drivers/block/virtio_blk.c

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