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/iov.h"
  16 #include "qemu/error-report.h"
  17 #include "trace.h"
  18 #include "hw/block/block.h"
  19 #include "sysemu/blockdev.h"
  20 #include "hw/virtio/virtio-blk.h"
  21 #include "dataplane/virtio-blk.h"
  22 #include "migration/migration.h"
  23 #include "block/scsi.h"
  24 #ifdef __linux__
  25 # include <scsi/sg.h>
  26 #endif
  27 #include "hw/virtio/virtio-bus.h"
  28 #include "hw/virtio/virtio-access.h"
  29
  30 VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
  31 {
  32     VirtIOBlockReq *req = g_slice_new(VirtIOBlockReq);
  33     req->dev = s;
  34     req->qiov.size = 0;
  35     req->next = NULL;
  36     return req;
  37 }
  38
  39 void virtio_blk_free_request(VirtIOBlockReq *req)
  40 {
  41     if (req) {
  42         g_slice_free(VirtIOBlockReq, req);
  43     }
  44 }
  45
  46 static void virtio_blk_complete_request(VirtIOBlockReq *req,
  47                                         unsigned char status)
  48 {
  49     VirtIOBlock *s = req->dev;
  50     VirtIODevice *vdev = VIRTIO_DEVICE(s);
  51
  52     trace_virtio_blk_req_complete(req, status);
  53
  54     stb_p(&req->in->status, status);
  55     virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
  56     virtio_notify(vdev, s->vq);
  57 }
  58
  59 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
  60 {
  61     req->dev->complete_request(req, status);
  62 }
  63
  64 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
  65     bool is_read)
  66 {
  67     BlockErrorAction action = bdrv_get_error_action(req->dev->bs, is_read, error);
  68     VirtIOBlock *s = req->dev;
  69
  70     if (action == BLOCK_ERROR_ACTION_STOP) {
  71         req->next = s->rq;
  72         s->rq = req;
  73     } else if (action == BLOCK_ERROR_ACTION_REPORT) {
  74         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
  75         block_acct_done(bdrv_get_stats(s->bs), &req->acct);
  76         virtio_blk_free_request(req);
  77     }
  78
  79     bdrv_error_action(s->bs, action, is_read, error);
  80     return action != BLOCK_ERROR_ACTION_IGNORE;
  81 }
  82
  83 static void virtio_blk_rw_complete(void *opaque, int ret)
  84 {
  85     VirtIOBlockReq *req = opaque;
  86
  87     trace_virtio_blk_rw_complete(req, ret);
  88
  89     if (ret) {
  90         int p = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type);
  91         bool is_read = !(p & VIRTIO_BLK_T_OUT);
  92         if (virtio_blk_handle_rw_error(req, -ret, is_read))
  93             return;
  94     }
  95
  96     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
  97     block_acct_done(bdrv_get_stats(req->dev->bs), &req->acct);
  98     virtio_blk_free_request(req);
  99 }
 100
 101 static void virtio_blk_flush_complete(void *opaque, int ret)
 102 {
 103     VirtIOBlockReq *req = opaque;
 104
 105     if (ret) {
 106         if (virtio_blk_handle_rw_error(req, -ret, 0)) {
 107             return;
 108         }
 109     }
 110
 111     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 112     block_acct_done(bdrv_get_stats(req->dev->bs), &req->acct);
 113     virtio_blk_free_request(req);
 114 }
 115
 116 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 117 {
 118     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 119
 120     if (!virtqueue_pop(s->vq, &req->elem)) {
 121         virtio_blk_free_request(req);
 122         return NULL;
 123     }
 124
 125     return req;
 126 }
 127
 128 int virtio_blk_handle_scsi_req(VirtIOBlock *blk,
 129                                VirtQueueElement *elem)
 130 {
 131     int status = VIRTIO_BLK_S_OK;
 132     struct virtio_scsi_inhdr *scsi = NULL;
 133     VirtIODevice *vdev = VIRTIO_DEVICE(blk);
 134
 135 #ifdef __linux__
 136     int i;
 137     struct sg_io_hdr hdr;
 138 #endif
 139
 140     /*
 141      * We require at least one output segment each for the virtio_blk_outhdr
 142      * and the SCSI command block.
