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