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 <scsi/scsi_cmnd.h>
  15 #include <linux/idr.h>
  16 #include <linux/blk-mq.h>
  17 #include <linux/blk-mq-virtio.h>
  18 #include <linux/numa.h>
  19
  20 #define PART_BITS 4
  21 #define VQ_NAME_LEN 16
  22 #define MAX_DISCARD_SEGMENTS 256u
  23
  24 static int major;
  25 static DEFINE_IDA(vd_index_ida);
  26
  27 static struct workqueue_struct *virtblk_wq;
  28
  29 struct virtio_blk_vq {
  30         struct virtqueue *vq;
  31         spinlock_t lock;
  32         char name[VQ_NAME_LEN];
  33 } ____cacheline_aligned_in_smp;
  34
  35 struct virtio_blk {
  36         struct virtio_device *vdev;
  37
  38         /* The disk structure for the kernel. */
  39         struct gendisk *disk;
  40
  41         /* Block layer tags. */
  42         struct blk_mq_tag_set tag_set;
  43
  44         /* Process context for config space updates */
  45         struct work_struct config_work;
  46
  47         /* What host tells us, plus 2 for header & tailer. */
  48         unsigned int sg_elems;
  49
  50         /* Ida index - used to track minor number allocations. */
  51         int index;
  52
  53         /* num of vqs */
  54         int num_vqs;
  55         struct virtio_blk_vq *vqs;
  56 };
  57
  58 struct virtblk_req {
  59 #ifdef CONFIG_VIRTIO_BLK_SCSI
  60         struct scsi_request sreq;       /* for SCSI passthrough, must be first */
  61         u8 sense[SCSI_SENSE_BUFFERSIZE];
  62         struct virtio_scsi_inhdr in_hdr;
  63 #endif
  64         struct virtio_blk_outhdr out_hdr;
  65         u8 status;
  66         struct scatterlist sg[];
  67 };
  68
  69 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
  70 {
  71         switch (vbr->status) {
  72         case VIRTIO_BLK_S_OK:
  73                 return BLK_STS_OK;
  74         case VIRTIO_BLK_S_UNSUPP:
  75                 return BLK_STS_NOTSUPP;
  76         default:
  77                 return BLK_STS_IOERR;
  78         }
  79 }
  80
  81 /*
  82  * If this is a packet command we need a couple of additional headers.  Behind
  83  * the normal outhdr we put a segment with the scsi command block, and before
  84  * the normal inhdr we put the sense data and the inhdr with additional status
  85  * information.
  86  */
  87 #ifdef CONFIG_VIRTIO_BLK_SCSI
  88 static int virtblk_add_req_scsi(struct virtqueue *vq, struct virtblk_req *vbr,
  89                 struct scatterlist *data_sg, bool have_data)
  90 {
  91         struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
  92         unsigned int num_out = 0, num_in = 0;
  93
  94         sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
  95         sgs[num_out++] = &hdr;
  96         sg_init_one(&cmd, vbr->sreq.cmd, vbr->sreq.cmd_len);
  97         sgs[num_out++] = &cmd;
  98
  99         if (have_data) {
 100                 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
 101                         sgs[num_out++] = data_sg;
 102                 else
 103                         sgs[num_out + num_in++] = data_sg;
 104         }
 105
 106         sg_init_one(&sense, vbr->sense, SCSI_SENSE_BUFFERSIZE);
 107         sgs[num_out + num_in++] = &sense;
 108         sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
 109         sgs[num_out + num_in++] = &inhdr;
 110         sg_init_one(&status, &vbr->status, sizeof(vbr->status));
 111         sgs[num_out + num_in++] = &status;
 112
 113         return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 114 }
 115
 116 static inline void virtblk_scsi_request_done(struct request *req)
 117 {
 118         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 119         struct virtio_blk *vblk = req->q->queuedata;
 120         struct scsi_request *sreq = &vbr->sreq;
 121
 122         sreq->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual);
 123         sreq->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len);
 124         sreq->result = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors);
 125 }
 126
 127 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
 128                              unsigned int cmd, unsigned long data)
 129 {
 130         struct gendisk *disk = bdev->bd_disk;
 131         struct virtio_blk *vblk = disk->private_data;
 132
 133         /*
 134          * Only allow the generic SCSI ioctls if the host can support it.
