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-report.h"
  16 #include "trace.h"
  17 #include "hw/block-common.h"
  18 #include "sysemu/blockdev.h"
  19 #include "hw/virtio-blk.h"
  20 #include "hw/scsi-defs.h"
  21 #ifdef __linux__
  22 # include <scsi/sg.h>
  23 #endif
  24 #include "hw/virtio-bus.h"
  25
  26 typedef struct VirtIOBlockReq
  27 {
  28     VirtIOBlock *dev;
  29     VirtQueueElement elem;
  30     struct virtio_blk_inhdr *in;
  31     struct virtio_blk_outhdr *out;
  32     struct virtio_scsi_inhdr *scsi;
  33     QEMUIOVector qiov;
  34     struct VirtIOBlockReq *next;
  35     BlockAcctCookie acct;
  36 } VirtIOBlockReq;
  37
  38 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
  39 {
  40     VirtIOBlock *s = req->dev;
  41     VirtIODevice *vdev = VIRTIO_DEVICE(s);
  42
  43     trace_virtio_blk_req_complete(req, status);
  44
  45     stb_p(&req->in->status, status);
  46     virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
  47     virtio_notify(vdev, s->vq);
  48 }
  49
  50 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
  51     bool is_read)
  52 {
  53     BlockErrorAction action = bdrv_get_error_action(req->dev->bs, is_read, error);
  54     VirtIOBlock *s = req->dev;
  55
  56     if (action == BDRV_ACTION_STOP) {
  57         req->next = s->rq;
  58         s->rq = req;
  59     } else if (action == BDRV_ACTION_REPORT) {
  60         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
  61         bdrv_acct_done(s->bs, &req->acct);
  62         g_free(req);
  63     }
  64
  65     bdrv_error_action(s->bs, action, is_read, error);
  66     return action != BDRV_ACTION_IGNORE;
  67 }
  68
  69 static void virtio_blk_rw_complete(void *opaque, int ret)
  70 {
  71     VirtIOBlockReq *req = opaque;
  72
  73     trace_virtio_blk_rw_complete(req, ret);
  74
  75     if (ret) {
  76         bool is_read = !(ldl_p(&req->out->type) & VIRTIO_BLK_T_OUT);
  77         if (virtio_blk_handle_rw_error(req, -ret, is_read))
  78             return;
  79     }
  80
  81     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
  82     bdrv_acct_done(req->dev->bs, &req->acct);
  83     g_free(req);
  84 }
  85
  86 static void virtio_blk_flush_complete(void *opaque, int ret)
  87 {
  88     VirtIOBlockReq *req = opaque;
  89
  90     if (ret) {
  91         if (virtio_blk_handle_rw_error(req, -ret, 0)) {
  92             return;
  93         }
  94     }
  95
  96     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
  97     bdrv_acct_done(req->dev->bs, &req->acct);
  98     g_free(req);
  99 }
 100
 101 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
 102 {
 103     VirtIOBlockReq *req = g_malloc(sizeof(*req));
 104     req->dev = s;
 105     req->qiov.size = 0;
 106     req->next = NULL;
 107     return req;
 108 }
 109
 110 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 111 {
 112     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 113
 114     if (req != NULL) {
 115         if (!virtqueue_pop(s->vq, &req->elem)) {
 116             g_free(req);
 117             return NULL;
 118         }
 119     }
 120
 121     return req;
 122 }
 123
 124 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 125 {
 126 #ifdef __linux__
 127     int ret;
 128     int i;
 129 #endif
 130     int status = VIRTIO_BLK_S_OK;
 131
 132     /*
 133      * We require at least one output segment each for the virtio_blk_outhdr
 134      * and the SCSI command block.
 135      *
 136      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 137      * and the sense buffer pointer in the input segments.
 138      */
 139     if (req->elem.out_num < 2 || req->elem.in_num < 3) {
 140         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 141         g_free(req);
 142         return;
 143     }
 144
 145     /*
 146      * The scsi inhdr is placed in the second-to-last input segment, just
 147      * before the regular inhdr.
