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 /* The maximum number of sg elements that fit into a virtqueue */
  25 #define VIRTIO_BLK_MAX_SG_ELEMS 32768
  26
  27 #ifdef CONFIG_ARCH_NO_SG_CHAIN
  28 #define VIRTIO_BLK_INLINE_SG_CNT        0
  29 #else
  30 #define VIRTIO_BLK_INLINE_SG_CNT        2
  31 #endif
  32
  33 static unsigned int num_request_queues;
  34 module_param(num_request_queues, uint, 0644);
  35 MODULE_PARM_DESC(num_request_queues,
  36                  "Limit the number of request queues to use for blk device. "
  37                  "0 for no limit. "
  38                  "Values > nr_cpu_ids truncated to nr_cpu_ids.");
  39
  40 static int major;
  41 static DEFINE_IDA(vd_index_ida);
  42
  43 static struct workqueue_struct *virtblk_wq;
  44
  45 struct virtio_blk_vq {
  46         struct virtqueue *vq;
  47         spinlock_t lock;
  48         char name[VQ_NAME_LEN];
  49 } ____cacheline_aligned_in_smp;
  50
  51 struct virtio_blk {
  52         /*
  53          * This mutex must be held by anything that may run after
  54          * virtblk_remove() sets vblk->vdev to NULL.
  55          *
  56          * blk-mq, virtqueue processing, and sysfs attribute code paths are
  57          * shut down before vblk->vdev is set to NULL and therefore do not need
  58          * to hold this mutex.
  59          */
  60         struct mutex vdev_mutex;
  61         struct virtio_device *vdev;
  62
  63         /* The disk structure for the kernel. */
  64         struct gendisk *disk;
  65
  66         /* Block layer tags. */
  67         struct blk_mq_tag_set tag_set;
  68
  69         /* Process context for config space updates */
  70         struct work_struct config_work;
  71
  72         /*
  73          * Tracks references from block_device_operations open/release and
  74          * virtio_driver probe/remove so this object can be freed once no
  75          * longer in use.
  76          */
  77         refcount_t refs;
  78
  79         /* What host tells us, plus 2 for header & tailer. */
  80         unsigned int sg_elems;
  81
  82         /* Ida index - used to track minor number allocations. */
  83         int index;
  84
  85         /* num of vqs */
  86         int num_vqs;
  87         struct virtio_blk_vq *vqs;
  88 };
  89
  90 struct virtblk_req {
  91         struct virtio_blk_outhdr out_hdr;
  92         u8 status;
  93         struct sg_table sg_table;
  94         struct scatterlist sg[];
  95 };
  96
  97 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
  98 {
  99         switch (vbr->status) {
 100         case VIRTIO_BLK_S_OK:
 101                 return BLK_STS_OK;
 102         case VIRTIO_BLK_S_UNSUPP:
 103                 return BLK_STS_NOTSUPP;
 104         default:
 105                 return BLK_STS_IOERR;
 106         }
 107 }
 108
 109 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
 110                 struct scatterlist *data_sg, bool have_data)
 111 {
 112         struct scatterlist hdr, status, *sgs[3];
 113         unsigned int num_out = 0, num_in = 0;
 114
 115         sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
 116         sgs[num_out++] = &hdr;
 117
 118         if (have_data) {
 119                 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
 120                         sgs[num_out++] = data_sg;
 121                 else
 122                         sgs[num_out + num_in++] = data_sg;
 123         }
 124
 125         sg_init_one(&status, &vbr->status, sizeof(vbr->status));
 126         sgs[num_out + num_in++] = &status;
 127
 128         return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 129 }
 130
 131 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
 132 {
 133         unsigned short segments = blk_rq_nr_discard_segments(req);
 134         unsigned short n = 0;
 135         struct virtio_blk_discard_write_zeroes *range;
 136         struct bio *bio;
 137         u32 flags = 0;
 138
 139         if (unmap)
 140                 flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
 141
 142         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 143         if (!range)
 144                 return -ENOMEM;
 145
 146         /*
 147          * Single max discard segment means multi-range discard isn't
 148          * supported, and block layer only runs contiguity merge like
 149          * normal RW request. So we can't reply on bio for retrieving
 150          * each range info.