 135          */
 136         if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
 137                 return -ENOTTY;
 138
 139         return scsi_cmd_blk_ioctl(bdev, mode, cmd,
 140                                   (void __user *)data);
 141 }
 142 #else
 143 static inline int virtblk_add_req_scsi(struct virtqueue *vq,
 144                 struct virtblk_req *vbr, struct scatterlist *data_sg,
 145                 bool have_data)
 146 {
 147         return -EIO;
 148 }
 149 static inline void virtblk_scsi_request_done(struct request *req)
 150 {
 151 }
 152 #define virtblk_ioctl   NULL
 153 #endif /* CONFIG_VIRTIO_BLK_SCSI */
 154
 155 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
 156                 struct scatterlist *data_sg, bool have_data)
 157 {
 158         struct scatterlist hdr, status, *sgs[3];
 159         unsigned int num_out = 0, num_in = 0;
 160
 161         sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
 162         sgs[num_out++] = &hdr;
 163
 164         if (have_data) {
 165                 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
 166                         sgs[num_out++] = data_sg;
 167                 else
 168                         sgs[num_out + num_in++] = data_sg;
 169         }
 170
 171         sg_init_one(&status, &vbr->status, sizeof(vbr->status));
 172         sgs[num_out + num_in++] = &status;
 173
 174         return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 175 }
 176
 177 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
 178 {
 179         unsigned short segments = blk_rq_nr_discard_segments(req);
 180         unsigned short n = 0;
 181         struct virtio_blk_discard_write_zeroes *range;
 182         struct bio *bio;
 183         u32 flags = 0;
 184
 185         if (unmap)
 186                 flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
 187
 188         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 189         if (!range)
 190                 return -ENOMEM;
 191
 192         __rq_for_each_bio(bio, req) {
 193                 u64 sector = bio->bi_iter.bi_sector;
 194                 u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
 195
 196                 range[n].flags = cpu_to_le32(flags);
 197                 range[n].num_sectors = cpu_to_le32(num_sectors);
 198                 range[n].sector = cpu_to_le64(sector);
 199                 n++;
 200         }
 201
 202         req->special_vec.bv_page = virt_to_page(range);
 203         req->special_vec.bv_offset = offset_in_page(range);
 204         req->special_vec.bv_len = sizeof(*range) * segments;
 205         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 206
 207         return 0;
 208 }
 209
 210 static inline void virtblk_request_done(struct request *req)
 211 {
 212         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 213
 214         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 215                 kfree(page_address(req->special_vec.bv_page) +
 216                       req->special_vec.bv_offset);
 217         }
 218
 219         switch (req_op(req)) {
 220         case REQ_OP_SCSI_IN:
 221         case REQ_OP_SCSI_OUT:
 222                 virtblk_scsi_request_done(req);
 223                 break;
 224         }
 225
 226         blk_mq_end_request(req, virtblk_result(vbr));
 227 }
 228
 229 static void virtblk_done(struct virtqueue *vq)
 230 {
 231         struct virtio_blk *vblk = vq->vdev->priv;
 232         bool req_done = false;
 233         int qid = vq->index;
 234         struct virtblk_req *vbr;
 235         unsigned long flags;
 236         unsigned int len;
 237
 238         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 239         do {
 240                 virtqueue_disable_cb(vq);
 241                 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
 242                         struct request *req = blk_mq_rq_from_pdu(vbr);
 243
 244                         blk_mq_complete_request(req);
 245                         req_done = true;
 246                 }
 247                 if (unlikely(virtqueue_is_broken(vq)))
 248                         break;
 249         } while (!virtqueue_enable_cb(vq));
 250
 251         /* In case queue is stopped waiting for more buffers. */
 252         if (req_done)
 253                 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 254         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 255 }
 256
 257 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
 258 {
 259         struct virtio_blk *vblk = hctx->queue->queuedata;
 260         struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 261         bool kick;
 262
 263         spin_lock_irq(&vq->lock);
 264         kick = virtqueue_kick_prepare(vq->vq);
 265         spin_unlock_irq(&vq->lock);
 266
 267         if (kick)
 268                 virtqueue_notify(vq->vq);
 269 }
 270
 271 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 272                            const struct blk_mq_queue_data *bd)
 273 {
 274         struct virtio_blk *vblk = hctx->queue->queuedata;
 275         struct request *req = bd->rq;
 276         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 277         unsigned long flags;
 278         unsigned int num;
 279         int qid = hctx->queue_num;
