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 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
  22 # include "dataplane/virtio-blk.h"
  23 # include "migration/migration.h"
  24 #endif
  25 #include "block/scsi.h"
  26 #ifdef __linux__
  27 # include <scsi/sg.h>
  28 #endif
  29 #include "hw/virtio/virtio-bus.h"
  30 #include "hw/virtio/virtio-access.h"
  31
  32 VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
  33 {
  34     VirtIOBlockReq *req = g_slice_new(VirtIOBlockReq);
  35     req->dev = s;
  36     req->qiov.size = 0;
  37     req->next = NULL;
  38     return req;
  39 }
  40
  41 void virtio_blk_free_request(VirtIOBlockReq *req)
  42 {
  43     if (req) {
  44         g_slice_free(VirtIOBlockReq, req);
  45     }
  46 }
  47
  48 static void virtio_blk_complete_request(VirtIOBlockReq *req,
  49                                         unsigned char status)
  50 {
  51     VirtIOBlock *s = req->dev;
  52     VirtIODevice *vdev = VIRTIO_DEVICE(s);
  53
  54     trace_virtio_blk_req_complete(req, status);
  55
  56     stb_p(&req->in->status, status);
  57     virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
  58     virtio_notify(vdev, s->vq);
  59 }
  60
  61 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
  62 {
  63     req->dev->complete_request(req, status);
  64 }
  65
  66 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
  67     bool is_read)
  68 {
  69     BlockErrorAction action = bdrv_get_error_action(req->dev->bs, is_read, error);
  70     VirtIOBlock *s = req->dev;
  71
  72     if (action == BLOCK_ERROR_ACTION_STOP) {
  73         req->next = s->rq;
  74         s->rq = req;
  75     } else if (action == BLOCK_ERROR_ACTION_REPORT) {
  76         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
  77         block_acct_done(bdrv_get_stats(s->bs), &req->acct);
  78         virtio_blk_free_request(req);
  79     }
  80
  81     bdrv_error_action(s->bs, action, is_read, error);
  82     return action != BLOCK_ERROR_ACTION_IGNORE;
  83 }
  84
  85 static void virtio_blk_rw_complete(void *opaque, int ret)
  86 {
  87     VirtIOBlockReq *req = opaque;
  88
  89     trace_virtio_blk_rw_complete(req, ret);
  90
  91     if (ret) {
  92         int p = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type);
  93         bool is_read = !(p & VIRTIO_BLK_T_OUT);
  94         if (virtio_blk_handle_rw_error(req, -ret, is_read))
  95             return;
  96     }
  97
  98     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
  99     block_acct_done(bdrv_get_stats(req->dev->bs), &req->acct);
 100     virtio_blk_free_request(req);
 101 }
 102
 103 static void virtio_blk_flush_complete(void *opaque, int ret)
 104 {
 105     VirtIOBlockReq *req = opaque;
 106
 107     if (ret) {
 108         if (virtio_blk_handle_rw_error(req, -ret, 0)) {
 109             return;
 110         }
 111     }
 112
 113     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 114     block_acct_done(bdrv_get_stats(req->dev->bs), &req->acct);
 115     virtio_blk_free_request(req);
 116 }
 117
 118 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 119 {
 120     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 121
 122     if (!virtqueue_pop(s->vq, &req->elem)) {
 123         virtio_blk_free_request(req);
 124         return NULL;
 125     }
 126
 127     return req;
 128 }
 129
 130 int virtio_blk_handle_scsi_req(VirtIOBlock *blk,
 131                                VirtQueueElement *elem)
 132 {
 133     int status = VIRTIO_BLK_S_OK;
 134     struct virtio_scsi_inhdr *scsi = NULL;
 135     VirtIODevice *vdev = VIRTIO_DEVICE(blk);
 136
 137 #ifdef __linux__
 138     int i;
 139     struct sg_io_hdr hdr;
 140 #endif
 141
 142     /*
 143      * We require at least one output segment each for the virtio_blk_outhdr
 144      * and the SCSI command block.
 145      *
 146      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 147      * and the sense buffer pointer in the input segments.
 148      */
 149     if (elem->out_num < 2 || elem->in_num < 3) {
 150         status = VIRTIO_BLK_S_IOERR;
 151         goto fail;
 152     }
 153
 154     /*
 155      * The scsi inhdr is placed in the second-to-last input segment, just
 156      * before the regular inhdr.
