drivers/block/virtio_blk.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 //#define DEBUG
   3 #include <linux/spinlock.h>
   4 #include <linux/slab.h>
   5 #include <linux/blkdev.h>
   6 #include <linux/hdreg.h>
   7 #include <linux/module.h>
   8 #include <linux/mutex.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/virtio.h>
  11 #include <linux/virtio_blk.h>
  12 #include <linux/scatterlist.h>
  13 #include <linux/string_helpers.h>
  14 #include <linux/idr.h>
  15 #include <linux/blk-mq.h>
  16 #include <linux/blk-mq-virtio.h>
  17 #include <linux/numa.h>
  18
  19 #define PART_BITS 4
  20 #define VQ_NAME_LEN 16
  21 #define MAX_DISCARD_SEGMENTS 256u
  22
  23 static int major;
  24 static DEFINE_IDA(vd_index_ida);
  25
  26 static struct workqueue_struct *virtblk_wq;
  27
  28 struct virtio_blk_vq {
  29         struct virtqueue *vq;
  30         spinlock_t lock;
  31         char name[VQ_NAME_LEN];
  32 } ____cacheline_aligned_in_smp;
  33
  34 struct virtio_blk {
  35         struct virtio_device *vdev;
  36
  37         /* The disk structure for the kernel. */
  38         struct gendisk *disk;
  39
  40         /* Block layer tags. */
  41         struct blk_mq_tag_set tag_set;
  42
  43         /* Process context for config space updates */
  44         struct work_struct config_work;
  45
  46         /* What host tells us, plus 2 for header & tailer. */
  47         unsigned int sg_elems;
  48
  49         /* Ida index - used to track minor number allocations. */
  50         int index;
  51
  52         /* num of vqs */
  53         int num_vqs;
  54         struct virtio_blk_vq *vqs;
  55 };
  56
  57 struct virtblk_req {
  58         struct virtio_blk_outhdr out_hdr;
  59         u8 status;
  60         struct scatterlist sg[];
  61 };
  62
  63 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
  64 {
  65         switch (vbr->status) {
  66         case VIRTIO_BLK_S_OK:
  67                 return BLK_STS_OK;
  68         case VIRTIO_BLK_S_UNSUPP:
  69                 return BLK_STS_NOTSUPP;
  70         default:
  71                 return BLK_STS_IOERR;
  72         }
  73 }
  74
  75 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
  76                 struct scatterlist *data_sg, bool have_data)
  77 {
  78         struct scatterlist hdr, status, *sgs[3];
  79         unsigned int num_out = 0, num_in = 0;
  80
  81         sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
  82         sgs[num_out++] = &hdr;
  83
  84         if (have_data) {
  85                 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
  86                         sgs[num_out++] = data_sg;
  87                 else
  88                         sgs[num_out + num_in++] = data_sg;
  89         }
  90
  91         sg_init_one(&status, &vbr->status, sizeof(vbr->status));
  92         sgs[num_out + num_in++] = &status;
  93
  94         return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
  95 }
  96
  97 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
  98 {
  99         unsigned short segments = blk_rq_nr_discard_segments(req);
 100         unsigned short n = 0;
 101         struct virtio_blk_discard_write_zeroes *range;
 102         struct bio *bio;
 103         u32 flags = 0;
 104
 105         if (unmap)
 106                 flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
 107
 108         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 109         if (!range)
 110                 return -ENOMEM;
 111
 112         __rq_for_each_bio(bio, req) {
 113                 u64 sector = bio->bi_iter.bi_sector;
 114                 u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
 115
 116                 range[n].flags = cpu_to_le32(flags);
 117                 range[n].num_sectors = cpu_to_le32(num_sectors);
 118                 range[n].sector = cpu_to_le64(sector);
 119                 n++;
 120         }
 121
 122         req->special_vec.bv_page = virt_to_page(range);
 123         req->special_vec.bv_offset = offset_in_page(range);
 124         req->special_vec.bv_len = sizeof(*range) * segments;
 125         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 126
 127         return 0;
 128 }
 129
 130 static inline void virtblk_request_done(struct request *req)
 131 {
 132         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 133
 134         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 135                 kfree(page_address(req->special_vec.bv_page) +
