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 #define VQ_NAME_LEN 16
  19
  20 static int major;
  21 static DEFINE_IDA(vd_index_ida);
  22
  23 static struct workqueue_struct *virtblk_wq;
  24
  25 struct virtio_blk_vq {
  26         struct virtqueue *vq;
  27         spinlock_t lock;
  28         char name[VQ_NAME_LEN];
  29 } ____cacheline_aligned_in_smp;
  30
  31 struct virtio_blk
  32 {
  33         struct virtio_device *vdev;
  34
  35         /* The disk structure for the kernel. */
  36         struct gendisk *disk;
  37
  38         /* Block layer tags. */
  39         struct blk_mq_tag_set tag_set;
  40
  41         /* Process context for config space updates */
  42         struct work_struct config_work;
  43
  44         /* Lock for config space updates */
  45         struct mutex config_lock;
  46
  47         /* enable config space updates */
  48         bool config_enable;
  49
  50         /* What host tells us, plus 2 for header & tailer. */
  51         unsigned int sg_elems;
  52
  53         /* Ida index - used to track minor number allocations. */
  54         int index;
  55
  56         /* num of vqs */
  57         int num_vqs;
  58         struct virtio_blk_vq *vqs;
  59 };
  60
  61 struct virtblk_req
  62 {
  63         struct request *req;
  64         struct virtio_blk_outhdr out_hdr;
  65         struct virtio_scsi_inhdr in_hdr;
  66         u8 status;
  67         struct scatterlist sg[];
  68 };
  69
  70 static inline int virtblk_result(struct virtblk_req *vbr)
  71 {
  72         switch (vbr->status) {
  73         case VIRTIO_BLK_S_OK:
  74                 return 0;
  75         case VIRTIO_BLK_S_UNSUPP:
  76                 return -ENOTTY;
  77         default:
  78                 return -EIO;
  79         }
  80 }
  81
  82 static int __virtblk_add_req(struct virtqueue *vq,
  83                              struct virtblk_req *vbr,
  84                              struct scatterlist *data_sg,
  85                              bool have_data)
  86 {
  87         struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
  88         unsigned int num_out = 0, num_in = 0;
  89         int type = vbr->out_hdr.type & ~VIRTIO_BLK_T_OUT;
  90
  91         sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
  92         sgs[num_out++] = &hdr;
  93
  94         /*
  95          * If this is a packet command we need a couple of additional headers.
  96          * Behind the normal outhdr we put a segment with the scsi command
  97          * block, and before the normal inhdr we put the sense data and the
  98          * inhdr with additional status information.
  99          */
 100         if (type == VIRTIO_BLK_T_SCSI_CMD) {
 101                 sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len);
 102                 sgs[num_out++] = &cmd;
 103         }
 104
 105         if (have_data) {
 106                 if (vbr->out_hdr.type & VIRTIO_BLK_T_OUT)
 107                         sgs[num_out++] = data_sg;
 108                 else
 109                         sgs[num_out + num_in++] = data_sg;
 110         }
 111
 112         if (type == VIRTIO_BLK_T_SCSI_CMD) {
 113                 sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
 114                 sgs[num_out + num_in++] = &sense;
 115                 sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
 116                 sgs[num_out + num_in++] = &inhdr;
 117         }
 118
 119         sg_init_one(&status, &vbr->status, sizeof(vbr->status));
 120         sgs[num_out + num_in++] = &status;
 121
 122         return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 123 }
 124
 125 static inline void virtblk_request_done(struct request *req)
 126 {
 127         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 128         int error = virtblk_result(vbr);
 129
 130         if (req->cmd_type == REQ_TYPE_BLOCK_PC) {
 131                 req->resid_len = vbr->in_hdr.residual;
 132                 req->sense_len = vbr->in_hdr.sense_len;
 133                 req->errors = vbr->in_hdr.errors;
 134         } else if (req->cmd_type == REQ_TYPE_SPECIAL) {
 135                 req->errors = (error != 0);
 136         }
 137
 138         blk_mq_end_io(req, error);
 139 }
 140
 141 static void virtblk_done(struct virtqueue *vq)
 142 {
 143         struct virtio_blk *vblk = vq->vdev->priv;
 144         bool req_done = false;
 145         int qid = vq->index;
 146         struct virtblk_req *vbr;
 147         unsigned long flags;
 148         unsigned int len;
 149
 150         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 151         do {
 152                 virtqueue_disable_cb(vq);
 153                 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
 154                         blk_mq_complete_request(vbr->req);
 155                         req_done = true;
