drivers/block/virtio_blk.c

   1 //#define DEBUG
   2 #include <linux/spinlock.h>
   3 #include <linux/slab.h>
   4 #include <linux/blkdev.h>
   5 #include <linux/hdreg.h>
   6 #include <linux/module.h>
   7 #include <linux/mutex.h>
   8 #include <linux/interrupt.h>
   9 #include <linux/virtio.h>
  10 #include <linux/virtio_blk.h>
  11 #include <linux/scatterlist.h>
  12 #include <linux/string_helpers.h>
  13 #include <scsi/scsi_cmnd.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 #ifdef CONFIG_VIRTIO_BLK_SCSI
  59         struct scsi_request sreq;       /* for SCSI passthrough, must be first */
  60         u8 sense[SCSI_SENSE_BUFFERSIZE];
  61         struct virtio_scsi_inhdr in_hdr;
  62 #endif
  63         struct virtio_blk_outhdr out_hdr;
  64         u8 status;
  65         struct scatterlist sg[];
  66 };
  67
  68 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
  69 {
  70         switch (vbr->status) {
  71         case VIRTIO_BLK_S_OK:
  72                 return BLK_STS_OK;
  73         case VIRTIO_BLK_S_UNSUPP:
  74                 return BLK_STS_NOTSUPP;
  75         default:
  76                 return BLK_STS_IOERR;
  77         }
  78 }
  79
  80 /*
  81  * If this is a packet command we need a couple of additional headers.  Behind
  82  * the normal outhdr we put a segment with the scsi command block, and before
  83  * the normal inhdr we put the sense data and the inhdr with additional status
  84  * information.
  85  */
  86 #ifdef CONFIG_VIRTIO_BLK_SCSI
  87 static int virtblk_add_req_scsi(struct virtqueue *vq, struct virtblk_req *vbr,
  88                 struct scatterlist *data_sg, bool have_data)
  89 {
  90         struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
  91         unsigned int num_out = 0, num_in = 0;
  92
  93         sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
  94         sgs[num_out++] = &hdr;
  95         sg_init_one(&cmd, vbr->sreq.cmd, vbr->sreq.cmd_len);
  96         sgs[num_out++] = &cmd;
  97
  98         if (have_data) {
  99                 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
 100                         sgs[num_out++] = data_sg;
 101                 else
 102                         sgs[num_out + num_in++] = data_sg;
 103         }
 104
 105         sg_init_one(&sense, vbr->sense, SCSI_SENSE_BUFFERSIZE);
 106         sgs[num_out + num_in++] = &sense;
 107         sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
 108         sgs[num_out + num_in++] = &inhdr;
 109         sg_init_one(&status, &vbr->status, sizeof(vbr->status));
 110         sgs[num_out + num_in++] = &status;
 111
 112         return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 113 }
 114
 115 static inline void virtblk_scsi_request_done(struct request *req)
 116 {
 117         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 118         struct virtio_blk *vblk = req->q->queuedata;
 119         struct scsi_request *sreq = &vbr->sreq;
 120
 121         sreq->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual);
 122         sreq->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len);
 123         sreq->result = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors);
 124 }
 125
 126 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
 127                              unsigned int cmd, unsigned long data)
 128 {
 129         struct gendisk *disk = bdev->bd_disk;
 130         struct virtio_blk *vblk = disk->private_data;
 131
 132         /*
 133          * Only allow the generic SCSI ioctls if the host can support it.