 136                       req->special_vec.bv_offset);
 137         }
 138
 139         blk_mq_end_request(req, virtblk_result(vbr));
 140 }
 141
 142 static void virtblk_done(struct virtqueue *vq)
 143 {
 144         struct virtio_blk *vblk = vq->vdev->priv;
 145         bool req_done = false;
 146         int qid = vq->index;
 147         struct virtblk_req *vbr;
 148         unsigned long flags;
 149         unsigned int len;
 150
 151         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 152         do {
 153                 virtqueue_disable_cb(vq);
 154                 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
 155                         struct request *req = blk_mq_rq_from_pdu(vbr);
 156
 157                         blk_mq_complete_request(req);
 158                         req_done = true;
 159                 }
 160                 if (unlikely(virtqueue_is_broken(vq)))
 161                         break;
 162         } while (!virtqueue_enable_cb(vq));
 163
 164         /* In case queue is stopped waiting for more buffers. */
 165         if (req_done)
 166                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 167         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 168 }
 169
 170 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
 171 {
 172         struct virtio_blk *vblk = hctx->queue->queuedata;
 173         struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 174         bool kick;
 175
 176         spin_lock_irq(&vq->lock);
 177         kick = virtqueue_kick_prepare(vq->vq);
 178         spin_unlock_irq(&vq->lock);
 179
 180         if (kick)
 181                 virtqueue_notify(vq->vq);
 182 }
 183
 184 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 185                            const struct blk_mq_queue_data *bd)
 186 {
 187         struct virtio_blk *vblk = hctx->queue->queuedata;
 188         struct request *req = bd->rq;
 189         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 190         unsigned long flags;
 191         unsigned int num;
 192         int qid = hctx->queue_num;
 193         int err;
 194         bool notify = false;
 195         bool unmap = false;
 196         u32 type;
 197
 198         BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 199
 200         switch (req_op(req)) {
 201         case REQ_OP_READ:
 202         case REQ_OP_WRITE:
 203                 type = 0;
 204                 break;
 205         case REQ_OP_FLUSH:
 206                 type = VIRTIO_BLK_T_FLUSH;
 207                 break;
 208         case REQ_OP_DISCARD:
 209                 type = VIRTIO_BLK_T_DISCARD;
 210                 break;
 211         case REQ_OP_WRITE_ZEROES:
 212                 type = VIRTIO_BLK_T_WRITE_ZEROES;
 213                 unmap = !(req->cmd_flags & REQ_NOUNMAP);
 214                 break;
 215         case REQ_OP_DRV_IN:
 216                 type = VIRTIO_BLK_T_GET_ID;
 217                 break;
 218         default:
 219                 WARN_ON_ONCE(1);
 220                 return BLK_STS_IOERR;
 221         }
 222
 223         vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
 224         vbr->out_hdr.sector = type ?
 225                 0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
 226         vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
 227
 228         blk_mq_start_request(req);
 229
 230         if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
 231                 err = virtblk_setup_discard_write_zeroes(req, unmap);
 232                 if (err)
 233                         return BLK_STS_RESOURCE;
 234         }
 235
 236         num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
 237         if (num) {
 238                 if (rq_data_dir(req) == WRITE)
 239                         vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
 240                 else
 241                         vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
 242         }
 243
 244         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 245         err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
 246         if (err) {
 247                 virtqueue_kick(vblk->vqs[qid].vq);
 248                 /* Don't stop the queue if -ENOMEM: we may have failed to
 249                  * bounce the buffer due to global resource outage.
