hw/virtio-blk.c

   1 /*
   2  * Virtio Block Device
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "qemu-common.h"
  15 #include "qemu-error.h"
  16 #include "trace.h"
  17 #include "hw/block-common.h"
  18 #include "blockdev.h"
  19 #include "virtio-blk.h"
  20 #include "scsi-defs.h"
  21 #ifdef __linux__
  22 # include <scsi/sg.h>
  23 #endif
  24
  25 typedef struct VirtIOBlock
  26 {
  27     VirtIODevice vdev;
  28     BlockDriverState *bs;
  29     VirtQueue *vq;
  30     void *rq;
  31     QEMUBH *bh;
  32     BlockConf *conf;
  33     VirtIOBlkConf *blk;
  34     unsigned short sector_mask;
  35     DeviceState *qdev;
  36 } VirtIOBlock;
  37
  38 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
  39 {
  40     return (VirtIOBlock *)vdev;
  41 }
  42
  43 typedef struct VirtIOBlockReq
  44 {
  45     VirtIOBlock *dev;
  46     VirtQueueElement elem;
  47     struct virtio_blk_inhdr *in;
  48     struct virtio_blk_outhdr *out;
  49     struct virtio_scsi_inhdr *scsi;
  50     QEMUIOVector qiov;
  51     struct VirtIOBlockReq *next;
  52     BlockAcctCookie acct;
  53 } VirtIOBlockReq;
  54
  55 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
  56 {
  57     VirtIOBlock *s = req->dev;
  58
  59     trace_virtio_blk_req_complete(req, status);
  60
  61     stb_p(&req->in->status, status);
  62     virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
  63     virtio_notify(&s->vdev, s->vq);
  64 }
  65
  66 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
  67     bool is_read)
  68 {
  69     BlockErrorAction action = bdrv_get_error_action(req->dev->bs, is_read, error);
  70     VirtIOBlock *s = req->dev;
  71
  72     if (action == BDRV_ACTION_STOP) {
  73         req->next = s->rq;
  74         s->rq = req;
  75     } else if (action == BDRV_ACTION_REPORT) {
  76         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
  77         bdrv_acct_done(s->bs, &req->acct);
  78         g_free(req);
  79     }
  80
  81     bdrv_error_action(s->bs, action, is_read, error);
  82     return action != BDRV_ACTION_IGNORE;
  83 }
  84
  85 static void virtio_blk_rw_complete(void *opaque, int ret)
  86 {
  87     VirtIOBlockReq *req = opaque;
  88
  89     trace_virtio_blk_rw_complete(req, ret);
  90
  91     if (ret) {
  92         bool is_read = !(ldl_p(&req->out->type) & VIRTIO_BLK_T_OUT);
  93         if (virtio_blk_handle_rw_error(req, -ret, is_read))
  94             return;
  95     }
  96
  97     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
  98     bdrv_acct_done(req->dev->bs, &req->acct);
  99     g_free(req);
 100 }
 101
 102 static void virtio_blk_flush_complete(void *opaque, int ret)
 103 {
 104     VirtIOBlockReq *req = opaque;
 105
 106     if (ret) {
 107         if (virtio_blk_handle_rw_error(req, -ret, 0)) {
 108             return;
 109         }
 110     }
 111
 112     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 113     bdrv_acct_done(req->dev->bs, &req->acct);
 114     g_free(req);
 115 }
 116
 117 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
 118 {
 119     VirtIOBlockReq *req = g_malloc(sizeof(*req));
 120     req->dev = s;
 121     req->qiov.size = 0;
 122     req->next = NULL;
 123     return req;
 124 }
 125
 126 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 127 {
 128     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 129
 130     if (req != NULL) {
 131         if (!virtqueue_pop(s->vq, &req->elem)) {
 132             g_free(req);
 133             return NULL;
 134         }
 135     }
 136
 137     return req;
 138 }
 139
 140 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 141 {
 142 #ifdef __linux__
 143     int ret;
 144     int i;
 145 #endif
 146     int status = VIRTIO_BLK_S_OK;
 147
 148     /*
 149      * We require at least one output segment each for the virtio_blk_outhdr
 150      * and the SCSI command block.