 151          */
 152         if (queue_max_discard_segments(req->q) == 1) {
 153                 range[0].flags = cpu_to_le32(flags);
 154                 range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
 155                 range[0].sector = cpu_to_le64(blk_rq_pos(req));
 156                 n = 1;
 157         } else {
 158                 __rq_for_each_bio(bio, req) {
 159                         u64 sector = bio->bi_iter.bi_sector;
 160                         u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
 161
 162                         range[n].flags = cpu_to_le32(flags);
 163                         range[n].num_sectors = cpu_to_le32(num_sectors);
 164                         range[n].sector = cpu_to_le64(sector);
 165                         n++;
 166                 }
 167         }
 168
 169         WARN_ON_ONCE(n != segments);
 170
 171         req->special_vec.bv_page = virt_to_page(range);
 172         req->special_vec.bv_offset = offset_in_page(range);
 173         req->special_vec.bv_len = sizeof(*range) * segments;
 174         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 175
 176         return 0;
 177 }
 178
 179 static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
 180 {
 181         if (blk_rq_nr_phys_segments(req))
 182                 sg_free_table_chained(&vbr->sg_table,
 183                                       VIRTIO_BLK_INLINE_SG_CNT);
 184 }
 185
 186 static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
 187                 struct virtblk_req *vbr)
 188 {
 189         int err;
 190
 191         if (!blk_rq_nr_phys_segments(req))
 192                 return 0;
 193
 194         vbr->sg_table.sgl = vbr->sg;
 195         err = sg_alloc_table_chained(&vbr->sg_table,
 196                                      blk_rq_nr_phys_segments(req),
 197                                      vbr->sg_table.sgl,
 198                                      VIRTIO_BLK_INLINE_SG_CNT);
 199         if (unlikely(err))
 200                 return -ENOMEM;
 201
 202         return blk_rq_map_sg(hctx->queue, req, vbr->sg_table.sgl);
 203 }
 204
 205 static void virtblk_cleanup_cmd(struct request *req)
 206 {
 207         if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
 208                 kfree(bvec_virt(&req->special_vec));
 209 }
 210
 211 static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 212                                       struct request *req,
 213                                       struct virtblk_req *vbr)
 214 {
 215         bool unmap = false;
 216         u32 type;
 217
 218         vbr->out_hdr.sector = 0;
 219
 220         switch (req_op(req)) {
 221         case REQ_OP_READ:
 222                 type = VIRTIO_BLK_T_IN;
 223                 vbr->out_hdr.sector = cpu_to_virtio64(vdev,
 224                                                       blk_rq_pos(req));
 225                 break;
 226         case REQ_OP_WRITE:
 227                 type = VIRTIO_BLK_T_OUT;
 228                 vbr->out_hdr.sector = cpu_to_virtio64(vdev,
 229                                                       blk_rq_pos(req));
 230                 break;
 231         case REQ_OP_FLUSH:
 232                 type = VIRTIO_BLK_T_FLUSH;
 233                 break;
 234         case REQ_OP_DISCARD:
 235                 type = VIRTIO_BLK_T_DISCARD;
 236                 break;
 237         case REQ_OP_WRITE_ZEROES:
 238                 type = VIRTIO_BLK_T_WRITE_ZEROES;
 239                 unmap = !(req->cmd_flags & REQ_NOUNMAP);
 240                 break;
 241         case REQ_OP_DRV_IN:
 242                 type = VIRTIO_BLK_T_GET_ID;
 243                 break;
 244         default:
 245                 WARN_ON_ONCE(1);
 246                 return BLK_STS_IOERR;
 247         }
 248
 249         vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
 250         vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
 251
 252         if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
 253                 if (virtblk_setup_discard_write_zeroes(req, unmap))
 254                         return BLK_STS_RESOURCE;
 255         }
 256
 257         return 0;
 258 }
 259
 260 static inline void virtblk_request_done(struct request *req)
 261 {
 262         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 263
 264         virtblk_unmap_data(req, vbr);
 265         virtblk_cleanup_cmd(req);
 266         blk_mq_end_request(req, virtblk_result(vbr));
 267 }
 268
 269 static void virtblk_done(struct virtqueue *vq)
 270 {
 271         struct virtio_blk *vblk = vq->vdev->priv;
 272         bool req_done = false;
 273         int qid = vq->index;
 274         struct virtblk_req *vbr;
 275         unsigned long flags;
 276         unsigned int len;
 277
 278         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 279         do {
 280                 virtqueue_disable_cb(vq);
 281                 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