 250                  */
 251                 if (err == -ENOSPC)
 252                         blk_mq_stop_hw_queue(hctx);
 253                 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 254                 switch (err) {
 255                 case -ENOSPC:
 256                         return BLK_STS_DEV_RESOURCE;
 257                 case -ENOMEM:
 258                         return BLK_STS_RESOURCE;
 259                 default:
 260                         return BLK_STS_IOERR;
 261                 }
 262         }
 263
 264         if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
 265                 notify = true;
 266         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 267
 268         if (notify)
 269                 virtqueue_notify(vblk->vqs[qid].vq);
 270         return BLK_STS_OK;
 271 }
 272
 273 /* return id (s/n) string for *disk to *id_str
 274  */
 275 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 276 {
 277         struct virtio_blk *vblk = disk->private_data;
 278         struct request_queue *q = vblk->disk->queue;
 279         struct request *req;
 280         int err;
 281
 282         req = blk_get_request(q, REQ_OP_DRV_IN, 0);
 283         if (IS_ERR(req))
 284                 return PTR_ERR(req);
 285
 286         err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
 287         if (err)
 288                 goto out;
 289
 290         blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
 291         err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
 292 out:
 293         blk_put_request(req);
 294         return err;
 295 }
 296
 297 /* We provide getgeo only to please some old bootloader/partitioning tools */
 298 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 299 {
 300         struct virtio_blk *vblk = bd->bd_disk->private_data;
 301
 302         /* see if the host passed in geometry config */
 303         if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 304                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 305                              geometry.cylinders, &geo->cylinders);
 306                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 307                              geometry.heads, &geo->heads);
 308                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 309                              geometry.sectors, &geo->sectors);
 310         } else {
 311                 /* some standard values, similar to sd */
 312                 geo->heads = 1 << 6;
 313                 geo->sectors = 1 << 5;
 314                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 315         }
 316         return 0;
 317 }
 318
 319 static const struct block_device_operations virtblk_fops = {
 320         .owner  = THIS_MODULE,
 321         .getgeo = virtblk_getgeo,
 322 };
 323
 324 static int index_to_minor(int index)
 325 {
 326         return index << PART_BITS;
 327 }
 328
 329 static int minor_to_index(int minor)
 330 {
 331         return minor >> PART_BITS;
 332 }
 333
 334 static ssize_t serial_show(struct device *dev,
 335                            struct device_attribute *attr, char *buf)
 336 {
 337         struct gendisk *disk = dev_to_disk(dev);
 338         int err;
 339
 340         /* sysfs gives us a PAGE_SIZE buffer */
 341         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 342
 343         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 344         err = virtblk_get_id(disk, buf);
 345         if (!err)
 346                 return strlen(buf);
 347
 348         if (err == -EIO) /* Unsupported? Make it empty. */
 349                 return 0;
 350
 351         return err;
 352 }
 353
 354 static DEVICE_ATTR_RO(serial);
 355
 356 /* The queue's logical block size must be set before calling this */
 357 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
 358 {
 359         struct virtio_device *vdev = vblk->vdev;
 360         struct request_queue *q = vblk->disk->queue;
 361         char cap_str_2[10], cap_str_10[10];
 362         unsigned long long nblocks;
 363         u64 capacity;
 364
 365         /* Host must always specify the capacity. */
 366         virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
 367
 368         /* If capacity is too big, truncate with warning. */
 369         if ((sector_t)capacity != capacity) {
 370                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 371                          (unsigned long long)capacity);
 372                 capacity = (sector_t)-1;
 373         }
 374
 375         nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
 376
 377         string_get_size(nblocks, queue_logical_block_size(q),
 378                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 379         string_get_size(nblocks, queue_logical_block_size(q),
 380                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 381
 382         dev_notice(&vdev->dev,
 383                    "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
 384                    vblk->disk->disk_name,
 385                    resize ? "new size: " : "",
 386                    nblocks,