 143      *
 144      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 145      * and the sense buffer pointer in the input segments.
 146      */
 147     if (elem->out_num < 2 || elem->in_num < 3) {
 148         status = VIRTIO_BLK_S_IOERR;
 149         goto fail;
 150     }
 151
 152     /*
 153      * The scsi inhdr is placed in the second-to-last input segment, just
 154      * before the regular inhdr.
 155      */
 156     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
 157
 158     if (!blk->blk.scsi) {
 159         status = VIRTIO_BLK_S_UNSUPP;
 160         goto fail;
 161     }
 162
 163     /*
 164      * No support for bidirection commands yet.
 165      */
 166     if (elem->out_num > 2 && elem->in_num > 3) {
 167         status = VIRTIO_BLK_S_UNSUPP;
 168         goto fail;
 169     }
 170
 171 #ifdef __linux__
 172     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 173     hdr.interface_id = 'S';
 174     hdr.cmd_len = elem->out_sg[1].iov_len;
 175     hdr.cmdp = elem->out_sg[1].iov_base;
 176     hdr.dxfer_len = 0;
 177
 178     if (elem->out_num > 2) {
 179         /*
 180          * If there are more than the minimally required 2 output segments
 181          * there is write payload starting from the third iovec.
 182          */
 183         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 184         hdr.iovec_count = elem->out_num - 2;
 185
 186         for (i = 0; i < hdr.iovec_count; i++)
 187             hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
 188
 189         hdr.dxferp = elem->out_sg + 2;
 190
 191     } else if (elem->in_num > 3) {
 192         /*
 193          * If we have more than 3 input segments the guest wants to actually
 194          * read data.
 195          */
 196         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 197         hdr.iovec_count = elem->in_num - 3;
 198         for (i = 0; i < hdr.iovec_count; i++)
 199             hdr.dxfer_len += elem->in_sg[i].iov_len;
 200
 201         hdr.dxferp = elem->in_sg;
 202     } else {
 203         /*
 204          * Some SCSI commands don't actually transfer any data.
 205          */
 206         hdr.dxfer_direction = SG_DXFER_NONE;
 207     }
 208
 209     hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
 210     hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
 211
 212     status = bdrv_ioctl(blk->bs, SG_IO, &hdr);
 213     if (status) {
 214         status = VIRTIO_BLK_S_UNSUPP;
 215         goto fail;
 216     }
 217
 218     /*
 219      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
 220      * clear the masked_status field [hence status gets cleared too, see
 221      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
 222      * status has occurred.  However they do set DRIVER_SENSE in driver_status
 223      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
 224      */
 225     if (hdr.status == 0 && hdr.sb_len_wr > 0) {
 226         hdr.status = CHECK_CONDITION;
 227     }
 228
 229     virtio_stl_p(vdev, &scsi->errors,
 230                  hdr.status | (hdr.msg_status << 8) |
 231                  (hdr.host_status << 16) | (hdr.driver_status << 24));
 232     virtio_stl_p(vdev, &scsi->residual, hdr.resid);
 233     virtio_stl_p(vdev, &scsi->sense_len, hdr.sb_len_wr);
 234     virtio_stl_p(vdev, &scsi->data_len, hdr.dxfer_len);
 235
 236     return status;
 237 #else
 238     abort();
 239 #endif
 240
 241 fail:
 242     /* Just put anything nonzero so that the ioctl fails in the guest.  */
 243     if (scsi) {
 244         virtio_stl_p(vdev, &scsi->errors, 255);
 245     }
 246     return status;
 247 }
 248
 249 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 250 {
 251     int status;
 252
 253     status = virtio_blk_handle_scsi_req(req->dev, &req->elem);
 254     virtio_blk_req_complete(req, status);
 255     virtio_blk_free_request(req);
 256 }
 257
 258 void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
 259 {
 260     int i, ret;
 261
 262     if (!mrb->num_writes) {
 263         return;
 264     }
 265
 266     ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
 267     if (ret != 0) {
 268         for (i = 0; i < mrb->num_writes; i++) {
 269             if (mrb->blkreq[i].error) {
 270                 virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
 271             }
 272         }
 273     }
 274
 275     mrb->num_writes = 0;
 276 }
 277
 278 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 279 {
 280     block_acct_start(bdrv_get_stats(req->dev->bs), &req->acct, 0,
 281                      BLOCK_ACCT_FLUSH);
 282
 283     /*
 284      * Make sure all outstanding writes are posted to the backing device.
