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                 blk_mq_stop_hw_queue(hctx);
 249                 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 250                 /* Out of mem doesn't actually happen, since we fall back
 251                  * to direct descriptors */
 252                 if (err == -ENOMEM || err == -ENOSPC)
 253                         return BLK_STS_DEV_RESOURCE;
 254                 return BLK_STS_IOERR;
 255         }
 256
 257         if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
 258                 notify = true;
 259         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 260
 261         if (notify)
 262                 virtqueue_notify(vblk->vqs[qid].vq);
 263         return BLK_STS_OK;
 264 }
 265
 266 /* return id (s/n) string for *disk to *id_str
 267  */
 268 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 269 {
 270         struct virtio_blk *vblk = disk->private_data;
 271         struct request_queue *q = vblk->disk->queue;
 272         struct request *req;
 273         int err;
 274
 275         req = blk_get_request(q, REQ_OP_DRV_IN, 0);
 276         if (IS_ERR(req))
 277                 return PTR_ERR(req);
 278
 279         err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
 280         if (err)
 281                 goto out;
 282
 283         blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
 284         err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
 285 out:
 286         blk_put_request(req);
 287         return err;
 288 }
 289
 290 /* We provide getgeo only to please some old bootloader/partitioning tools */
 291 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 292 {
 293         struct virtio_blk *vblk = bd->bd_disk->private_data;
 294
 295         /* see if the host passed in geometry config */
 296         if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 297                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 298                              geometry.cylinders, &geo->cylinders);
 299                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 300                              geometry.heads, &geo->heads);
 301                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 302                              geometry.sectors, &geo->sectors);
 303         } else {
 304                 /* some standard values, similar to sd */
 305                 geo->heads = 1 << 6;
 306                 geo->sectors = 1 << 5;
 307                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 308         }
 309         return 0;
 310 }
 311
 312 static const struct block_device_operations virtblk_fops = {
 313         .owner  = THIS_MODULE,
 314         .getgeo = virtblk_getgeo,
 315 };
 316
 317 static int index_to_minor(int index)
 318 {
 319         return index << PART_BITS;
 320 }
 321
 322 static int minor_to_index(int minor)
 323 {
 324         return minor >> PART_BITS;
 325 }
 326
 327 static ssize_t serial_show(struct device *dev,
 328                            struct device_attribute *attr, char *buf)
 329 {
 330         struct gendisk *disk = dev_to_disk(dev);
 331         int err;
 332
 333         /* sysfs gives us a PAGE_SIZE buffer */
 334         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 335
 336         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 337         err = virtblk_get_id(disk, buf);
 338         if (!err)
 339                 return strlen(buf);
 340
 341         if (err == -EIO) /* Unsupported? Make it empty. */
 342                 return 0;
 343
 344         return err;
 345 }
 346
 347 static DEVICE_ATTR_RO(serial);
 348
 349 /* The queue's logical block size must be set before calling this */
 350 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
 351 {
 352         struct virtio_device *vdev = vblk->vdev;
 353         struct request_queue *q = vblk->disk->queue;
 354         char cap_str_2[10], cap_str_10[10];
 355         unsigned long long nblocks;
 356         u64 capacity;
 357
 358         /* Host must always specify the capacity. */
 359         virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
 360
 361         /* If capacity is too big, truncate with warning. */
 362         if ((sector_t)capacity != capacity) {
 363                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 364                          (unsigned long long)capacity);
 365                 capacity = (sector_t)-1;
 366         }
 367
 368         nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
 369
 370         string_get_size(nblocks, queue_logical_block_size(q),
 371                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 372         string_get_size(nblocks, queue_logical_block_size(q),
 373                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 374
 375         dev_notice(&vdev->dev,
 376                    "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
 377                    vblk->disk->disk_name,
 378                    resize ? "new size: " : "",
 379                    nblocks,
 380                    queue_logical_block_size(q),
 381                    cap_str_10,
