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 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
  31 {
  32     VirtIOBlockReq *req = g_slice_new0(VirtIOBlockReq);
  33     req->dev = s;
  34     req->elem = g_slice_new0(VirtQueueElement);
  35     return req;
  36 }
  37
  38 static void virtio_blk_free_request(VirtIOBlockReq *req)
  39 {
  40     if (req) {
  41         g_slice_free(VirtQueueElement, req->elem);
  42         g_slice_free(VirtIOBlockReq, req);
  43     }
  44 }
  45
  46 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
  47 {
  48     VirtIOBlock *s = req->dev;
  49     VirtIODevice *vdev = VIRTIO_DEVICE(s);
  50
  51     trace_virtio_blk_req_complete(req, status);
  52
  53     stb_p(&req->in->status, status);
  54     virtqueue_push(s->vq, req->elem, req->qiov.size + sizeof(*req->in));
  55     virtio_notify(vdev, s->vq);
  56 }
  57
  58 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
  59     bool is_read)
  60 {
  61     BlockErrorAction action = bdrv_get_error_action(req->dev->bs, is_read, error);
  62     VirtIOBlock *s = req->dev;
  63
  64     if (action == BLOCK_ERROR_ACTION_STOP) {
  65         req->next = s->rq;
  66         s->rq = req;
  67     } else if (action == BLOCK_ERROR_ACTION_REPORT) {
  68         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
  69         bdrv_acct_done(s->bs, &req->acct);
  70         virtio_blk_free_request(req);
  71     }
  72
  73     bdrv_error_action(s->bs, action, is_read, error);
  74     return action != BLOCK_ERROR_ACTION_IGNORE;
  75 }
  76
  77 static void virtio_blk_rw_complete(void *opaque, int ret)
  78 {
  79     VirtIOBlockReq *req = opaque;
  80
  81     trace_virtio_blk_rw_complete(req, ret);
  82
  83     if (ret) {
  84         bool is_read = !(ldl_p(&req->out->type) & VIRTIO_BLK_T_OUT);
  85         if (virtio_blk_handle_rw_error(req, -ret, is_read))
  86             return;
  87     }
  88
  89     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
  90     bdrv_acct_done(req->dev->bs, &req->acct);
  91     virtio_blk_free_request(req);
  92 }
  93
  94 static void virtio_blk_flush_complete(void *opaque, int ret)
  95 {
  96     VirtIOBlockReq *req = opaque;
  97
  98     if (ret) {
  99         if (virtio_blk_handle_rw_error(req, -ret, 0)) {
 100             return;
 101         }
 102     }
 103
 104     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 105     bdrv_acct_done(req->dev->bs, &req->acct);
 106     virtio_blk_free_request(req);
 107 }
 108
 109 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 110 {
 111     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 112
 113     if (!virtqueue_pop(s->vq, req->elem)) {
 114         virtio_blk_free_request(req);
 115         return NULL;
 116     }
 117
 118     return req;
 119 }
 120
 121 int virtio_blk_handle_scsi_req(VirtIOBlock *blk,
 122                                VirtQueueElement *elem)
 123 {
 124     int status = VIRTIO_BLK_S_OK;
 125     struct virtio_scsi_inhdr *scsi = NULL;
 126 #ifdef __linux__
 127     int i;
 128     struct sg_io_hdr hdr;
 129 #endif
 130
 131     /*
 132      * We require at least one output segment each for the virtio_blk_outhdr
 133      * and the SCSI command block.
 134      *
 135      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 136      * and the sense buffer pointer in the input segments.
 137      */
 138     if (elem->out_num < 2 || elem->in_num < 3) {
 139         status = VIRTIO_BLK_S_IOERR;
 140         goto fail;
 141     }
 142
 143     /*
 144      * The scsi inhdr is placed in the second-to-last input segment, just
 145      * before the regular inhdr.
 146      */
 147     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
 148
 149     if (!blk->blk.scsi) {
 150         status = VIRTIO_BLK_S_UNSUPP;
 151         goto fail;
 152     }
 153
 154     /*
 155      * No support for bidirection commands yet.
 156      */
 157     if (elem->out_num > 2 && elem->in_num > 3) {
 158         status = VIRTIO_BLK_S_UNSUPP;
 159         goto fail;
 160     }
 161
 162 #ifdef __linux__
 163     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 164     hdr.interface_id = 'S';
 165     hdr.cmd_len = elem->out_sg[1].iov_len;
 166     hdr.cmdp = elem->out_sg[1].iov_base;
 167     hdr.dxfer_len = 0;
 168
 169     if (elem->out_num > 2) {
 170         /*
 171          * If there are more than the minimally required 2 output segments
 172          * there is write payload starting from the third iovec.