 156                 }
 157                 if (unlikely(virtqueue_is_broken(vq)))
 158                         break;
 159         } while (!virtqueue_enable_cb(vq));
 160
 161         /* In case queue is stopped waiting for more buffers. */
 162         if (req_done)
 163                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 164         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 165 }
 166
 167 static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req)
 168 {
 169         struct virtio_blk *vblk = hctx->queue->queuedata;
 170         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 171         unsigned long flags;
 172         unsigned int num;
 173         int qid = hctx->queue_num;
 174         const bool last = (req->cmd_flags & REQ_END) != 0;
 175         int err;
 176         bool notify = false;
 177
 178         BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 179
 180         vbr->req = req;
 181         if (req->cmd_flags & REQ_FLUSH) {
 182                 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
 183                 vbr->out_hdr.sector = 0;
 184                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 185         } else {
 186                 switch (req->cmd_type) {
 187                 case REQ_TYPE_FS:
 188                         vbr->out_hdr.type = 0;
 189                         vbr->out_hdr.sector = blk_rq_pos(vbr->req);
 190                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 191                         break;
 192                 case REQ_TYPE_BLOCK_PC:
 193                         vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
 194                         vbr->out_hdr.sector = 0;
 195                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 196                         break;
 197                 case REQ_TYPE_SPECIAL:
 198                         vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
 199                         vbr->out_hdr.sector = 0;
 200                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 201                         break;
 202                 default:
 203                         /* We don't put anything else in the queue. */
 204                         BUG();
 205                 }
 206         }
 207
 208         num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg);
 209         if (num) {
 210                 if (rq_data_dir(vbr->req) == WRITE)
 211                         vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 212                 else
 213                         vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 214         }
 215
 216         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 217         err = __virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
 218         if (err) {
 219                 virtqueue_kick(vblk->vqs[qid].vq);
 220                 blk_mq_stop_hw_queue(hctx);
 221                 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 222                 /* Out of mem doesn't actually happen, since we fall back
 223                  * to direct descriptors */
 224                 if (err == -ENOMEM || err == -ENOSPC)
 225                         return BLK_MQ_RQ_QUEUE_BUSY;
 226                 return BLK_MQ_RQ_QUEUE_ERROR;
 227         }
 228
 229         if (last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
 230                 notify = true;
 231         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 232
 233         if (notify)
 234                 virtqueue_notify(vblk->vqs[qid].vq);
 235         return BLK_MQ_RQ_QUEUE_OK;
 236 }
 237
 238 /* return id (s/n) string for *disk to *id_str
 239  */
 240 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 241 {
 242         struct virtio_blk *vblk = disk->private_data;
 243         struct request *req;
 244         struct bio *bio;
 245         int err;
 246
 247         bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
 248                            GFP_KERNEL);
 249         if (IS_ERR(bio))
 250                 return PTR_ERR(bio);
 251
 252         req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
 253         if (IS_ERR(req)) {
 254                 bio_put(bio);
 255                 return PTR_ERR(req);
 256         }
 257
 258         req->cmd_type = REQ_TYPE_SPECIAL;
 259         err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
 260         blk_put_request(req);
 261
 262         return err;
 263 }
 264
 265 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
 266                              unsigned int cmd, unsigned long data)
 267 {
 268         struct gendisk *disk = bdev->bd_disk;
 269         struct virtio_blk *vblk = disk->private_data;
 270
 271         /*
 272          * Only allow the generic SCSI ioctls if the host can support it.