 151      *
 152      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 153      * and the sense buffer pointer in the input segments.
 154      */
 155     if (req->elem.out_num < 2 || req->elem.in_num < 3) {
 156         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 157         g_free(req);
 158         return;
 159     }
 160
 161     /*
 162      * The scsi inhdr is placed in the second-to-last input segment, just
 163      * before the regular inhdr.
 164      */
 165     req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
 166
 167     if (!req->dev->blk->scsi) {
 168         status = VIRTIO_BLK_S_UNSUPP;
 169         goto fail;
 170     }
 171
 172     /*
 173      * No support for bidirection commands yet.
 174      */
 175     if (req->elem.out_num > 2 && req->elem.in_num > 3) {
 176         status = VIRTIO_BLK_S_UNSUPP;
 177         goto fail;
 178     }
 179
 180 #ifdef __linux__
 181     struct sg_io_hdr hdr;
 182     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 183     hdr.interface_id = 'S';
 184     hdr.cmd_len = req->elem.out_sg[1].iov_len;
 185     hdr.cmdp = req->elem.out_sg[1].iov_base;
 186     hdr.dxfer_len = 0;
 187
 188     if (req->elem.out_num > 2) {
 189         /*
 190          * If there are more than the minimally required 2 output segments
 191          * there is write payload starting from the third iovec.
 192          */
 193         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 194         hdr.iovec_count = req->elem.out_num - 2;
 195
 196         for (i = 0; i < hdr.iovec_count; i++)
 197             hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
 198
 199         hdr.dxferp = req->elem.out_sg + 2;
 200
 201     } else if (req->elem.in_num > 3) {
 202         /*
 203          * If we have more than 3 input segments the guest wants to actually
 204          * read data.
 205          */
 206         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 207         hdr.iovec_count = req->elem.in_num - 3;
 208         for (i = 0; i < hdr.iovec_count; i++)
 209             hdr.dxfer_len += req->elem.in_sg[i].iov_len;
 210
 211         hdr.dxferp = req->elem.in_sg;
 212     } else {
 213         /*
 214          * Some SCSI commands don't actually transfer any data.
 215          */
 216         hdr.dxfer_direction = SG_DXFER_NONE;
 217     }
 218
 219     hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
 220     hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
 221
 222     ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
 223     if (ret) {
 224         status = VIRTIO_BLK_S_UNSUPP;
 225         goto fail;
 226     }
 227
 228     /*
 229      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
 230      * clear the masked_status field [hence status gets cleared too, see
 231      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
 232      * status has occurred.  However they do set DRIVER_SENSE in driver_status
 233      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
 234      */
 235     if (hdr.status == 0 && hdr.sb_len_wr > 0) {
 236         hdr.status = CHECK_CONDITION;
 237     }
 238
 239     stl_p(&req->scsi->errors,
 240           hdr.status | (hdr.msg_status << 8) |
 241           (hdr.host_status << 16) | (hdr.driver_status << 24));
 242     stl_p(&req->scsi->residual, hdr.resid);
 243     stl_p(&req->scsi->sense_len, hdr.sb_len_wr);
 244     stl_p(&req->scsi->data_len, hdr.dxfer_len);
 245
 246     virtio_blk_req_complete(req, status);
 247     g_free(req);
 248     return;
 249 #else
 250     abort();
 251 #endif
 252
 253 fail:
 254     /* Just put anything nonzero so that the ioctl fails in the guest.  */
 255     stl_p(&req->scsi->errors, 255);
 256     virtio_blk_req_complete(req, status);
 257     g_free(req);
 258 }
 259
 260 typedef struct MultiReqBuffer {
 261     BlockRequest        blkreq[32];
 262     unsigned int        num_writes;
 263 } MultiReqBuffer;
 264
 265 static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
 266 {
 267     int i, ret;
 268
 269     if (!mrb->num_writes) {
 270         return;
 271     }
 272
 273     ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
 274     if (ret != 0) {
 275         for (i = 0; i < mrb->num_writes; i++) {
 276             if (mrb->blkreq[i].error) {
 277                 virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
 278             }
 279         }
 280     }
 281
 282     mrb->num_writes = 0;
 283 }
 284
 285 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 286 {
 287     bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH);
 288
 289     /*
 290      * Make sure all outstanding writes are posted to the backing device.