 173          */
 174         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 175         hdr.iovec_count = elem->out_num - 2;
 176
 177         for (i = 0; i < hdr.iovec_count; i++)
 178             hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
 179
 180         hdr.dxferp = elem->out_sg + 2;
 181
 182     } else if (elem->in_num > 3) {
 183         /*
 184          * If we have more than 3 input segments the guest wants to actually
 185          * read data.
 186          */
 187         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 188         hdr.iovec_count = elem->in_num - 3;
 189         for (i = 0; i < hdr.iovec_count; i++)
 190             hdr.dxfer_len += elem->in_sg[i].iov_len;
 191
 192         hdr.dxferp = elem->in_sg;
 193     } else {
 194         /*
 195          * Some SCSI commands don't actually transfer any data.
 196          */
 197         hdr.dxfer_direction = SG_DXFER_NONE;
 198     }
 199
 200     hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
 201     hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
 202
 203     status = bdrv_ioctl(blk->bs, SG_IO, &hdr);
 204     if (status) {
 205         status = VIRTIO_BLK_S_UNSUPP;
 206         goto fail;
 207     }
 208
 209     /*
 210      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
 211      * clear the masked_status field [hence status gets cleared too, see
 212      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
 213      * status has occurred.  However they do set DRIVER_SENSE in driver_status
 214      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
 215      */
 216     if (hdr.status == 0 && hdr.sb_len_wr > 0) {
 217         hdr.status = CHECK_CONDITION;
 218     }
 219
 220     stl_p(&scsi->errors,
 221           hdr.status | (hdr.msg_status << 8) |
 222           (hdr.host_status << 16) | (hdr.driver_status << 24));
 223     stl_p(&scsi->residual, hdr.resid);
 224     stl_p(&scsi->sense_len, hdr.sb_len_wr);
 225     stl_p(&scsi->data_len, hdr.dxfer_len);
 226
 227     return status;
 228 #else
 229     abort();
 230 #endif
 231
 232 fail:
 233     /* Just put anything nonzero so that the ioctl fails in the guest.  */
 234     if (scsi) {
 235         stl_p(&scsi->errors, 255);
 236     }
 237     return status;
 238 }
 239
 240 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 241 {
 242     int status;
 243
 244     status = virtio_blk_handle_scsi_req(req->dev, req->elem);
 245     virtio_blk_req_complete(req, status);
 246     virtio_blk_free_request(req);
 247 }
 248
 249 typedef struct MultiReqBuffer {
 250     BlockRequest        blkreq[32];
 251     unsigned int        num_writes;
 252 } MultiReqBuffer;
 253
 254 static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
 255 {
 256     int i, ret;
 257
 258     if (!mrb->num_writes) {
 259         return;
 260     }
 261
 262     ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
 263     if (ret != 0) {
 264         for (i = 0; i < mrb->num_writes; i++) {
 265             if (mrb->blkreq[i].error) {
 266                 virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
 267             }
 268         }
 269     }
 270
 271     mrb->num_writes = 0;
 272 }
 273
 274 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 275 {
 276     bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH);
 277
 278     /*
 279      * Make sure all outstanding writes are posted to the backing device.