 273          */
 274         if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
 275                 return -ENOTTY;
 276
 277         return scsi_cmd_blk_ioctl(bdev, mode, cmd,
 278                                   (void __user *)data);
 279 }
 280
 281 /* We provide getgeo only to please some old bootloader/partitioning tools */
 282 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 283 {
 284         struct virtio_blk *vblk = bd->bd_disk->private_data;
 285
 286         /* see if the host passed in geometry config */
 287         if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 288                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 289                              geometry.cylinders, &geo->cylinders);
 290                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 291                              geometry.heads, &geo->heads);
 292                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 293                              geometry.sectors, &geo->sectors);
 294         } else {
 295                 /* some standard values, similar to sd */
 296                 geo->heads = 1 << 6;
 297                 geo->sectors = 1 << 5;
 298                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 299         }
 300         return 0;
 301 }
 302
 303 static const struct block_device_operations virtblk_fops = {
 304         .ioctl  = virtblk_ioctl,
 305         .owner  = THIS_MODULE,
 306         .getgeo = virtblk_getgeo,
 307 };
 308
 309 static int index_to_minor(int index)
 310 {
 311         return index << PART_BITS;
 312 }
 313
 314 static int minor_to_index(int minor)
 315 {
 316         return minor >> PART_BITS;
 317 }
 318
 319 static ssize_t virtblk_serial_show(struct device *dev,
 320                                 struct device_attribute *attr, char *buf)
 321 {
 322         struct gendisk *disk = dev_to_disk(dev);
 323         int err;
 324
 325         /* sysfs gives us a PAGE_SIZE buffer */
 326         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 327
 328         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 329         err = virtblk_get_id(disk, buf);
 330         if (!err)
 331                 return strlen(buf);
 332
 333         if (err == -EIO) /* Unsupported? Make it empty. */
 334                 return 0;
 335
 336         return err;
 337 }
 338 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
 339
 340 static void virtblk_config_changed_work(struct work_struct *work)
 341 {
 342         struct virtio_blk *vblk =
 343                 container_of(work, struct virtio_blk, config_work);
 344         struct virtio_device *vdev = vblk->vdev;
 345         struct request_queue *q = vblk->disk->queue;
 346         char cap_str_2[10], cap_str_10[10];
 347         char *envp[] = { "RESIZE=1", NULL };
 348         u64 capacity, size;
 349
 350         mutex_lock(&vblk->config_lock);
 351         if (!vblk->config_enable)
 352                 goto done;
 353
 354         /* Host must always specify the capacity. */
 355         virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
 356
 357         /* If capacity is too big, truncate with warning. */
 358         if ((sector_t)capacity != capacity) {
 359                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 360                          (unsigned long long)capacity);
 361                 capacity = (sector_t)-1;
 362         }
 363
 364         size = capacity * queue_logical_block_size(q);
 365         string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 366         string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 367
 368         dev_notice(&vdev->dev,
 369                   "new size: %llu %d-byte logical blocks (%s/%s)\n",
 370                   (unsigned long long)capacity,
 371                   queue_logical_block_size(q),
 372                   cap_str_10, cap_str_2);
 373
 374         set_capacity(vblk->disk, capacity);
 375         revalidate_disk(vblk->disk);
 376         kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
 377 done:
 378         mutex_unlock(&vblk->config_lock);
 379 }
 380
 381 static void virtblk_config_changed(struct virtio_device *vdev)
 382 {
 383         struct virtio_blk *vblk = vdev->priv;
 384
 385         queue_work(virtblk_wq, &vblk->config_work);
 386 }
 387
 388 static int init_vq(struct virtio_blk *vblk)
 389 {
 390         int err = 0;
 391         int i;
 392         vq_callback_t **callbacks;
 393         const char **names;
 394         struct virtqueue **vqs;
 395         unsigned short num_vqs;