 148      */
 149     req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
 150
 151     if (!req->dev->blk.scsi) {
 152         status = VIRTIO_BLK_S_UNSUPP;
 153         goto fail;
 154     }
 155
 156     /*
 157      * No support for bidirection commands yet.
 158      */
 159     if (req->elem.out_num > 2 && req->elem.in_num > 3) {
 160         status = VIRTIO_BLK_S_UNSUPP;
 161         goto fail;
 162     }
 163
 164 #ifdef __linux__
 165     struct sg_io_hdr hdr;
 166     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 167     hdr.interface_id = 'S';
 168     hdr.cmd_len = req->elem.out_sg[1].iov_len;
 169     hdr.cmdp = req->elem.out_sg[1].iov_base;
 170     hdr.dxfer_len = 0;
 171
 172     if (req->elem.out_num > 2) {
 173         /*
 174          * If there are more than the minimally required 2 output segments
 175          * there is write payload starting from the third iovec.
 176          */
 177         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 178         hdr.iovec_count = req->elem.out_num - 2;
 179
 180         for (i = 0; i < hdr.iovec_count; i++)
 181             hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
 182
 183         hdr.dxferp = req->elem.out_sg + 2;
 184
 185     } else if (req->elem.in_num > 3) {
 186         /*
 187          * If we have more than 3 input segments the guest wants to actually
 188          * read data.
 189          */
 190         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 191         hdr.iovec_count = req->elem.in_num - 3;
 192         for (i = 0; i < hdr.iovec_count; i++)
 193             hdr.dxfer_len += req->elem.in_sg[i].iov_len;
 194
 195         hdr.dxferp = req->elem.in_sg;
 196     } else {
 197         /*
 198          * Some SCSI commands don't actually transfer any data.
 199          */
 200         hdr.dxfer_direction = SG_DXFER_NONE;
 201     }
 202
 203     hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
 204     hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
 205
 206     ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
 207     if (ret) {
 208         status = VIRTIO_BLK_S_UNSUPP;
 209         goto fail;
 210     }
 211
 212     /*
 213      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
 214      * clear the masked_status field [hence status gets cleared too, see
 215      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
 216      * status has occurred.  However they do set DRIVER_SENSE in driver_status
 217      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
 218      */
 219     if (hdr.status == 0 && hdr.sb_len_wr > 0) {
 220         hdr.status = CHECK_CONDITION;
 221     }
 222
 223     stl_p(&req->scsi->errors,
 224           hdr.status | (hdr.msg_status << 8) |
 225           (hdr.host_status << 16) | (hdr.driver_status << 24));
 226     stl_p(&req->scsi->residual, hdr.resid);
 227     stl_p(&req->scsi->sense_len, hdr.sb_len_wr);
 228     stl_p(&req->scsi->data_len, hdr.dxfer_len);
 229
 230     virtio_blk_req_complete(req, status);
 231     g_free(req);
 232     return;
 233 #else
 234     abort();
 235 #endif
 236
 237 fail:
 238     /* Just put anything nonzero so that the ioctl fails in the guest.  */
 239     stl_p(&req->scsi->errors, 255);
 240     virtio_blk_req_complete(req, status);
 241     g_free(req);
 242 }
 243
 244 typedef struct MultiReqBuffer {
 245     BlockRequest        blkreq[32];
 246     unsigned int        num_writes;
 247 } MultiReqBuffer;
 248
 249 static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
 250 {
 251     int i, ret;
 252
 253     if (!mrb->num_writes) {
 254         return;
 255     }
 256
 257     ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
 258     if (ret != 0) {
 259         for (i = 0; i < mrb->num_writes; i++) {
 260             if (mrb->blkreq[i].error) {
 261                 virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
 262             }
 263         }
 264     }
 265
 266     mrb->num_writes = 0;
 267 }
 268
 269 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 270 {
 271     bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH);
 272
 273     /*
 274      * Make sure all outstanding writes are posted to the backing device.
