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 # include "migration/migration.h"
  23 #endif
  24 #include "block/scsi.h"
  25 #ifdef __linux__
  26 # include <scsi/sg.h>
  27 #endif
  28 #include "hw/virtio/virtio-bus.h"
  29
  30 typedef struct VirtIOBlockReq
  31 {
  32     VirtIOBlock *dev;
  33     VirtQueueElement elem;
  34     struct virtio_blk_inhdr *in;
  35     struct virtio_blk_outhdr *out;
  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 int virtio_blk_handle_scsi_req(VirtIOBlock *blk,
 128                                VirtQueueElement *elem)
 129 {
 130     int status = VIRTIO_BLK_S_OK;
 131     struct virtio_scsi_inhdr *scsi = NULL;
 132 #ifdef __linux__
 133     int i;
 134     struct sg_io_hdr hdr;
 135 #endif
 136
 137     /*
 138      * We require at least one output segment each for the virtio_blk_outhdr
 139      * and the SCSI command block.
 140      *
 141      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 142      * and the sense buffer pointer in the input segments.
 143      */
 144     if (elem->out_num < 2 || elem->in_num < 3) {
 145         status = VIRTIO_BLK_S_IOERR;
 146         goto fail;
 147     }
 148
 149     /*
 150      * The scsi inhdr is placed in the second-to-last input segment, just
 151      * before the regular inhdr.
 152      */
 153     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
 154
 155     if (!blk->blk.scsi) {
 156         status = VIRTIO_BLK_S_UNSUPP;
 157         goto fail;
 158     }
 159
 160     /*
 161      * No support for bidirection commands yet.
 162      */
 163     if (elem->out_num > 2 && elem->in_num > 3) {
 164         status = VIRTIO_BLK_S_UNSUPP;
 165         goto fail;
 166     }
 167
 168 #ifdef __linux__
 169     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 170     hdr.interface_id = 'S';
 171     hdr.cmd_len = elem->out_sg[1].iov_len;
 172     hdr.cmdp = elem->out_sg[1].iov_base;
 173     hdr.dxfer_len = 0;
 174
 175     if (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 = elem->out_num - 2;
 182
 183         for (i = 0; i < hdr.iovec_count; i++)
 184             hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
 185
 186         hdr.dxferp = elem->out_sg + 2;
 187
 188     } else if (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 = elem->in_num - 3;
 195         for (i = 0; i < hdr.iovec_count; i++)
 196             hdr.dxfer_len += elem->in_sg[i].iov_len;
 197
 198         hdr.dxferp = 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 = elem->in_sg[elem->in_num - 3].iov_base;
 207     hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
 208
 209     status = bdrv_ioctl(blk->bs, SG_IO, &hdr);
 210     if (status) {
 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(&scsi->errors,
 227           hdr.status | (hdr.msg_status << 8) |
 228           (hdr.host_status << 16) | (hdr.driver_status << 24));
 229     stl_p(&scsi->residual, hdr.resid);
 230     stl_p(&scsi->sense_len, hdr.sb_len_wr);
 231     stl_p(&scsi->data_len, hdr.dxfer_len);
 232
 233     return status;
 234 #else
 235     abort();
 236 #endif
 237
 238 fail:
 239     /* Just put anything nonzero so that the ioctl fails in the guest.  */
 240     if (scsi) {
 241         stl_p(&scsi->errors, 255);
 242     }
 243     return status;
 244 }
 245
 246 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 247 {
 248     int status;
 249
 250     status = virtio_blk_handle_scsi_req(req->dev, &req->elem);
 251     virtio_blk_req_complete(req, status);
 252     g_free(req);
 253 }
 254
 255 typedef struct MultiReqBuffer {
 256     BlockRequest        blkreq[32];
 257     unsigned int        num_writes;
 258 } MultiReqBuffer;
 259
 260 static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
 261 {
 262     int i, ret;
 263
 264     if (!mrb->num_writes) {
 265         return;
 266     }
 267
 268     ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
 269     if (ret != 0) {
 270         for (i = 0; i < mrb->num_writes; i++) {
 271             if (mrb->blkreq[i].error) {
 272                 virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
 273             }
 274         }
 275     }
 276
 277     mrb->num_writes = 0;
 278 }
 279
 280 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 281 {
 282     bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH);
 283
 284     /*
 285      * Make sure all outstanding writes are posted to the backing device.
