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 #include <linux/blk-mq.h>
  15 #include <linux/numa.h>
  16
  17 #define PART_BITS 4
  18
  19 static int major;
  20 static DEFINE_IDA(vd_index_ida);
  21
  22 static struct workqueue_struct *virtblk_wq;
  23
  24 struct virtio_blk
  25 {
  26         struct virtio_device *vdev;
  27         struct virtqueue *vq;
  28         spinlock_t vq_lock;
  29
  30         /* The disk structure for the kernel. */
  31         struct gendisk *disk;
  32
  33         /* Block layer tags. */
  34         struct blk_mq_tag_set tag_set;
  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
  52 struct virtblk_req
  53 {
  54         struct request *req;
  55         struct virtio_blk_outhdr out_hdr;
  56         struct virtio_scsi_inhdr in_hdr;
  57         u8 status;
  58         struct scatterlist sg[];
  59 };
  60
  61 static inline int virtblk_result(struct virtblk_req *vbr)
  62 {
  63         switch (vbr->status) {
  64         case VIRTIO_BLK_S_OK:
  65                 return 0;
  66         case VIRTIO_BLK_S_UNSUPP:
  67                 return -ENOTTY;
  68         default:
  69                 return -EIO;
  70         }
  71 }
  72
  73 static int __virtblk_add_req(struct virtqueue *vq,
  74                              struct virtblk_req *vbr,
  75                              struct scatterlist *data_sg,
  76                              bool have_data)
  77 {
  78         struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
  79         unsigned int num_out = 0, num_in = 0;
  80         int type = vbr->out_hdr.type & ~VIRTIO_BLK_T_OUT;
  81
  82         sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
  83         sgs[num_out++] = &hdr;
  84
  85         /*
  86          * If this is a packet command we need a couple of additional headers.
  87          * Behind the normal outhdr we put a segment with the scsi command
  88          * block, and before the normal inhdr we put the sense data and the
  89          * inhdr with additional status information.
  90          */
  91         if (type == VIRTIO_BLK_T_SCSI_CMD) {
  92                 sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len);
  93                 sgs[num_out++] = &cmd;
  94         }
  95
  96         if (have_data) {
  97                 if (vbr->out_hdr.type & VIRTIO_BLK_T_OUT)
  98                         sgs[num_out++] = data_sg;
  99                 else
 100                         sgs[num_out + num_in++] = data_sg;
 101         }
 102
 103         if (type == VIRTIO_BLK_T_SCSI_CMD) {
 104                 sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
 105                 sgs[num_out + num_in++] = &sense;
 106                 sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
 107                 sgs[num_out + num_in++] = &inhdr;
 108         }
 109
 110         sg_init_one(&status, &vbr->status, sizeof(vbr->status));
 111         sgs[num_out + num_in++] = &status;
 112
 113         return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 114 }
 115
 116 static inline void virtblk_request_done(struct request *req)
 117 {
 118         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 119         int error = virtblk_result(vbr);
 120
 121         if (req->cmd_type == REQ_TYPE_BLOCK_PC) {
 122                 req->resid_len = vbr->in_hdr.residual;
 123                 req->sense_len = vbr->in_hdr.sense_len;
 124                 req->errors = vbr->in_hdr.errors;
 125         } else if (req->cmd_type == REQ_TYPE_SPECIAL) {
 126                 req->errors = (error != 0);
 127         }
 128
 129         blk_mq_end_io(req, error);
 130 }
 131
 132 static void virtblk_done(struct virtqueue *vq)
 133 {
 134         struct virtio_blk *vblk = vq->vdev->priv;
 135         bool req_done = false;
 136         struct virtblk_req *vbr;
 137         unsigned long flags;
 138         unsigned int len;
 139
 140         spin_lock_irqsave(&vblk->vq_lock, flags);
 141         do {
 142                 virtqueue_disable_cb(vq);
 143                 while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
 144                         blk_mq_complete_request(vbr->req);
 145                         req_done = true;
 146                 }
 147                 if (unlikely(virtqueue_is_broken(vq)))
 148                         break;
 149         } while (!virtqueue_enable_cb(vq));
 150
 151         /* In case queue is stopped waiting for more buffers. */