 134          */
 135         if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
 136                 return -ENOTTY;
 137
 138         return scsi_cmd_blk_ioctl(bdev, mode, cmd,
 139                                   (void __user *)data);
 140 }
 141 #else
 142 static inline int virtblk_add_req_scsi(struct virtqueue *vq,
 143                 struct virtblk_req *vbr, struct scatterlist *data_sg,
 144                 bool have_data)
 145 {
 146         return -EIO;
 147 }
 148 static inline void virtblk_scsi_request_done(struct request *req)
 149 {
 150 }
 151 #define virtblk_ioctl   NULL
 152 #endif /* CONFIG_VIRTIO_BLK_SCSI */
 153
 154 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
 155                 struct scatterlist *data_sg, bool have_data)
 156 {
 157         struct scatterlist hdr, status, *sgs[3];
 158         unsigned int num_out = 0, num_in = 0;
 159
 160         sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
 161         sgs[num_out++] = &hdr;
 162
 163         if (have_data) {
 164                 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
 165                         sgs[num_out++] = data_sg;
 166                 else
 167                         sgs[num_out + num_in++] = data_sg;
 168         }
 169
 170         sg_init_one(&status, &vbr->status, sizeof(vbr->status));
 171         sgs[num_out + num_in++] = &status;
 172
 173         return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 174 }
 175
 176 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
 177 {
 178         unsigned short segments = blk_rq_nr_discard_segments(req);
 179         unsigned short n = 0;
 180         struct virtio_blk_discard_write_zeroes *range;
 181         struct bio *bio;
 182         u32 flags = 0;
 183
 184         if (unmap)
 185                 flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
 186
 187         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 188         if (!range)
 189                 return -ENOMEM;
 190
 191         __rq_for_each_bio(bio, req) {
 192                 u64 sector = bio->bi_iter.bi_sector;
 193                 u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
 194
 195                 range[n].flags = cpu_to_le32(flags);
 196                 range[n].num_sectors = cpu_to_le32(num_sectors);
 197                 range[n].sector = cpu_to_le64(sector);
 198                 n++;
 199         }
 200
 201         req->special_vec.bv_page = virt_to_page(range);
 202         req->special_vec.bv_offset = offset_in_page(range);
 203         req->special_vec.bv_len = sizeof(*range) * segments;
 204         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 205
 206         return 0;
 207 }
 208
 209 static inline void virtblk_request_done(struct request *req)
 210 {
 211         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 212
 213         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 214                 kfree(page_address(req->special_vec.bv_page) +
 215                       req->special_vec.bv_offset);
 216         }
 217
 218         switch (req_op(req)) {
 219         case REQ_OP_SCSI_IN:
 220         case REQ_OP_SCSI_OUT:
 221                 virtblk_scsi_request_done(req);
 222                 break;
 223         }
 224
 225         blk_mq_end_request(req, virtblk_result(vbr));
 226 }
 227
 228 static void virtblk_done(struct virtqueue *vq)
 229 {
 230         struct virtio_blk *vblk = vq->vdev->priv;
 231         bool req_done = false;
 232         int qid = vq->index;
 233         struct virtblk_req *vbr;
 234         unsigned long flags;
 235         unsigned int len;
 236
 237         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 238         do {
 239                 virtqueue_disable_cb(vq);
 240                 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
 241                         struct request *req = blk_mq_rq_from_pdu(vbr);
 242
 243                         blk_mq_complete_request(req);
 244                         req_done = true;
 245                 }
 246                 if (unlikely(virtqueue_is_broken(vq)))
 247                         break;
 248         } while (!virtqueue_enable_cb(vq));
 249
 250         /* In case queue is stopped waiting for more buffers. */
 251         if (req_done)
 252                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 253         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 254 }
 255
 256 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
 257 {
 258         struct virtio_blk *vblk = hctx->queue->queuedata;
 259         struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 260         bool kick;
 261
 262         spin_lock_irq(&vq->lock);
 263         kick = virtqueue_kick_prepare(vq->vq);
 264         spin_unlock_irq(&vq->lock);
 265
 266         if (kick)
 267                 virtqueue_notify(vq->vq);
 268 }
 269
 270 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 271                            const struct blk_mq_queue_data *bd)
 272 {
 273         struct virtio_blk *vblk = hctx->queue->queuedata;
 274         struct request *req = bd->rq;
 275         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 276         unsigned long flags;
 277         unsigned int num;
 278         int qid = hctx->queue_num;
 279         int err;
 280         bool notify = false;
 281         bool unmap = false;
 282         u32 type;
 283
 284         BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 285
 286         switch (req_op(req)) {
 287         case REQ_OP_READ:
 288         case REQ_OP_WRITE:
 289                 type = 0;
 290                 break;
 291         case REQ_OP_FLUSH:
 292                 type = VIRTIO_BLK_T_FLUSH;
 293                 break;
 294         case REQ_OP_DISCARD:
 295                 type = VIRTIO_BLK_T_DISCARD;
 296                 break;
 297         case REQ_OP_WRITE_ZEROES:
 298                 type = VIRTIO_BLK_T_WRITE_ZEROES;
 299                 unmap = !(req->cmd_flags & REQ_NOUNMAP);
 300                 break;
 301         case REQ_OP_SCSI_IN:
 302         case REQ_OP_SCSI_OUT:
 303                 type = VIRTIO_BLK_T_SCSI_CMD;
 304                 break;
 305         case REQ_OP_DRV_IN:
 306                 type = VIRTIO_BLK_T_GET_ID;
 307                 break;
 308         default:
 309                 WARN_ON_ONCE(1);
 310                 return BLK_STS_IOERR;
 311         }
 312
 313         vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
 314         vbr->out_hdr.sector = type ?