 291      */
 292     virtio_submit_multiwrite(req->dev->bs, mrb);
 293     bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 294 }
 295
 296 static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 297 {
 298     BlockRequest *blkreq;
 299     uint64_t sector;
 300
 301     sector = ldq_p(&req->out->sector);
 302
 303     bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE);
 304
 305     trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
 306
 307     if (sector & req->dev->sector_mask) {
 308         virtio_blk_rw_complete(req, -EIO);
 309         return;
 310     }
 311     if (req->qiov.size % req->dev->conf->logical_block_size) {
 312         virtio_blk_rw_complete(req, -EIO);
 313         return;
 314     }
 315
 316     if (mrb->num_writes == 32) {
 317         virtio_submit_multiwrite(req->dev->bs, mrb);
 318     }
 319
 320     blkreq = &mrb->blkreq[mrb->num_writes];
 321     blkreq->sector = sector;
 322     blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
 323     blkreq->qiov = &req->qiov;
 324     blkreq->cb = virtio_blk_rw_complete;
 325     blkreq->opaque = req;
 326     blkreq->error = 0;
 327
 328     mrb->num_writes++;
 329 }
 330
 331 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 332 {
 333     uint64_t sector;
 334
 335     sector = ldq_p(&req->out->sector);
 336
 337     bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ);
 338
 339     trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512);
 340
 341     if (sector & req->dev->sector_mask) {
 342         virtio_blk_rw_complete(req, -EIO);
 343         return;
 344     }
 345     if (req->qiov.size % req->dev->conf->logical_block_size) {
 346         virtio_blk_rw_complete(req, -EIO);
 347         return;
 348     }
 349     bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
 350                    req->qiov.size / BDRV_SECTOR_SIZE,
 351                    virtio_blk_rw_complete, req);
 352 }
 353
 354 static void virtio_blk_handle_request(VirtIOBlockReq *req,
 355     MultiReqBuffer *mrb)
 356 {
 357     uint32_t type;
 358
 359     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 360         error_report("virtio-blk missing headers");
 361         exit(1);
 362     }
 363
 364     if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 365         req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 366         error_report("virtio-blk header not in correct element");
 367         exit(1);
 368     }
 369
 370     req->out = (void *)req->elem.out_sg[0].iov_base;
 371     req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 372
 373     type = ldl_p(&req->out->type);
 374
 375     if (type & VIRTIO_BLK_T_FLUSH) {
 376         virtio_blk_handle_flush(req, mrb);
 377     } else if (type & VIRTIO_BLK_T_SCSI_CMD) {
 378         virtio_blk_handle_scsi(req);
 379     } else if (type & VIRTIO_BLK_T_GET_ID) {
 380         VirtIOBlock *s = req->dev;
 381
 382         /*
 383          * NB: per existing s/n string convention the string is
 384          * terminated by '\0' only when shorter than buffer.
 385          */
 386         strncpy(req->elem.in_sg[0].iov_base,
 387                 s->blk->serial ? s->blk->serial : "",
 388                 MIN(req->elem.in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES));
 389         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 390         g_free(req);
 391     } else if (type & VIRTIO_BLK_T_OUT) {
 392         qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 393                                  req->elem.out_num - 1);
 394         virtio_blk_handle_write(req, mrb);
 395     } else {
 396         qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 397                                  req->elem.in_num - 1);
 398         virtio_blk_handle_read(req);
 399     }
 400 }
 401
 402 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 403 {
 404     VirtIOBlock *s = to_virtio_blk(vdev);
 405     VirtIOBlockReq *req;
 406     MultiReqBuffer mrb = {
 407         .num_writes = 0,
 408     };
 409
 410     while ((req = virtio_blk_get_request(s))) {
 411         virtio_blk_handle_request(req, &mrb);
 412     }
 413
 414     virtio_submit_multiwrite(s->bs, &mrb);
 415
 416     /*
 417      * FIXME: Want to check for completions before returning to guest mode,
 418      * so cached reads and writes are reported as quickly as possible. But
 419      * that should be done in the generic block layer.