 286      */
 287     virtio_submit_multiwrite(req->dev->bs, mrb);
 288     bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 289 }
 290
 291 static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 292 {
 293     BlockRequest *blkreq;
 294     uint64_t sector;
 295
 296     sector = ldq_p(&req->out->sector);
 297
 298     bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE);
 299
 300     trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
 301
 302     if (sector & req->dev->sector_mask) {
 303         virtio_blk_rw_complete(req, -EIO);
 304         return;
 305     }
 306     if (req->qiov.size % req->dev->conf->logical_block_size) {
 307         virtio_blk_rw_complete(req, -EIO);
 308         return;
 309     }
 310
 311     if (mrb->num_writes == 32) {
 312         virtio_submit_multiwrite(req->dev->bs, mrb);
 313     }
 314
 315     blkreq = &mrb->blkreq[mrb->num_writes];
 316     blkreq->sector = sector;
 317     blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
 318     blkreq->qiov = &req->qiov;
 319     blkreq->cb = virtio_blk_rw_complete;
 320     blkreq->opaque = req;
 321     blkreq->error = 0;
 322
 323     mrb->num_writes++;
 324 }
 325
 326 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 327 {
 328     uint64_t sector;
 329
 330     sector = ldq_p(&req->out->sector);
 331
 332     bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ);
 333
 334     trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512);
 335
 336     if (sector & req->dev->sector_mask) {
 337         virtio_blk_rw_complete(req, -EIO);
 338         return;
 339     }
 340     if (req->qiov.size % req->dev->conf->logical_block_size) {
 341         virtio_blk_rw_complete(req, -EIO);
 342         return;
 343     }
 344     bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
 345                    req->qiov.size / BDRV_SECTOR_SIZE,
 346                    virtio_blk_rw_complete, req);
 347 }
 348
 349 static void virtio_blk_handle_request(VirtIOBlockReq *req,
 350     MultiReqBuffer *mrb)
 351 {
 352     uint32_t type;
 353
 354     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 355         error_report("virtio-blk missing headers");
 356         exit(1);
 357     }
 358
 359     if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 360         req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 361         error_report("virtio-blk header not in correct element");
 362         exit(1);
 363     }
 364
 365     req->out = (void *)req->elem.out_sg[0].iov_base;
 366     req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 367
 368     type = ldl_p(&req->out->type);
 369
 370     if (type & VIRTIO_BLK_T_FLUSH) {
 371         virtio_blk_handle_flush(req, mrb);
 372     } else if (type & VIRTIO_BLK_T_SCSI_CMD) {
 373         virtio_blk_handle_scsi(req);
 374     } else if (type & VIRTIO_BLK_T_GET_ID) {
 375         VirtIOBlock *s = req->dev;
 376
 377         /*
 378          * NB: per existing s/n string convention the string is
 379          * terminated by '\0' only when shorter than buffer.
 380          */
 381         strncpy(req->elem.in_sg[0].iov_base,
 382                 s->blk.serial ? s->blk.serial : "",
 383                 MIN(req->elem.in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES));
 384         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 385         g_free(req);
 386     } else if (type & VIRTIO_BLK_T_OUT) {
 387         qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 388                                  req->elem.out_num - 1);
 389         virtio_blk_handle_write(req, mrb);
 390     } else if (type == VIRTIO_BLK_T_IN || type == VIRTIO_BLK_T_BARRIER) {
 391         /* VIRTIO_BLK_T_IN is 0, so we can't just & it. */
 392         qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 393                                  req->elem.in_num - 1);
 394         virtio_blk_handle_read(req);
 395     } else {
 396         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 397         g_free(req);
 398     }
 399 }
 400
 401 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 402 {
 403     VirtIOBlock *s = VIRTIO_BLK(vdev);
 404     VirtIOBlockReq *req;
 405     MultiReqBuffer mrb = {
 406         .num_writes = 0,
 407     };
 408
 409 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 410     /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
 411      * dataplane here instead of waiting for .set_status().
 412      */
 413     if (s->dataplane) {
 414         virtio_blk_data_plane_start(s->dataplane);
 415         return;
 416     }
 417 #endif
 418
 419     while ((req = virtio_blk_get_request(s))) {
 420         virtio_blk_handle_request(req, &mrb);
 421     }
 422
 423     virtio_submit_multiwrite(s->bs, &mrb);
 424
 425     /*
 426      * FIXME: Want to check for completions before returning to guest mode,
 427      * so cached reads and writes are reported as quickly as possible. But
 428      * that should be done in the generic block layer.