 315                 0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
 316         vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
 317
 318         blk_mq_start_request(req);
 319
 320         if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
 321                 err = virtblk_setup_discard_write_zeroes(req, unmap);
 322                 if (err)
 323                         return BLK_STS_RESOURCE;
 324         }
 325
 326         num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
 327         if (num) {
 328                 if (rq_data_dir(req) == WRITE)
 329                         vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
 330                 else
 331                         vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
 332         }
 333
 334         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 335         if (blk_rq_is_scsi(req))
 336                 err = virtblk_add_req_scsi(vblk->vqs[qid].vq, vbr, vbr->sg, num);
 337         else
 338                 err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
 339         if (err) {
 340                 virtqueue_kick(vblk->vqs[qid].vq);
 341                 blk_mq_stop_hw_queue(hctx);
 342                 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 343                 /* Out of mem doesn't actually happen, since we fall back
 344                  * to direct descriptors */
 345                 if (err == -ENOMEM || err == -ENOSPC)
 346                         return BLK_STS_DEV_RESOURCE;
 347                 return BLK_STS_IOERR;
 348         }
 349
 350         if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
 351                 notify = true;
 352         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 353
 354         if (notify)
 355                 virtqueue_notify(vblk->vqs[qid].vq);
 356         return BLK_STS_OK;
 357 }
 358
 359 /* return id (s/n) string for *disk to *id_str
 360  */
 361 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 362 {
 363         struct virtio_blk *vblk = disk->private_data;
 364         struct request_queue *q = vblk->disk->queue;
 365         struct request *req;
 366         int err;
 367
 368         req = blk_get_request(q, REQ_OP_DRV_IN, 0);
 369         if (IS_ERR(req))
 370                 return PTR_ERR(req);
 371
 372         err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
 373         if (err)
 374                 goto out;
 375
 376         blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
 377         err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
 378 out:
 379         blk_put_request(req);
 380         return err;
 381 }
 382
 383 /* We provide getgeo only to please some old bootloader/partitioning tools */
 384 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 385 {
 386         struct virtio_blk *vblk = bd->bd_disk->private_data;
 387
 388         /* see if the host passed in geometry config */
 389         if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 390                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 391                              geometry.cylinders, &geo->cylinders);
 392                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 393                              geometry.heads, &geo->heads);
 394                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 395                              geometry.sectors, &geo->sectors);
 396         } else {
 397                 /* some standard values, similar to sd */
 398                 geo->heads = 1 << 6;
 399                 geo->sectors = 1 << 5;
 400                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 401         }
 402         return 0;
 403 }
 404
 405 static const struct block_device_operations virtblk_fops = {
 406         .ioctl  = virtblk_ioctl,
 407         .owner  = THIS_MODULE,
 408         .getgeo = virtblk_getgeo,
 409 };
 410
 411 static int index_to_minor(int index)
 412 {
 413         return index << PART_BITS;
 414 }
 415
 416 static int minor_to_index(int minor)
 417 {
 418         return minor >> PART_BITS;
 419 }
 420
 421 static ssize_t serial_show(struct device *dev,
 422                            struct device_attribute *attr, char *buf)
 423 {
 424         struct gendisk *disk = dev_to_disk(dev);
 425         int err;
 426
 427         /* sysfs gives us a PAGE_SIZE buffer */
 428         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 429
 430         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 431         err = virtblk_get_id(disk, buf);