 157      */
 158     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
 159
 160     if (!blk->blk.scsi) {
 161         status = VIRTIO_BLK_S_UNSUPP;
 162         goto fail;
 163     }
 164
 165     /*
 166      * No support for bidirection commands yet.
 167      */
 168     if (elem->out_num > 2 && elem->in_num > 3) {
 169         status = VIRTIO_BLK_S_UNSUPP;
 170         goto fail;
 171     }
 172
 173 #ifdef __linux__
 174     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 175     hdr.interface_id = 'S';
 176     hdr.cmd_len = elem->out_sg[1].iov_len;
 177     hdr.cmdp = elem->out_sg[1].iov_base;
 178     hdr.dxfer_len = 0;
 179
 180     if (elem->out_num > 2) {
 181         /*
 182          * If there are more than the minimally required 2 output segments
 183          * there is write payload starting from the third iovec.
 184          */
 185         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 186         hdr.iovec_count = elem->out_num - 2;
 187
 188         for (i = 0; i < hdr.iovec_count; i++)
 189             hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
 190
 191         hdr.dxferp = elem->out_sg + 2;
 192
 193     } else if (elem->in_num > 3) {
 194         /*
 195          * If we have more than 3 input segments the guest wants to actually
 196          * read data.
 197          */
 198         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 199         hdr.iovec_count = elem->in_num - 3;
 200         for (i = 0; i < hdr.iovec_count; i++)
 201             hdr.dxfer_len += elem->in_sg[i].iov_len;
 202
 203         hdr.dxferp = elem->in_sg;
 204     } else {
 205         /*
 206          * Some SCSI commands don't actually transfer any data.
 207          */
 208         hdr.dxfer_direction = SG_DXFER_NONE;
 209     }
 210
 211     hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
 212     hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
 213
 214     status = bdrv_ioctl(blk->bs, SG_IO, &hdr);
 215     if (status) {
 216         status = VIRTIO_BLK_S_UNSUPP;
 217         goto fail;
 218     }
 219
 220     /*
 221      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
 222      * clear the masked_status field [hence status gets cleared too, see
 223      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
 224      * status has occurred.  However they do set DRIVER_SENSE in driver_status
 225      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
 226      */
 227     if (hdr.status == 0 && hdr.sb_len_wr > 0) {
 228         hdr.status = CHECK_CONDITION;
 229     }
 230
 231     virtio_stl_p(vdev, &scsi->errors,
 232                  hdr.status | (hdr.msg_status << 8) |
 233                  (hdr.host_status << 16) | (hdr.driver_status << 24));
 234     virtio_stl_p(vdev, &scsi->residual, hdr.resid);
 235     virtio_stl_p(vdev, &scsi->sense_len, hdr.sb_len_wr);
 236     virtio_stl_p(vdev, &scsi->data_len, hdr.dxfer_len);
 237
 238     return status;
 239 #else
 240     abort();
 241 #endif
 242
 243 fail:
 244     /* Just put anything nonzero so that the ioctl fails in the guest.  */
 245     if (scsi) {
 246         virtio_stl_p(vdev, &scsi->errors, 255);
 247     }
 248     return status;
 249 }
 250
 251 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 252 {
 253     int status;
 254
 255     status = virtio_blk_handle_scsi_req(req->dev, &req->elem);
 256     virtio_blk_req_complete(req, status);
 257     virtio_blk_free_request(req);
 258 }
 259
 260 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     block_acct_start(bdrv_get_stats(req->dev->bs), &req->acct, 0,
 283                      BLOCK_ACCT_FLUSH);
 284
 285     /*
 286      * Make sure all outstanding writes are posted to the backing device.