 285      */
 286     virtio_submit_multiwrite(req->dev->bs, mrb);
 287     bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 288 }
 289
 290 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
 291                                      uint64_t sector, size_t size)
 292 {
 293     uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
 294     uint64_t total_sectors;
 295
 296     if (sector & dev->sector_mask) {
 297         return false;
 298     }
 299     if (size % dev->conf->logical_block_size) {
 300         return false;
 301     }
 302     bdrv_get_geometry(dev->bs, &total_sectors);
 303     if (sector > total_sectors || nb_sectors > total_sectors - sector) {
 304         return false;
 305     }
 306     return true;
 307 }
 308
 309 static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 310 {
 311     BlockRequest *blkreq;
 312     uint64_t sector;
 313
 314     sector = virtio_ldq_p(VIRTIO_DEVICE(req->dev), &req->out.sector);
 315
 316     trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
 317
 318     if (!virtio_blk_sect_range_ok(req->dev, sector, req->qiov.size)) {
 319         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 320         virtio_blk_free_request(req);
 321         return;
 322     }
 323
 324     block_acct_start(bdrv_get_stats(req->dev->bs), &req->acct, req->qiov.size,
 325                      BLOCK_ACCT_WRITE);
 326
 327     if (mrb->num_writes == 32) {
 328         virtio_submit_multiwrite(req->dev->bs, mrb);
 329     }
 330
 331     blkreq = &mrb->blkreq[mrb->num_writes];
 332     blkreq->sector = sector;
 333     blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
 334     blkreq->qiov = &req->qiov;
 335     blkreq->cb = virtio_blk_rw_complete;
 336     blkreq->opaque = req;
 337     blkreq->error = 0;
 338
 339     mrb->num_writes++;
 340 }
 341
 342 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 343 {
 344     uint64_t sector;
 345
 346     sector = virtio_ldq_p(VIRTIO_DEVICE(req->dev), &req->out.sector);
 347
 348     trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512);
 349
 350     if (!virtio_blk_sect_range_ok(req->dev, sector, req->qiov.size)) {
 351         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 352         virtio_blk_free_request(req);
 353         return;
 354     }
 355
 356     block_acct_start(bdrv_get_stats(req->dev->bs), &req->acct, req->qiov.size,
 357                      BLOCK_ACCT_READ);
 358     bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
 359                    req->qiov.size / BDRV_SECTOR_SIZE,
 360                    virtio_blk_rw_complete, req);
 361 }
 362
 363 void virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 364 {
 365     uint32_t type;
 366     struct iovec *in_iov = req->elem.in_sg;
 367     struct iovec *iov = req->elem.out_sg;
 368     unsigned in_num = req->elem.in_num;
 369     unsigned out_num = req->elem.out_num;
 370
 371     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 372         error_report("virtio-blk missing headers");
 373         exit(1);
 374     }
 375
 376     if (unlikely(iov_to_buf(iov, out_num, 0, &req->out,
 377                             sizeof(req->out)) != sizeof(req->out))) {
 378         error_report("virtio-blk request outhdr too short");
 379         exit(1);
 380     }
 381
 382     iov_discard_front(&iov, &out_num, sizeof(req->out));
 383
 384     if (in_num < 1 ||
 385         in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
 386         error_report("virtio-blk request inhdr too short");
 387         exit(1);
 388     }
 389
 390     req->in = (void *)in_iov[in_num - 1].iov_base
 391               + in_iov[in_num - 1].iov_len
 392               - sizeof(struct virtio_blk_inhdr);
 393     iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr));
 394
 395     type = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type);
 396
 397     if (type & VIRTIO_BLK_T_FLUSH) {
 398         virtio_blk_handle_flush(req, mrb);
 399     } else if (type & VIRTIO_BLK_T_SCSI_CMD) {
 400         virtio_blk_handle_scsi(req);
 401     } else if (type & VIRTIO_BLK_T_GET_ID) {
 402         VirtIOBlock *s = req->dev;
 403
 404         /*
 405          * NB: per existing s/n string convention the string is
 406          * terminated by '\0' only when shorter than buffer.