 396         struct virtio_device *vdev = vblk->vdev;
 397
 398         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
 399                                    struct virtio_blk_config, num_queues,
 400                                    &num_vqs);
 401         if (err)
 402                 num_vqs = 1;
 403
 404         vblk->vqs = kmalloc(sizeof(*vblk->vqs) * num_vqs, GFP_KERNEL);
 405         if (!vblk->vqs) {
 406                 err = -ENOMEM;
 407                 goto out;
 408         }
 409
 410         names = kmalloc(sizeof(*names) * num_vqs, GFP_KERNEL);
 411         if (!names)
 412                 goto err_names;
 413
 414         callbacks = kmalloc(sizeof(*callbacks) * num_vqs, GFP_KERNEL);
 415         if (!callbacks)
 416                 goto err_callbacks;
 417
 418         vqs = kmalloc(sizeof(*vqs) * num_vqs, GFP_KERNEL);
 419         if (!vqs)
 420                 goto err_vqs;
 421
 422         for (i = 0; i < num_vqs; i++) {
 423                 callbacks[i] = virtblk_done;
 424                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
 425                 names[i] = vblk->vqs[i].name;
 426         }
 427
 428         /* Discover virtqueues and write information to configuration.  */
 429         err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names);
 430         if (err)
 431                 goto err_find_vqs;
 432
 433         for (i = 0; i < num_vqs; i++) {
 434                 spin_lock_init(&vblk->vqs[i].lock);
 435                 vblk->vqs[i].vq = vqs[i];
 436         }
 437         vblk->num_vqs = num_vqs;
 438
 439  err_find_vqs:
 440         kfree(vqs);
 441  err_vqs:
 442         kfree(callbacks);
 443  err_callbacks:
 444         kfree(names);
 445  err_names:
 446         if (err)
 447                 kfree(vblk->vqs);
 448  out:
 449         return err;
 450 }
 451
 452 /*
 453  * Legacy naming scheme used for virtio devices.  We are stuck with it for
 454  * virtio blk but don't ever use it for any new driver.
 455  */
 456 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
 457 {
 458         const int base = 'z' - 'a' + 1;
 459         char *begin = buf + strlen(prefix);
 460         char *end = buf + buflen;
 461         char *p;
 462         int unit;
 463
 464         p = end - 1;
 465         *p = '\0';
 466         unit = base;
 467         do {
 468                 if (p == begin)
 469                         return -EINVAL;
 470                 *--p = 'a' + (index % unit);
 471                 index = (index / unit) - 1;
 472         } while (index >= 0);
 473
 474         memmove(begin, p, end - p);
 475         memcpy(buf, prefix, strlen(prefix));
 476
 477         return 0;
 478 }
 479
 480 static int virtblk_get_cache_mode(struct virtio_device *vdev)
 481 {
 482         u8 writeback;
 483         int err;
 484
 485         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
 486                                    struct virtio_blk_config, wce,
 487                                    &writeback);
 488         if (err)
 489                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE);
 490
 491         return writeback;
 492 }
 493
 494 static void virtblk_update_cache_mode(struct virtio_device *vdev)
 495 {
 496         u8 writeback = virtblk_get_cache_mode(vdev);
 497         struct virtio_blk *vblk = vdev->priv;
 498
 499         if (writeback)
 500                 blk_queue_flush(vblk->disk->queue, REQ_FLUSH);
 501         else
 502                 blk_queue_flush(vblk->disk->queue, 0);
 503
 504         revalidate_disk(vblk->disk);
 505 }
 506
 507 static const char *const virtblk_cache_types[] = {
 508         "write through", "write back"
 509 };
 510
 511 static ssize_t
 512 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
 513                          const char *buf, size_t count)
 514 {
 515         struct gendisk *disk = dev_to_disk(dev);
 516         struct virtio_blk *vblk = disk->private_data;
 517         struct virtio_device *vdev = vblk->vdev;
 518         int i;
 519
 520         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
 521         for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
 522                 if (sysfs_streq(buf, virtblk_cache_types[i]))
 523                         break;
 524
 525         if (i < 0)
 526                 return -EINVAL;
 527
 528         virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
 529         virtblk_update_cache_mode(vdev);
 530         return count;
 531 }
 532
 533 static ssize_t
 534 virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