 282                         struct request *req = blk_mq_rq_from_pdu(vbr);
 283
 284                         if (likely(!blk_should_fake_timeout(req->q)))
 285                                 blk_mq_complete_request(req);
 286                         req_done = true;
 287                 }
 288                 if (unlikely(virtqueue_is_broken(vq)))
 289                         break;
 290         } while (!virtqueue_enable_cb(vq));
 291
 292         /* In case queue is stopped waiting for more buffers. */
 293         if (req_done)
 294                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 295         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 296 }
 297
 298 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
 299 {
 300         struct virtio_blk *vblk = hctx->queue->queuedata;
 301         struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 302         bool kick;
 303
 304         spin_lock_irq(&vq->lock);
 305         kick = virtqueue_kick_prepare(vq->vq);
 306         spin_unlock_irq(&vq->lock);
 307
 308         if (kick)
 309                 virtqueue_notify(vq->vq);
 310 }
 311
 312 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 313                            const struct blk_mq_queue_data *bd)
 314 {
 315         struct virtio_blk *vblk = hctx->queue->queuedata;
 316         struct request *req = bd->rq;
 317         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 318         unsigned long flags;
 319         int num;
 320         int qid = hctx->queue_num;
 321         bool notify = false;
 322         blk_status_t status;
 323         int err;
 324
 325         BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 326
 327         status = virtblk_setup_cmd(vblk->vdev, req, vbr);
 328         if (unlikely(status))
 329                 return status;
 330
 331         blk_mq_start_request(req);
 332
 333         num = virtblk_map_data(hctx, req, vbr);
 334         if (unlikely(num < 0)) {
 335                 virtblk_cleanup_cmd(req);
 336                 return BLK_STS_RESOURCE;
 337         }
 338
 339         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 340         err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg_table.sgl, num);
 341         if (err) {
 342                 virtqueue_kick(vblk->vqs[qid].vq);
 343                 /* Don't stop the queue if -ENOMEM: we may have failed to
 344                  * bounce the buffer due to global resource outage.
 345                  */
 346                 if (err == -ENOSPC)
 347                         blk_mq_stop_hw_queue(hctx);
 348                 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 349                 virtblk_unmap_data(req, vbr);
 350                 virtblk_cleanup_cmd(req);
 351                 switch (err) {
 352                 case -ENOSPC:
 353                         return BLK_STS_DEV_RESOURCE;
 354                 case -ENOMEM:
 355                         return BLK_STS_RESOURCE;
 356                 default:
 357                         return BLK_STS_IOERR;
 358                 }
 359         }
 360
 361         if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
 362                 notify = true;
 363         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 364
 365         if (notify)
 366                 virtqueue_notify(vblk->vqs[qid].vq);
 367         return BLK_STS_OK;
 368 }
 369
 370 /* return id (s/n) string for *disk to *id_str
 371  */
 372 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 373 {
 374         struct virtio_blk *vblk = disk->private_data;
 375         struct request_queue *q = vblk->disk->queue;
 376         struct request *req;
 377         int err;
 378
 379         req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
 380         if (IS_ERR(req))
 381                 return PTR_ERR(req);
 382
 383         err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
 384         if (err)
 385                 goto out;
 386
 387         blk_execute_rq(vblk->disk, req, false);
 388         err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
 389 out:
 390         blk_mq_free_request(req);
 391         return err;
 392 }
 393
 394 static void virtblk_get(struct virtio_blk *vblk)
 395 {
 396         refcount_inc(&vblk->refs);
 397 }
 398
 399 static void virtblk_put(struct virtio_blk *vblk)
 400 {
 401         if (refcount_dec_and_test(&vblk->refs)) {
 402                 ida_simple_remove(&vd_index_ida, vblk->index);
 403                 mutex_destroy(&vblk->vdev_mutex);
 404                 kfree(vblk);
 405         }
 406 }
 407
 408 static int virtblk_open(struct block_device *bd, fmode_t mode)
 409 {
 410         struct virtio_blk *vblk = bd->bd_disk->private_data;
 411         int ret = 0;
 412
 413         mutex_lock(&vblk->vdev_mutex);
 414
 415         if (vblk->vdev)
 416                 virtblk_get(vblk);
 417         else
 418                 ret = -ENXIO;
 419
