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