 429      */
 430 }
 431
 432 static void virtio_blk_dma_restart_bh(void *opaque)
 433 {
 434     VirtIOBlock *s = opaque;
 435     VirtIOBlockReq *req = s->rq;
 436     MultiReqBuffer mrb = {
 437         .num_writes = 0,
 438     };
 439
 440     qemu_bh_delete(s->bh);
 441     s->bh = NULL;
 442
 443     s->rq = NULL;
 444
 445     while (req) {
 446         virtio_blk_handle_request(req, &mrb);
 447         req = req->next;
 448     }
 449
 450     virtio_submit_multiwrite(s->bs, &mrb);
 451 }
 452
 453 static void virtio_blk_dma_restart_cb(void *opaque, int running,
 454                                       RunState state)
 455 {
 456     VirtIOBlock *s = opaque;
 457
 458     if (!running) {
 459         return;
 460     }
 461
 462     if (!s->bh) {
 463         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 464         qemu_bh_schedule(s->bh);
 465     }
 466 }
 467
 468 static void virtio_blk_reset(VirtIODevice *vdev)
 469 {
 470     VirtIOBlock *s = VIRTIO_BLK(vdev);
 471
 472 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 473     if (s->dataplane) {
 474         virtio_blk_data_plane_stop(s->dataplane);
 475     }
 476 #endif
 477
 478     /*
 479      * This should cancel pending requests, but can't do nicely until there
 480      * are per-device request lists.
 481      */
 482     bdrv_drain_all();
 483     bdrv_set_enable_write_cache(s->bs, s->original_wce);
 484 }
 485
 486 /* coalesce internal state, copy to pci i/o region 0
 487  */
 488 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 489 {
 490     VirtIOBlock *s = VIRTIO_BLK(vdev);
 491     struct virtio_blk_config blkcfg;
 492     uint64_t capacity;
 493     int blk_size = s->conf->logical_block_size;
 494
 495     bdrv_get_geometry(s->bs, &capacity);
 496     memset(&blkcfg, 0, sizeof(blkcfg));
 497     stq_raw(&blkcfg.capacity, capacity);
 498     stl_raw(&blkcfg.seg_max, 128 - 2);
 499     stw_raw(&blkcfg.cylinders, s->conf->cyls);
 500     stl_raw(&blkcfg.blk_size, blk_size);
 501     stw_raw(&blkcfg.min_io_size, s->conf->min_io_size / blk_size);
 502     stw_raw(&blkcfg.opt_io_size, s->conf->opt_io_size / blk_size);
 503     blkcfg.heads = s->conf->heads;
 504     /*
 505      * We must ensure that the block device capacity is a multiple of
 506      * the logical block size. If that is not the case, let's use
 507      * sector_mask to adopt the geometry to have a correct picture.
 508      * For those devices where the capacity is ok for the given geometry
 509      * we don't touch the sector value of the geometry, since some devices
 510      * (like s390 dasd) need a specific value. Here the capacity is already
 511      * cyls*heads*secs*blk_size and the sector value is not block size
 512      * divided by 512 - instead it is the amount of blk_size blocks
 513      * per track (cylinder).
 514      */
 515     if (bdrv_getlength(s->bs) /  s->conf->heads / s->conf->secs % blk_size) {
 516         blkcfg.sectors = s->conf->secs & ~s->sector_mask;
 517     } else {
 518         blkcfg.sectors = s->conf->secs;
 519     }
 520     blkcfg.size_max = 0;
 521     blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
 522     blkcfg.alignment_offset = 0;
 523     blkcfg.wce = bdrv_enable_write_cache(s->bs);
 524     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
 525 }
 526
 527 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
 528 {
 529     VirtIOBlock *s = VIRTIO_BLK(vdev);
 530     struct virtio_blk_config blkcfg;
 531
 532     memcpy(&blkcfg, config, sizeof(blkcfg));
 533
 534     aio_context_acquire(bdrv_get_aio_context(s->bs));
 535     bdrv_set_enable_write_cache(s->bs, blkcfg.wce != 0);
 536     aio_context_release(bdrv_get_aio_context(s->bs));
 537 }
 538
 539 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 540 {
 541     VirtIOBlock *s = VIRTIO_BLK(vdev);
 542
 543     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 544     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 545     features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
