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