 280         int err;
 281         bool notify = false;
 282         bool unmap = false;
 283         u32 type;
 284
 285         BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 286
 287         switch (req_op(req)) {
 288         case REQ_OP_READ:
 289         case REQ_OP_WRITE:
 290                 type = 0;
 291                 break;
 292         case REQ_OP_FLUSH:
 293                 type = VIRTIO_BLK_T_FLUSH;
 294                 break;
 295         case REQ_OP_DISCARD:
 296                 type = VIRTIO_BLK_T_DISCARD;
 297                 break;
 298         case REQ_OP_WRITE_ZEROES:
 299                 type = VIRTIO_BLK_T_WRITE_ZEROES;
 300                 unmap = !(req->cmd_flags & REQ_NOUNMAP);
 301                 break;
 302         case REQ_OP_SCSI_IN:
 303         case REQ_OP_SCSI_OUT:
 304                 type = VIRTIO_BLK_T_SCSI_CMD;
 305                 break;
 306         case REQ_OP_DRV_IN:
 307                 type = VIRTIO_BLK_T_GET_ID;
 308                 break;
 309         default:
 310                 WARN_ON_ONCE(1);
 311                 return BLK_STS_IOERR;
 312         }
 313
 314         vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
 315         vbr->out_hdr.sector = type ?
 316                 0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
 317         vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
 318
 319         blk_mq_start_request(req);
 320
 321         if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
 322                 err = virtblk_setup_discard_write_zeroes(req, unmap);
 323                 if (err)
 324                         return BLK_STS_RESOURCE;
 325         }
 326
 327         num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
 328         if (num) {
 329                 if (rq_data_dir(req) == WRITE)
 330                         vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
 331                 else
 332                         vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
 333         }
 334
 335         spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 336         if (blk_rq_is_scsi(req))
 337                 err = virtblk_add_req_scsi(vblk->vqs[qid].vq, vbr, vbr->sg, num);
 338         else
 339                 err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
 340         if (err) {
 341                 virtqueue_kick(vblk->vqs[qid].vq);
 342                 blk_mq_stop_hw_queue(hctx);
 343                 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 344                 /* Out of mem doesn't actually happen, since we fall back
 345                  * to direct descriptors */
 346                 if (err == -ENOMEM || err == -ENOSPC)
 347                         return BLK_STS_DEV_RESOURCE;
 348                 return BLK_STS_IOERR;
 349         }
 350
 351         if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
 352                 notify = true;
 353         spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 354
 355         if (notify)
 356                 virtqueue_notify(vblk->vqs[qid].vq);
 357         return BLK_STS_OK;
 358 }
 359
 360 /* return id (s/n) string for *disk to *id_str
 361  */
 362 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 363 {
 364         struct virtio_blk *vblk = disk->private_data;
 365         struct request_queue *q = vblk->disk->queue;
 366         struct request *req;
 367         int err;
 368
 369         req = blk_get_request(q, REQ_OP_DRV_IN, 0);
 370         if (IS_ERR(req))
 371                 return PTR_ERR(req);
 372
 373         err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
 374         if (err)
 375                 goto out;
 376
 377         blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
 378         err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
 379 out:
 380         blk_put_request(req);
 381         return err;
 382 }
 383
 384 /* We provide getgeo only to please some old bootloader/partitioning tools */
 385 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 386 {
 387         struct virtio_blk *vblk = bd->bd_disk->private_data;
 388
 389         /* see if the host passed in geometry config */
 390         if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 391                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 392                              geometry.cylinders, &geo->cylinders);
 393                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 394                              geometry.heads, &geo->heads);
 395                 virtio_cread(vblk->vdev, struct virtio_blk_config,
 396                              geometry.sectors, &geo->sectors);
 397         } else {
 398                 /* some standard values, similar to sd */
 399                 geo->heads = 1 << 6;
 400                 geo->sectors = 1 << 5;
 401                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 402         }
 403         return 0;
 404 }
 405
 406 static const struct block_device_operations virtblk_fops = {
 407         .ioctl  = virtblk_ioctl,
 408 #ifdef CONFIG_COMPAT
 409         .compat_ioctl = blkdev_compat_ptr_ioctl,
 410 #endif
 411         .owner  = THIS_MODULE,
 412         .getgeo = virtblk_getgeo,
 413 };
 414