 432         if (!err)
 433                 return strlen(buf);
 434
 435         if (err == -EIO) /* Unsupported? Make it empty. */
 436                 return 0;
 437
 438         return err;
 439 }
 440
 441 static DEVICE_ATTR_RO(serial);
 442
 443 /* The queue's logical block size must be set before calling this */
 444 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
 445 {
 446         struct virtio_device *vdev = vblk->vdev;
 447         struct request_queue *q = vblk->disk->queue;
 448         char cap_str_2[10], cap_str_10[10];
 449         unsigned long long nblocks;
 450         u64 capacity;
 451
 452         /* Host must always specify the capacity. */
 453         virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
 454
 455         /* If capacity is too big, truncate with warning. */
 456         if ((sector_t)capacity != capacity) {
 457                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 458                          (unsigned long long)capacity);
 459                 capacity = (sector_t)-1;
 460         }
 461
 462         nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
 463
 464         string_get_size(nblocks, queue_logical_block_size(q),
 465                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 466         string_get_size(nblocks, queue_logical_block_size(q),
 467                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 468
 469         dev_notice(&vdev->dev,
 470                    "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
 471                    vblk->disk->disk_name,
 472                    resize ? "new size: " : "",
 473                    nblocks,
 474                    queue_logical_block_size(q),
 475                    cap_str_10,
 476                    cap_str_2);
 477
 478         set_capacity(vblk->disk, capacity);
 479 }
 480
 481 static void virtblk_config_changed_work(struct work_struct *work)
 482 {
 483         struct virtio_blk *vblk =
 484                 container_of(work, struct virtio_blk, config_work);
 485         char *envp[] = { "RESIZE=1", NULL };
 486
 487         virtblk_update_capacity(vblk, true);
 488         revalidate_disk(vblk->disk);
 489         kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
 490 }
 491
 492 static void virtblk_config_changed(struct virtio_device *vdev)
 493 {
 494         struct virtio_blk *vblk = vdev->priv;
 495
 496         queue_work(virtblk_wq, &vblk->config_work);
 497 }
 498
 499 static int init_vq(struct virtio_blk *vblk)
 500 {
 501         int err;
 502         int i;
 503         vq_callback_t **callbacks;
 504         const char **names;
 505         struct virtqueue **vqs;
 506         unsigned short num_vqs;
 507         struct virtio_device *vdev = vblk->vdev;
 508         struct irq_affinity desc = { 0, };
 509
 510         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
 511                                    struct virtio_blk_config, num_queues,
 512                                    &num_vqs);
 513         if (err)
 514                 num_vqs = 1;
 515
 516         vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
 517         if (!vblk->vqs)
 518                 return -ENOMEM;
 519
 520         names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
 521         callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
 522         vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
 523         if (!names || !callbacks || !vqs) {
 524                 err = -ENOMEM;
 525                 goto out;
 526         }
 527
 528         for (i = 0; i < num_vqs; i++) {
 529                 callbacks[i] = virtblk_done;
 530                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
 531                 names[i] = vblk->vqs[i].name;
 532         }
 533
 534         /* Discover virtqueues and write information to configuration.  */
 535         err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
 536         if (err)
 537                 goto out;
 538
 539         for (i = 0; i < num_vqs; i++) {
 540                 spin_lock_init(&vblk->vqs[i].lock);
 541                 vblk->vqs[i].vq = vqs[i];
 542         }
 543         vblk->num_vqs = num_vqs;
 544
 545 out:
 546         kfree(vqs);
 547         kfree(callbacks);
 548         kfree(names);
 549         if (err)
 550                 kfree(vblk->vqs);
 551         return err;
 552 }
 553
 554 /*
 555  * Legacy naming scheme used for virtio devices.  We are stuck with it for
 556  * virtio blk but don't ever use it for any new driver.
