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