 415 static int index_to_minor(int index)
 416 {
 417         return index << PART_BITS;
 418 }
 419
 420 static int minor_to_index(int minor)
 421 {
 422         return minor >> PART_BITS;
 423 }
 424
 425 static ssize_t serial_show(struct device *dev,
 426                            struct device_attribute *attr, char *buf)
 427 {
 428         struct gendisk *disk = dev_to_disk(dev);
 429         int err;
 430
 431         /* sysfs gives us a PAGE_SIZE buffer */
 432         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 433
 434         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 435         err = virtblk_get_id(disk, buf);
 436         if (!err)
 437                 return strlen(buf);
 438
 439         if (err == -EIO) /* Unsupported? Make it empty. */
 440                 return 0;
 441
 442         return err;
 443 }
 444
 445 static DEVICE_ATTR_RO(serial);
 446
 447 /* The queue's logical block size must be set before calling this */
 448 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
 449 {
 450         struct virtio_device *vdev = vblk->vdev;
 451         struct request_queue *q = vblk->disk->queue;
 452         char cap_str_2[10], cap_str_10[10];
 453         unsigned long long nblocks;
 454         u64 capacity;
 455
 456         /* Host must always specify the capacity. */
 457         virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
 458
 459         /* If capacity is too big, truncate with warning. */
 460         if ((sector_t)capacity != capacity) {
 461                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 462                          (unsigned long long)capacity);
 463                 capacity = (sector_t)-1;
 464         }
 465
 466         nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
 467
 468         string_get_size(nblocks, queue_logical_block_size(q),
 469                         STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 470         string_get_size(nblocks, queue_logical_block_size(q),
 471                         STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 472
 473         dev_notice(&vdev->dev,
 474                    "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
 475                    vblk->disk->disk_name,
 476                    resize ? "new size: " : "",
 477                    nblocks,
 478                    queue_logical_block_size(q),
 479                    cap_str_10,
 480                    cap_str_2);
 481
 482         set_capacity(vblk->disk, capacity);
 483 }
 484
 485 static void virtblk_config_changed_work(struct work_struct *work)
 486 {
 487         struct virtio_blk *vblk =
 488                 container_of(work, struct virtio_blk, config_work);
 489         char *envp[] = { "RESIZE=1", NULL };
 490
 491         virtblk_update_capacity(vblk, true);
 492         revalidate_disk(vblk->disk);
 493         kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
 494 }
 495
 496 static void virtblk_config_changed(struct virtio_device *vdev)
 497 {
 498         struct virtio_blk *vblk = vdev->priv;
 499
 500         queue_work(virtblk_wq, &vblk->config_work);
 501 }
 502
 503 static int init_vq(struct virtio_blk *vblk)
 504 {
 505         int err;
 506         int i;
 507         vq_callback_t **callbacks;
 508         const char **names;
 509         struct virtqueue **vqs;
 510         unsigned short num_vqs;
 511         struct virtio_device *vdev = vblk->vdev;
 512         struct irq_affinity desc = { 0, };
 513
 514         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
 515                                    struct virtio_blk_config, num_queues,
 516                                    &num_vqs);
 517         if (err)
 518                 num_vqs = 1;
 519
 520         num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
 521
 522         vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
 523         if (!vblk->vqs)
 524                 return -ENOMEM;
 525
 526         names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
 527         callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
 528         vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
 529         if (!names || !callbacks || !vqs) {
 530                 err = -ENOMEM;
 531                 goto out;
 532         }
 533
 534         for (i = 0; i < num_vqs; i++) {
 535                 callbacks[i] = virtblk_done;
 536                 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
 537                 names[i] = vblk->vqs[i].name;
 538         }
 539
 540         /* Discover virtqueues and write information to configuration.  */
 541         err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
 542         if (err)
 543                 goto out;
 544
 545         for (i = 0; i < num_vqs; i++) {
 546                 spin_lock_init(&vblk->vqs[i].lock);
 547                 vblk->vqs[i].vq = vqs[i];
 548         }
 549         vblk->num_vqs = num_vqs;
 550
 551 out:
 552         kfree(vqs);
 553         kfree(callbacks);
 554         kfree(names);
 555         if (err)
 556                 kfree(vblk->vqs);
 557         return err;
 558 }
 559
 560 /*
 561  * Legacy naming scheme used for virtio devices.  We are stuck with it for
 562  * virtio blk but don't ever use it for any new driver.