 152         if (req_done)
 153                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 154         spin_unlock_irqrestore(&vblk->vq_lock, flags);
 155 }
 156
 157 static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req)
 158 {
 159         struct virtio_blk *vblk = hctx->queue->queuedata;
 160         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 161         unsigned long flags;
 162         unsigned int num;
 163         const bool last = (req->cmd_flags & REQ_END) != 0;
 164         int err;
 165         bool notify = false;
 166
 167         BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 168
 169         vbr->req = req;
 170         if (req->cmd_flags & REQ_FLUSH) {
 171                 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
 172                 vbr->out_hdr.sector = 0;
 173                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 174         } else {
 175                 switch (req->cmd_type) {
 176                 case REQ_TYPE_FS:
 177                         vbr->out_hdr.type = 0;
 178                         vbr->out_hdr.sector = blk_rq_pos(vbr->req);
 179                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 180                         break;
 181                 case REQ_TYPE_BLOCK_PC:
 182                         vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
 183                         vbr->out_hdr.sector = 0;
 184                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 185                         break;
 186                 case REQ_TYPE_SPECIAL:
 187                         vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
 188                         vbr->out_hdr.sector = 0;
 189                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 190                         break;
 191                 default:
 192                         /* We don't put anything else in the queue. */
 193                         BUG();
 194                 }
 195         }
 196
 197         num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg);
 198         if (num) {
 199                 if (rq_data_dir(vbr->req) == WRITE)
 200                         vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 201                 else
 202                         vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 203         }
 204
 205         spin_lock_irqsave(&vblk->vq_lock, flags);
 206         err = __virtblk_add_req(vblk->vq, vbr, vbr->sg, num);
 207         if (err) {
 208                 virtqueue_kick(vblk->vq);
 209                 blk_mq_stop_hw_queue(hctx);
 210                 spin_unlock_irqrestore(&vblk->vq_lock, flags);
 211                 /* Out of mem doesn't actually happen, since we fall back
 212                  * to direct descriptors */
 213                 if (err == -ENOMEM || err == -ENOSPC)
 214                         return BLK_MQ_RQ_QUEUE_BUSY;
 215                 return BLK_MQ_RQ_QUEUE_ERROR;
 216         }
 217
 218         if (last && virtqueue_kick_prepare(vblk->vq))
 219                 notify = true;
 220         spin_unlock_irqrestore(&vblk->vq_lock, flags);
 221
 222         if (notify)
 223                 virtqueue_notify(vblk->vq);
 224         return BLK_MQ_RQ_QUEUE_OK;
 225 }
 226
 227 /* return id (s/n) string for *disk to *id_str
 228  */
 229 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 230 {
 231         struct virtio_blk *vblk = disk->private_data;
 232         struct request *req;
 233         struct bio *bio;
 234         int err;
 235
 236         bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
 237                            GFP_KERNEL);
 238         if (IS_ERR(bio))
 239                 return PTR_ERR(bio);
 240
 241         req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
 242         if (IS_ERR(req)) {
 243                 bio_put(bio);
 244                 return PTR_ERR(req);
 245         }
 246
 247         req->cmd_type = REQ_TYPE_SPECIAL;
 248         err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
 249         blk_put_request(req);
 250
 251         return err;
 252 }
 253
 254 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
 255                              unsigned int cmd, unsigned long data)
 256 {
 257         struct gendisk *disk = bdev->bd_disk;
 258         struct virtio_blk *vblk = disk->private_data;
 259
 260         /*
 261          * Only allow the generic SCSI ioctls if the host can support it.