 387                    queue_logical_block_size(q),
 388                    cap_str_10,
 389                    cap_str_2);
 390
 391         set_capacity_revalidate_and_notify(vblk->disk, capacity, true);
 392 }
 393
 394 static void virtblk_config_changed_work(struct work_struct *work)
 395 {
 396         struct virtio_blk *vblk =
 397                 container_of(work, struct virtio_blk, config_work);
 398
 399         virtblk_update_capacity(vblk, true);
 400 }
 401
 402 static void virtblk_config_changed(struct virtio_device *vdev)
 403 {
 404         struct virtio_blk *vblk = vdev->priv;
 405
 406         queue_work(virtblk_wq, &vblk->config_work);
 407 }
 408
 409 static int init_vq(struct virtio_blk *vblk)
 410 {
 411         int err;
 412         int i;
 413         vq_callback_t **callbacks;
 414         const char **names;
 415         struct virtqueue **vqs;
 416         unsigned short num_vqs;
 417         struct virtio_device *vdev = vblk->vdev;
 418         struct irq_affinity desc = { 0, };
 419
 420         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
 421                                    struct virtio_blk_config, num_queues,
 422                                    &num_vqs);
 423         if (err)
 424                 num_vqs = 1;
 425
 426         num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
 427
 428         vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
 429         if (!vblk->vqs)
 430                 return -ENOMEM;
 431
 432         names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
 433         callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
 434         vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
 435         if (!names || !callbacks || !vqs) {
 436                 err = -ENOMEM;
 437                 goto out;
 438         }
 439
 440         for (i = 0; i < num_vqs; i++) {
 441                 callbacks[i] = virtblk_done;
 442                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
 443                 names[i] = vblk->vqs[i].name;
 444         }
 445
 446         /* Discover virtqueues and write information to configuration.  */
 447         err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
 448         if (err)
 449                 goto out;
 450
 451         for (i = 0; i < num_vqs; i++) {
 452                 spin_lock_init(&vblk->vqs[i].lock);
 453                 vblk->vqs[i].vq = vqs[i];
 454         }
 455         vblk->num_vqs = num_vqs;
 456
 457 out:
 458         kfree(vqs);
 459         kfree(callbacks);
 460         kfree(names);
 461         if (err)
 462                 kfree(vblk->vqs);
 463         return err;
 464 }
 465
 466 /*
 467  * Legacy naming scheme used for virtio devices.  We are stuck with it for
 468  * virtio blk but don't ever use it for any new driver.
 469  */
 470 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
 471 {
 472         const int base = 'z' - 'a' + 1;
 473         char *begin = buf + strlen(prefix);
 474         char *end = buf + buflen;
 475         char *p;
 476         int unit;
 477
 478         p = end - 1;
 479         *p = '\0';
 480         unit = base;
 481         do {
 482                 if (p == begin)
 483                         return -EINVAL;
 484                 *--p = 'a' + (index % unit);
 485                 index = (index / unit) - 1;
 486         } while (index >= 0);
 487
 488         memmove(begin, p, end - p);
 489         memcpy(buf, prefix, strlen(prefix));
 490
 491         return 0;
 492 }
 493
 494 static int virtblk_get_cache_mode(struct virtio_device *vdev)
 495 {
 496         u8 writeback;
 497         int err;
 498
 499         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
 500                                    struct virtio_blk_config, wce,
 501                                    &writeback);
 502
 503         /*
 504          * If WCE is not configurable and flush is not available,
 505          * assume no writeback cache is in use.
 506          */
 507         if (err)
 508                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
 509
 510         return writeback;
 511 }
 512
 513 static void virtblk_update_cache_mode(struct virtio_device *vdev)
 514 {
 515         u8 writeback = virtblk_get_cache_mode(vdev);
 516         struct virtio_blk *vblk = vdev->priv;
 517
 518         blk_queue_write_cache(vblk->disk->queue, writeback, false);
 519         revalidate_disk(vblk->disk);
 520 }
 521
 522 static const char *const virtblk_cache_types[] = {
 523         "write through", "write back"
 524 };
 525
 526 static ssize_t
 527 cache_type_store(struct device *dev, struct device_attribute *attr,
 528                  const char *buf, size_t count)
 529 {
 530         struct gendisk *disk = dev_to_disk(dev);
 531         struct virtio_blk *vblk = disk->private_data;
 532         struct virtio_device *vdev = vblk->vdev;
 533         int i;
 534
 535         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
 536         i = sysfs_match_string(virtblk_cache_types, buf);
 537         if (i < 0)
 538                 return i;
 539
 540         virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