 420         mutex_unlock(&vblk->vdev_mutex);
 421         return ret;
 422 }
 423
 424 static void virtblk_release(struct gendisk *disk, fmode_t mode)
 425 {
 426         struct virtio_blk *vblk = disk->private_data;
 427
 428         virtblk_put(vblk);
 429 }
 430
 431 /* We provide getgeo only to please some old bootloader/partitioning tools */
 432 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 433 {
 434         struct virtio_blk *vblk = bd->bd_disk->private_data;
 435         int ret = 0;
 436
 437         mutex_lock(&vblk->vdev_mutex);
 438
 439         if (!vblk->vdev) {
 440                 ret = -ENXIO;
 441                 goto out;
 442         }
 443
 444         /* see if the host passed in geometry config */
 445         if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 446                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 447                              geometry.cylinders, &geo->cylinders);
 448                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 449                              geometry.heads, &geo->heads);
 450                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 451                              geometry.sectors, &geo->sectors);
 452         } else {
 453                 /* some standard values, similar to sd */
 454                 geo->heads = 1 << 6;
 455                 geo->sectors = 1 << 5;
 456                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 457         }
 458 out:
 459         mutex_unlock(&vblk->vdev_mutex);
 460         return ret;
 461 }
 462
 463 static const struct block_device_operations virtblk_fops = {
 464         .owner  = THIS_MODULE,
 465         .open = virtblk_open,
 466         .release = virtblk_release,
 467         .getgeo = virtblk_getgeo,
 468 };
 469
 470 static int index_to_minor(int index)
 471 {
 472         return index << PART_BITS;
 473 }
 474
 475 static int minor_to_index(int minor)
 476 {
 477         return minor >> PART_BITS;
 478 }
 479
 480 static ssize_t serial_show(struct device *dev,
 481                            struct device_attribute *attr, char *buf)
 482 {
 483         struct gendisk *disk = dev_to_disk(dev);
 484         int err;
 485
 486         /* sysfs gives us a PAGE_SIZE buffer */
 487         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 488
 489         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 490         err = virtblk_get_id(disk, buf);
 491         if (!err)
 492                 return strlen(buf);
 493
 494         if (err == -EIO) /* Unsupported? Make it empty. */
 495                 return 0;
 496
 497         return err;
 498 }
 499
 500 static DEVICE_ATTR_RO(serial);
 501
 502 /* The queue's logical block size must be set before calling this */
 503 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
 504 {
 505         struct virtio_device *vdev = vblk->vdev;
 506         struct request_queue *q = vblk->disk->queue;
 507         char cap_str_2[10], cap_str_10[10];
 508         unsigned long long nblocks;
 509         u64 capacity;
 510
 511         /* Host must always specify the capacity. */
 512         virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
 513
 514         nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
 515
 516         string_get_size(nblocks, queue_logical_block_size(q),
 517                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 518         string_get_size(nblocks, queue_logical_block_size(q),
 519                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 520
 521         dev_notice(&vdev->dev,
 522                    "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
 523                    vblk->disk->disk_name,
 524                    resize ? "new size: " : "",
 525                    nblocks,
 526                    queue_logical_block_size(q),
 527                    cap_str_10,
 528                    cap_str_2);
 529
 530         set_capacity_and_notify(vblk->disk, capacity);
 531 }
 532
 533 static void virtblk_config_changed_work(struct work_struct *work)
 534 {
 535         struct virtio_blk *vblk =
 536                 container_of(work, struct virtio_blk, config_work);
 537
 538         virtblk_update_capacity(vblk, true);
 539 }
 540
 541 static void virtblk_config_changed(struct virtio_device *vdev)
 542 {
 543         struct virtio_blk *vblk = vdev->priv;
 544
 545         queue_work(virtblk_wq, &vblk->config_work);
 546 }
 547
 548 static int init_vq(struct virtio_blk *vblk)
 549 {
 550         int err;
 551         int i;
 552         vq_callback_t **callbacks;
 553         const char **names;
 554         struct virtqueue **vqs;
 555         unsigned short num_vqs;
 556         struct virtio_device *vdev = vblk->vdev;
 557         struct irq_affinity desc = { 0, };
 558
 559         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
 560                                    struct virtio_blk_config, num_queues,