 280      */
 281     virtio_submit_multiwrite(req->dev->bs, mrb);
 282     bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 283 }
 284
 285 static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 286 {
 287     BlockRequest *blkreq;
 288     uint64_t sector;
 289
 290     sector = ldq_p(&req->out->sector);
 291
 292     bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE);
 293
 294     trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
 295
 296     if (sector & req->dev->sector_mask) {
 297         virtio_blk_rw_complete(req, -EIO);
 298         return;
 299     }
 300     if (req->qiov.size % req->dev->conf->logical_block_size) {
 301         virtio_blk_rw_complete(req, -EIO);
 302         return;
 303     }
 304
 305     if (mrb->num_writes == 32) {
 306         virtio_submit_multiwrite(req->dev->bs, mrb);
 307     }
 308
 309     blkreq = &mrb->blkreq[mrb->num_writes];
 310     blkreq->sector = sector;
 311     blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
 312     blkreq->qiov = &req->qiov;
 313     blkreq->cb = virtio_blk_rw_complete;
 314     blkreq->opaque = req;
 315     blkreq->error = 0;
 316
 317     mrb->num_writes++;
 318 }
 319
 320 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 321 {
 322     uint64_t sector;
 323
 324     sector = ldq_p(&req->out->sector);
 325
 326     bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ);
 327
 328     trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512);
 329
 330     if (sector & req->dev->sector_mask) {
 331         virtio_blk_rw_complete(req, -EIO);
 332         return;
 333     }
 334     if (req->qiov.size % req->dev->conf->logical_block_size) {
 335         virtio_blk_rw_complete(req, -EIO);
 336         return;
 337     }
 338     bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
 339                    req->qiov.size / BDRV_SECTOR_SIZE,
 340                    virtio_blk_rw_complete, req);
 341 }
 342
 343 static void virtio_blk_handle_request(VirtIOBlockReq *req,
 344     MultiReqBuffer *mrb)
 345 {
 346     uint32_t type;
 347
 348     if (req->elem->out_num < 1 || req->elem->in_num < 1) {
 349         error_report("virtio-blk missing headers");
 350         exit(1);
 351     }
 352
 353     if (req->elem->out_sg[0].iov_len < sizeof(*req->out) ||
 354         req->elem->in_sg[req->elem->in_num - 1].iov_len < sizeof(*req->in)) {
 355         error_report("virtio-blk header not in correct element");
 356         exit(1);
 357     }
 358
 359     req->out = (void *)req->elem->out_sg[0].iov_base;
 360     req->in = (void *)req->elem->in_sg[req->elem->in_num - 1].iov_base;
 361
 362     type = ldl_p(&req->out->type);
 363
 364     if (type & VIRTIO_BLK_T_FLUSH) {
 365         virtio_blk_handle_flush(req, mrb);
 366     } else if (type & VIRTIO_BLK_T_SCSI_CMD) {
 367         virtio_blk_handle_scsi(req);
 368     } else if (type & VIRTIO_BLK_T_GET_ID) {
 369         VirtIOBlock *s = req->dev;
 370
 371         /*
 372          * NB: per existing s/n string convention the string is
 373          * terminated by '\0' only when shorter than buffer.
 374          */
 375         strncpy(req->elem->in_sg[0].iov_base,
 376                 s->blk.serial ? s->blk.serial : "",
 377                 MIN(req->elem->in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES));
 378         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 379         virtio_blk_free_request(req);
 380     } else if (type & VIRTIO_BLK_T_OUT) {
 381         qemu_iovec_init_external(&req->qiov, &req->elem->out_sg[1],
 382                                  req->elem->out_num - 1);
 383         virtio_blk_handle_write(req, mrb);
 384     } else if (type == VIRTIO_BLK_T_IN || type == VIRTIO_BLK_T_BARRIER) {
 385         /* VIRTIO_BLK_T_IN is 0, so we can't just & it. */
 386         qemu_iovec_init_external(&req->qiov, &req->elem->in_sg[0],
 387                                  req->elem->in_num - 1);
 388         virtio_blk_handle_read(req);
 389     } else {
 390         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 391         virtio_blk_free_request(req);
 392     }
 393 }
 394
 395 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 396 {
 397     VirtIOBlock *s = VIRTIO_BLK(vdev);
 398     VirtIOBlockReq *req;
 399     MultiReqBuffer mrb = {
 400         .num_writes = 0,
 401     };
 402
 403 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 404     /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
 405      * dataplane here instead of waiting for .set_status().
 406      */
 407     if (s->dataplane) {
 408         virtio_blk_data_plane_start(s->dataplane);
 409         return;
 410     }
 411 #endif
 412
 413     while ((req = virtio_blk_get_request(s))) {
 414         virtio_blk_handle_request(req, &mrb);
 415     }
 416
 417     virtio_submit_multiwrite(s->bs, &mrb);
 418
 419     /*
 420      * FIXME: Want to check for completions before returning to guest mode,
 421      * so cached reads and writes are reported as quickly as possible. But
 422      * that should be done in the generic block layer.