 262          */
 263         if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
 264                 return -ENOTTY;
 265
 266         return scsi_cmd_blk_ioctl(bdev, mode, cmd,
 267                                   (void __user *)data);
 268 }
 269
 270 /* We provide getgeo only to please some old bootloader/partitioning tools */
 271 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 272 {
 273         struct virtio_blk *vblk = bd->bd_disk->private_data;
 274
 275         /* see if the host passed in geometry config */
 276         if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 277                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 278                              geometry.cylinders, &geo->cylinders);
 279                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 280                              geometry.heads, &geo->heads);
 281                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 282                              geometry.sectors, &geo->sectors);
 283         } else {
 284                 /* some standard values, similar to sd */
 285                 geo->heads = 1 << 6;
 286                 geo->sectors = 1 << 5;
 287                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 288         }
 289         return 0;
 290 }
 291
 292 static const struct block_device_operations virtblk_fops = {
 293         .ioctl  = virtblk_ioctl,
 294         .owner  = THIS_MODULE,
 295         .getgeo = virtblk_getgeo,
 296 };
 297
 298 static int index_to_minor(int index)
 299 {
 300         return index << PART_BITS;
 301 }
 302
 303 static int minor_to_index(int minor)
 304 {
 305         return minor >> PART_BITS;
 306 }
 307
 308 static ssize_t virtblk_serial_show(struct device *dev,
 309                                 struct device_attribute *attr, char *buf)
 310 {
 311         struct gendisk *disk = dev_to_disk(dev);
 312         int err;
 313
 314         /* sysfs gives us a PAGE_SIZE buffer */
 315         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 316
 317         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 318         err = virtblk_get_id(disk, buf);
 319         if (!err)
 320                 return strlen(buf);
 321
 322         if (err == -EIO) /* Unsupported? Make it empty. */
 323                 return 0;
 324
 325         return err;
 326 }
 327 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
 328
 329 static void virtblk_config_changed_work(struct work_struct *work)
 330 {
 331         struct virtio_blk *vblk =
 332                 container_of(work, struct virtio_blk, config_work);
 333         struct virtio_device *vdev = vblk->vdev;
 334         struct request_queue *q = vblk->disk->queue;
 335         char cap_str_2[10], cap_str_10[10];
 336         char *envp[] = { "RESIZE=1", NULL };
 337         u64 capacity, size;
 338
 339         mutex_lock(&vblk->config_lock);
 340         if (!vblk->config_enable)
 341                 goto done;
 342
 343         /* Host must always specify the capacity. */
 344         virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
 345
 346         /* If capacity is too big, truncate with warning. */
 347         if ((sector_t)capacity != capacity) {
 348                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 349                          (unsigned long long)capacity);
 350                 capacity = (sector_t)-1;
 351         }
 352
 353         size = capacity * queue_logical_block_size(q);
 354         string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 355         string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 356
 357         dev_notice(&vdev->dev,
 358                   "new size: %llu %d-byte logical blocks (%s/%s)\n",
 359                   (unsigned long long)capacity,
 360                   queue_logical_block_size(q),
 361                   cap_str_10, cap_str_2);
 362
 363         set_capacity(vblk->disk, capacity);
 364         revalidate_disk(vblk->disk);
 365         kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
 366 done:
 367         mutex_unlock(&vblk->config_lock);
 368 }
 369
 370 static void virtblk_config_changed(struct virtio_device *vdev)
 371 {
 372         struct virtio_blk *vblk = vdev->priv;
 373
 374         queue_work(virtblk_wq, &vblk->config_work);
 375 }
 376
 377 static int init_vq(struct virtio_blk *vblk)
 378 {
 379         int err = 0;
 380
 381         /* We expect one virtqueue, for output. */
 382         vblk->vq = virtio_find_single_vq(vblk->vdev, virtblk_done, "requests");
 383         if (IS_ERR(vblk->vq))
 384                 err = PTR_ERR(vblk->vq);
 385
 386         return err;
 387 }
 388
 389 /*
 390  * Legacy naming scheme used for virtio devices.  We are stuck with it for
 391  * virtio blk but don't ever use it for any new driver.