 535                          char *buf)
 536 {
 537         struct gendisk *disk = dev_to_disk(dev);
 538         struct virtio_blk *vblk = disk->private_data;
 539         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
 540
 541         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
 542         return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
 543 }
 544
 545 static const struct device_attribute dev_attr_cache_type_ro =
 546         __ATTR(cache_type, S_IRUGO,
 547                virtblk_cache_type_show, NULL);
 548 static const struct device_attribute dev_attr_cache_type_rw =
 549         __ATTR(cache_type, S_IRUGO|S_IWUSR,
 550                virtblk_cache_type_show, virtblk_cache_type_store);
 551
 552 static int virtblk_init_request(void *data, struct request *rq,
 553                 unsigned int hctx_idx, unsigned int request_idx,
 554                 unsigned int numa_node)
 555 {
 556         struct virtio_blk *vblk = data;
 557         struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
 558
 559         sg_init_table(vbr->sg, vblk->sg_elems);
 560         return 0;
 561 }
 562
 563 static struct blk_mq_ops virtio_mq_ops = {
 564         .queue_rq       = virtio_queue_rq,
 565         .map_queue      = blk_mq_map_queue,
 566         .complete       = virtblk_request_done,
 567         .init_request   = virtblk_init_request,
 568 };
 569
 570 static unsigned int virtblk_queue_depth;
 571 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
 572
 573 static int virtblk_probe(struct virtio_device *vdev)
 574 {
 575         struct virtio_blk *vblk;
 576         struct request_queue *q;
 577         int err, index;
 578
 579         u64 cap;
 580         u32 v, blk_size, sg_elems, opt_io_size;
 581         u16 min_io_size;
 582         u8 physical_block_exp, alignment_offset;
 583
 584         err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
 585                              GFP_KERNEL);
 586         if (err < 0)
 587                 goto out;
 588         index = err;
 589
 590         /* We need to know how many segments before we allocate. */
 591         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
 592                                    struct virtio_blk_config, seg_max,
 593                                    &sg_elems);
 594
 595         /* We need at least one SG element, whatever they say. */
 596         if (err || !sg_elems)
 597                 sg_elems = 1;
 598
 599         /* We need an extra sg elements at head and tail. */
 600         sg_elems += 2;
 601         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
 602         if (!vblk) {
 603                 err = -ENOMEM;
 604                 goto out_free_index;
 605         }
 606
 607         vblk->vdev = vdev;
 608         vblk->sg_elems = sg_elems;
 609         mutex_init(&vblk->config_lock);
 610
 611         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 612         vblk->config_enable = true;
 613
 614         err = init_vq(vblk);
 615         if (err)
 616                 goto out_free_vblk;
 617
 618         /* FIXME: How many partitions?  How long is a piece of string? */
 619         vblk->disk = alloc_disk(1 << PART_BITS);
 620         if (!vblk->disk) {
 621                 err = -ENOMEM;
 622                 goto out_free_vq;
 623         }
 624
 625         /* Default queue sizing is to fill the ring. */
 626         if (!virtblk_queue_depth) {
 627                 virtblk_queue_depth = vblk->vqs[0].vq->num_free;
 628                 /* ... but without indirect descs, we use 2 descs per req */
 629                 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
 630                         virtblk_queue_depth /= 2;
 631         }
 632
 633         memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
 634         vblk->tag_set.ops = &virtio_mq_ops;
 635         vblk->tag_set.queue_depth = virtblk_queue_depth;
 636         vblk->tag_set.numa_node = NUMA_NO_NODE;
 637         vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 638         vblk->tag_set.cmd_size =
 639                 sizeof(struct virtblk_req) +
 640                 sizeof(struct scatterlist) * sg_elems;
 641         vblk->tag_set.driver_data = vblk;
 642         vblk->tag_set.nr_hw_queues = vblk->num_vqs;
 643
 644         err = blk_mq_alloc_tag_set(&vblk->tag_set);
 645         if (err)
 646                 goto out_put_disk;
 647
 648         q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set);
 649         if (!q) {
 650                 err = -ENOMEM;
 651                 goto out_free_tags;