 407          */
 408         const char *serial = s->blk.serial ? s->blk.serial : "";
 409         size_t size = MIN(strlen(serial) + 1,
 410                           MIN(iov_size(in_iov, in_num),
 411                               VIRTIO_BLK_ID_BYTES));
 412         iov_from_buf(in_iov, in_num, 0, serial, size);
 413         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 414         virtio_blk_free_request(req);
 415     } else if (type & VIRTIO_BLK_T_OUT) {
 416         qemu_iovec_init_external(&req->qiov, iov, out_num);
 417         virtio_blk_handle_write(req, mrb);
 418     } else if (type == VIRTIO_BLK_T_IN || type == VIRTIO_BLK_T_BARRIER) {
 419         /* VIRTIO_BLK_T_IN is 0, so we can't just & it. */
 420         qemu_iovec_init_external(&req->qiov, in_iov, in_num);
 421         virtio_blk_handle_read(req);
 422     } else {
 423         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 424         virtio_blk_free_request(req);
 425     }
 426 }
 427
 428 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 429 {
 430     VirtIOBlock *s = VIRTIO_BLK(vdev);
 431     VirtIOBlockReq *req;
 432     MultiReqBuffer mrb = {
 433         .num_writes = 0,
 434     };
 435
 436     /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
 437      * dataplane here instead of waiting for .set_status().
 438      */
 439     if (s->dataplane) {
 440         virtio_blk_data_plane_start(s->dataplane);
 441         return;
 442     }
 443
 444     while ((req = virtio_blk_get_request(s))) {
 445         virtio_blk_handle_request(req, &mrb);
 446     }
 447
 448     virtio_submit_multiwrite(s->bs, &mrb);
 449
 450     /*
 451      * FIXME: Want to check for completions before returning to guest mode,
 452      * so cached reads and writes are reported as quickly as possible. But
 453      * that should be done in the generic block layer.
 454      */
 455 }
 456
 457 static void virtio_blk_dma_restart_bh(void *opaque)
 458 {
 459     VirtIOBlock *s = opaque;
 460     VirtIOBlockReq *req = s->rq;
 461     MultiReqBuffer mrb = {
 462         .num_writes = 0,
 463     };
 464
 465     qemu_bh_delete(s->bh);
 466     s->bh = NULL;
 467
 468     s->rq = NULL;
 469
 470     while (req) {
 471         VirtIOBlockReq *next = req->next;
 472         virtio_blk_handle_request(req, &mrb);
 473         req = next;
 474     }
 475
 476     virtio_submit_multiwrite(s->bs, &mrb);
 477 }
 478
 479 static void virtio_blk_dma_restart_cb(void *opaque, int running,
 480                                       RunState state)
 481 {
 482     VirtIOBlock *s = opaque;
 483
 484     if (!running) {
 485         return;
 486     }
 487
 488     if (!s->bh) {
 489         s->bh = aio_bh_new(bdrv_get_aio_context(s->blk.conf.bs),
 490                            virtio_blk_dma_restart_bh, s);
 491         qemu_bh_schedule(s->bh);
 492     }
 493 }
 494
 495 static void virtio_blk_reset(VirtIODevice *vdev)
 496 {
 497     VirtIOBlock *s = VIRTIO_BLK(vdev);
 498
 499     if (s->dataplane) {
 500         virtio_blk_data_plane_stop(s->dataplane);
 501     }
 502
 503     /*
 504      * This should cancel pending requests, but can't do nicely until there
 505      * are per-device request lists.