 392  */
 393 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
 394 {
 395         const int base = 'z' - 'a' + 1;
 396         char *begin = buf + strlen(prefix);
 397         char *end = buf + buflen;
 398         char *p;
 399         int unit;
 400
 401         p = end - 1;
 402         *p = '\0';
 403         unit = base;
 404         do {
 405                 if (p == begin)
 406                         return -EINVAL;
 407                 *--p = 'a' + (index % unit);
 408                 index = (index / unit) - 1;
 409         } while (index >= 0);
 410
 411         memmove(begin, p, end - p);
 412         memcpy(buf, prefix, strlen(prefix));
 413
 414         return 0;
 415 }
 416
 417 static int virtblk_get_cache_mode(struct virtio_device *vdev)
 418 {
 419         u8 writeback;
 420         int err;
 421
 422         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
 423                                    struct virtio_blk_config, wce,
 424                                    &writeback);
 425         if (err)
 426                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE);
 427
 428         return writeback;
 429 }
 430
 431 static void virtblk_update_cache_mode(struct virtio_device *vdev)
 432 {
 433         u8 writeback = virtblk_get_cache_mode(vdev);
 434         struct virtio_blk *vblk = vdev->priv;
 435
 436         if (writeback)
 437                 blk_queue_flush(vblk->disk->queue, REQ_FLUSH);
 438         else
 439                 blk_queue_flush(vblk->disk->queue, 0);
 440
 441         revalidate_disk(vblk->disk);
 442 }
 443
 444 static const char *const virtblk_cache_types[] = {
 445         "write through", "write back"
 446 };
 447
 448 static ssize_t
 449 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
 450                          const char *buf, size_t count)
 451 {
 452         struct gendisk *disk = dev_to_disk(dev);
 453         struct virtio_blk *vblk = disk->private_data;
 454         struct virtio_device *vdev = vblk->vdev;
 455         int i;
 456
 457         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
 458         for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
 459                 if (sysfs_streq(buf, virtblk_cache_types[i]))
 460                         break;
 461
 462         if (i < 0)
 463                 return -EINVAL;
 464
 465         virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
 466         virtblk_update_cache_mode(vdev);
 467         return count;
 468 }
 469
 470 static ssize_t
 471 virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
 472                          char *buf)
 473 {
 474         struct gendisk *disk = dev_to_disk(dev);
 475         struct virtio_blk *vblk = disk->private_data;
 476         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
 477
 478         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
 479         return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
 480 }
 481
 482 static const struct device_attribute dev_attr_cache_type_ro =
 483         __ATTR(cache_type, S_IRUGO,
 484                virtblk_cache_type_show, NULL);
 485 static const struct device_attribute dev_attr_cache_type_rw =
 486         __ATTR(cache_type, S_IRUGO|S_IWUSR,
 487                virtblk_cache_type_show, virtblk_cache_type_store);
 488
 489 static int virtblk_init_request(void *data, struct request *rq,
 490                 unsigned int hctx_idx, unsigned int request_idx,
 491                 unsigned int numa_node)
 492 {
 493         struct virtio_blk *vblk = data;
 494         struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
 495
 496         sg_init_table(vbr->sg, vblk->sg_elems);
 497         return 0;
 498 }
 499
 500 static struct blk_mq_ops virtio_mq_ops = {
 501         .queue_rq       = virtio_queue_rq,
 502         .map_queue      = blk_mq_map_queue,
 503         .complete       = virtblk_request_done,
 504         .init_request   = virtblk_init_request,
 505 };
 506
 507 static unsigned int virtblk_queue_depth;
 508 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
 509
 510 static int virtblk_probe(struct virtio_device *vdev)
 511 {
 512         struct virtio_blk *vblk;
 513         struct request_queue *q;
 514         int err, index;
 515
 516         u64 cap;
 517         u32 v, blk_size, sg_elems, opt_io_size;
 518         u16 min_io_size;
 519         u8 physical_block_exp, alignment_offset;
 520
 521         err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
 522                              GFP_KERNEL);
 523         if (err < 0)
 524                 goto out;
 525         index = err;
 526
 527         /* We need to know how many segments before we allocate. */
 528         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
 529                                    struct virtio_blk_config, seg_max,
 530                                    &sg_elems);
 531
 532         /* We need at least one SG element, whatever they say. */
 533         if (err || !sg_elems)
 534                 sg_elems = 1;
 535
 536         /* We need an extra sg elements at head and tail. */
 537         sg_elems += 2;
 538         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
 539         if (!vblk) {
 540                 err = -ENOMEM;
 541                 goto out_free_index;
 542         }
 543
 544         vblk->vdev = vdev;
 545         vblk->sg_elems = sg_elems;
 546         mutex_init(&vblk->config_lock);