 382                    cap_str_2);
 383
 384         set_capacity(vblk->disk, capacity);
 385 }
 386
 387 static void virtblk_config_changed_work(struct work_struct *work)
 388 {
 389         struct virtio_blk *vblk =
 390                 container_of(work, struct virtio_blk, config_work);
 391         char *envp[] = { "RESIZE=1", NULL };
 392
 393         virtblk_update_capacity(vblk, true);
 394         revalidate_disk(vblk->disk);
 395         kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
 396 }
 397
 398 static void virtblk_config_changed(struct virtio_device *vdev)
 399 {
 400         struct virtio_blk *vblk = vdev->priv;
 401
 402         queue_work(virtblk_wq, &vblk->config_work);
 403 }
 404
 405 static int init_vq(struct virtio_blk *vblk)
 406 {
 407         int err;
 408         int i;
 409         vq_callback_t **callbacks;
 410         const char **names;
 411         struct virtqueue **vqs;
 412         unsigned short num_vqs;
 413         struct virtio_device *vdev = vblk->vdev;
 414         struct irq_affinity desc = { 0, };
 415
 416         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
 417                                    struct virtio_blk_config, num_queues,
 418                                    &num_vqs);
 419         if (err)
 420                 num_vqs = 1;
 421
 422         num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
 423
 424         vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
 425         if (!vblk->vqs)
 426                 return -ENOMEM;
 427
 428         names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
 429         callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
 430         vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
 431         if (!names || !callbacks || !vqs) {
 432                 err = -ENOMEM;
 433                 goto out;
 434         }
 435
 436         for (i = 0; i < num_vqs; i++) {
 437                 callbacks[i] = virtblk_done;
 438                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
 439                 names[i] = vblk->vqs[i].name;
 440         }
 441
 442         /* Discover virtqueues and write information to configuration.  */
 443         err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
 444         if (err)
 445                 goto out;
 446
 447         for (i = 0; i < num_vqs; i++) {
 448                 spin_lock_init(&vblk->vqs[i].lock);
 449                 vblk->vqs[i].vq = vqs[i];
 450         }
 451         vblk->num_vqs = num_vqs;
 452
 453 out:
 454         kfree(vqs);
 455         kfree(callbacks);
 456         kfree(names);
 457         if (err)
 458                 kfree(vblk->vqs);
 459         return err;
 460 }
 461
 462 /*
 463  * Legacy naming scheme used for virtio devices.  We are stuck with it for
 464  * virtio blk but don't ever use it for any new driver.
 465  */
 466 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
 467 {
 468         const int base = 'z' - 'a' + 1;
 469         char *begin = buf + strlen(prefix);
 470         char *end = buf + buflen;
 471         char *p;
 472         int unit;
 473
 474         p = end - 1;
 475         *p = '\0';
 476         unit = base;
 477         do {
 478                 if (p == begin)
 479                         return -EINVAL;
 480                 *--p = 'a' + (index % unit);
 481                 index = (index / unit) - 1;
 482         } while (index >= 0);
 483
 484         memmove(begin, p, end - p);
 485         memcpy(buf, prefix, strlen(prefix));
 486
 487         return 0;
 488 }
 489
 490 static int virtblk_get_cache_mode(struct virtio_device *vdev)
 491 {
 492         u8 writeback;
 493         int err;
 494
 495         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
 496                                    struct virtio_blk_config, wce,
 497                                    &writeback);
 498
 499         /*
 500          * If WCE is not configurable and flush is not available,
 501          * assume no writeback cache is in use.
 502          */
 503         if (err)
 504                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
 505
 506         return writeback;
 507 }
 508
 509 static void virtblk_update_cache_mode(struct virtio_device *vdev)
 510 {
 511         u8 writeback = virtblk_get_cache_mode(vdev);
 512         struct virtio_blk *vblk = vdev->priv;
 513
 514         blk_queue_write_cache(vblk->disk->queue, writeback, false);
 515         revalidate_disk(vblk->disk);
 516 }
 517
 518 static const char *const virtblk_cache_types[] = {
 519         "write through", "write back"
 520 };
 521
 522 static ssize_t
 523 cache_type_store(struct device *dev, struct device_attribute *attr,
 524                  const char *buf, size_t count)
 525 {
 526         struct gendisk *disk = dev_to_disk(dev);
 527         struct virtio_blk *vblk = disk->private_data;
 528         struct virtio_device *vdev = vblk->vdev;
 529         int i;
 530
 531         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
 532         i = sysfs_match_string(virtblk_cache_types, buf);
 533         if (i < 0)
 534                 return i;
 535