 541         virtblk_update_cache_mode(vdev);
 542         return count;
 543 }
 544
 545 static ssize_t
 546 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 547 {
 548         struct gendisk *disk = dev_to_disk(dev);
 549         struct virtio_blk *vblk = disk->private_data;
 550         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
 551
 552         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
 553         return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
 554 }
 555
 556 static DEVICE_ATTR_RW(cache_type);
 557
 558 static struct attribute *virtblk_attrs[] = {
 559         &dev_attr_serial.attr,
 560         &dev_attr_cache_type.attr,
 561         NULL,
 562 };
 563
 564 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
 565                 struct attribute *a, int n)
 566 {
 567         struct device *dev = container_of(kobj, struct device, kobj);
 568         struct gendisk *disk = dev_to_disk(dev);
 569         struct virtio_blk *vblk = disk->private_data;
 570         struct virtio_device *vdev = vblk->vdev;
 571
 572         if (a == &dev_attr_cache_type.attr &&
 573             !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
 574                 return S_IRUGO;
 575
 576         return a->mode;
 577 }
 578
 579 static const struct attribute_group virtblk_attr_group = {
 580         .attrs = virtblk_attrs,
 581         .is_visible = virtblk_attrs_are_visible,
 582 };
 583
 584 static const struct attribute_group *virtblk_attr_groups[] = {
 585         &virtblk_attr_group,
 586         NULL,
 587 };
 588
 589 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
 590                 unsigned int hctx_idx, unsigned int numa_node)
 591 {
 592         struct virtio_blk *vblk = set->driver_data;
 593         struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
 594
 595         sg_init_table(vbr->sg, vblk->sg_elems);
 596         return 0;
 597 }
 598
 599 static int virtblk_map_queues(struct blk_mq_tag_set *set)
 600 {
 601         struct virtio_blk *vblk = set->driver_data;
 602
 603         return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
 604                                         vblk->vdev, 0);
 605 }
 606
 607 static const struct blk_mq_ops virtio_mq_ops = {
 608         .queue_rq       = virtio_queue_rq,
 609         .commit_rqs     = virtio_commit_rqs,
 610         .complete       = virtblk_request_done,
 611         .init_request   = virtblk_init_request,
 612         .map_queues     = virtblk_map_queues,
 613 };
 614
 615 static unsigned int virtblk_queue_depth;
 616 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
 617
 618 static int virtblk_probe(struct virtio_device *vdev)
 619 {
 620         struct virtio_blk *vblk;
 621         struct request_queue *q;
 622         int err, index;
 623
 624         u32 v, blk_size, max_size, sg_elems, opt_io_size;
 625         u16 min_io_size;
 626         u8 physical_block_exp, alignment_offset;
 627
 628         if (!vdev->config->get) {
 629                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
 630                         __func__);
 631                 return -EINVAL;
 632         }
 633
 634         err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
 635                              GFP_KERNEL);
 636         if (err < 0)
 637                 goto out;
 638         index = err;
 639
 640         /* We need to know how many segments before we allocate. */
 641         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
 642                                    struct virtio_blk_config, seg_max,
 643                                    &sg_elems);
 644
 645         /* We need at least one SG element, whatever they say. */
 646         if (err || !sg_elems)
 647                 sg_elems = 1;
 648
 649         /* We need an extra sg elements at head and tail. */
 650         sg_elems += 2;
 651         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
 652         if (!vblk) {
 653                 err = -ENOMEM;
 654                 goto out_free_index;
 655         }
 656
 657         vblk->vdev = vdev;
 658         vblk->sg_elems = sg_elems;
 659
 660         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 661
 662         err = init_vq(vblk);
 663         if (err)
 664                 goto out_free_vblk;
 665
 666         /* FIXME: How many partitions?  How long is a piece of string? */
 667         vblk->disk = alloc_disk(1 << PART_BITS);
 668         if (!vblk->disk) {
 669                 err = -ENOMEM;
 670                 goto out_free_vq;
 671         }
 672
 673         /* Default queue sizing is to fill the ring. */
 674         if (!virtblk_queue_depth) {
 675                 virtblk_queue_depth = vblk->vqs[0].vq->num_free;
 676                 /* ... but without indirect descs, we use 2 descs per req */
 677                 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
 678                         virtblk_queue_depth /= 2;
 679         }
 680
 681         memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
 682         vblk->tag_set.ops = &virtio_mq_ops;
 683         vblk->tag_set.queue_depth = virtblk_queue_depth;