 423      */
 424 }
 425
 426 static void virtio_blk_dma_restart_bh(void *opaque)
 427 {
 428     VirtIOBlock *s = opaque;
 429     VirtIOBlockReq *req = s->rq;
 430     MultiReqBuffer mrb = {
 431         .num_writes = 0,
 432     };
 433
 434     qemu_bh_delete(s->bh);
 435     s->bh = NULL;
 436
 437     s->rq = NULL;
 438
 439     while (req) {
 440         virtio_blk_handle_request(req, &mrb);
 441         req = req->next;
 442     }
 443
 444     virtio_submit_multiwrite(s->bs, &mrb);
 445 }
 446
 447 static void virtio_blk_dma_restart_cb(void *opaque, int running,
 448                                       RunState state)
 449 {
 450     VirtIOBlock *s = opaque;
 451
 452     if (!running) {
 453         return;
 454     }
 455
 456     if (!s->bh) {
 457         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 458         qemu_bh_schedule(s->bh);
 459     }
 460 }
 461
 462 static void virtio_blk_reset(VirtIODevice *vdev)
 463 {
 464     VirtIOBlock *s = VIRTIO_BLK(vdev);
 465
 466 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 467     if (s->dataplane) {
 468         virtio_blk_data_plane_stop(s->dataplane);
 469     }
 470 #endif
 471
 472     /*
 473      * This should cancel pending requests, but can't do nicely until there
 474      * are per-device request lists.
 475      */
 476     bdrv_drain_all();
 477     bdrv_set_enable_write_cache(s->bs, s->original_wce);
 478 }
 479
 480 /* coalesce internal state, copy to pci i/o region 0
 481  */
 482 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 483 {
 484     VirtIOBlock *s = VIRTIO_BLK(vdev);
 485     struct virtio_blk_config blkcfg;
 486     uint64_t capacity;
 487     int blk_size = s->conf->logical_block_size;
 488
 489     bdrv_get_geometry(s->bs, &capacity);
 490     memset(&blkcfg, 0, sizeof(blkcfg));
 491     stq_p(&blkcfg.capacity, capacity);
 492     stl_p(&blkcfg.seg_max, 128 - 2);
 493     stw_p(&blkcfg.cylinders, s->conf->cyls);
 494     stl_p(&blkcfg.blk_size, blk_size);
 495     stw_p(&blkcfg.min_io_size, s->conf->min_io_size / blk_size);
 496     stw_p(&blkcfg.opt_io_size, s->conf->opt_io_size / blk_size);
 497     blkcfg.heads = s->conf->heads;
 498     /*
 499      * We must ensure that the block device capacity is a multiple of
 500      * the logical block size. If that is not the case, let's use
 501      * sector_mask to adopt the geometry to have a correct picture.
 502      * For those devices where the capacity is ok for the given geometry
 503      * we don't touch the sector value of the geometry, since some devices
 504      * (like s390 dasd) need a specific value. Here the capacity is already
 505      * cyls*heads*secs*blk_size and the sector value is not block size
 506      * divided by 512 - instead it is the amount of blk_size blocks
 507      * per track (cylinder).
 508      */
 509     if (bdrv_getlength(s->bs) /  s->conf->heads / s->conf->secs % blk_size) {
 510         blkcfg.sectors = s->conf->secs & ~s->sector_mask;
 511     } else {
 512         blkcfg.sectors = s->conf->secs;
 513     }
 514     blkcfg.size_max = 0;
 515     blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
 516     blkcfg.alignment_offset = 0;
 517     blkcfg.wce = bdrv_enable_write_cache(s->bs);
 518     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
 519 }
 520
 521 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
 522 {
 523     VirtIOBlock *s = VIRTIO_BLK(vdev);
 524     struct virtio_blk_config blkcfg;
 525
 526     memcpy(&blkcfg, config, sizeof(blkcfg));
 527
 528     aio_context_acquire(bdrv_get_aio_context(s->bs));
 529     bdrv_set_enable_write_cache(s->bs, blkcfg.wce != 0);
 530     aio_context_release(bdrv_get_aio_context(s->bs));
 531 }
 532
 533 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 534 {
 535     VirtIOBlock *s = VIRTIO_BLK(vdev);
 536
 537     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 538     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 539     features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
 540     features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
 541     features |= (1 << VIRTIO_BLK_F_SCSI);
 542
 543     if (s->blk.config_wce) {
 544         features |= (1 << VIRTIO_BLK_F_CONFIG_WCE);
 545     }
 546     if (bdrv_enable_write_cache(s->bs))
 547         features |= (1 << VIRTIO_BLK_F_WCE);
 548
 549     if (bdrv_is_read_only(s->bs))
 550         features |= 1 << VIRTIO_BLK_F_RO;
 551
 552     return features;
 553 }
 554
 555 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
 556 {
 557     VirtIOBlock *s = VIRTIO_BLK(vdev);
 558     uint32_t features;
 559
 560 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 561     if (s->dataplane && !(status & (VIRTIO_CONFIG_S_DRIVER |
 562                                     VIRTIO_CONFIG_S_DRIVER_OK))) {
 563         virtio_blk_data_plane_stop(s->dataplane);
 564     }
 565 #endif
 566
 567     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 568         return;
 569     }
 570
 571     features = vdev->guest_features;
 572
 573     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
 574      * cache flushes.  Thus, the "auto writethrough" behavior is never
 575      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
 576      * Leaving it enabled would break the following sequence:
 577      *
 578      *     Guest started with "-drive cache=writethrough"
 579      *     Guest sets status to 0
 580      *     Guest sets DRIVER bit in status field
 581      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
 582      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
 583      *     Guest writes 1 to the WCE configuration field (writeback mode)
 584      *     Guest sets DRIVER_OK bit in status field
 585      *
 586      * s->bs would erroneously be placed in writethrough mode.