 563  */
 564 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
 565 {
 566         const int base = 'z' - 'a' + 1;
 567         char *begin = buf + strlen(prefix);
 568         char *end = buf + buflen;
 569         char *p;
 570         int unit;
 571
 572         p = end - 1;
 573         *p = '\0';
 574         unit = base;
 575         do {
 576                 if (p == begin)
 577                         return -EINVAL;
 578                 *--p = 'a' + (index % unit);
 579                 index = (index / unit) - 1;
 580         } while (index >= 0);
 581
 582         memmove(begin, p, end - p);
 583         memcpy(buf, prefix, strlen(prefix));
 584
 585         return 0;
 586 }
 587
 588 static int virtblk_get_cache_mode(struct virtio_device *vdev)
 589 {
 590         u8 writeback;
 591         int err;
 592
 593         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
 594                                    struct virtio_blk_config, wce,
 595                                    &writeback);
 596
 597         /*
 598          * If WCE is not configurable and flush is not available,
 599          * assume no writeback cache is in use.
 600          */
 601         if (err)
 602                 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
 603
 604         return writeback;
 605 }
 606
 607 static void virtblk_update_cache_mode(struct virtio_device *vdev)
 608 {
 609         u8 writeback = virtblk_get_cache_mode(vdev);
 610         struct virtio_blk *vblk = vdev->priv;
 611
 612         blk_queue_write_cache(vblk->disk->queue, writeback, false);
 613         revalidate_disk(vblk->disk);
 614 }
 615
 616 static const char *const virtblk_cache_types[] = {
 617         "write through", "write back"
 618 };
 619
 620 static ssize_t
 621 cache_type_store(struct device *dev, struct device_attribute *attr,
 622                  const char *buf, size_t count)
 623 {
 624         struct gendisk *disk = dev_to_disk(dev);
 625         struct virtio_blk *vblk = disk->private_data;
 626         struct virtio_device *vdev = vblk->vdev;
 627         int i;
 628
 629         BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
 630         i = sysfs_match_string(virtblk_cache_types, buf);
 631         if (i < 0)
 632                 return i;
 633
 634         virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
 635         virtblk_update_cache_mode(vdev);
 636         return count;
 637 }
 638
 639 static ssize_t
 640 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
 641 {
 642         struct gendisk *disk = dev_to_disk(dev);
 643         struct virtio_blk *vblk = disk->private_data;
 644         u8 writeback = virtblk_get_cache_mode(vblk->vdev);
 645
 646         BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
 647         return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
 648 }
 649
 650 static DEVICE_ATTR_RW(cache_type);
 651
 652 static struct attribute *virtblk_attrs[] = {
 653         &dev_attr_serial.attr,
 654         &dev_attr_cache_type.attr,
 655         NULL,
 656 };
 657
 658 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
 659                 struct attribute *a, int n)
 660 {
 661         struct device *dev = container_of(kobj, struct device, kobj);
 662         struct gendisk *disk = dev_to_disk(dev);
 663         struct virtio_blk *vblk = disk->private_data;
 664         struct virtio_device *vdev = vblk->vdev;
 665
 666         if (a == &dev_attr_cache_type.attr &&
 667             !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
 668                 return S_IRUGO;
 669
 670         return a->mode;
 671 }
 672
 673 static const struct attribute_group virtblk_attr_group = {
 674         .attrs = virtblk_attrs,
 675         .is_visible = virtblk_attrs_are_visible,
 676 };
 677
 678 static const struct attribute_group *virtblk_attr_groups[] = {
 679         &virtblk_attr_group,
 680         NULL,
 681 };
 682
 683 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
 684                 unsigned int hctx_idx, unsigned int numa_node)
 685 {
 686         struct virtio_blk *vblk = set->driver_data;
 687         struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
 688
 689 #ifdef CONFIG_VIRTIO_BLK_SCSI
 690         vbr->sreq.sense = vbr->sense;
 691 #endif
 692         sg_init_table(vbr->sg, vblk->sg_elems);
 693         return 0;
 694 }
 695
 696 static int virtblk_map_queues(struct blk_mq_tag_set *set)
 697 {
 698         struct virtio_blk *vblk = set->driver_data;
 699
 700         return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