 561                                    &num_vqs);
 562         if (err)
 563                 num_vqs = 1;
 564         if (!err && !num_vqs) {
 565                 dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
 566                 return -EINVAL;
 567         }
 568
 569         num_vqs = min_t(unsigned int,
 570                         min_not_zero(num_request_queues, nr_cpu_ids),
 571                         num_vqs);
 572
 573         vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
 574         if (!vblk->vqs)
 575                 return -ENOMEM;
 576
 577         names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
 578         callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
 579         vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
 580         if (!names || !callbacks || !vqs) {
 581                 err = -ENOMEM;
 582                 goto out;
 583         }
 584
 585         for (i = 0; i < num_vqs; i++) {
 586                 callbacks[i] = virtblk_done;
 587                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
 588                 names[i] = vblk->vqs[i].name;
 589         }
 590
 591         /* Discover virtqueues and write information to configuration.  */
 592         err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
 593         if (err)
 594                 goto out;
 595
 596         for (i = 0; i < num_vqs; i++) {
 597                 spin_lock_init(&vblk->vqs[i].lock);
 598                 vblk->vqs[i].vq = vqs[i];
 599         }
 600         vblk->num_vqs = num_vqs;
 601
 602 out:
 603         kfree(vqs);
 604         kfree(callbacks);
 605         kfree(names);
 606         if (err)
 607                 kfree(vblk->vqs);
 608         return err;
 609 }
 610
 611 /*
 612  * Legacy naming scheme used for virtio devices.  We are stuck with it for
 613  * virtio blk but don't ever use it for any new driver.
 614  */
 615 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
 616 {
 617         const int base = 'z' - 'a' + 1;
 618         char *begin = buf + strlen(prefix);
 619         char *end = buf + buflen;
 620         char *p;
 621         int unit;
 622
 623         p = end - 1;
 624         *p = '\0';
 625         unit = base;
 626         do {
 627                 if (p == begin)
 628                         return -EINVAL;
 629                 *--p = 'a' + (index % unit);
 630                 index = (index / unit) - 1;
 631         } while (index >= 0);
 632
 633         memmove(begin, p, end - p);
 634         memcpy(buf, prefix, strlen(prefix));
 635
 636         return 0;
 637 }
 638
 639 static int virtblk_get_cache_mode(struct virtio_device *vdev)
 640 {
 641         u8 writeback;
 642         int err;
 643
 644         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
 645                                    struct virtio_blk_config, wce,
 646                                    &writeback);
 647
 648         /*
 649          * If WCE is not configurable and flush is not available,
 650          * assume no writeback cache is in use.
 651          */
 652         if (err)
 653                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
 654
 655         return writeback;
 656 }
 657
 658 static void virtblk_update_cache_mode(struct virtio_device *vdev)
 659 {
 660         u8 writeback = virtblk_get_cache_mode(vdev);
 661         struct virtio_blk *vblk = vdev->priv;
 662
 663         blk_queue_write_cache(vblk->disk->queue, writeback, false);
 664 }
 665
 666 static const char *const virtblk_cache_types[] = {
 667         "write through", "write back"
 668 };
 669
 670 static ssize_t
 671 cache_type_store(struct device *dev, struct device_attribute *attr,
 672                  const char *buf, size_t count)
 673 {
 674         struct gendisk *disk = dev_to_disk(dev);
 675         struct virtio_blk *vblk = disk->private_data;
 676         struct virtio_device *vdev = vblk->vdev;
 677         int i;
 678
 679         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
 680         i = sysfs_match_string(virtblk_cache_types, buf);
 681         if (i < 0)
 682                 return i;
 683
 684         virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
 685         virtblk_update_cache_mode(vdev);
 686         return count;
 687 }
 688
 689 static ssize_t
 690 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 691 {
 692         struct gendisk *disk = dev_to_disk(dev);
 693         struct virtio_blk *vblk = disk->private_data;
 694         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
 695
 696         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
 697         return sysfs_emit(buf, "%s\n", virtblk_cache_types[writeback]);
 698 }
 699
 700 static DEVICE_ATTR_RW(cache_type);
 701
 702 static struct attribute *virtblk_attrs[] = {
 703         &dev_attr_serial.attr,
 704         &dev_attr_cache_type.attr,
 705         NULL,
 706 };
 707