 557  */
 558 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
 559 {
 560         const int base = 'z' - 'a' + 1;
 561         char *begin = buf + strlen(prefix);
 562         char *end = buf + buflen;
 563         char *p;
 564         int unit;
 565
 566         p = end - 1;
 567         *p = '\0';
 568         unit = base;
 569         do {
 570                 if (p == begin)
 571                         return -EINVAL;
 572                 *--p = 'a' + (index % unit);
 573                 index = (index / unit) - 1;
 574         } while (index >= 0);
 575
 576         memmove(begin, p, end - p);
 577         memcpy(buf, prefix, strlen(prefix));
 578
 579         return 0;
 580 }
 581
 582 static int virtblk_get_cache_mode(struct virtio_device *vdev)
 583 {
 584         u8 writeback;
 585         int err;
 586
 587         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
 588                                    struct virtio_blk_config, wce,
 589                                    &writeback);
 590
 591         /*
 592          * If WCE is not configurable and flush is not available,
 593          * assume no writeback cache is in use.
 594          */
 595         if (err)
 596                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
 597
 598         return writeback;
 599 }
 600
 601 static void virtblk_update_cache_mode(struct virtio_device *vdev)
 602 {
 603         u8 writeback = virtblk_get_cache_mode(vdev);
 604         struct virtio_blk *vblk = vdev->priv;
 605
 606         blk_queue_write_cache(vblk->disk->queue, writeback, false);
 607         revalidate_disk(vblk->disk);
 608 }
 609
 610 static const char *const virtblk_cache_types[] = {
 611         "write through", "write back"
 612 };
 613
 614 static ssize_t
 615 cache_type_store(struct device *dev, struct device_attribute *attr,
 616                  const char *buf, size_t count)
 617 {
 618         struct gendisk *disk = dev_to_disk(dev);
 619         struct virtio_blk *vblk = disk->private_data;
 620         struct virtio_device *vdev = vblk->vdev;
 621         int i;
 622
 623         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
 624         i = sysfs_match_string(virtblk_cache_types, buf);
 625         if (i < 0)
 626                 return i;
 627
 628         virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
 629         virtblk_update_cache_mode(vdev);
 630         return count;
 631 }
 632
 633 static ssize_t
 634 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 635 {
 636         struct gendisk *disk = dev_to_disk(dev);
 637         struct virtio_blk *vblk = disk->private_data;
 638         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
 639
 640         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
 641         return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
 642 }
 643
 644 static DEVICE_ATTR_RW(cache_type);
 645
 646 static struct attribute *virtblk_attrs[] = {
 647         &dev_attr_serial.attr,
 648         &dev_attr_cache_type.attr,
 649         NULL,
 650 };
 651
 652 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
 653                 struct attribute *a, int n)
 654 {
 655         struct device *dev = container_of(kobj, struct device, kobj);
 656         struct gendisk *disk = dev_to_disk(dev);
 657         struct virtio_blk *vblk = disk->private_data;
 658         struct virtio_device *vdev = vblk->vdev;
 659
 660         if (a == &dev_attr_cache_type.attr &&
 661             !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
 662                 return S_IRUGO;
 663
 664         return a->mode;
 665 }
 666
 667 static const struct attribute_group virtblk_attr_group = {
 668         .attrs = virtblk_attrs,
 669         .is_visible = virtblk_attrs_are_visible,
 670 };
 671
 672 static const struct attribute_group *virtblk_attr_groups[] = {
 673         &virtblk_attr_group,
 674         NULL,
 675 };
 676
 677 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
 678                 unsigned int hctx_idx, unsigned int numa_node)
 679 {
 680         struct virtio_blk *vblk = set->driver_data;
 681         struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
 682
 683 #ifdef CONFIG_VIRTIO_BLK_SCSI
 684         vbr->sreq.sense = vbr->sense;
 685 #endif
 686         sg_init_table(vbr->sg, vblk->sg_elems);
 687         return 0;
 688 }
 689
 690 static int virtblk_map_queues(struct blk_mq_tag_set *set)
 691 {
 692         struct virtio_blk *vblk = set->driver_data;
 693
 694         return blk_mq_virtio_map_queues(&set->map[0], vblk->vdev, 0);