 701                                         vblk->vdev, 0);
 702 }
 703
 704 #ifdef CONFIG_VIRTIO_BLK_SCSI
 705 static void virtblk_initialize_rq(struct request *req)
 706 {
 707         struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 708
 709         scsi_req_init(&vbr->sreq);
 710 }
 711 #endif
 712
 713 static const struct blk_mq_ops virtio_mq_ops = {
 714         .queue_rq       = virtio_queue_rq,
 715         .commit_rqs     = virtio_commit_rqs,
 716         .complete       = virtblk_request_done,
 717         .init_request   = virtblk_init_request,
 718 #ifdef CONFIG_VIRTIO_BLK_SCSI
 719         .initialize_rq_fn = virtblk_initialize_rq,
 720 #endif
 721         .map_queues     = virtblk_map_queues,
 722 };
 723
 724 static unsigned int virtblk_queue_depth;
 725 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
 726
 727 static int virtblk_probe(struct virtio_device *vdev)
 728 {
 729         struct virtio_blk *vblk;
 730         struct request_queue *q;
 731         int err, index;
 732
 733         u32 v, blk_size, max_size, sg_elems, opt_io_size;
 734         u16 min_io_size;
 735         u8 physical_block_exp, alignment_offset;
 736
 737         if (!vdev->config->get) {
 738                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
 739                         __func__);
 740                 return -EINVAL;
 741         }
 742
 743         err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
 744                              GFP_KERNEL);
 745         if (err < 0)
 746                 goto out;
 747         index = err;
 748
 749         /* We need to know how many segments before we allocate. */
 750         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
 751                                    struct virtio_blk_config, seg_max,
 752                                    &sg_elems);
 753
 754         /* We need at least one SG element, whatever they say. */
 755         if (err || !sg_elems)
 756                 sg_elems = 1;
 757
 758         /* We need an extra sg elements at head and tail. */
 759         sg_elems += 2;
 760         vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
 761         if (!vblk) {
 762                 err = -ENOMEM;
 763                 goto out_free_index;
 764         }
 765
 766         vblk->vdev = vdev;
 767         vblk->sg_elems = sg_elems;
 768
 769         INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
 770
 771         err = init_vq(vblk);
 772         if (err)
 773                 goto out_free_vblk;
 774
 775         /* FIXME: How many partitions?  How long is a piece of string? */
 776         vblk->disk = alloc_disk(1 << PART_BITS);
 777         if (!vblk->disk) {
 778                 err = -ENOMEM;
 779                 goto out_free_vq;
 780         }
 781
 782         /* Default queue sizing is to fill the ring. */
 783         if (!virtblk_queue_depth) {
 784                 virtblk_queue_depth = vblk->vqs[0].vq->num_free;
 785                 /* ... but without indirect descs, we use 2 descs per req */
 786                 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
 787                         virtblk_queue_depth /= 2;
 788         }
 789
 790         memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
 791         vblk->tag_set.ops = &virtio_mq_ops;
 792         vblk->tag_set.queue_depth = virtblk_queue_depth;
 793         vblk->tag_set.numa_node = NUMA_NO_NODE;
 794         vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 795         vblk->tag_set.cmd_size =
 796                 sizeof(struct virtblk_req) +
 797                 sizeof(struct scatterlist) * sg_elems;
 798         vblk->tag_set.driver_data = vblk;
 799         vblk->tag_set.nr_hw_queues = vblk->num_vqs;
 800
 801         err = blk_mq_alloc_tag_set(&vblk->tag_set);
 802         if (err)
 803                 goto out_put_disk;
 804
 805         q = blk_mq_init_queue(&vblk->tag_set);
 806         if (IS_ERR(q)) {
 807                 err = -ENOMEM;
 808                 goto out_free_tags;
 809         }
 810         vblk->disk->queue = q;
 811
 812         q->queuedata = vblk;
 813
 814         virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
 815
 816         vblk->disk->major = major;
 817         vblk->disk->first_minor = index_to_minor(index);
 818         vblk->disk->private_data = vblk;
 819         vblk->disk->fops = &virtblk_fops;
 820         vblk->disk->flags |= GENHD_FL_EXT_DEVT;
 821         vblk->index = index;
 822
 823         /* configure queue flush support */
 824         virtblk_update_cache_mode(vdev);
 825