 275      */
 276     virtio_submit_multiwrite(req->dev->bs, mrb);
 277     bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 278 }
 279
 280 static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 281 {
 282     BlockRequest *blkreq;
 283     uint64_t sector;
 284
 285     sector = ldq_p(&req->out->sector);
 286
 287     bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE);
 288
 289     trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
 290
 291     if (sector & req->dev->sector_mask) {
 292         virtio_blk_rw_complete(req, -EIO);
 293         return;
 294     }
 295     if (req->qiov.size % req->dev->conf->logical_block_size) {
 296         virtio_blk_rw_complete(req, -EIO);
 297         return;
 298     }
 299
 300     if (mrb->num_writes == 32) {
 301         virtio_submit_multiwrite(req->dev->bs, mrb);
 302     }
 303
 304     blkreq = &mrb->blkreq[mrb->num_writes];
 305     blkreq->sector = sector;
 306     blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
 307     blkreq->qiov = &req->qiov;
 308     blkreq->cb = virtio_blk_rw_complete;
 309     blkreq->opaque = req;
 310     blkreq->error = 0;
 311
 312     mrb->num_writes++;
 313 }
 314
 315 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 316 {
 317     uint64_t sector;
 318
 319     sector = ldq_p(&req->out->sector);
 320
 321     bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ);
 322
 323     trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512);
 324
 325     if (sector & req->dev->sector_mask) {
 326         virtio_blk_rw_complete(req, -EIO);
 327         return;
 328     }
 329     if (req->qiov.size % req->dev->conf->logical_block_size) {
 330         virtio_blk_rw_complete(req, -EIO);
 331         return;
 332     }
 333     bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
 334                    req->qiov.size / BDRV_SECTOR_SIZE,
 335                    virtio_blk_rw_complete, req);
 336 }
 337
 338 static void virtio_blk_handle_request(VirtIOBlockReq *req,
 339     MultiReqBuffer *mrb)
 340 {
 341     uint32_t type;
 342
 343     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 344         error_report("virtio-blk missing headers");
 345         exit(1);
 346     }
 347
 348     if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 349         req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 350         error_report("virtio-blk header not in correct element");
 351         exit(1);
 352     }
 353
 354     req->out = (void *)req->elem.out_sg[0].iov_base;
 355     req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 356
 357     type = ldl_p(&req->out->type);
 358
 359     if (type & VIRTIO_BLK_T_FLUSH) {
 360         virtio_blk_handle_flush(req, mrb);
 361     } else if (type & VIRTIO_BLK_T_SCSI_CMD) {
 362         virtio_blk_handle_scsi(req);
 363     } else if (type & VIRTIO_BLK_T_GET_ID) {
 364         VirtIOBlock *s = req->dev;
 365
 366         /*
 367          * NB: per existing s/n string convention the string is
 368          * terminated by '\0' only when shorter than buffer.
 369          */
 370         strncpy(req->elem.in_sg[0].iov_base,
 371                 s->blk.serial ? s->blk.serial : "",
 372                 MIN(req->elem.in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES));
 373         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 374         g_free(req);
 375     } else if (type & VIRTIO_BLK_T_OUT) {
 376         qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 377                                  req->elem.out_num - 1);
 378         virtio_blk_handle_write(req, mrb);
 379     } else if (type == VIRTIO_BLK_T_IN || type == VIRTIO_BLK_T_BARRIER) {
 380         /* VIRTIO_BLK_T_IN is 0, so we can't just & it. */
 381         qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 382                                  req->elem.in_num - 1);
 383         virtio_blk_handle_read(req);
 384     } else {
 385         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 386         g_free(req);
 387     }
 388 }
 389
 390 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 391 {
 392     VirtIOBlock *s = VIRTIO_BLK(vdev);
 393     VirtIOBlockReq *req;
 394     MultiReqBuffer mrb = {
 395         .num_writes = 0,
 396     };
 397
 398 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 399     /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
 400      * dataplane here instead of waiting for .set_status().