 536         virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
 537         virtblk_update_cache_mode(vdev);
 538         return count;
 539 }
 540
 541 static ssize_t
 542 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 543 {
 544         struct gendisk *disk = dev_to_disk(dev);
 545         struct virtio_blk *vblk = disk->private_data;
 546         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
 547
 548         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
 549         return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
 550 }
 551
 552 static DEVICE_ATTR_RW(cache_type);
 553
 554 static struct attribute *virtblk_attrs[] = {
 555         &dev_attr_serial.attr,
 556         &dev_attr_cache_type.attr,
 557         NULL,
 558 };
 559
 560 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
 561                 struct attribute *a, int n)
 562 {
 563         struct device *dev = container_of(kobj, struct device, kobj);
 564         struct gendisk *disk = dev_to_disk(dev);
 565         struct virtio_blk *vblk = disk->private_data;
 566         struct virtio_device *vdev = vblk->vdev;
 567
 568         if (a == &dev_attr_cache_type.attr &&
 569             !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
 570                 return S_IRUGO;
 571
 572         return a->mode;
 573 }
 574
 575 static const struct attribute_group virtblk_attr_group = {
 576         .attrs = virtblk_attrs,
 577         .is_visible = virtblk_attrs_are_visible,
 578 };
 579
 580 static const struct attribute_group *virtblk_attr_groups[] = {
 581         &virtblk_attr_group,
 582         NULL,
 583 };
 584
 585 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
 586                 unsigned int hctx_idx, unsigned int numa_node)
 587 {
 588         struct virtio_blk *vblk = set->driver_data;
 589         struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
 590
 591         sg_init_table(vbr->sg, vblk->sg_elems);
 592         return 0;
 593 }
 594
 595 static int virtblk_map_queues(struct blk_mq_tag_set *set)
 596 {
 597         struct virtio_blk *vblk = set->driver_data;
 598
 599         return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
 600                                         vblk->vdev, 0);
 601 }
 602
 603 static const struct blk_mq_ops virtio_mq_ops = {
 604         .queue_rq       = virtio_queue_rq,
 605         .commit_rqs     = virtio_commit_rqs,
 606         .complete       = virtblk_request_done,
 607         .init_request   = virtblk_init_request,
 608         .map_queues     = virtblk_map_queues,
 609 };
 610
 611 static unsigned int virtblk_queue_depth;
 612 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
 613
 614 static int virtblk_probe(struct virtio_device *vdev)
 615 {
 616         struct virtio_blk *vblk;
 617         struct request_queue *q;
 618         int err, index;
 619
 620         u32 v, blk_size, max_size, sg_elems, opt_io_size;
 621         u16 min_io_size;
 622         u8 physical_block_exp, alignment_offset;
 623
 624         if (!vdev->config->get) {
 625                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
 626                         __func__);
 627                 return -EINVAL;
 628         }
 629
 630         err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
 631                              GFP_KERNEL);
 632         if (err < 0)
 633                 goto out;
 634         index = err;
 635
 636         /* We need to know how many segments before we allocate. */
 637         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
 638                                    struct virtio_blk_config, seg_max,
 639                                    &sg_elems);
 640
 641         /* We need at least one SG element, whatever they say. */
 642         if (err || !sg_elems)
 643                 sg_elems = 1;
 644
 645         /* We need an extra sg elements at head and tail. */
 646         sg_elems += 2;
 647         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
 648         if (!vblk) {
 649                 err = -ENOMEM;
 650                 goto out_free_index;
 651         }
 652
 653         vblk->vdev = vdev;
 654         vblk->sg_elems = sg_elems;
 655
 656         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 657
 658         err = init_vq(vblk);
 659         if (err)
 660                 goto out_free_vblk;
 661
 662         /* FIXME: How many partitions?  How long is a piece of string? */
 663         vblk->disk = alloc_disk(1 << PART_BITS);
 664         if (!vblk->disk) {
 665                 err = -ENOMEM;
 666                 goto out_free_vq;
 667         }
 668
 669         /* Default queue sizing is to fill the ring. */
 670         if (!virtblk_queue_depth) {
 671                 virtblk_queue_depth = vblk->vqs[0].vq->num_free;
 672                 /* ... but without indirect descs, we use 2 descs per req */
 673                 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
 674                         virtblk_queue_depth /= 2;
 675         }
 676
 677         memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
 678         vblk->tag_set.ops = &virtio_mq_ops;