 506      */
 507     bdrv_drain_all();
 508     bdrv_set_enable_write_cache(s->bs, s->original_wce);
 509 }
 510
 511 /* coalesce internal state, copy to pci i/o region 0
 512  */
 513 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 514 {
 515     VirtIOBlock *s = VIRTIO_BLK(vdev);
 516     struct virtio_blk_config blkcfg;
 517     uint64_t capacity;
 518     int blk_size = s->conf->logical_block_size;
 519
 520     bdrv_get_geometry(s->bs, &capacity);
 521     memset(&blkcfg, 0, sizeof(blkcfg));
 522     virtio_stq_p(vdev, &blkcfg.capacity, capacity);
 523     virtio_stl_p(vdev, &blkcfg.seg_max, 128 - 2);
 524     virtio_stw_p(vdev, &blkcfg.cylinders, s->conf->cyls);
 525     virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
 526     virtio_stw_p(vdev, &blkcfg.min_io_size, s->conf->min_io_size / blk_size);
 527     virtio_stw_p(vdev, &blkcfg.opt_io_size, s->conf->opt_io_size / blk_size);
 528     blkcfg.heads = s->conf->heads;
 529     /*
 530      * We must ensure that the block device capacity is a multiple of
 531      * the logical block size. If that is not the case, let's use
 532      * sector_mask to adopt the geometry to have a correct picture.
 533      * For those devices where the capacity is ok for the given geometry
 534      * we don't touch the sector value of the geometry, since some devices
 535      * (like s390 dasd) need a specific value. Here the capacity is already
 536      * cyls*heads*secs*blk_size and the sector value is not block size
 537      * divided by 512 - instead it is the amount of blk_size blocks
 538      * per track (cylinder).
 539      */
 540     if (bdrv_getlength(s->bs) /  s->conf->heads / s->conf->secs % blk_size) {
 541         blkcfg.sectors = s->conf->secs & ~s->sector_mask;
 542     } else {
 543         blkcfg.sectors = s->conf->secs;
 544     }
 545     blkcfg.size_max = 0;
 546     blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
 547     blkcfg.alignment_offset = 0;
 548     blkcfg.wce = bdrv_enable_write_cache(s->bs);
 549     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
 550 }
 551
 552 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
 553 {
 554     VirtIOBlock *s = VIRTIO_BLK(vdev);
 555     struct virtio_blk_config blkcfg;
 556
 557     memcpy(&blkcfg, config, sizeof(blkcfg));
 558
 559     aio_context_acquire(bdrv_get_aio_context(s->bs));
 560     bdrv_set_enable_write_cache(s->bs, blkcfg.wce != 0);
 561     aio_context_release(bdrv_get_aio_context(s->bs));
 562 }
 563
 564 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 565 {
 566     VirtIOBlock *s = VIRTIO_BLK(vdev);
 567
 568     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 569     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 570     features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
 571     features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
 572     features |= (1 << VIRTIO_BLK_F_SCSI);
 573
 574     if (s->blk.config_wce) {
 575         features |= (1 << VIRTIO_BLK_F_CONFIG_WCE);
 576     }
 577     if (bdrv_enable_write_cache(s->bs))
 578         features |= (1 << VIRTIO_BLK_F_WCE);
 579
 580     if (bdrv_is_read_only(s->bs))
 581         features |= 1 << VIRTIO_BLK_F_RO;
 582
 583     return features;
 584 }
 585
 586 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
 587 {
 588     VirtIOBlock *s = VIRTIO_BLK(vdev);
 589     uint32_t features;
 590
 591     if (s->dataplane && !(status & (VIRTIO_CONFIG_S_DRIVER |
 592                                     VIRTIO_CONFIG_S_DRIVER_OK))) {
 593         virtio_blk_data_plane_stop(s->dataplane);
 594     }
 595
 596     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 597         return;
 598     }
 599
 600     features = vdev->guest_features;
 601
 602     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
 603      * cache flushes.  Thus, the "auto writethrough" behavior is never
 604      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
 605      * Leaving it enabled would break the following sequence:
 606      *
 607      *     Guest started with "-drive cache=writethrough"
 608      *     Guest sets status to 0
 609      *     Guest sets DRIVER bit in status field
 610      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
 611      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
 612      *     Guest writes 1 to the WCE configuration field (writeback mode)
 613      *     Guest sets DRIVER_OK bit in status field
 614      *
 615      * s->bs would erroneously be placed in writethrough mode.