 287      */
 288     virtio_submit_multiwrite(req->dev->bs, mrb);
 289     bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 290 }
 291
 292 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
 293                                      uint64_t sector, size_t size)
 294 {
 295     uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
 296     uint64_t total_sectors;
 297
 298     if (sector & dev->sector_mask) {
 299         return false;
 300     }
 301     if (size % dev->conf->logical_block_size) {
 302         return false;
 303     }
 304     bdrv_get_geometry(dev->bs, &total_sectors);
 305     if (sector > total_sectors || nb_sectors > total_sectors - sector) {
 306         return false;
 307     }
 308     return true;
 309 }
 310
 311 static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 312 {
 313     BlockRequest *blkreq;
 314     uint64_t sector;
 315
 316     sector = virtio_ldq_p(VIRTIO_DEVICE(req->dev), &req->out.sector);
 317
 318     trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
 319
 320     if (!virtio_blk_sect_range_ok(req->dev, sector, req->qiov.size)) {
 321         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 322         virtio_blk_free_request(req);
 323         return;
 324     }
 325
 326     block_acct_start(bdrv_get_stats(req->dev->bs), &req->acct, req->qiov.size,
 327                      BLOCK_ACCT_WRITE);
 328
 329     if (mrb->num_writes == 32) {
 330         virtio_submit_multiwrite(req->dev->bs, mrb);
 331     }
 332
 333     blkreq = &mrb->blkreq[mrb->num_writes];
 334     blkreq->sector = sector;
 335     blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
 336     blkreq->qiov = &req->qiov;
 337     blkreq->cb = virtio_blk_rw_complete;
 338     blkreq->opaque = req;
 339     blkreq->error = 0;
 340
 341     mrb->num_writes++;
 342 }
 343
 344 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 345 {
 346     uint64_t sector;
 347
 348     sector = virtio_ldq_p(VIRTIO_DEVICE(req->dev), &req->out.sector);
 349
 350     trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512);
 351
 352     if (!virtio_blk_sect_range_ok(req->dev, sector, req->qiov.size)) {
 353         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 354         virtio_blk_free_request(req);
 355         return;
 356     }
 357
 358     block_acct_start(bdrv_get_stats(req->dev->bs), &req->acct, req->qiov.size,
 359                      BLOCK_ACCT_READ);
 360     bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
 361                    req->qiov.size / BDRV_SECTOR_SIZE,
 362                    virtio_blk_rw_complete, req);
 363 }
 364
 365 void virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 366 {
 367     uint32_t type;
 368     struct iovec *in_iov = req->elem.in_sg;
 369     struct iovec *iov = req->elem.out_sg;
 370     unsigned in_num = req->elem.in_num;
 371     unsigned out_num = req->elem.out_num;
 372
 373     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 374         error_report("virtio-blk missing headers");
 375         exit(1);
 376     }
 377
 378     if (unlikely(iov_to_buf(iov, out_num, 0, &req->out,
 379                             sizeof(req->out)) != sizeof(req->out))) {
 380         error_report("virtio-blk request outhdr too short");
 381         exit(1);
 382     }
 383
 384     iov_discard_front(&iov, &out_num, sizeof(req->out));
 385
 386     if (in_num < 1 ||
 387         in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
 388         error_report("virtio-blk request inhdr too short");
 389         exit(1);
 390     }
 391
 392     req->in = (void *)in_iov[in_num - 1].iov_base
 393               + in_iov[in_num - 1].iov_len
 394               - sizeof(struct virtio_blk_inhdr);
 395     iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr));
 396
 397     type = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type);
 398
 399     if (type & VIRTIO_BLK_T_FLUSH) {
 400         virtio_blk_handle_flush(req, mrb);
 401     } else if (type & VIRTIO_BLK_T_SCSI_CMD) {
 402         virtio_blk_handle_scsi(req);
 403     } else if (type & VIRTIO_BLK_T_GET_ID) {
 404         VirtIOBlock *s = req->dev;
 405
 406         /*
 407          * NB: per existing s/n string convention the string is
 408          * terminated by '\0' only when shorter than buffer.
 409          */
 410         const char *serial = s->blk.serial ? s->blk.serial : "";
 411         size_t size = MIN(strlen(serial) + 1,
 412                           MIN(iov_size(in_iov, in_num),
 413                               VIRTIO_BLK_ID_BYTES));
 414         iov_from_buf(in_iov, in_num, 0, serial, size);
 415         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 416         virtio_blk_free_request(req);
 417     } else if (type & VIRTIO_BLK_T_OUT) {
 418         qemu_iovec_init_external(&req->qiov, iov, out_num);
 419         virtio_blk_handle_write(req, mrb);
 420     } else if (type == VIRTIO_BLK_T_IN || type == VIRTIO_BLK_T_BARRIER) {
 421         /* VIRTIO_BLK_T_IN is 0, so we can't just & it. */
 422         qemu_iovec_init_external(&req->qiov, in_iov, in_num);
 423         virtio_blk_handle_read(req);
 424     } else {
 425         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 426         virtio_blk_free_request(req);
 427     }
 428 }
 429
 430 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 431 {
 432     VirtIOBlock *s = VIRTIO_BLK(vdev);
 433     VirtIOBlockReq *req;
 434     MultiReqBuffer mrb = {
 435         .num_writes = 0,
 436     };
 437
 438 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 439     /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
 440      * dataplane here instead of waiting for .set_status().