 826         /* If disk is read-only in the host, the guest should obey */
 827         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 828                 set_disk_ro(vblk->disk, 1);
 829
 830         /* We can handle whatever the host told us to handle. */
 831         blk_queue_max_segments(q, vblk->sg_elems-2);
 832
 833         /* No real sector limit. */
 834         blk_queue_max_hw_sectors(q, -1U);
 835
 836         max_size = virtio_max_dma_size(vdev);
 837
 838         /* Host can optionally specify maximum segment size and number of
 839          * segments. */
 840         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
 841                                    struct virtio_blk_config, size_max, &v);
 842         if (!err)
 843                 max_size = min(max_size, v);
 844
 845         blk_queue_max_segment_size(q, max_size);
 846
 847         /* Host can optionally specify the block size of the device */
 848         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
 849                                    struct virtio_blk_config, blk_size,
 850                                    &blk_size);
 851         if (!err)
 852                 blk_queue_logical_block_size(q, blk_size);
 853         else
 854                 blk_size = queue_logical_block_size(q);
 855
 856         /* Use topology information if available */
 857         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 858                                    struct virtio_blk_config, physical_block_exp,
 859                                    &physical_block_exp);
 860         if (!err && physical_block_exp)
 861                 blk_queue_physical_block_size(q,
 862                                 blk_size * (1 << physical_block_exp));
 863
 864         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 865                                    struct virtio_blk_config, alignment_offset,
 866                                    &alignment_offset);
 867         if (!err && alignment_offset)
 868                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 869
 870         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 871                                    struct virtio_blk_config, min_io_size,
 872                                    &min_io_size);
 873         if (!err && min_io_size)
 874                 blk_queue_io_min(q, blk_size * min_io_size);
 875
 876         err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
 877                                    struct virtio_blk_config, opt_io_size,
 878                                    &opt_io_size);
 879         if (!err && opt_io_size)
 880                 blk_queue_io_opt(q, blk_size * opt_io_size);
 881
 882         if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
 883                 q->limits.discard_granularity = blk_size;
 884
 885                 virtio_cread(vdev, struct virtio_blk_config,
 886                              discard_sector_alignment, &v);
 887                 q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
 888
 889                 virtio_cread(vdev, struct virtio_blk_config,
 890                              max_discard_sectors, &v);
 891                 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
 892
 893                 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
 894                              &v);
 895                 blk_queue_max_discard_segments(q,
 896                                                min_not_zero(v,
 897                                                             MAX_DISCARD_SEGMENTS));
 898
 899                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 900         }
 901
 902         if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
 903                 virtio_cread(vdev, struct virtio_blk_config,
 904                              max_write_zeroes_sectors, &v);
 905                 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
 906         }
 907
 908         virtblk_update_capacity(vblk, false);
 909         virtio_device_ready(vdev);
 910
 911         device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
 912         return 0;
 913
 914 out_free_tags:
 915         blk_mq_free_tag_set(&vblk->tag_set);
 916 out_put_disk:
 917         put_disk(vblk->disk);
 918 out_free_vq:
 919         vdev->config->del_vqs(vdev);
 920 out_free_vblk:
 921         kfree(vblk);
 922 out_free_index:
 923         ida_simple_remove(&vd_index_ida, index);
 924 out:
 925         return err;
 926 }
 927
 928 static void virtblk_remove(struct virtio_device *vdev)
 929 {