 684         vblk->tag_set.numa_node = NUMA_NO_NODE;
 685         vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 686         vblk->tag_set.cmd_size =
 687                 sizeof(struct virtblk_req) +
 688                 sizeof(struct scatterlist) * sg_elems;
 689         vblk->tag_set.driver_data = vblk;
 690         vblk->tag_set.nr_hw_queues = vblk->num_vqs;
 691
 692         err = blk_mq_alloc_tag_set(&vblk->tag_set);
 693         if (err)
 694                 goto out_put_disk;
 695
 696         q = blk_mq_init_queue(&vblk->tag_set);
 697         if (IS_ERR(q)) {
 698                 err = -ENOMEM;
 699                 goto out_free_tags;
 700         }
 701         vblk->disk->queue = q;
 702
 703         q->queuedata = vblk;
 704
 705         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
 706
 707         vblk->disk->major = major;
 708         vblk->disk->first_minor = index_to_minor(index);
 709         vblk->disk->private_data = vblk;
 710         vblk->disk->fops = &virtblk_fops;
 711         vblk->disk->flags |= GENHD_FL_EXT_DEVT;
 712         vblk->index = index;
 713
 714         /* configure queue flush support */
 715         virtblk_update_cache_mode(vdev);
 716
 717         /* If disk is read-only in the host, the guest should obey */
 718         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 719                 set_disk_ro(vblk->disk, 1);
 720
 721         /* We can handle whatever the host told us to handle. */
 722         blk_queue_max_segments(q, vblk->sg_elems-2);
 723
 724         /* No real sector limit. */
 725         blk_queue_max_hw_sectors(q, -1U);
 726
 727         max_size = virtio_max_dma_size(vdev);
 728
 729         /* Host can optionally specify maximum segment size and number of
 730          * segments. */
 731         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
 732                                    struct virtio_blk_config, size_max, &v);
 733         if (!err)
 734                 max_size = min(max_size, v);
 735
 736         blk_queue_max_segment_size(q, max_size);
 737
 738         /* Host can optionally specify the block size of the device */
 739         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
 740                                    struct virtio_blk_config, blk_size,
 741                                    &blk_size);
 742         if (!err)
 743                 blk_queue_logical_block_size(q, blk_size);
 744         else
 745                 blk_size = queue_logical_block_size(q);
 746
 747         /* Use topology information if available */
 748         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 749                                    struct virtio_blk_config, physical_block_exp,
 750                                    &physical_block_exp);
 751         if (!err && physical_block_exp)
 752                 blk_queue_physical_block_size(q,
 753                                 blk_size * (1 << physical_block_exp));
 754
 755         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 756                                    struct virtio_blk_config, alignment_offset,
 757                                    &alignment_offset);
 758         if (!err && alignment_offset)
 759                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 760
 761         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 762                                    struct virtio_blk_config, min_io_size,
 763                                    &min_io_size);
 764         if (!err && min_io_size)
 765                 blk_queue_io_min(q, blk_size * min_io_size);
 766
 767         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 768                                    struct virtio_blk_config, opt_io_size,
 769                                    &opt_io_size);
 770         if (!err && opt_io_size)
 771                 blk_queue_io_opt(q, blk_size * opt_io_size);
 772
 773         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
 774                 q->limits.discard_granularity = blk_size;
 775
 776                 virtio_cread(vdev, struct virtio_blk_config,
 777                              discard_sector_alignment, &v);
 778                 q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
 779
 780                 virtio_cread(vdev, struct virtio_blk_config,
 781                              max_discard_sectors, &v);
 782                 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
 783
 784                 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
 785                              &v);
 786                 blk_queue_max_discard_segments(q,
 787                                                min_not_zero(v,
 788                                                             MAX_DISCARD_SEGMENTS));
 789
 790                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 791         }
 792
 793         if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
 794                 virtio_cread(vdev, struct virtio_blk_config,
 795                              max_write_zeroes_sectors, &v);
 796                 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