 546     features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
 547     features |= (1 << VIRTIO_BLK_F_SCSI);
 548
 549     if (s->blk.config_wce) {
 550         features |= (1 << VIRTIO_BLK_F_CONFIG_WCE);
 551     }
 552     if (bdrv_enable_write_cache(s->bs))
 553         features |= (1 << VIRTIO_BLK_F_WCE);
 554
 555     if (bdrv_is_read_only(s->bs))
 556         features |= 1 << VIRTIO_BLK_F_RO;
 557
 558     return features;
 559 }
 560
 561 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
 562 {
 563     VirtIOBlock *s = VIRTIO_BLK(vdev);
 564     uint32_t features;
 565
 566 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 567     if (s->dataplane && !(status & (VIRTIO_CONFIG_S_DRIVER |
 568                                     VIRTIO_CONFIG_S_DRIVER_OK))) {
 569         virtio_blk_data_plane_stop(s->dataplane);
 570     }
 571 #endif
 572
 573     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 574         return;
 575     }
 576
 577     features = vdev->guest_features;
 578
 579     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
 580      * cache flushes.  Thus, the "auto writethrough" behavior is never
 581      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
 582      * Leaving it enabled would break the following sequence:
 583      *
 584      *     Guest started with "-drive cache=writethrough"
 585      *     Guest sets status to 0
 586      *     Guest sets DRIVER bit in status field
 587      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
 588      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
 589      *     Guest writes 1 to the WCE configuration field (writeback mode)
 590      *     Guest sets DRIVER_OK bit in status field
 591      *
 592      * s->bs would erroneously be placed in writethrough mode.
 593      */
 594     if (!(features & (1 << VIRTIO_BLK_F_CONFIG_WCE))) {
 595         aio_context_acquire(bdrv_get_aio_context(s->bs));
 596         bdrv_set_enable_write_cache(s->bs,
 597                                     !!(features & (1 << VIRTIO_BLK_F_WCE)));
 598         aio_context_release(bdrv_get_aio_context(s->bs));
 599     }
 600 }
 601
 602 static void virtio_blk_save(QEMUFile *f, void *opaque)
 603 {
 604     VirtIOBlock *s = opaque;
 605     VirtIODevice *vdev = VIRTIO_DEVICE(s);
 606     VirtIOBlockReq *req = s->rq;
 607
 608     virtio_save(vdev, f);
 609
 610     while (req) {
 611         qemu_put_sbyte(f, 1);
 612         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 613         req = req->next;
 614     }
 615     qemu_put_sbyte(f, 0);
 616 }
 617
 618 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 619 {
 620     VirtIOBlock *s = opaque;
 621     VirtIODevice *vdev = VIRTIO_DEVICE(s);
 622     int ret;
 623
 624     if (version_id != 2)
 625         return -EINVAL;
 626
 627     ret = virtio_load(vdev, f);
 628     if (ret) {
 629         return ret;
 630     }
 631
 632     while (qemu_get_sbyte(f)) {
 633         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 634         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 635         req->next = s->rq;
 636         s->rq = req;
 637
 638         virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr,
 639             req->elem.in_num, 1);
 640         virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr,
 641             req->elem.out_num, 0);
 642     }
 643
 644     return 0;
 645 }
 646
 647 static void virtio_blk_resize(void *opaque)
 648 {
 649     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
 650
 651     virtio_notify_config(vdev);
 652 }
 653
 654 static const BlockDevOps virtio_block_ops = {
 655     .resize_cb = virtio_blk_resize,
 656 };
 657
 658 void virtio_blk_set_conf(DeviceState *dev, VirtIOBlkConf *blk)
 659 {
 660     VirtIOBlock *s = VIRTIO_BLK(dev);
 661     memcpy(&(s->blk), blk, sizeof(struct VirtIOBlkConf));
 662 }
 663
 664 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 665 /* Disable dataplane thread during live migration since it does not
 666  * update the dirty memory bitmap yet.