 695 }
 696
 697 #ifdef CONFIG_VIRTIO_BLK_SCSI
 698 static void virtblk_initialize_rq(struct request *req)
 699 {
 700         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 701
 702         scsi_req_init(&vbr->sreq);
 703 }
 704 #endif
 705
 706 static const struct blk_mq_ops virtio_mq_ops = {
 707         .queue_rq       = virtio_queue_rq,
 708         .commit_rqs     = virtio_commit_rqs,
 709         .complete       = virtblk_request_done,
 710         .init_request   = virtblk_init_request,
 711 #ifdef CONFIG_VIRTIO_BLK_SCSI
 712         .initialize_rq_fn = virtblk_initialize_rq,
 713 #endif
 714         .map_queues     = virtblk_map_queues,
 715 };
 716
 717 static unsigned int virtblk_queue_depth;
 718 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
 719
 720 static int virtblk_probe(struct virtio_device *vdev)
 721 {
 722         struct virtio_blk *vblk;
 723         struct request_queue *q;
 724         int err, index;
 725
 726         u32 v, blk_size, sg_elems, opt_io_size;
 727         u16 min_io_size;
 728         u8 physical_block_exp, alignment_offset;
 729
 730         if (!vdev->config->get) {
 731                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
 732                         __func__);
 733                 return -EINVAL;
 734         }
 735
 736         err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
 737                              GFP_KERNEL);
 738         if (err < 0)
 739                 goto out;
 740         index = err;
 741
 742         /* We need to know how many segments before we allocate. */
 743         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
 744                                    struct virtio_blk_config, seg_max,
 745                                    &sg_elems);
 746
 747         /* We need at least one SG element, whatever they say. */
 748         if (err || !sg_elems)
 749                 sg_elems = 1;
 750
 751         /* We need an extra sg elements at head and tail. */
 752         sg_elems += 2;
 753         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
 754         if (!vblk) {
 755                 err = -ENOMEM;
 756                 goto out_free_index;
 757         }
 758
 759         vblk->vdev = vdev;
 760         vblk->sg_elems = sg_elems;
 761
 762         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 763
 764         err = init_vq(vblk);
 765         if (err)
 766                 goto out_free_vblk;
 767
 768         /* FIXME: How many partitions?  How long is a piece of string? */
 769         vblk->disk = alloc_disk(1 << PART_BITS);
 770         if (!vblk->disk) {
 771                 err = -ENOMEM;
 772                 goto out_free_vq;
 773         }
 774
 775         /* Default queue sizing is to fill the ring. */
 776         if (!virtblk_queue_depth) {
 777                 virtblk_queue_depth = vblk->vqs[0].vq->num_free;
 778                 /* ... but without indirect descs, we use 2 descs per req */
 779                 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
 780                         virtblk_queue_depth /= 2;
 781         }
 782
 783         memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
 784         vblk->tag_set.ops = &virtio_mq_ops;
 785         vblk->tag_set.queue_depth = virtblk_queue_depth;
 786         vblk->tag_set.numa_node = NUMA_NO_NODE;
 787         vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 788         vblk->tag_set.cmd_size =
 789                 sizeof(struct virtblk_req) +
 790                 sizeof(struct scatterlist) * sg_elems;
 791         vblk->tag_set.driver_data = vblk;
 792         vblk->tag_set.nr_hw_queues = vblk->num_vqs;
 793
 794         err = blk_mq_alloc_tag_set(&vblk->tag_set);
 795         if (err)
 796                 goto out_put_disk;
 797
 798         q = blk_mq_init_queue(&vblk->tag_set);
 799         if (IS_ERR(q)) {
 800                 err = -ENOMEM;
 801                 goto out_free_tags;
 802         }
 803         vblk->disk->queue = q;
 804
 805         q->queuedata = vblk;
 806
 807         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
 808
 809         vblk->disk->major = major;
 810         vblk->disk->first_minor = index_to_minor(index);
 811         vblk->disk->private_data = vblk;
 812         vblk->disk->fops = &virtblk_fops;
 813         vblk->disk->flags |= GENHD_FL_EXT_DEVT;
 814         vblk->index = index;
 815
 816         /* configure queue flush support */
 817         virtblk_update_cache_mode(vdev);
 818