 587      */
 588     if (!(features & (1 << VIRTIO_BLK_F_CONFIG_WCE))) {
 589         aio_context_acquire(bdrv_get_aio_context(s->bs));
 590         bdrv_set_enable_write_cache(s->bs,
 591                                     !!(features & (1 << VIRTIO_BLK_F_WCE)));
 592         aio_context_release(bdrv_get_aio_context(s->bs));
 593     }
 594 }
 595
 596 static void virtio_blk_save(QEMUFile *f, void *opaque)
 597 {
 598     VirtIOBlock *s = opaque;
 599     VirtIODevice *vdev = VIRTIO_DEVICE(s);
 600     VirtIOBlockReq *req = s->rq;
 601
 602     virtio_save(vdev, f);
 603
 604     while (req) {
 605         qemu_put_sbyte(f, 1);
 606         qemu_put_buffer(f, (unsigned char *)req->elem,
 607                         sizeof(VirtQueueElement));
 608         req = req->next;
 609     }
 610     qemu_put_sbyte(f, 0);
 611 }
 612
 613 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 614 {
 615     VirtIOBlock *s = opaque;
 616     VirtIODevice *vdev = VIRTIO_DEVICE(s);
 617     int ret;
 618
 619     if (version_id != 2)
 620         return -EINVAL;
 621
 622     ret = virtio_load(vdev, f);
 623     if (ret) {
 624         return ret;
 625     }
 626
 627     while (qemu_get_sbyte(f)) {
 628         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 629         qemu_get_buffer(f, (unsigned char *)req->elem,
 630                         sizeof(VirtQueueElement));
 631         req->next = s->rq;
 632         s->rq = req;
 633
 634         virtqueue_map_sg(req->elem->in_sg, req->elem->in_addr,
 635             req->elem->in_num, 1);
 636         virtqueue_map_sg(req->elem->out_sg, req->elem->out_addr,
 637             req->elem->out_num, 0);
 638     }
 639
 640     return 0;
 641 }
 642
 643 static void virtio_blk_resize(void *opaque)
 644 {
 645     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
 646
 647     virtio_notify_config(vdev);
 648 }
 649
 650 static const BlockDevOps virtio_block_ops = {
 651     .resize_cb = virtio_blk_resize,
 652 };
 653
 654 void virtio_blk_set_conf(DeviceState *dev, VirtIOBlkConf *blk)
 655 {
 656     VirtIOBlock *s = VIRTIO_BLK(dev);
 657     memcpy(&(s->blk), blk, sizeof(struct VirtIOBlkConf));
 658 }
 659
 660 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 661 /* Disable dataplane thread during live migration since it does not
 662  * update the dirty memory bitmap yet.