 797         }
 798
 799         virtblk_update_capacity(vblk, false);
 800         virtio_device_ready(vdev);
 801
 802         device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
 803         return 0;
 804
 805 out_free_tags:
 806         blk_mq_free_tag_set(&vblk->tag_set);
 807 out_put_disk:
 808         put_disk(vblk->disk);
 809 out_free_vq:
 810         vdev->config->del_vqs(vdev);
 811 out_free_vblk:
 812         kfree(vblk);
 813 out_free_index:
 814         ida_simple_remove(&vd_index_ida, index);
 815 out:
 816         return err;
 817 }
 818
 819 static void virtblk_remove(struct virtio_device *vdev)
 820 {
 821         struct virtio_blk *vblk = vdev->priv;
 822         int index = vblk->index;
 823         int refc;
 824
 825         /* Make sure no work handler is accessing the device. */
 826         flush_work(&vblk->config_work);
 827
 828         del_gendisk(vblk->disk);
 829         blk_cleanup_queue(vblk->disk->queue);
 830
 831         blk_mq_free_tag_set(&vblk->tag_set);
 832
 833         /* Stop all the virtqueues. */
 834         vdev->config->reset(vdev);
 835
 836         refc = kref_read(&disk_to_dev(vblk->disk)->kobj.kref);
 837         put_disk(vblk->disk);
 838         vdev->config->del_vqs(vdev);
 839         kfree(vblk->vqs);
 840         kfree(vblk);
 841
 842         /* Only free device id if we don't have any users */
 843         if (refc == 1)
 844                 ida_simple_remove(&vd_index_ida, index);
 845 }
 846
 847 #ifdef CONFIG_PM_SLEEP
 848 static int virtblk_freeze(struct virtio_device *vdev)
 849 {
 850         struct virtio_blk *vblk = vdev->priv;
 851
 852         /* Ensure we don't receive any more interrupts */
 853         vdev->config->reset(vdev);
 854
 855         /* Make sure no work handler is accessing the device. */
 856         flush_work(&vblk->config_work);
 857
 858         blk_mq_quiesce_queue(vblk->disk->queue);
 859
 860         vdev->config->del_vqs(vdev);
 861         return 0;
 862 }
 863
 864 static int virtblk_restore(struct virtio_device *vdev)
 865 {
 866         struct virtio_blk *vblk = vdev->priv;
 867         int ret;
 868
 869         ret = init_vq(vdev->priv);
 870         if (ret)
 871                 return ret;
 872
 873         virtio_device_ready(vdev);
 874
 875         blk_mq_unquiesce_queue(vblk->disk->queue);
 876         return 0;
 877 }
 878 #endif
 879
 880 static const struct virtio_device_id id_table[] = {
 881         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 882         { 0 },
 883 };
 884
 885 static unsigned int features_legacy[] = {
 886         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 887         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 888         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
 889         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
 890 }
 891 ;
 892 static unsigned int features[] = {
 893         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 894         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 895         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
 896         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
 897 };
 898
 899 static struct virtio_driver virtio_blk = {
 900         .feature_table                  = features,
 901         .feature_table_size             = ARRAY_SIZE(features),
 902         .feature_table_legacy           = features_legacy,
 903         .feature_table_size_legacy      = ARRAY_SIZE(features_legacy),
 904         .driver.name                    = KBUILD_MODNAME,
 905         .driver.owner                   = THIS_MODULE,
 906         .id_table                       = id_table,
 907         .probe                          = virtblk_probe,
 908         .remove                         = virtblk_remove,
 909         .config_changed                 = virtblk_config_changed,
 910 #ifdef CONFIG_PM_SLEEP
 911         .freeze                         = virtblk_freeze,
 912         .restore                        = virtblk_restore,
 913 #endif
 914 };
 915
 916 static int __init init(void)
 917 {
 918         int error;
 919
 920         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
 921         if (!virtblk_wq)
 922                 return -ENOMEM;
 923
 924         major = register_blkdev(0, "virtblk");
 925         if (major < 0) {
 926                 error = major;
 927                 goto out_destroy_workqueue;
 928         }
 929
 930         error = register_virtio_driver(&virtio_blk);
 931         if (error)
 932                 goto out_unregister_blkdev;
 933         return 0;
 934
 935 out_unregister_blkdev:
 936         unregister_blkdev(major, "virtblk");
 937 out_destroy_workqueue:
 938         destroy_workqueue(virtblk_wq);
 939         return error;
 940 }
 941
 942 static void __exit fini(void)
 943 {
 944         unregister_virtio_driver(&virtio_blk);
 945         unregister_blkdev(major, "virtblk");
 946         destroy_workqueue(virtblk_wq);
 947 }
 948 module_init(init);
 949 module_exit(fini);
 950
 951 MODULE_DEVICE_TABLE(virtio, id_table);
 952 MODULE_DESCRIPTION("Virtio block driver");
 953 MODULE_LICENSE("GPL");