 708 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
 709                 struct attribute *a, int n)
 710 {
 711         struct device *dev = kobj_to_dev(kobj);
 712         struct gendisk *disk = dev_to_disk(dev);
 713         struct virtio_blk *vblk = disk->private_data;
 714         struct virtio_device *vdev = vblk->vdev;
 715
 716         if (a == &dev_attr_cache_type.attr &&
 717             !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
 718                 return S_IRUGO;
 719
 720         return a->mode;
 721 }
 722
 723 static const struct attribute_group virtblk_attr_group = {
 724         .attrs = virtblk_attrs,
 725         .is_visible = virtblk_attrs_are_visible,
 726 };
 727
 728 static const struct attribute_group *virtblk_attr_groups[] = {
 729         &virtblk_attr_group,
 730         NULL,
 731 };
 732
 733 static int virtblk_map_queues(struct blk_mq_tag_set *set)
 734 {
 735         struct virtio_blk *vblk = set->driver_data;
 736
 737         return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
 738                                         vblk->vdev, 0);
 739 }
 740
 741 static const struct blk_mq_ops virtio_mq_ops = {
 742         .queue_rq       = virtio_queue_rq,
 743         .commit_rqs     = virtio_commit_rqs,
 744         .complete       = virtblk_request_done,
 745         .map_queues     = virtblk_map_queues,
 746 };
 747
 748 static unsigned int virtblk_queue_depth;
 749 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
 750
 751 static int virtblk_probe(struct virtio_device *vdev)
 752 {
 753         struct virtio_blk *vblk;
 754         struct request_queue *q;
 755         int err, index;
 756
 757         u32 v, blk_size, max_size, sg_elems, opt_io_size;
 758         u16 min_io_size;
 759         u8 physical_block_exp, alignment_offset;
 760         unsigned int queue_depth;
 761
 762         if (!vdev->config->get) {
 763                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
 764                         __func__);
 765                 return -EINVAL;
 766         }
 767
 768         err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
 769                              GFP_KERNEL);
 770         if (err < 0)
 771                 goto out;
 772         index = err;
 773
 774         /* We need to know how many segments before we allocate. */
 775         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
 776                                    struct virtio_blk_config, seg_max,
 777                                    &sg_elems);
 778
 779         /* We need at least one SG element, whatever they say. */
 780         if (err || !sg_elems)
 781                 sg_elems = 1;
 782
 783         /* Prevent integer overflows and honor max vq size */
 784         sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
 785
 786         /* We need extra sg elements at head and tail. */
 787         sg_elems += 2;
 788         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
 789         if (!vblk) {
 790                 err = -ENOMEM;
 791                 goto out_free_index;
 792         }
 793
 794         /* This reference is dropped in virtblk_remove(). */
 795         refcount_set(&vblk->refs, 1);
 796         mutex_init(&vblk->vdev_mutex);
 797
 798         vblk->vdev = vdev;
 799         vblk->sg_elems = sg_elems;
 800
 801         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 802
 803         err = init_vq(vblk);
 804         if (err)
 805                 goto out_free_vblk;
 806
 807         /* Default queue sizing is to fill the ring. */
 808         if (!virtblk_queue_depth) {
 809                 queue_depth = vblk->vqs[0].vq->num_free;
 810                 /* ... but without indirect descs, we use 2 descs per req */
 811                 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
 812                         queue_depth /= 2;
 813         } else {
 814                 queue_depth = virtblk_queue_depth;
 815         }
 816
 817         memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
 818         vblk->tag_set.ops = &virtio_mq_ops;
 819         vblk->tag_set.queue_depth = queue_depth;
 820         vblk->tag_set.numa_node = NUMA_NO_NODE;
 821         vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 822         vblk->tag_set.cmd_size =
 823                 sizeof(struct virtblk_req) +
 824                 sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
 825         vblk->tag_set.driver_data = vblk;
 826         vblk->tag_set.nr_hw_queues = vblk->num_vqs;
 827
 828         err = blk_mq_alloc_tag_set(&vblk->tag_set);
 829         if (err)
 830                 goto out_free_vq;
 831
 832         vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
 833         if (IS_ERR(vblk->disk)) {
 834                 err = PTR_ERR(vblk->disk);
 835                 goto out_free_tags;
 836         }
 837         q = vblk->disk->queue;
 838
 839         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