 930         struct virtio_blk *vblk = vdev->priv;
 931         int index = vblk->index;
 932         int refc;
 933
 934         /* Make sure no work handler is accessing the device. */
 935         flush_work(&vblk->config_work);
 936
 937         del_gendisk(vblk->disk);
 938         blk_cleanup_queue(vblk->disk->queue);
 939
 940         blk_mq_free_tag_set(&vblk->tag_set);
 941
 942         /* Stop all the virtqueues. */
 943         vdev->config->reset(vdev);
 944
 945         refc = kref_read(&disk_to_dev(vblk->disk)->kobj.kref);
 946         put_disk(vblk->disk);
 947         vdev->config->del_vqs(vdev);
 948         kfree(vblk->vqs);
 949         kfree(vblk);
 950
 951         /* Only free device id if we don't have any users */
 952         if (refc == 1)
 953                 ida_simple_remove(&vd_index_ida, index);
 954 }
 955
 956 #ifdef CONFIG_PM_SLEEP
 957 static int virtblk_freeze(struct virtio_device *vdev)
 958 {
 959         struct virtio_blk *vblk = vdev->priv;
 960
 961         /* Ensure we don't receive any more interrupts */
 962         vdev->config->reset(vdev);
 963
 964         /* Make sure no work handler is accessing the device. */
 965         flush_work(&vblk->config_work);
 966
 967         blk_mq_quiesce_queue(vblk->disk->queue);
 968
 969         vdev->config->del_vqs(vdev);
 970         return 0;
 971 }
 972
 973 static int virtblk_restore(struct virtio_device *vdev)
 974 {
 975         struct virtio_blk *vblk = vdev->priv;
 976         int ret;
 977
 978         ret = init_vq(vdev->priv);
 979         if (ret)
 980                 return ret;
 981
 982         virtio_device_ready(vdev);
 983
 984         blk_mq_unquiesce_queue(vblk->disk->queue);
 985         return 0;
 986 }
 987 #endif
 988
 989 static const struct virtio_device_id id_table[] = {
 990         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 991         { 0 },
 992 };
 993
 994 static unsigned int features_legacy[] = {
 995         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
 996         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 997 #ifdef CONFIG_VIRTIO_BLK_SCSI
 998         VIRTIO_BLK_F_SCSI,
 999 #endif
1000         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1001         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1002 }
1003 ;
1004 static unsigned int features[] = {
1005         VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1006         VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1007         VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1008         VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1009 };
1010
1011 static struct virtio_driver virtio_blk = {
1012         .feature_table                  = features,
1013         .feature_table_size             = ARRAY_SIZE(features),
1014         .feature_table_legacy           = features_legacy,
1015         .feature_table_size_legacy      = ARRAY_SIZE(features_legacy),
1016         .driver.name                    = KBUILD_MODNAME,
1017         .driver.owner                   = THIS_MODULE,
1018         .id_table                       = id_table,
1019         .probe                          = virtblk_probe,
1020         .remove                         = virtblk_remove,
1021         .config_changed                 = virtblk_config_changed,
1022 #ifdef CONFIG_PM_SLEEP
1023         .freeze                         = virtblk_freeze,
1024         .restore                        = virtblk_restore,
1025 #endif
1026 };
1027
1028 static int __init init(void)
1029 {
1030         int error;
1031
1032         virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1033         if (!virtblk_wq)
1034                 return -ENOMEM;
1035
1036         major = register_blkdev(0, "virtblk");
1037         if (major < 0) {
1038                 error = major;
1039                 goto out_destroy_workqueue;
1040         }
1041
1042         error = register_virtio_driver(&virtio_blk);
1043         if (error)
1044                 goto out_unregister_blkdev;
1045         return 0;
1046
1047 out_unregister_blkdev:
1048         unregister_blkdev(major, "virtblk");
1049 out_destroy_workqueue:
1050         destroy_workqueue(virtblk_wq);
1051         return error;
1052 }
1053
1054 static void __exit fini(void)
1055 {
1056         unregister_virtio_driver(&virtio_blk);
1057         unregister_blkdev(major, "virtblk");
1058         destroy_workqueue(virtblk_wq);
1059 }
1060 module_init(init);
1061 module_exit(fini);
1062
1063 MODULE_DEVICE_TABLE(virtio, id_table);
1064 MODULE_DESCRIPTION("Virtio block driver");
1065 MODULE_LICENSE("GPL");