 441      */
 442     if (s->dataplane) {
 443         virtio_blk_data_plane_start(s->dataplane);
 444         return;
 445     }
 446 #endif
 447
 448     while ((req = virtio_blk_get_request(s))) {
 449         virtio_blk_handle_request(req, &mrb);
 450     }
 451
 452     virtio_submit_multiwrite(s->bs, &mrb);
 453
 454     /*
 455      * FIXME: Want to check for completions before returning to guest mode,
 456      * so cached reads and writes are reported as quickly as possible. But
 457      * that should be done in the generic block layer.
 458      */
 459 }
 460
 461 static void virtio_blk_dma_restart_bh(void *opaque)
 462 {
 463     VirtIOBlock *s = opaque;
 464     VirtIOBlockReq *req = s->rq;
 465     MultiReqBuffer mrb = {
 466         .num_writes = 0,
 467     };
 468
 469     qemu_bh_delete(s->bh);
 470     s->bh = NULL;
 471
 472     s->rq = NULL;
 473
 474     while (req) {
 475         VirtIOBlockReq *next = req->next;
 476         virtio_blk_handle_request(req, &mrb);
 477         req = next;
 478     }
 479
 480     virtio_submit_multiwrite(s->bs, &mrb);
 481 }
 482
 483 static void virtio_blk_dma_restart_cb(void *opaque, int running,
 484                                       RunState state)
 485 {
 486     VirtIOBlock *s = opaque;
 487
 488     if (!running) {
 489         return;
 490     }
 491
 492     if (!s->bh) {
 493         s->bh = aio_bh_new(bdrv_get_aio_context(s->blk.conf.bs),
 494                            virtio_blk_dma_restart_bh, s);
 495         qemu_bh_schedule(s->bh);
 496     }
 497 }
 498
 499 static void virtio_blk_reset(VirtIODevice *vdev)
 500 {
 501     VirtIOBlock *s = VIRTIO_BLK(vdev);
 502
 503 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 504     if (s->dataplane) {
 505         virtio_blk_data_plane_stop(s->dataplane);
 506     }
 507 #endif
 508
 509     /*
 510      * This should cancel pending requests, but can't do nicely until there
 511      * are per-device request lists.
 512      */
 513     bdrv_drain_all();
 514     bdrv_set_enable_write_cache(s->bs, s->original_wce);
 515 }
 516
 517 /* coalesce internal state, copy to pci i/o region 0
 518  */
 519 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 520 {
 521     VirtIOBlock *s = VIRTIO_BLK(vdev);
 522     struct virtio_blk_config blkcfg;
 523     uint64_t capacity;
 524     int blk_size = s->conf->logical_block_size;
 525
 526     bdrv_get_geometry(s->bs, &capacity);
 527     memset(&blkcfg, 0, sizeof(blkcfg));
 528     virtio_stq_p(vdev, &blkcfg.capacity, capacity);
 529     virtio_stl_p(vdev, &blkcfg.seg_max, 128 - 2);
 530     virtio_stw_p(vdev, &blkcfg.cylinders, s->conf->cyls);
 531     virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
 532     virtio_stw_p(vdev, &blkcfg.min_io_size, s->conf->min_io_size / blk_size);
 533     virtio_stw_p(vdev, &blkcfg.opt_io_size, s->conf->opt_io_size / blk_size);
 534     blkcfg.heads = s->conf->heads;
 535     /*
 536      * We must ensure that the block device capacity is a multiple of
 537      * the logical block size. If that is not the case, let's use
 538      * sector_mask to adopt the geometry to have a correct picture.
 539      * For those devices where the capacity is ok for the given geometry
 540      * we don't touch the sector value of the geometry, since some devices
 541      * (like s390 dasd) need a specific value. Here the capacity is already
 542      * cyls*heads*secs*blk_size and the sector value is not block size
 543      * divided by 512 - instead it is the amount of blk_size blocks
 544      * per track (cylinder).