 401      */
 402     if (s->dataplane) {
 403         virtio_blk_data_plane_start(s->dataplane);
 404         return;
 405     }
 406 #endif
 407
 408     while ((req = virtio_blk_get_request(s))) {
 409         virtio_blk_handle_request(req, &mrb);
 410     }
 411
 412     virtio_submit_multiwrite(s->bs, &mrb);
 413
 414     /*
 415      * FIXME: Want to check for completions before returning to guest mode,
 416      * so cached reads and writes are reported as quickly as possible. But
 417      * that should be done in the generic block layer.
 418      */
 419 }
 420
 421 static void virtio_blk_dma_restart_bh(void *opaque)
 422 {
 423     VirtIOBlock *s = opaque;
 424     VirtIOBlockReq *req = s->rq;
 425     MultiReqBuffer mrb = {
 426         .num_writes = 0,
 427     };
 428
 429     qemu_bh_delete(s->bh);
 430     s->bh = NULL;
 431
 432     s->rq = NULL;
 433
 434     while (req) {
 435         virtio_blk_handle_request(req, &mrb);
 436         req = req->next;
 437     }
 438
 439     virtio_submit_multiwrite(s->bs, &mrb);
 440 }
 441
 442 static void virtio_blk_dma_restart_cb(void *opaque, int running,
 443                                       RunState state)
 444 {
 445     VirtIOBlock *s = opaque;
 446
 447     if (!running) {
 448         return;
 449     }
 450
 451     if (!s->bh) {
 452         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 453         qemu_bh_schedule(s->bh);
 454     }
 455 }
 456
 457 static void virtio_blk_reset(VirtIODevice *vdev)
 458 {
 459 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 460     VirtIOBlock *s = VIRTIO_BLK(vdev);
 461
 462     if (s->dataplane) {
 463         virtio_blk_data_plane_stop(s->dataplane);
 464     }
 465 #endif
 466
 467     /*
 468      * This should cancel pending requests, but can't do nicely until there
 469      * are per-device request lists.
 470      */
 471     bdrv_drain_all();
 472 }
 473
 474 /* coalesce internal state, copy to pci i/o region 0
 475  */
 476 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 477 {
 478     VirtIOBlock *s = VIRTIO_BLK(vdev);
 479     struct virtio_blk_config blkcfg;
 480     uint64_t capacity;
 481     int blk_size = s->conf->logical_block_size;
 482
 483     bdrv_get_geometry(s->bs, &capacity);
 484     memset(&blkcfg, 0, sizeof(blkcfg));
 485     stq_raw(&blkcfg.capacity, capacity);
 486     stl_raw(&blkcfg.seg_max, 128 - 2);
 487     stw_raw(&blkcfg.cylinders, s->conf->cyls);
 488     stl_raw(&blkcfg.blk_size, blk_size);
 489     stw_raw(&blkcfg.min_io_size, s->conf->min_io_size / blk_size);
 490     stw_raw(&blkcfg.opt_io_size, s->conf->opt_io_size / blk_size);
 491     blkcfg.heads = s->conf->heads;
 492     /*
 493      * We must ensure that the block device capacity is a multiple of
 494      * the logical block size. If that is not the case, lets use
 495      * sector_mask to adopt the geometry to have a correct picture.
 496      * For those devices where the capacity is ok for the given geometry
 497      * we dont touch the sector value of the geometry, since some devices
 498      * (like s390 dasd) need a specific value. Here the capacity is already
 499      * cyls*heads*secs*blk_size and the sector value is not block size
 500      * divided by 512 - instead it is the amount of blk_size blocks
 501      * per track (cylinder).