 420      */
 421 }
 422
 423 static void virtio_blk_dma_restart_bh(void *opaque)
 424 {
 425     VirtIOBlock *s = opaque;
 426     VirtIOBlockReq *req = s->rq;
 427     MultiReqBuffer mrb = {
 428         .num_writes = 0,
 429     };
 430
 431     qemu_bh_delete(s->bh);
 432     s->bh = NULL;
 433
 434     s->rq = NULL;
 435
 436     while (req) {
 437         virtio_blk_handle_request(req, &mrb);
 438         req = req->next;
 439     }
 440
 441     virtio_submit_multiwrite(s->bs, &mrb);
 442 }
 443
 444 static void virtio_blk_dma_restart_cb(void *opaque, int running,
 445                                       RunState state)
 446 {
 447     VirtIOBlock *s = opaque;
 448
 449     if (!running)
 450         return;
 451
 452     if (!s->bh) {
 453         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 454         qemu_bh_schedule(s->bh);
 455     }
 456 }
 457
 458 static void virtio_blk_reset(VirtIODevice *vdev)
 459 {
 460     /*
 461      * This should cancel pending requests, but can't do nicely until there
 462      * are per-device request lists.
 463      */
 464     bdrv_drain_all();
 465 }
 466
 467 /* coalesce internal state, copy to pci i/o region 0
 468  */
 469 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 470 {
 471     VirtIOBlock *s = to_virtio_blk(vdev);
 472     struct virtio_blk_config blkcfg;
 473     uint64_t capacity;
 474     int blk_size = s->conf->logical_block_size;
 475
 476     bdrv_get_geometry(s->bs, &capacity);
 477     memset(&blkcfg, 0, sizeof(blkcfg));
 478     stq_raw(&blkcfg.capacity, capacity);
 479     stl_raw(&blkcfg.seg_max, 128 - 2);
 480     stw_raw(&blkcfg.cylinders, s->conf->cyls);
 481     stl_raw(&blkcfg.blk_size, blk_size);
 482     stw_raw(&blkcfg.min_io_size, s->conf->min_io_size / blk_size);
 483     stw_raw(&blkcfg.opt_io_size, s->conf->opt_io_size / blk_size);
 484     blkcfg.heads = s->conf->heads;
 485     /*
 486      * We must ensure that the block device capacity is a multiple of
 487      * the logical block size. If that is not the case, lets use
 488      * sector_mask to adopt the geometry to have a correct picture.
 489      * For those devices where the capacity is ok for the given geometry
 490      * we dont touch the sector value of the geometry, since some devices
 491      * (like s390 dasd) need a specific value. Here the capacity is already
 492      * cyls*heads*secs*blk_size and the sector value is not block size
 493      * divided by 512 - instead it is the amount of blk_size blocks
 494      * per track (cylinder).