 819         /* If disk is read-only in the host, the guest should obey */
 820         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 821                 set_disk_ro(vblk->disk, 1);
 822
 823         /* We can handle whatever the host told us to handle. */
 824         blk_queue_max_segments(q, vblk->sg_elems-2);
 825
 826         /* No real sector limit. */
 827         blk_queue_max_hw_sectors(q, -1U);
 828
 829         /* Host can optionally specify maximum segment size and number of
 830          * segments. */
 831         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
 832                                    struct virtio_blk_config, size_max, &v);
 833         if (!err)
 834                 blk_queue_max_segment_size(q, v);
 835         else
 836                 blk_queue_max_segment_size(q, -1U);
 837
 838         /* Host can optionally specify the block size of the device */
 839         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
 840                                    struct virtio_blk_config, blk_size,
 841                                    &blk_size);
 842         if (!err)
 843                 blk_queue_logical_block_size(q, blk_size);
 844         else
 845                 blk_size = queue_logical_block_size(q);
 846
 847         /* Use topology information if available */
 848         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 849                                    struct virtio_blk_config, physical_block_exp,
 850                                    &physical_block_exp);
 851         if (!err && physical_block_exp)
 852                 blk_queue_physical_block_size(q,
 853                                 blk_size * (1 << physical_block_exp));
 854
 855         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 856                                    struct virtio_blk_config, alignment_offset,
 857                                    &alignment_offset);
 858         if (!err && alignment_offset)
 859                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 860
 861         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 862                                    struct virtio_blk_config, min_io_size,
 863                                    &min_io_size);
 864         if (!err && min_io_size)
 865                 blk_queue_io_min(q, blk_size * min_io_size);
 866
 867         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 868                                    struct virtio_blk_config, opt_io_size,
 869                                    &opt_io_size);
 870         if (!err && opt_io_size)
 871                 blk_queue_io_opt(q, blk_size * opt_io_size);
 872
 873         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
 874                 q->limits.discard_granularity = blk_size;
 875
 876                 virtio_cread(vdev, struct virtio_blk_config,
 877                              discard_sector_alignment, &v);
 878                 q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
 879
 880                 virtio_cread(vdev, struct virtio_blk_config,
 881                              max_discard_sectors, &v);
 882                 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
 883
 884                 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
 885                              &v);
 886                 blk_queue_max_discard_segments(q,
 887                                                min_not_zero(v,
 888                                                             MAX_DISCARD_SEGMENTS));
 889
 890                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 891         }
 892
 893         if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
 894                 virtio_cread(vdev, struct virtio_blk_config,
 895                              max_write_zeroes_sectors, &v);
 896                 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
 897         }
 898
 899         virtblk_update_capacity(vblk, false);
 900         virtio_device_ready(vdev);
 901
 902         device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
 903         return 0;
 904
 905 out_free_tags:
 906         blk_mq_free_tag_set(&vblk->tag_set);
 907 out_put_disk:
 908         put_disk(vblk->disk);
 909 out_free_vq:
 910         vdev->config->del_vqs(vdev);
 911 out_free_vblk:
 912         kfree(vblk);
 913 out_free_index:
 914         ida_simple_remove(&vd_index_ida, index);
 915 out:
 916         return err;
 917 }
 918
 919 static void virtblk_remove(struct virtio_device *vdev)
 920 {
 921         struct virtio_blk *vblk = vdev->priv;