 502      */
 503     if (bdrv_getlength(s->bs) /  s->conf->heads / s->conf->secs % blk_size) {
 504         blkcfg.sectors = s->conf->secs & ~s->sector_mask;
 505     } else {
 506         blkcfg.sectors = s->conf->secs;
 507     }
 508     blkcfg.size_max = 0;
 509     blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
 510     blkcfg.alignment_offset = 0;
 511     blkcfg.wce = bdrv_enable_write_cache(s->bs);
 512     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
 513 }
 514
 515 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
 516 {
 517     VirtIOBlock *s = VIRTIO_BLK(vdev);
 518     struct virtio_blk_config blkcfg;
 519
 520     memcpy(&blkcfg, config, sizeof(blkcfg));
 521     bdrv_set_enable_write_cache(s->bs, blkcfg.wce != 0);
 522 }
 523
 524 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 525 {
 526     VirtIOBlock *s = VIRTIO_BLK(vdev);
 527
 528     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 529     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 530     features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
 531     features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
 532     features |= (1 << VIRTIO_BLK_F_SCSI);
 533
 534     if (s->blk.config_wce) {
 535         features |= (1 << VIRTIO_BLK_F_CONFIG_WCE);
 536     }
 537     if (bdrv_enable_write_cache(s->bs))
 538         features |= (1 << VIRTIO_BLK_F_WCE);
 539
 540     if (bdrv_is_read_only(s->bs))
 541         features |= 1 << VIRTIO_BLK_F_RO;
 542
 543     return features;
 544 }
 545
 546 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
 547 {
 548     VirtIOBlock *s = VIRTIO_BLK(vdev);
 549     uint32_t features;
 550
 551 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 552     if (s->dataplane && !(status & (VIRTIO_CONFIG_S_DRIVER |
 553                                     VIRTIO_CONFIG_S_DRIVER_OK))) {
 554         virtio_blk_data_plane_stop(s->dataplane);
 555     }
 556 #endif
 557
 558     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 559         return;
 560     }
 561
 562     features = vdev->guest_features;
 563     bdrv_set_enable_write_cache(s->bs, !!(features & (1 << VIRTIO_BLK_F_WCE)));
 564 }
 565
 566 static void virtio_blk_save(QEMUFile *f, void *opaque)
 567 {
 568     VirtIOBlock *s = opaque;
 569     VirtIODevice *vdev = VIRTIO_DEVICE(s);
 570     VirtIOBlockReq *req = s->rq;
 571
 572     virtio_save(vdev, f);
 573
 574     while (req) {
 575         qemu_put_sbyte(f, 1);
 576         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 577         req = req->next;
 578     }
 579     qemu_put_sbyte(f, 0);
 580 }
 581
 582 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 583 {
 584     VirtIOBlock *s = opaque;
 585     VirtIODevice *vdev = VIRTIO_DEVICE(s);
 586     int ret;
 587
 588     if (version_id != 2)
 589         return -EINVAL;
 590
 591     ret = virtio_load(vdev, f);
 592     if (ret) {
 593         return ret;
 594     }
 595
 596     while (qemu_get_sbyte(f)) {
 597         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 598         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 599         req->next = s->rq;
 600         s->rq = req;
 601
 602         virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr,
 603             req->elem.in_num, 1);
 604         virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr,
 605             req->elem.out_num, 0);
 606     }
 607
 608     return 0;
 609 }
 610
 611 static void virtio_blk_resize(void *opaque)
 612 {
 613     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
 614
 615     virtio_notify_config(vdev);
 616 }
 617
 618 static const BlockDevOps virtio_block_ops = {
 619     .resize_cb = virtio_blk_resize,
 620 };
 621
 622 void virtio_blk_set_conf(DeviceState *dev, VirtIOBlkConf *blk)
 623 {