 616      */
 617     if (!(features & (1 << VIRTIO_BLK_F_CONFIG_WCE))) {
 618         aio_context_acquire(bdrv_get_aio_context(s->bs));
 619         bdrv_set_enable_write_cache(s->bs,
 620                                     !!(features & (1 << VIRTIO_BLK_F_WCE)));
 621         aio_context_release(bdrv_get_aio_context(s->bs));
 622     }
 623 }
 624
 625 static void virtio_blk_save(QEMUFile *f, void *opaque)
 626 {
 627     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
 628
 629     virtio_save(vdev, f);
 630 }
 631
 632 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
 633 {
 634     VirtIOBlock *s = VIRTIO_BLK(vdev);
 635     VirtIOBlockReq *req = s->rq;
 636
 637     while (req) {
 638         qemu_put_sbyte(f, 1);
 639         qemu_put_buffer(f, (unsigned char *)&req->elem,
 640                         sizeof(VirtQueueElement));
 641         req = req->next;
 642     }
 643     qemu_put_sbyte(f, 0);
 644 }
 645
 646 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 647 {
 648     VirtIOBlock *s = opaque;
 649     VirtIODevice *vdev = VIRTIO_DEVICE(s);
 650
 651     if (version_id != 2)
 652         return -EINVAL;
 653
 654     return virtio_load(vdev, f, version_id);
 655 }
 656
 657 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
 658                                   int version_id)
 659 {
 660     VirtIOBlock *s = VIRTIO_BLK(vdev);
 661
 662     while (qemu_get_sbyte(f)) {
 663         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 664         qemu_get_buffer(f, (unsigned char *)&req->elem,
 665                         sizeof(VirtQueueElement));
 666         req->next = s->rq;
 667         s->rq = req;
 668
 669         virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr,
 670             req->elem.in_num, 1);
 671         virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr,
 672             req->elem.out_num, 0);
 673     }
 674
 675     return 0;
 676 }
 677
 678 static void virtio_blk_resize(void *opaque)
 679 {
 680     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
 681
 682     virtio_notify_config(vdev);
 683 }
 684
 685 static const BlockDevOps virtio_block_ops = {
 686     .resize_cb = virtio_blk_resize,
 687 };
 688
 689 /* Disable dataplane thread during live migration since it does not
 690  * update the dirty memory bitmap yet.
 691  */
 692 static void virtio_blk_migration_state_changed(Notifier *notifier, void *data)
 693 {
 694     VirtIOBlock *s = container_of(notifier, VirtIOBlock,
 695                                   migration_state_notifier);
 696     MigrationState *mig = data;
 697     Error *err = NULL;
 698
 699     if (migration_in_setup(mig)) {
 700         if (!s->dataplane) {
 701             return;
 702         }
 703         virtio_blk_data_plane_destroy(s->dataplane);
 704         s->dataplane = NULL;
 705     } else if (migration_has_finished(mig) ||
 706                migration_has_failed(mig)) {
 707         if (s->dataplane) {
 708             return;
 709         }
 710         bdrv_drain_all(); /* complete in-flight non-dataplane requests */
 711         virtio_blk_data_plane_create(VIRTIO_DEVICE(s), &s->blk,
 712                                      &s->dataplane, &err);
 713         if (err != NULL) {
 714             error_report("%s", error_get_pretty(err));
 715             error_free(err);
 716         }
 717     }
 718 }
 719
 720 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
 721 {
 722     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 723     VirtIOBlock *s = VIRTIO_BLK(dev);
 724     VirtIOBlkConf *blk = &(s->blk);
 725     Error *err = NULL;
 726     static int virtio_blk_id;
 727
 728     if (!blk->conf.bs) {
 729         error_setg(errp, "drive property not set");
 730         return;
 731     }
 732     if (!bdrv_is_inserted(blk->conf.bs)) {
 733         error_setg(errp, "Device needs media, but drive is empty");
 734         return;
 735     }
 736
 737     blkconf_serial(&blk->conf, &blk->serial);
 738     s->original_wce = bdrv_enable_write_cache(blk->conf.bs);
