 889 {
 890         struct virtio_blk *vblk = vdev->priv;
 891         int index = vblk->index;
 892
 893         /* Prevent config work handler from accessing the device. */
 894         mutex_lock(&vblk->config_lock);
 895         vblk->config_enable = false;
 896         mutex_unlock(&vblk->config_lock);
 897
 898         del_gendisk(vblk->disk);
 899         blk_cleanup_queue(vblk->disk->queue);
 900
 901         /* Stop all the virtqueues. */
 902         vdev->config->reset(vdev);
 903
 904         flush_work(&vblk->config_work);
 905
 906         put_disk(vblk->disk);
 907         mempool_destroy(vblk->pool);
 908         vdev->config->del_vqs(vdev);
 909         kfree(vblk);
 910         ida_simple_remove(&vd_index_ida, index);
 911 }
 912
 913 #ifdef CONFIG_PM
 914 static int virtblk_freeze(struct virtio_device *vdev)
 915 {
 916         struct virtio_blk *vblk = vdev->priv;
 917
 918         /* Ensure we don't receive any more interrupts */
 919         vdev->config->reset(vdev);
 920
 921         /* Prevent config work handler from accessing the device. */
 922         mutex_lock(&vblk->config_lock);
 923         vblk->config_enable = false;
 924         mutex_unlock(&vblk->config_lock);
 925
 926         flush_work(&vblk->config_work);
 927
 928         spin_lock_irq(vblk->disk->queue->queue_lock);
 929         blk_stop_queue(vblk->disk->queue);
 930         spin_unlock_irq(vblk->disk->queue->queue_lock);
 931         blk_sync_queue(vblk->disk->queue);
 932
 933         vdev->config->del_vqs(vdev);
 934         return 0;
 935 }
 936
 937 static int virtblk_restore(struct virtio_device *vdev)
 938 {
 939         struct virtio_blk *vblk = vdev->priv;
 940         int ret;
 941
 942         vblk->config_enable = true;
 943         ret = init_vq(vdev->priv);
 944         if (!ret) {
 945                 spin_lock_irq(vblk->disk->queue->queue_lock);
 946                 blk_start_queue(vblk->disk->queue);
 947                 spin_unlock_irq(vblk->disk->queue->queue_lock);
 948         }
 949         return ret;
 950 }
 951 #endif
 952
 953 static const struct virtio_device_id id_table[] = {
 954         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 955         { 0 },
 956 };
 957
 958 static unsigned int features[] = {
 959         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 960         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
 961         VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE
 962 };
 963
 964 /*
 965  * virtio_blk causes spurious section mismatch warning by
 966  * simultaneously referring to a __devinit and a __devexit function.
 967  * Use __refdata to avoid this warning.
 968  */
 969 static struct virtio_driver __refdata virtio_blk = {
 970         .feature_table          = features,
 971         .feature_table_size     = ARRAY_SIZE(features),
 972         .driver.name            = KBUILD_MODNAME,
 973         .driver.owner           = THIS_MODULE,
 974         .id_table               = id_table,
 975         .probe                  = virtblk_probe,
 976         .remove                 = __devexit_p(virtblk_remove),
 977         .config_changed         = virtblk_config_changed,
 978 #ifdef CONFIG_PM
 979         .freeze                 = virtblk_freeze,
 980         .restore                = virtblk_restore,
 981 #endif
 982 };
 983
 984 static int __init init(void)
 985 {
 986         int error;
 987
 988         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
 989         if (!virtblk_wq)
 990                 return -ENOMEM;
 991
 992         major = register_blkdev(0, "virtblk");
 993         if (major < 0) {
 994                 error = major;
 995                 goto out_destroy_workqueue;
 996         }
 997
 998         error = register_virtio_driver(&virtio_blk);
 999         if (error)
1000                 goto out_unregister_blkdev;
1001         return 0;
1002
1003 out_unregister_blkdev:
1004         unregister_blkdev(major, "virtblk");
1005 out_destroy_workqueue:
1006         destroy_workqueue(virtblk_wq);
1007         return error;
1008 }
1009
1010 static void __exit fini(void)
1011 {
1012         unregister_blkdev(major, "virtblk");
1013         unregister_virtio_driver(&virtio_blk);
1014         destroy_workqueue(virtblk_wq);
1015 }
1016 module_init(init);
1017 module_exit(fini);
1018
1019 MODULE_DEVICE_TABLE(virtio, id_table);
1020 MODULE_DESCRIPTION("Virtio block driver");
1021 MODULE_LICENSE("GPL");