 840
 841         vblk->disk->major = major;
 842         vblk->disk->first_minor = index_to_minor(index);
 843         vblk->disk->minors = 1 << PART_BITS;
 844         vblk->disk->private_data = vblk;
 845         vblk->disk->fops = &virtblk_fops;
 846         vblk->disk->flags |= GENHD_FL_EXT_DEVT;
 847         vblk->index = index;
 848
 849         /* configure queue flush support */
 850         virtblk_update_cache_mode(vdev);
 851
 852         /* If disk is read-only in the host, the guest should obey */
 853         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 854                 set_disk_ro(vblk->disk, 1);
 855
 856         /* We can handle whatever the host told us to handle. */
 857         blk_queue_max_segments(q, vblk->sg_elems-2);
 858
 859         /* No real sector limit. */
 860         blk_queue_max_hw_sectors(q, -1U);
 861
 862         max_size = virtio_max_dma_size(vdev);
 863
 864         /* Host can optionally specify maximum segment size and number of
 865          * segments. */
 866         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
 867                                    struct virtio_blk_config, size_max, &v);
 868         if (!err)
 869                 max_size = min(max_size, v);
 870
 871         blk_queue_max_segment_size(q, max_size);
 872
 873         /* Host can optionally specify the block size of the device */
 874         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
 875                                    struct virtio_blk_config, blk_size,
 876                                    &blk_size);
 877         if (!err) {
 878                 err = blk_validate_block_size(blk_size);
 879                 if (err) {
 880                         dev_err(&vdev->dev,
 881                                 "virtio_blk: invalid block size: 0x%x\n",
 882                                 blk_size);
 883                         goto out_cleanup_disk;
 884                 }
 885
 886                 blk_queue_logical_block_size(q, blk_size);
 887         } else
 888                 blk_size = queue_logical_block_size(q);
 889
 890         /* Use topology information if available */
 891         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 892                                    struct virtio_blk_config, physical_block_exp,
 893                                    &physical_block_exp);
 894         if (!err && physical_block_exp)
 895                 blk_queue_physical_block_size(q,
 896                                 blk_size * (1 << physical_block_exp));
 897
 898         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 899                                    struct virtio_blk_config, alignment_offset,
 900                                    &alignment_offset);
 901         if (!err && alignment_offset)
 902                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 903
 904         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 905                                    struct virtio_blk_config, min_io_size,
 906                                    &min_io_size);
 907         if (!err && min_io_size)
 908                 blk_queue_io_min(q, blk_size * min_io_size);
 909
 910         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 911                                    struct virtio_blk_config, opt_io_size,
 912                                    &opt_io_size);
 913         if (!err && opt_io_size)
 914                 blk_queue_io_opt(q, blk_size * opt_io_size);
 915
 916         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
 917                 q->limits.discard_granularity = blk_size;
 918
 919                 virtio_cread(vdev, struct virtio_blk_config,
 920                              discard_sector_alignment, &v);
 921                 q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
 922
 923                 virtio_cread(vdev, struct virtio_blk_config,
 924                              max_discard_sectors, &v);
 925                 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
 926
 927                 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
 928                              &v);
 929                 blk_queue_max_discard_segments(q,
 930                                                min_not_zero(v,
 931                                                             MAX_DISCARD_SEGMENTS));
 932
 933                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 934         }
 935
 936         if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
 937                 virtio_cread(vdev, struct virtio_blk_config,
 938                              max_write_zeroes_sectors, &v);
 939                 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
 940         }
 941
 942         virtblk_update_capacity(vblk, false);
 943         virtio_device_ready(vdev);
 944
 945         err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
 946         if (err)