 679         vblk->tag_set.queue_depth = virtblk_queue_depth;
 680         vblk->tag_set.numa_node = NUMA_NO_NODE;
 681         vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 682         vblk->tag_set.cmd_size =
 683                 sizeof(struct virtblk_req) +
 684                 sizeof(struct scatterlist) * sg_elems;
 685         vblk->tag_set.driver_data = vblk;
 686         vblk->tag_set.nr_hw_queues = vblk->num_vqs;
 687
 688         err = blk_mq_alloc_tag_set(&vblk->tag_set);
 689         if (err)
 690                 goto out_put_disk;
 691
 692         q = blk_mq_init_queue(&vblk->tag_set);
 693         if (IS_ERR(q)) {
 694                 err = -ENOMEM;
 695                 goto out_free_tags;
 696         }
 697         vblk->disk->queue = q;
 698
 699         q->queuedata = vblk;
 700
 701         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
 702
 703         vblk->disk->major = major;
 704         vblk->disk->first_minor = index_to_minor(index);
 705         vblk->disk->private_data = vblk;
 706         vblk->disk->fops = &virtblk_fops;
 707         vblk->disk->flags |= GENHD_FL_EXT_DEVT;
 708         vblk->index = index;
 709
 710         /* configure queue flush support */
 711         virtblk_update_cache_mode(vdev);
 712
 713         /* If disk is read-only in the host, the guest should obey */
 714         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 715                 set_disk_ro(vblk->disk, 1);
 716
 717         /* We can handle whatever the host told us to handle. */
 718         blk_queue_max_segments(q, vblk->sg_elems-2);
 719
 720         /* No real sector limit. */
 721         blk_queue_max_hw_sectors(q, -1U);
 722
 723         max_size = virtio_max_dma_size(vdev);
 724
 725         /* Host can optionally specify maximum segment size and number of
 726          * segments. */
 727         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
 728                                    struct virtio_blk_config, size_max, &v);
 729         if (!err)
 730                 max_size = min(max_size, v);
 731
 732         blk_queue_max_segment_size(q, max_size);
 733
 734         /* Host can optionally specify the block size of the device */
 735         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
 736                                    struct virtio_blk_config, blk_size,
 737                                    &blk_size);
 738         if (!err)
 739                 blk_queue_logical_block_size(q, blk_size);
 740         else
 741                 blk_size = queue_logical_block_size(q);
 742
 743         /* Use topology information if available */
 744         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 745                                    struct virtio_blk_config, physical_block_exp,
 746                                    &physical_block_exp);
 747         if (!err && physical_block_exp)
 748                 blk_queue_physical_block_size(q,
 749                                 blk_size * (1 << physical_block_exp));
 750
 751         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 752                                    struct virtio_blk_config, alignment_offset,
 753                                    &alignment_offset);
 754         if (!err && alignment_offset)
 755                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 756
 757         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 758                                    struct virtio_blk_config, min_io_size,
 759                                    &min_io_size);
 760         if (!err && min_io_size)
 761                 blk_queue_io_min(q, blk_size * min_io_size);
 762
 763         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 764                                    struct virtio_blk_config, opt_io_size,
 765                                    &opt_io_size);
 766         if (!err && opt_io_size)
 767                 blk_queue_io_opt(q, blk_size * opt_io_size);
 768
 769         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
 770                 q->limits.discard_granularity = blk_size;
 771
 772                 virtio_cread(vdev, struct virtio_blk_config,
 773                              discard_sector_alignment, &v);
 774                 q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
 775
 776                 virtio_cread(vdev, struct virtio_blk_config,
 777                              max_discard_sectors, &v);
 778                 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
 779
 780                 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
 781                              &v);
 782                 blk_queue_max_discard_segments(q,
 783                                                min_not_zero(v,
 784                                                             MAX_DISCARD_SEGMENTS));
 785
 786                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 787         }
 788
 789         if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
 790                 virtio_cread(vdev, struct virtio_blk_config,
 791                              max_write_zeroes_sectors, &v);
 792                 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