 547
 548         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 549         vblk->config_enable = true;
 550
 551         err = init_vq(vblk);
 552         if (err)
 553                 goto out_free_vblk;
 554         spin_lock_init(&vblk->vq_lock);
 555
 556         /* FIXME: How many partitions?  How long is a piece of string? */
 557         vblk->disk = alloc_disk(1 << PART_BITS);
 558         if (!vblk->disk) {
 559                 err = -ENOMEM;
 560                 goto out_free_vq;
 561         }
 562
 563         /* Default queue sizing is to fill the ring. */
 564         if (!virtblk_queue_depth) {
 565                 virtblk_queue_depth = vblk->vq->num_free;
 566                 /* ... but without indirect descs, we use 2 descs per req */
 567                 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
 568                         virtblk_queue_depth /= 2;
 569         }
 570
 571         memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
 572         vblk->tag_set.ops = &virtio_mq_ops;
 573         vblk->tag_set.nr_hw_queues = 1;
 574         vblk->tag_set.queue_depth = virtblk_queue_depth;
 575         vblk->tag_set.numa_node = NUMA_NO_NODE;
 576         vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 577         vblk->tag_set.cmd_size =
 578                 sizeof(struct virtblk_req) +
 579                 sizeof(struct scatterlist) * sg_elems;
 580         vblk->tag_set.driver_data = vblk;
 581
 582         err = blk_mq_alloc_tag_set(&vblk->tag_set);
 583         if (err)
 584                 goto out_put_disk;
 585
 586         q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set);
 587         if (!q) {
 588                 err = -ENOMEM;
 589                 goto out_free_tags;
 590         }
 591
 592         q->queuedata = vblk;
 593
 594         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
 595
 596         vblk->disk->major = major;
 597         vblk->disk->first_minor = index_to_minor(index);
 598         vblk->disk->private_data = vblk;
 599         vblk->disk->fops = &virtblk_fops;
 600         vblk->disk->driverfs_dev = &vdev->dev;
 601         vblk->index = index;
 602
 603         /* configure queue flush support */
 604         virtblk_update_cache_mode(vdev);
 605
 606         /* If disk is read-only in the host, the guest should obey */
 607         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 608                 set_disk_ro(vblk->disk, 1);
 609
 610         /* Host must always specify the capacity. */
 611         virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
 612
 613         /* If capacity is too big, truncate with warning. */
 614         if ((sector_t)cap != cap) {
 615                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 616                          (unsigned long long)cap);
 617                 cap = (sector_t)-1;
 618         }
 619         set_capacity(vblk->disk, cap);
 620
 621         /* We can handle whatever the host told us to handle. */
 622         blk_queue_max_segments(q, vblk->sg_elems-2);
 623
 624         /* No need to bounce any requests */
 625         blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
 626
 627         /* No real sector limit. */
 628         blk_queue_max_hw_sectors(q, -1U);
 629
 630         /* Host can optionally specify maximum segment size and number of
 631          * segments. */
 632         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
 633                                    struct virtio_blk_config, size_max, &v);
 634         if (!err)
 635                 blk_queue_max_segment_size(q, v);
 636         else
 637                 blk_queue_max_segment_size(q, -1U);
 638
 639         /* Host can optionally specify the block size of the device */
 640         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
 641                                    struct virtio_blk_config, blk_size,
 642                                    &blk_size);
 643         if (!err)
 644                 blk_queue_logical_block_size(q, blk_size);
 645         else
 646                 blk_size = queue_logical_block_size(q);
 647
 648         /* Use topology information if available */
 649         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 650                                    struct virtio_blk_config, physical_block_exp,
 651                                    &physical_block_exp);
 652         if (!err && physical_block_exp)
 653                 blk_queue_physical_block_size(q,
 654                                 blk_size * (1 << physical_block_exp));
 655
 656         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 657                                    struct virtio_blk_config, alignment_offset,
 658                                    &alignment_offset);
 659         if (!err && alignment_offset)
 660                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 661
 662         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 663                                    struct virtio_blk_config, min_io_size,
 664                                    &min_io_size);
 665         if (!err && min_io_size)
 666                 blk_queue_io_min(q, blk_size * min_io_size);
 667
 668         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 669                                    struct virtio_blk_config, opt_io_size,
 670                                    &opt_io_size);