 947                 goto out_cleanup_disk;
 948
 949         return 0;
 950
 951 out_cleanup_disk:
 952         blk_cleanup_disk(vblk->disk);
 953 out_free_tags:
 954         blk_mq_free_tag_set(&vblk->tag_set);
 955 out_free_vq:
 956         vdev->config->del_vqs(vdev);
 957         kfree(vblk->vqs);
 958 out_free_vblk:
 959         kfree(vblk);
 960 out_free_index:
 961         ida_simple_remove(&vd_index_ida, index);
 962 out:
 963         return err;
 964 }
 965
 966 static void virtblk_remove(struct virtio_device *vdev)
 967 {
 968         struct virtio_blk *vblk = vdev->priv;
 969
 970         /* Make sure no work handler is accessing the device. */
 971         flush_work(&vblk->config_work);
 972
 973         del_gendisk(vblk->disk);
 974         blk_cleanup_disk(vblk->disk);
 975         blk_mq_free_tag_set(&vblk->tag_set);
 976
 977         mutex_lock(&vblk->vdev_mutex);
 978
 979         /* Stop all the virtqueues. */
 980         vdev->config->reset(vdev);
 981
 982         /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
 983         vblk->vdev = NULL;
 984
 985         vdev->config->del_vqs(vdev);
 986         kfree(vblk->vqs);
 987
 988         mutex_unlock(&vblk->vdev_mutex);
 989
 990         virtblk_put(vblk);
 991 }
 992
 993 #ifdef CONFIG_PM_SLEEP
 994 static int virtblk_freeze(struct virtio_device *vdev)
 995 {
 996         struct virtio_blk *vblk = vdev->priv;
 997
 998         /* Ensure we don't receive any more interrupts */
 999         vdev->config->reset(vdev);
1000
1001         /* Make sure no work handler is accessing the device. */
1002         flush_work(&vblk->config_work);
1003
1004         blk_mq_quiesce_queue(vblk->disk->queue);
1005
1006         vdev->config->del_vqs(vdev);
1007         kfree(vblk->vqs);
1008
1009         return 0;
1010 }
1011
1012 static int virtblk_restore(struct virtio_device *vdev)
1013 {
1014         struct virtio_blk *vblk = vdev->priv;
1015         int ret;
1016
1017         ret = init_vq(vdev->priv);
1018         if (ret)
1019                 return ret;
1020
1021         virtio_device_ready(vdev);
1022
1023         blk_mq_unquiesce_queue(vblk->disk->queue);
1024         return 0;
1025 }
1026 #endif
1027
1028 static const struct virtio_device_id id_table[] = {
1029         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1030         { 0 },
1031 };
1032
1033 static unsigned int features_legacy[] = {
1034         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1035         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1036         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1037         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1038 }
1039 ;
1040 static unsigned int features[] = {
1041         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1042         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1043         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1044         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1045 };
1046
1047 static struct virtio_driver virtio_blk = {
1048         .feature_table                  = features,
1049         .feature_table_size             = ARRAY_SIZE(features),
1050         .feature_table_legacy           = features_legacy,
1051         .feature_table_size_legacy      = ARRAY_SIZE(features_legacy),
1052         .driver.name                    = KBUILD_MODNAME,
1053         .driver.owner                   = THIS_MODULE,
1054         .id_table                       = id_table,
1055         .probe                          = virtblk_probe,
1056         .remove                         = virtblk_remove,
1057         .config_changed                 = virtblk_config_changed,
1058 #ifdef CONFIG_PM_SLEEP
1059         .freeze                         = virtblk_freeze,
1060         .restore                        = virtblk_restore,
1061 #endif
1062 };
1063
1064 static int __init init(void)
1065 {
1066         int error;
1067
1068         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1069         if (!virtblk_wq)
1070                 return -ENOMEM;
1071
1072         major = register_blkdev(0, "virtblk");
1073         if (major < 0) {
1074                 error = major;
1075                 goto out_destroy_workqueue;
1076         }
1077
1078         error = register_virtio_driver(&virtio_blk);
1079         if (error)
1080                 goto out_unregister_blkdev;
1081         return 0;
1082
1083 out_unregister_blkdev:
1084         unregister_blkdev(major, "virtblk");
1085 out_destroy_workqueue:
1086         destroy_workqueue(virtblk_wq);
1087         return error;
1088 }
1089
1090 static void __exit fini(void)
1091 {
1092         unregister_virtio_driver(&virtio_blk);
1093         unregister_blkdev(major, "virtblk");
1094         destroy_workqueue(virtblk_wq);
1095 }
1096 module_init(init);
1097 module_exit(fini);
1098
1099 MODULE_DEVICE_TABLE(virtio, id_table);
1100 MODULE_DESCRIPTION("Virtio block driver");
1101 MODULE_LICENSE("GPL");