 545      */
 546     if (bdrv_getlength(s->bs) /  s->conf->heads / s->conf->secs % blk_size) {
 547         blkcfg.sectors = s->conf->secs & ~s->sector_mask;
 548     } else {
 549         blkcfg.sectors = s->conf->secs;
 550     }
 551     blkcfg.size_max = 0;
 552     blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
 553     blkcfg.alignment_offset = 0;
 554     blkcfg.wce = bdrv_enable_write_cache(s->bs);
 555     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
 556 }
 557
 558 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
 559 {
 560     VirtIOBlock *s = VIRTIO_BLK(vdev);
 561     struct virtio_blk_config blkcfg;
 562
 563     memcpy(&blkcfg, config, sizeof(blkcfg));
 564
 565     aio_context_acquire(bdrv_get_aio_context(s->bs));
 566     bdrv_set_enable_write_cache(s->bs, blkcfg.wce != 0);
 567     aio_context_release(bdrv_get_aio_context(s->bs));
 568 }
 569
 570 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 571 {
 572     VirtIOBlock *s = VIRTIO_BLK(vdev);
 573
 574     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 575     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 576     features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
 577     features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
 578     features |= (1 << VIRTIO_BLK_F_SCSI);
 579
 580     if (s->blk.config_wce) {
 581         features |= (1 << VIRTIO_BLK_F_CONFIG_WCE);
 582     }
 583     if (bdrv_enable_write_cache(s->bs))
 584         features |= (1 << VIRTIO_BLK_F_WCE);
 585
 586     if (bdrv_is_read_only(s->bs))
 587         features |= 1 << VIRTIO_BLK_F_RO;
 588
 589     return features;
 590 }
 591
 592 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
 593 {
 594     VirtIOBlock *s = VIRTIO_BLK(vdev);
 595     uint32_t features;
 596
 597 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 598     if (s->dataplane && !(status & (VIRTIO_CONFIG_S_DRIVER |
 599                                     VIRTIO_CONFIG_S_DRIVER_OK))) {
 600         virtio_blk_data_plane_stop(s->dataplane);
 601     }
 602 #endif
 603
 604     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 605         return;
 606     }
 607
 608     features = vdev->guest_features;
 609
 610     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
 611      * cache flushes.  Thus, the "auto writethrough" behavior is never
 612      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
 613      * Leaving it enabled would break the following sequence:
 614      *
 615      *     Guest started with "-drive cache=writethrough"
 616      *     Guest sets status to 0
 617      *     Guest sets DRIVER bit in status field
 618      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
 619      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
 620      *     Guest writes 1 to the WCE configuration field (writeback mode)
 621      *     Guest sets DRIVER_OK bit in status field
 622      *
 623      * s->bs would erroneously be placed in writethrough mode.
 624      */
 625     if (!(features & (1 << VIRTIO_BLK_F_CONFIG_WCE))) {
 626         aio_context_acquire(bdrv_get_aio_context(s->bs));
 627         bdrv_set_enable_write_cache(s->bs,
 628                                     !!(features & (1 << VIRTIO_BLK_F_WCE)));
 629         aio_context_release(bdrv_get_aio_context(s->bs));
 630     }
 631 }
 632
 633 static void virtio_blk_save(QEMUFile *f, void *opaque)
 634 {
 635     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
 636
 637     virtio_save(vdev, f);
 638 }
 639
 640 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
 641 {
 642     VirtIOBlock *s = VIRTIO_BLK(vdev);
 643     VirtIOBlockReq *req = s->rq;
 644
 645     while (req) {
 646         qemu_put_sbyte(f, 1);
 647         qemu_put_buffer(f, (unsigned char *)&req->elem,
 648                         sizeof(VirtQueueElement));
 649         req = req->next;
 650     }
 651     qemu_put_sbyte(f, 0);
 652 }
 653
 654 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 655 {
 656     VirtIOBlock *s = opaque;
 657     VirtIODevice *vdev = VIRTIO_DEVICE(s);
 658
 659     if (version_id != 2)
 660         return -EINVAL;
 661
 662     return virtio_load(vdev, f, version_id);
 663 }
 664
 665 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
 666                                   int version_id)
 667 {
 668     VirtIOBlock *s = VIRTIO_BLK(vdev);
 669
 670     while (qemu_get_sbyte(f)) {
 671         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 672         qemu_get_buffer(f, (unsigned char *)&req->elem,
 673                         sizeof(VirtQueueElement));
 674         req->next = s->rq;
 675         s->rq = req;
 676
 677         virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr,
 678             req->elem.in_num, 1);
 679         virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr,
 680             req->elem.out_num, 0);
 681     }
 682
 683     return 0;
 684 }
 685
 686 static void virtio_blk_resize(void *opaque)
 687 {
 688     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
 689
 690     virtio_notify_config(vdev);
 691 }
 692
 693 static const BlockDevOps virtio_block_ops = {
 694     .resize_cb = virtio_blk_resize,
 695 };
 696
 697 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 698 /* Disable dataplane thread during live migration since it does not
 699  * update the dirty memory bitmap yet.