 652         }
 653
 654         q->queuedata = vblk;
 655
 656         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
 657
 658         vblk->disk->major = major;
 659         vblk->disk->first_minor = index_to_minor(index);
 660         vblk->disk->private_data = vblk;
 661         vblk->disk->fops = &virtblk_fops;
 662         vblk->disk->driverfs_dev = &vdev->dev;
 663         vblk->index = index;
 664
 665         /* configure queue flush support */
 666         virtblk_update_cache_mode(vdev);
 667
 668         /* If disk is read-only in the host, the guest should obey */
 669         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 670                 set_disk_ro(vblk->disk, 1);
 671
 672         /* Host must always specify the capacity. */
 673         virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
 674
 675         /* If capacity is too big, truncate with warning. */
 676         if ((sector_t)cap != cap) {
 677                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 678                          (unsigned long long)cap);
 679                 cap = (sector_t)-1;
 680         }
 681         set_capacity(vblk->disk, cap);
 682
 683         /* We can handle whatever the host told us to handle. */
 684         blk_queue_max_segments(q, vblk->sg_elems-2);
 685
 686         /* No need to bounce any requests */
 687         blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
 688
 689         /* No real sector limit. */
 690         blk_queue_max_hw_sectors(q, -1U);
 691
 692         /* Host can optionally specify maximum segment size and number of
 693          * segments. */
 694         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
 695                                    struct virtio_blk_config, size_max, &v);
 696         if (!err)
 697                 blk_queue_max_segment_size(q, v);
 698         else
 699                 blk_queue_max_segment_size(q, -1U);
 700
 701         /* Host can optionally specify the block size of the device */
 702         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
 703                                    struct virtio_blk_config, blk_size,
 704                                    &blk_size);
 705         if (!err)
 706                 blk_queue_logical_block_size(q, blk_size);
 707         else
 708                 blk_size = queue_logical_block_size(q);
 709
 710         /* Use topology information if available */
 711         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 712                                    struct virtio_blk_config, physical_block_exp,
 713                                    &physical_block_exp);
 714         if (!err && physical_block_exp)
 715                 blk_queue_physical_block_size(q,
 716                                 blk_size * (1 << physical_block_exp));
 717
 718         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 719                                    struct virtio_blk_config, alignment_offset,
 720                                    &alignment_offset);
 721         if (!err && alignment_offset)
 722                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 723
 724         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 725                                    struct virtio_blk_config, min_io_size,
 726                                    &min_io_size);
 727         if (!err && min_io_size)
 728                 blk_queue_io_min(q, blk_size * min_io_size);
 729
 730         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 731                                    struct virtio_blk_config, opt_io_size,
 732                                    &opt_io_size);
 733         if (!err && opt_io_size)
 734                 blk_queue_io_opt(q, blk_size * opt_io_size);
 735
 736         add_disk(vblk->disk);
 737         err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
 738         if (err)
 739                 goto out_del_disk;
 740
 741         if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
 742                 err = device_create_file(disk_to_dev(vblk->disk),
 743                                          &dev_attr_cache_type_rw);
 744         else
 745                 err = device_create_file(disk_to_dev(vblk->disk),
 746                                          &dev_attr_cache_type_ro);
 747         if (err)
 748                 goto out_del_disk;
 749         return 0;
 750
 751 out_del_disk:
 752         del_gendisk(vblk->disk);
 753         blk_cleanup_queue(vblk->disk->queue);
 754 out_free_tags:
 755         blk_mq_free_tag_set(&vblk->tag_set);