 922         int index = vblk->index;
 923         int refc;
 924
 925         /* Make sure no work handler is accessing the device. */
 926         flush_work(&vblk->config_work);
 927
 928         del_gendisk(vblk->disk);
 929         blk_cleanup_queue(vblk->disk->queue);
 930
 931         blk_mq_free_tag_set(&vblk->tag_set);
 932
 933         /* Stop all the virtqueues. */
 934         vdev->config->reset(vdev);
 935
 936         refc = kref_read(&disk_to_dev(vblk->disk)->kobj.kref);
 937         put_disk(vblk->disk);
 938         vdev->config->del_vqs(vdev);
 939         kfree(vblk->vqs);
 940         kfree(vblk);
 941
 942         /* Only free device id if we don't have any users */
 943         if (refc == 1)
 944                 ida_simple_remove(&vd_index_ida, index);
 945 }
 946
 947 #ifdef CONFIG_PM_SLEEP
 948 static int virtblk_freeze(struct virtio_device *vdev)
 949 {
 950         struct virtio_blk *vblk = vdev->priv;
 951
 952         /* Ensure we don't receive any more interrupts */
 953         vdev->config->reset(vdev);
 954
 955         /* Make sure no work handler is accessing the device. */
 956         flush_work(&vblk->config_work);
 957
 958         blk_mq_quiesce_queue(vblk->disk->queue);
 959
 960         vdev->config->del_vqs(vdev);
 961         return 0;
 962 }
 963
 964 static int virtblk_restore(struct virtio_device *vdev)
 965 {
 966         struct virtio_blk *vblk = vdev->priv;
 967         int ret;
 968
 969         ret = init_vq(vdev->priv);
 970         if (ret)
 971                 return ret;
 972
 973         virtio_device_ready(vdev);
 974
 975         blk_mq_unquiesce_queue(vblk->disk->queue);
 976         return 0;
 977 }
 978 #endif
 979
 980 static const struct virtio_device_id id_table[] = {
 981         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 982         { 0 },
 983 };
 984
 985 static unsigned int features_legacy[] = {
 986         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 987         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 988 #ifdef CONFIG_VIRTIO_BLK_SCSI
 989         VIRTIO_BLK_F_SCSI,
 990 #endif
 991         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
 992         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
 993 }
 994 ;
 995 static unsigned int features[] = {
 996         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 997         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 998         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
 999         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1000 };
1001
1002 static struct virtio_driver virtio_blk = {
1003         .feature_table                  = features,
1004         .feature_table_size             = ARRAY_SIZE(features),
1005         .feature_table_legacy           = features_legacy,
1006         .feature_table_size_legacy      = ARRAY_SIZE(features_legacy),
1007         .driver.name                    = KBUILD_MODNAME,
1008         .driver.owner                   = THIS_MODULE,
1009         .id_table                       = id_table,
1010         .probe                          = virtblk_probe,
1011         .remove                         = virtblk_remove,
1012         .config_changed                 = virtblk_config_changed,
1013 #ifdef CONFIG_PM_SLEEP
1014         .freeze                         = virtblk_freeze,
1015         .restore                        = virtblk_restore,
1016 #endif
1017 };
1018
1019 static int __init init(void)
1020 {
1021         int error;
1022
1023         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1024         if (!virtblk_wq)
1025                 return -ENOMEM;
1026
1027         major = register_blkdev(0, "virtblk");
1028         if (major < 0) {
1029                 error = major;
1030                 goto out_destroy_workqueue;
1031         }
1032
1033         error = register_virtio_driver(&virtio_blk);
1034         if (error)
1035                 goto out_unregister_blkdev;
1036         return 0;
1037
1038 out_unregister_blkdev:
1039         unregister_blkdev(major, "virtblk");
1040 out_destroy_workqueue:
1041         destroy_workqueue(virtblk_wq);
1042         return error;
1043 }
1044
1045 static void __exit fini(void)
1046 {
1047         unregister_virtio_driver(&virtio_blk);
1048         unregister_blkdev(major, "virtblk");
1049         destroy_workqueue(virtblk_wq);
1050 }
1051 module_init(init);
1052 module_exit(fini);
1053
1054 MODULE_DEVICE_TABLE(virtio, id_table);
1055 MODULE_DESCRIPTION("Virtio block driver");
1056 MODULE_LICENSE("GPL");