 663  */
 664 static void virtio_blk_migration_state_changed(Notifier *notifier, void *data)
 665 {
 666     VirtIOBlock *s = container_of(notifier, VirtIOBlock,
 667                                   migration_state_notifier);
 668     MigrationState *mig = data;
 669     Error *err = NULL;
 670
 671     if (migration_in_setup(mig)) {
 672         if (!s->dataplane) {
 673             return;
 674         }
 675         virtio_blk_data_plane_destroy(s->dataplane);
 676         s->dataplane = NULL;
 677     } else if (migration_has_finished(mig) ||
 678                migration_has_failed(mig)) {
 679         if (s->dataplane) {
 680             return;
 681         }
 682         bdrv_drain_all(); /* complete in-flight non-dataplane requests */
 683         virtio_blk_data_plane_create(VIRTIO_DEVICE(s), &s->blk,
 684                                      &s->dataplane, &err);
 685         if (err != NULL) {
 686             error_report("%s", error_get_pretty(err));
 687             error_free(err);
 688         }
 689     }
 690 }
 691 #endif /* CONFIG_VIRTIO_BLK_DATA_PLANE */
 692
 693 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
 694 {
 695     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 696     VirtIOBlock *s = VIRTIO_BLK(dev);
 697     VirtIOBlkConf *blk = &(s->blk);
 698 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 699     Error *err = NULL;
 700 #endif
 701     static int virtio_blk_id;
 702
 703     if (!blk->conf.bs) {
 704         error_setg(errp, "drive property not set");
 705         return;
 706     }
 707     if (!bdrv_is_inserted(blk->conf.bs)) {
 708         error_setg(errp, "Device needs media, but drive is empty");
 709         return;
 710     }
 711
 712     blkconf_serial(&blk->conf, &blk->serial);
 713     s->original_wce = bdrv_enable_write_cache(blk->conf.bs);
 714     if (blkconf_geometry(&blk->conf, NULL, 65535, 255, 255) < 0) {
 715         error_setg(errp, "Error setting geometry");
 716         return;
 717     }
 718
 719     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK,
 720                 sizeof(struct virtio_blk_config));
 721
 722     s->bs = blk->conf.bs;
 723     s->conf = &blk->conf;
 724     s->rq = NULL;
 725     s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
 726
 727     s->vq = virtio_add_queue(vdev, 128, virtio_blk_handle_output);
 728 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 729     virtio_blk_data_plane_create(vdev, blk, &s->dataplane, &err);
 730     if (err != NULL) {
 731         error_propagate(errp, err);
 732         virtio_cleanup(vdev);
 733         return;
 734     }
 735     s->migration_state_notifier.notify = virtio_blk_migration_state_changed;
 736     add_migration_state_change_notifier(&s->migration_state_notifier);
 737 #endif
 738
 739     s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 740     register_savevm(dev, "virtio-blk", virtio_blk_id++, 2,
 741                     virtio_blk_save, virtio_blk_load, s);
 742     bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
 743     bdrv_set_guest_block_size(s->bs, s->conf->logical_block_size);
 744
 745     bdrv_iostatus_enable(s->bs);
 746
 747     add_boot_device_path(s->conf->bootindex, dev, "/disk@0,0");
 748 }
 749
 750 static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp)
 751 {
 752     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 753     VirtIOBlock *s = VIRTIO_BLK(dev);
 754
 755 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 756     remove_migration_state_change_notifier(&s->migration_state_notifier);
 757     virtio_blk_data_plane_destroy(s->dataplane);
 758     s->dataplane = NULL;
 759 #endif
 760     qemu_del_vm_change_state_handler(s->change);
 761     unregister_savevm(dev, "virtio-blk", s);
 762     blockdev_mark_auto_del(s->bs);
 763     virtio_cleanup(vdev);
 764 }
 765
 766 static Property virtio_blk_properties[] = {
 767     DEFINE_VIRTIO_BLK_PROPERTIES(VirtIOBlock, blk),
 768     DEFINE_PROP_END_OF_LIST(),
 769 };
 770
 771 static void virtio_blk_class_init(ObjectClass *klass, void *data)
 772 {
 773     DeviceClass *dc = DEVICE_CLASS(klass);
 774     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
 775
 776     dc->props = virtio_blk_properties;
 777     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
 778     vdc->realize = virtio_blk_device_realize;
 779     vdc->unrealize = virtio_blk_device_unrealize;
 780     vdc->get_config = virtio_blk_update_config;
 781     vdc->set_config = virtio_blk_set_config;
 782     vdc->get_features = virtio_blk_get_features;
 783     vdc->set_status = virtio_blk_set_status;
 784     vdc->reset = virtio_blk_reset;
 785 }
 786
 787 static const TypeInfo virtio_device_info = {
 788     .name = TYPE_VIRTIO_BLK,
 789     .parent = TYPE_VIRTIO_DEVICE,
 790     .instance_size = sizeof(VirtIOBlock),
 791     .class_init = virtio_blk_class_init,
 792 };
 793
 794 static void virtio_register_types(void)
 795 {
 796     type_register_static(&virtio_device_info);
 797 }
 798
 799 type_init(virtio_register_types)