 793         }
 794
 795         virtblk_update_capacity(vblk, false);
 796         virtio_device_ready(vdev);
 797
 798         device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
 799         return 0;
 800
 801 out_free_tags:
 802         blk_mq_free_tag_set(&vblk->tag_set);
 803 out_put_disk:
 804         put_disk(vblk->disk);
 805 out_free_vq:
 806         vdev->config->del_vqs(vdev);
 807 out_free_vblk:
 808         kfree(vblk);
 809 out_free_index:
 810         ida_simple_remove(&vd_index_ida, index);
 811 out:
 812         return err;
 813 }
 814
 815 static void virtblk_remove(struct virtio_device *vdev)
 816 {
 817         struct virtio_blk *vblk = vdev->priv;
 818         int index = vblk->index;
 819         int refc;
 820
 821         /* Make sure no work handler is accessing the device. */
 822         flush_work(&vblk->config_work);
 823
 824         del_gendisk(vblk->disk);
 825         blk_cleanup_queue(vblk->disk->queue);
 826
 827         blk_mq_free_tag_set(&vblk->tag_set);
 828
 829         /* Stop all the virtqueues. */
 830         vdev->config->reset(vdev);
 831
 832         refc = kref_read(&disk_to_dev(vblk->disk)->kobj.kref);
 833         put_disk(vblk->disk);
 834         vdev->config->del_vqs(vdev);
 835         kfree(vblk->vqs);
 836         kfree(vblk);
 837
 838         /* Only free device id if we don't have any users */
 839         if (refc == 1)
 840                 ida_simple_remove(&vd_index_ida, index);
 841 }
 842
 843 #ifdef CONFIG_PM_SLEEP
 844 static int virtblk_freeze(struct virtio_device *vdev)
 845 {
 846         struct virtio_blk *vblk = vdev->priv;
 847
 848         /* Ensure we don't receive any more interrupts */
 849         vdev->config->reset(vdev);
 850
 851         /* Make sure no work handler is accessing the device. */
 852         flush_work(&vblk->config_work);
 853
 854         blk_mq_quiesce_queue(vblk->disk->queue);
 855
 856         vdev->config->del_vqs(vdev);
 857         return 0;
 858 }
 859
 860 static int virtblk_restore(struct virtio_device *vdev)
 861 {
 862         struct virtio_blk *vblk = vdev->priv;
 863         int ret;
 864
 865         ret = init_vq(vdev->priv);
 866         if (ret)
 867                 return ret;
 868
 869         virtio_device_ready(vdev);
 870
 871         blk_mq_unquiesce_queue(vblk->disk->queue);
 872         return 0;
 873 }
 874 #endif
 875
 876 static const struct virtio_device_id id_table[] = {
 877         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 878         { 0 },
 879 };
 880
 881 static unsigned int features_legacy[] = {
 882         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 883         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 884         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
 885         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
 886 }
 887 ;
 888 static unsigned int features[] = {
 889         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 890         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 891         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
 892         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
 893 };
 894
 895 static struct virtio_driver virtio_blk = {
 896         .feature_table                  = features,
 897         .feature_table_size             = ARRAY_SIZE(features),
 898         .feature_table_legacy           = features_legacy,
 899         .feature_table_size_legacy      = ARRAY_SIZE(features_legacy),
 900         .driver.name                    = KBUILD_MODNAME,
 901         .driver.owner                   = THIS_MODULE,
 902         .id_table                       = id_table,
 903         .probe                          = virtblk_probe,
 904         .remove                         = virtblk_remove,
 905         .config_changed                 = virtblk_config_changed,
 906 #ifdef CONFIG_PM_SLEEP
 907         .freeze                         = virtblk_freeze,
 908         .restore                        = virtblk_restore,
 909 #endif
 910 };
 911
 912 static int __init init(void)
 913 {
 914         int error;
 915
 916         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
 917         if (!virtblk_wq)
 918                 return -ENOMEM;
 919
 920         major = register_blkdev(0, "virtblk");
 921         if (major < 0) {
 922                 error = major;
 923                 goto out_destroy_workqueue;
 924         }
 925
 926         error = register_virtio_driver(&virtio_blk);
 927         if (error)
 928                 goto out_unregister_blkdev;
 929         return 0;
 930
 931 out_unregister_blkdev:
 932         unregister_blkdev(major, "virtblk");
 933 out_destroy_workqueue:
 934         destroy_workqueue(virtblk_wq);
 935         return error;
 936 }
 937
 938 static void __exit fini(void)
 939 {
 940         unregister_virtio_driver(&virtio_blk);
 941         unregister_blkdev(major, "virtblk");
 942         destroy_workqueue(virtblk_wq);
 943 }
 944 module_init(init);
 945 module_exit(fini);
 946
 947 MODULE_DEVICE_TABLE(virtio, id_table);
 948 MODULE_DESCRIPTION("Virtio block driver");
 949 MODULE_LICENSE("GPL");