 756 out_put_disk:
 757         put_disk(vblk->disk);
 758 out_free_vq:
 759         vdev->config->del_vqs(vdev);
 760 out_free_vblk:
 761         kfree(vblk);
 762 out_free_index:
 763         ida_simple_remove(&vd_index_ida, index);
 764 out:
 765         return err;
 766 }
 767
 768 static void virtblk_remove(struct virtio_device *vdev)
 769 {
 770         struct virtio_blk *vblk = vdev->priv;
 771         int index = vblk->index;
 772         int refc;
 773
 774         /* Prevent config work handler from accessing the device. */
 775         mutex_lock(&vblk->config_lock);
 776         vblk->config_enable = false;
 777         mutex_unlock(&vblk->config_lock);
 778
 779         del_gendisk(vblk->disk);
 780         blk_cleanup_queue(vblk->disk->queue);
 781
 782         blk_mq_free_tag_set(&vblk->tag_set);
 783
 784         /* Stop all the virtqueues. */
 785         vdev->config->reset(vdev);
 786
 787         flush_work(&vblk->config_work);
 788
 789         refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount);
 790         put_disk(vblk->disk);
 791         vdev->config->del_vqs(vdev);
 792         kfree(vblk->vqs);
 793         kfree(vblk);
 794
 795         /* Only free device id if we don't have any users */
 796         if (refc == 1)
 797                 ida_simple_remove(&vd_index_ida, index);
 798 }
 799
 800 #ifdef CONFIG_PM_SLEEP
 801 static int virtblk_freeze(struct virtio_device *vdev)
 802 {
 803         struct virtio_blk *vblk = vdev->priv;
 804
 805         /* Ensure we don't receive any more interrupts */
 806         vdev->config->reset(vdev);
 807
 808         /* Prevent config work handler from accessing the device. */
 809         mutex_lock(&vblk->config_lock);
 810         vblk->config_enable = false;
 811         mutex_unlock(&vblk->config_lock);
 812
 813         flush_work(&vblk->config_work);
 814
 815         blk_mq_stop_hw_queues(vblk->disk->queue);
 816
 817         vdev->config->del_vqs(vdev);
 818         return 0;
 819 }
 820
 821 static int virtblk_restore(struct virtio_device *vdev)
 822 {
 823         struct virtio_blk *vblk = vdev->priv;
 824         int ret;
 825
 826         vblk->config_enable = true;
 827         ret = init_vq(vdev->priv);
 828         if (!ret)
 829                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 830
 831         return ret;
 832 }
 833 #endif
 834
 835 static const struct virtio_device_id id_table[] = {
 836         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 837         { 0 },
 838 };
 839
 840 static unsigned int features[] = {
 841         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 842         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
 843         VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
 844         VIRTIO_BLK_F_MQ,
 845 };
 846
 847 static struct virtio_driver virtio_blk = {
 848         .feature_table          = features,
 849         .feature_table_size     = ARRAY_SIZE(features),
 850         .driver.name            = KBUILD_MODNAME,
 851         .driver.owner           = THIS_MODULE,
 852         .id_table               = id_table,
 853         .probe                  = virtblk_probe,
 854         .remove                 = virtblk_remove,
 855         .config_changed         = virtblk_config_changed,
 856 #ifdef CONFIG_PM_SLEEP
 857         .freeze                 = virtblk_freeze,
 858         .restore                = virtblk_restore,
 859 #endif
 860 };
 861
 862 static int __init init(void)
 863 {
 864         int error;
 865
 866         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
 867         if (!virtblk_wq)
 868                 return -ENOMEM;
 869
 870         major = register_blkdev(0, "virtblk");
 871         if (major < 0) {
 872                 error = major;
 873                 goto out_destroy_workqueue;
 874         }
 875
 876         error = register_virtio_driver(&virtio_blk);
 877         if (error)
 878                 goto out_unregister_blkdev;
 879         return 0;
 880
 881 out_unregister_blkdev:
 882         unregister_blkdev(major, "virtblk");
 883 out_destroy_workqueue:
 884         destroy_workqueue(virtblk_wq);
 885         return error;
 886 }
 887
 888 static void __exit fini(void)
 889 {
 890         unregister_blkdev(major, "virtblk");
 891         unregister_virtio_driver(&virtio_blk);
 892         destroy_workqueue(virtblk_wq);
 893 }
 894 module_init(init);
 895 module_exit(fini);
 896
 897 MODULE_DEVICE_TABLE(virtio, id_table);
 898 MODULE_DESCRIPTION("Virtio block driver");
 899 MODULE_LICENSE("GPL");