 624     VirtIOBlock *s = VIRTIO_BLK(dev);
 625     memcpy(&(s->blk), blk, sizeof(struct VirtIOBlkConf));
 626 }
 627
 628 static int virtio_blk_device_init(VirtIODevice *vdev)
 629 {
 630     DeviceState *qdev = DEVICE(vdev);
 631     VirtIOBlock *s = VIRTIO_BLK(vdev);
 632     VirtIOBlkConf *blk = &(s->blk);
 633     static int virtio_blk_id;
 634
 635     if (!blk->conf.bs) {
 636         error_report("drive property not set");
 637         return -1;
 638     }
 639     if (!bdrv_is_inserted(blk->conf.bs)) {
 640         error_report("Device needs media, but drive is empty");
 641         return -1;
 642     }
 643
 644     blkconf_serial(&blk->conf, &blk->serial);
 645     if (blkconf_geometry(&blk->conf, NULL, 65535, 255, 255) < 0) {
 646         return -1;
 647     }
 648
 649     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK,
 650                 sizeof(struct virtio_blk_config));
 651
 652     vdev->get_config = virtio_blk_update_config;
 653     vdev->set_config = virtio_blk_set_config;
 654     vdev->get_features = virtio_blk_get_features;
 655     vdev->set_status = virtio_blk_set_status;
 656     vdev->reset = virtio_blk_reset;
 657     s->bs = blk->conf.bs;
 658     s->conf = &blk->conf;
 659     memcpy(&(s->blk), blk, sizeof(struct VirtIOBlkConf));
 660     s->rq = NULL;
 661     s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
 662
 663     s->vq = virtio_add_queue(vdev, 128, virtio_blk_handle_output);
 664 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 665     if (!virtio_blk_data_plane_create(vdev, blk, &s->dataplane)) {
 666         virtio_cleanup(vdev);
 667         return -1;
 668     }
 669 #endif
 670
 671     s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 672     s->qdev = qdev;
 673     register_savevm(qdev, "virtio-blk", virtio_blk_id++, 2,
 674                     virtio_blk_save, virtio_blk_load, s);
 675     bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
 676     bdrv_set_buffer_alignment(s->bs, s->conf->logical_block_size);
 677
 678     bdrv_iostatus_enable(s->bs);
 679
 680     add_boot_device_path(s->conf->bootindex, qdev, "/disk@0,0");
 681     return 0;
 682 }
 683
 684 static int virtio_blk_device_exit(DeviceState *dev)
 685 {
 686     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 687     VirtIOBlock *s = VIRTIO_BLK(dev);
 688 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 689     virtio_blk_data_plane_destroy(s->dataplane);
 690     s->dataplane = NULL;
 691 #endif
 692     qemu_del_vm_change_state_handler(s->change);
 693     unregister_savevm(s->qdev, "virtio-blk", s);
 694     blockdev_mark_auto_del(s->bs);
 695     virtio_common_cleanup(vdev);
 696     return 0;
 697 }
 698
 699 static Property virtio_blk_properties[] = {
 700     DEFINE_VIRTIO_BLK_PROPERTIES(VirtIOBlock, blk),
 701     DEFINE_PROP_END_OF_LIST(),
 702 };
 703
 704 static void virtio_blk_class_init(ObjectClass *klass, void *data)
 705 {
 706     DeviceClass *dc = DEVICE_CLASS(klass);
 707     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
 708     dc->exit = virtio_blk_device_exit;
 709     dc->props = virtio_blk_properties;
 710     vdc->init = virtio_blk_device_init;
 711     vdc->get_config = virtio_blk_update_config;
 712     vdc->set_config = virtio_blk_set_config;
 713     vdc->get_features = virtio_blk_get_features;
 714     vdc->set_status = virtio_blk_set_status;
 715     vdc->reset = virtio_blk_reset;
 716 }
 717
 718 static const TypeInfo virtio_device_info = {
 719     .name = TYPE_VIRTIO_BLK,
 720     .parent = TYPE_VIRTIO_DEVICE,
 721     .instance_size = sizeof(VirtIOBlock),
 722     .class_init = virtio_blk_class_init,
 723 };
 724
 725 static void virtio_register_types(void)
 726 {
 727     type_register_static(&virtio_device_info);
 728 }
 729
 730 type_init(virtio_register_types)