 671         if (!err && opt_io_size)
 672                 blk_queue_io_opt(q, blk_size * opt_io_size);
 673
 674         add_disk(vblk->disk);
 675         err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
 676         if (err)
 677                 goto out_del_disk;
 678
 679         if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
 680                 err = device_create_file(disk_to_dev(vblk->disk),
 681                                          &dev_attr_cache_type_rw);
 682         else
 683                 err = device_create_file(disk_to_dev(vblk->disk),
 684                                          &dev_attr_cache_type_ro);
 685         if (err)
 686                 goto out_del_disk;
 687         return 0;
 688
 689 out_del_disk:
 690         del_gendisk(vblk->disk);
 691         blk_cleanup_queue(vblk->disk->queue);
 692 out_free_tags:
 693         blk_mq_free_tag_set(&vblk->tag_set);
 694 out_put_disk:
 695         put_disk(vblk->disk);
 696 out_free_vq:
 697         vdev->config->del_vqs(vdev);
 698 out_free_vblk:
 699         kfree(vblk);
 700 out_free_index:
 701         ida_simple_remove(&vd_index_ida, index);
 702 out:
 703         return err;
 704 }
 705
 706 static void virtblk_remove(struct virtio_device *vdev)
 707 {
 708         struct virtio_blk *vblk = vdev->priv;
 709         int index = vblk->index;
 710         int refc;
 711
 712         /* Prevent config work handler from accessing the device. */
 713         mutex_lock(&vblk->config_lock);
 714         vblk->config_enable = false;
 715         mutex_unlock(&vblk->config_lock);
 716
 717         del_gendisk(vblk->disk);
 718         blk_cleanup_queue(vblk->disk->queue);
 719
 720         blk_mq_free_tag_set(&vblk->tag_set);
 721
 722         /* Stop all the virtqueues. */
 723         vdev->config->reset(vdev);
 724
 725         flush_work(&vblk->config_work);
 726
 727         refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount);
 728         put_disk(vblk->disk);
 729         vdev->config->del_vqs(vdev);
 730         kfree(vblk);
 731
 732         /* Only free device id if we don't have any users */
 733         if (refc == 1)
 734                 ida_simple_remove(&vd_index_ida, index);
 735 }
 736
 737 #ifdef CONFIG_PM_SLEEP
 738 static int virtblk_freeze(struct virtio_device *vdev)
 739 {
 740         struct virtio_blk *vblk = vdev->priv;
 741
 742         /* Ensure we don't receive any more interrupts */
 743         vdev->config->reset(vdev);
 744
 745         /* Prevent config work handler from accessing the device. */
 746         mutex_lock(&vblk->config_lock);
 747         vblk->config_enable = false;
 748         mutex_unlock(&vblk->config_lock);
 749
 750         flush_work(&vblk->config_work);
 751
 752         blk_mq_stop_hw_queues(vblk->disk->queue);
 753
 754         vdev->config->del_vqs(vdev);
 755         return 0;
 756 }
 757
 758 static int virtblk_restore(struct virtio_device *vdev)
 759 {
 760         struct virtio_blk *vblk = vdev->priv;
 761         int ret;
 762
 763         vblk->config_enable = true;
 764         ret = init_vq(vdev->priv);
 765         if (!ret)
 766                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 767
 768         return ret;
 769 }
 770 #endif
 771
 772 static const struct virtio_device_id id_table[] = {
 773         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 774         { 0 },
 775 };
 776
 777 static unsigned int features[] = {
 778         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 779         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
 780         VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE
 781 };
 782
 783 static struct virtio_driver virtio_blk = {
 784         .feature_table          = features,
 785         .feature_table_size     = ARRAY_SIZE(features),
 786         .driver.name            = KBUILD_MODNAME,
 787         .driver.owner           = THIS_MODULE,
 788         .id_table               = id_table,
 789         .probe                  = virtblk_probe,
 790         .remove                 = virtblk_remove,
 791         .config_changed         = virtblk_config_changed,
 792 #ifdef CONFIG_PM_SLEEP
 793         .freeze                 = virtblk_freeze,
 794         .restore                = virtblk_restore,
 795 #endif
 796 };
 797
 798 static int __init init(void)
 799 {
 800         int error;
 801
 802         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
 803         if (!virtblk_wq)
 804                 return -ENOMEM;
 805
 806         major = register_blkdev(0, "virtblk");
 807         if (major < 0) {
 808                 error = major;
 809                 goto out_destroy_workqueue;
 810         }
 811
 812         error = register_virtio_driver(&virtio_blk);
 813         if (error)
 814                 goto out_unregister_blkdev;
 815         return 0;
 816
 817 out_unregister_blkdev:
 818         unregister_blkdev(major, "virtblk");
 819 out_destroy_workqueue:
 820         destroy_workqueue(virtblk_wq);
 821         return error;
 822 }
 823
 824 static void __exit fini(void)
 825 {
 826         unregister_blkdev(major, "virtblk");
 827         unregister_virtio_driver(&virtio_blk);
 828         destroy_workqueue(virtblk_wq);
 829 }
 830 module_init(init);
 831 module_exit(fini);
 832
 833 MODULE_DEVICE_TABLE(virtio, id_table);
 834 MODULE_DESCRIPTION("Virtio block driver");
 835 MODULE_LICENSE("GPL");