 495      */
 496     if (bdrv_getlength(s->bs) /  s->conf->heads / s->conf->secs % blk_size) {
 497         blkcfg.sectors = s->conf->secs & ~s->sector_mask;
 498     } else {
 499         blkcfg.sectors = s->conf->secs;
 500     }
 501     blkcfg.size_max = 0;
 502     blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
 503     blkcfg.alignment_offset = 0;
 504     blkcfg.wce = bdrv_enable_write_cache(s->bs);
 505     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
 506 }
 507
 508 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
 509 {
 510     VirtIOBlock *s = to_virtio_blk(vdev);
 511     struct virtio_blk_config blkcfg;
 512
 513     memcpy(&blkcfg, config, sizeof(blkcfg));
 514     bdrv_set_enable_write_cache(s->bs, blkcfg.wce != 0);
 515 }
 516
 517 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 518 {
 519     VirtIOBlock *s = to_virtio_blk(vdev);
 520
 521     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 522     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 523     features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
 524     features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
 525     features |= (1 << VIRTIO_BLK_F_SCSI);
 526
 527     if (bdrv_enable_write_cache(s->bs))
 528         features |= (1 << VIRTIO_BLK_F_WCE);
 529
 530     if (bdrv_is_read_only(s->bs))
 531         features |= 1 << VIRTIO_BLK_F_RO;
 532
 533     return features;
 534 }
 535
 536 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
 537 {
 538     VirtIOBlock *s = to_virtio_blk(vdev);
 539     uint32_t features;
 540
 541     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 542         return;
 543     }
 544
 545     features = vdev->guest_features;
 546     bdrv_set_enable_write_cache(s->bs, !!(features & (1 << VIRTIO_BLK_F_WCE)));
 547 }
 548
 549 static void virtio_blk_save(QEMUFile *f, void *opaque)
 550 {
 551     VirtIOBlock *s = opaque;
 552     VirtIOBlockReq *req = s->rq;
 553
 554     virtio_save(&s->vdev, f);
 555
 556     while (req) {
 557         qemu_put_sbyte(f, 1);
 558         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 559         req = req->next;
 560     }
 561     qemu_put_sbyte(f, 0);
 562 }
 563
 564 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 565 {
 566     VirtIOBlock *s = opaque;
 567     int ret;
 568
 569     if (version_id != 2)
 570         return -EINVAL;
 571
 572     ret = virtio_load(&s->vdev, f);
 573     if (ret) {
 574         return ret;
 575     }
 576
 577     while (qemu_get_sbyte(f)) {
 578         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 579         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 580         req->next = s->rq;
 581         s->rq = req;
 582
 583         virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr,
 584             req->elem.in_num, 1);
 585         virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr,
 586             req->elem.out_num, 0);
 587     }
 588
 589     return 0;
 590 }
 591
 592 static void virtio_blk_resize(void *opaque)
 593 {
 594     VirtIOBlock *s = opaque;
 595
 596     virtio_notify_config(&s->vdev);
 597 }
 598
 599 static const BlockDevOps virtio_block_ops = {
 600     .resize_cb = virtio_blk_resize,
 601 };
 602
 603 VirtIODevice *virtio_blk_init(DeviceState *dev, VirtIOBlkConf *blk)
 604 {
 605     VirtIOBlock *s;
 606     static int virtio_blk_id;
 607
 608     if (!blk->conf.bs) {
 609         error_report("drive property not set");
 610         return NULL;
 611     }
 612     if (!bdrv_is_inserted(blk->conf.bs)) {
 613         error_report("Device needs media, but drive is empty");
 614         return NULL;
 615     }
 616
 617     blkconf_serial(&blk->conf, &blk->serial);
 618     if (blkconf_geometry(&blk->conf, NULL, 65535, 255, 255) < 0) {
 619         return NULL;
 620     }
 621
 622     s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
 623                                           sizeof(struct virtio_blk_config),
 624                                           sizeof(VirtIOBlock));
 625
 626     s->vdev.get_config = virtio_blk_update_config;
 627     s->vdev.set_config = virtio_blk_set_config;
 628     s->vdev.get_features = virtio_blk_get_features;
 629     s->vdev.set_status = virtio_blk_set_status;
 630     s->vdev.reset = virtio_blk_reset;
 631     s->bs = blk->conf.bs;
 632     s->conf = &blk->conf;
 633     s->blk = blk;
 634     s->rq = NULL;
 635     s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
 636
 637     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 638
 639     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 640     s->qdev = dev;
 641     register_savevm(dev, "virtio-blk", virtio_blk_id++, 2,
 642                     virtio_blk_save, virtio_blk_load, s);
 643     bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
 644     bdrv_set_buffer_alignment(s->bs, s->conf->logical_block_size);
 645
 646     bdrv_iostatus_enable(s->bs);
 647     add_boot_device_path(s->conf->bootindex, dev, "/disk@0,0");
 648
 649     return &s->vdev;
 650 }
 651
 652 void virtio_blk_exit(VirtIODevice *vdev)
 653 {
 654     VirtIOBlock *s = to_virtio_blk(vdev);
 655     unregister_savevm(s->qdev, "virtio-blk", s);
 656     blockdev_mark_auto_del(s->bs);
 657     virtio_cleanup(vdev);
 658 }