 667  */
 668 static void virtio_blk_migration_state_changed(Notifier *notifier, void *data)
 669 {
 670     VirtIOBlock *s = container_of(notifier, VirtIOBlock,
 671                                   migration_state_notifier);
 672     MigrationState *mig = data;
 673     Error *err = NULL;
 674
 675     if (migration_in_setup(mig)) {
 676         if (!s->dataplane) {
 677             return;
 678         }
 679         virtio_blk_data_plane_destroy(s->dataplane);
 680         s->dataplane = NULL;
 681     } else if (migration_has_finished(mig) ||
 682                migration_has_failed(mig)) {
 683         if (s->dataplane) {
 684             return;
 685         }
 686         bdrv_drain_all(); /* complete in-flight non-dataplane requests */
 687         virtio_blk_data_plane_create(VIRTIO_DEVICE(s), &s->blk,
 688                                      &s->dataplane, &err);
 689         if (err != NULL) {
 690             error_report("%s", error_get_pretty(err));
 691             error_free(err);
 692         }
 693     }
 694 }
 695 #endif /* CONFIG_VIRTIO_BLK_DATA_PLANE */
 696
 697 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
 698 {
 699     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 700     VirtIOBlock *s = VIRTIO_BLK(dev);
 701     VirtIOBlkConf *blk = &(s->blk);
 702 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 703     Error *err = NULL;
 704 #endif
 705     static int virtio_blk_id;
 706
 707     if (!blk->conf.bs) {
 708         error_setg(errp, "drive property not set");
 709         return;
 710     }
 711     if (!bdrv_is_inserted(blk->conf.bs)) {
 712         error_setg(errp, "Device needs media, but drive is empty");
 713         return;
 714     }
 715
 716     blkconf_serial(&blk->conf, &blk->serial);
 717     s->original_wce = bdrv_enable_write_cache(blk->conf.bs);
 718     if (blkconf_geometry(&blk->conf, NULL, 65535, 255, 255) < 0) {
 719         error_setg(errp, "Error setting geometry");
 720         return;
 721     }
 722
 723     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK,
 724                 sizeof(struct virtio_blk_config));
 725
 726     s->bs = blk->conf.bs;
 727     s->conf = &blk->conf;
 728     s->rq = NULL;
 729     s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
 730
 731     s->vq = virtio_add_queue(vdev, 128, virtio_blk_handle_output);
 732 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 733     virtio_blk_data_plane_create(vdev, blk, &s->dataplane, &err);
 734     if (err != NULL) {
 735         error_propagate(errp, err);
 736         virtio_cleanup(vdev);
 737         return;
 738     }
 739     s->migration_state_notifier.notify = virtio_blk_migration_state_changed;
 740     add_migration_state_change_notifier(&s->migration_state_notifier);
 741 #endif
 742
 743     s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 744     register_savevm(dev, "virtio-blk", virtio_blk_id++, 2,
 745                     virtio_blk_save, virtio_blk_load, s);
 746     bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
 747     bdrv_set_guest_block_size(s->bs, s->conf->logical_block_size);
 748
 749     bdrv_iostatus_enable(s->bs);
 750
 751     add_boot_device_path(s->conf->bootindex, dev, "/disk@0,0");
 752 }
 753
 754 static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp)
 755 {
 756     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 757     VirtIOBlock *s = VIRTIO_BLK(dev);
 758
 759 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 760     remove_migration_state_change_notifier(&s->migration_state_notifier);
 761     virtio_blk_data_plane_destroy(s->dataplane);
 762     s->dataplane = NULL;
 763 #endif
 764     qemu_del_vm_change_state_handler(s->change);
 765     unregister_savevm(dev, "virtio-blk", s);
 766     blockdev_mark_auto_del(s->bs);
 767     virtio_cleanup(vdev);
 768 }
 769
 770 static Property virtio_blk_properties[] = {
 771     DEFINE_VIRTIO_BLK_PROPERTIES(VirtIOBlock, blk),
 772     DEFINE_PROP_END_OF_LIST(),
 773 };
 774
 775 static void virtio_blk_class_init(ObjectClass *klass, void *data)
 776 {
 777     DeviceClass *dc = DEVICE_CLASS(klass);
 778     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
 779
 780     dc->props = virtio_blk_properties;
 781     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
 782     vdc->realize = virtio_blk_device_realize;
 783     vdc->unrealize = virtio_blk_device_unrealize;
 784     vdc->get_config = virtio_blk_update_config;
 785     vdc->set_config = virtio_blk_set_config;
 786     vdc->get_features = virtio_blk_get_features;
 787     vdc->set_status = virtio_blk_set_status;
 788     vdc->reset = virtio_blk_reset;
 789 }
 790
 791 static const TypeInfo virtio_device_info = {
 792     .name = TYPE_VIRTIO_BLK,
 793     .parent = TYPE_VIRTIO_DEVICE,
 794     .instance_size = sizeof(VirtIOBlock),
 795     .class_init = virtio_blk_class_init,
 796 };
 797
 798 static void virtio_register_types(void)
 799 {
 800     type_register_static(&virtio_device_info);
 801 }
 802
 803 type_init(virtio_register_types)