 739     blkconf_geometry(&blk->conf, NULL, 65535, 255, 255, &err);
 740     if (err) {
 741         error_propagate(errp, err);
 742         return;
 743     }
 744
 745     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK,
 746                 sizeof(struct virtio_blk_config));
 747
 748     s->bs = blk->conf.bs;
 749     s->conf = &blk->conf;
 750     s->rq = NULL;
 751     s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
 752
 753     s->vq = virtio_add_queue(vdev, 128, virtio_blk_handle_output);
 754     s->complete_request = virtio_blk_complete_request;
 755     virtio_blk_data_plane_create(vdev, blk, &s->dataplane, &err);
 756     if (err != NULL) {
 757         error_propagate(errp, err);
 758         virtio_cleanup(vdev);
 759         return;
 760     }
 761     s->migration_state_notifier.notify = virtio_blk_migration_state_changed;
 762     add_migration_state_change_notifier(&s->migration_state_notifier);
 763
 764     s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 765     register_savevm(dev, "virtio-blk", virtio_blk_id++, 2,
 766                     virtio_blk_save, virtio_blk_load, s);
 767     bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
 768     bdrv_set_guest_block_size(s->bs, s->conf->logical_block_size);
 769
 770     bdrv_iostatus_enable(s->bs);
 771
 772     add_boot_device_path(s->conf->bootindex, dev, "/disk@0,0");
 773 }
 774
 775 static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp)
 776 {
 777     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 778     VirtIOBlock *s = VIRTIO_BLK(dev);
 779
 780     remove_migration_state_change_notifier(&s->migration_state_notifier);
 781     virtio_blk_data_plane_destroy(s->dataplane);
 782     s->dataplane = NULL;
 783     qemu_del_vm_change_state_handler(s->change);
 784     unregister_savevm(dev, "virtio-blk", s);
 785     blockdev_mark_auto_del(s->bs);
 786     virtio_cleanup(vdev);
 787 }
 788
 789 static void virtio_blk_instance_init(Object *obj)
 790 {
 791     VirtIOBlock *s = VIRTIO_BLK(obj);
 792
 793     object_property_add_link(obj, "iothread", TYPE_IOTHREAD,
 794                              (Object **)&s->blk.iothread,
 795                              qdev_prop_allow_set_link_before_realize,
 796                              OBJ_PROP_LINK_UNREF_ON_RELEASE, NULL);
 797 }
 798
 799 static Property virtio_blk_properties[] = {
 800     DEFINE_BLOCK_PROPERTIES(VirtIOBlock, blk.conf),
 801     DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, blk.conf),
 802     DEFINE_PROP_STRING("serial", VirtIOBlock, blk.serial),
 803     DEFINE_PROP_BIT("config-wce", VirtIOBlock, blk.config_wce, 0, true),
 804 #ifdef __linux__
 805     DEFINE_PROP_BIT("scsi", VirtIOBlock, blk.scsi, 0, true),
 806 #endif
 807     DEFINE_PROP_BIT("x-data-plane", VirtIOBlock, blk.data_plane, 0, false),
 808     DEFINE_PROP_END_OF_LIST(),
 809 };
 810
 811 static void virtio_blk_class_init(ObjectClass *klass, void *data)
 812 {
 813     DeviceClass *dc = DEVICE_CLASS(klass);
 814     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
 815
 816     dc->props = virtio_blk_properties;
 817     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
 818     vdc->realize = virtio_blk_device_realize;
 819     vdc->unrealize = virtio_blk_device_unrealize;
 820     vdc->get_config = virtio_blk_update_config;
 821     vdc->set_config = virtio_blk_set_config;
 822     vdc->get_features = virtio_blk_get_features;
 823     vdc->set_status = virtio_blk_set_status;
 824     vdc->reset = virtio_blk_reset;
 825     vdc->save = virtio_blk_save_device;
 826     vdc->load = virtio_blk_load_device;
 827 }
 828
 829 static const TypeInfo virtio_device_info = {
 830     .name = TYPE_VIRTIO_BLK,
 831     .parent = TYPE_VIRTIO_DEVICE,
 832     .instance_size = sizeof(VirtIOBlock),
 833     .instance_init = virtio_blk_instance_init,
 834     .class_init = virtio_blk_class_init,
 835 };
 836
 837 static void virtio_register_types(void)
 838 {
 839     type_register_static(&virtio_device_info);
 840 }
 841
 842 type_init(virtio_register_types)