 700  */
 701 static void virtio_blk_migration_state_changed(Notifier *notifier, void *data)
 702 {
 703     VirtIOBlock *s = container_of(notifier, VirtIOBlock,
 704                                   migration_state_notifier);
 705     MigrationState *mig = data;
 706     Error *err = NULL;
 707
 708     if (migration_in_setup(mig)) {
 709         if (!s->dataplane) {
 710             return;
 711         }
 712         virtio_blk_data_plane_destroy(s->dataplane);
 713         s->dataplane = NULL;
 714     } else if (migration_has_finished(mig) ||
 715                migration_has_failed(mig)) {
 716         if (s->dataplane) {
 717             return;
 718         }
 719         bdrv_drain_all(); /* complete in-flight non-dataplane requests */
 720         virtio_blk_data_plane_create(VIRTIO_DEVICE(s), &s->blk,
 721                                      &s->dataplane, &err);
 722         if (err != NULL) {
 723             error_report("%s", error_get_pretty(err));
 724             error_free(err);
 725         }
 726     }
 727 }
 728 #endif /* CONFIG_VIRTIO_BLK_DATA_PLANE */
 729
 730 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
 731 {
 732     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 733     VirtIOBlock *s = VIRTIO_BLK(dev);
 734     VirtIOBlkConf *blk = &(s->blk);
 735     Error *err = NULL;
 736     static int virtio_blk_id;
 737
 738     if (!blk->conf.bs) {
 739         error_setg(errp, "drive property not set");
 740         return;
 741     }
 742     if (!bdrv_is_inserted(blk->conf.bs)) {
 743         error_setg(errp, "Device needs media, but drive is empty");
 744         return;
 745     }
 746
 747     blkconf_serial(&blk->conf, &blk->serial);
 748     s->original_wce = bdrv_enable_write_cache(blk->conf.bs);
 749     blkconf_geometry(&blk->conf, NULL, 65535, 255, 255, &err);
 750     if (err) {
 751         error_propagate(errp, err);
 752         return;
 753     }
 754
 755     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK,
 756                 sizeof(struct virtio_blk_config));
 757
 758     s->bs = blk->conf.bs;
 759     s->conf = &blk->conf;
 760     s->rq = NULL;
 761     s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
 762
 763     s->vq = virtio_add_queue(vdev, 128, virtio_blk_handle_output);
 764     s->complete_request = virtio_blk_complete_request;
 765 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 766     virtio_blk_data_plane_create(vdev, blk, &s->dataplane, &err);
 767     if (err != NULL) {
 768         error_propagate(errp, err);
 769         virtio_cleanup(vdev);
 770         return;
 771     }
 772     s->migration_state_notifier.notify = virtio_blk_migration_state_changed;
 773     add_migration_state_change_notifier(&s->migration_state_notifier);
 774 #endif
 775
 776     s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 777     register_savevm(dev, "virtio-blk", virtio_blk_id++, 2,
 778                     virtio_blk_save, virtio_blk_load, s);
 779     bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
 780     bdrv_set_guest_block_size(s->bs, s->conf->logical_block_size);
 781
 782     bdrv_iostatus_enable(s->bs);
 783
 784     add_boot_device_path(s->conf->bootindex, dev, "/disk@0,0");
 785 }
 786
 787 static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp)
 788 {
 789     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
 790     VirtIOBlock *s = VIRTIO_BLK(dev);
 791
 792 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 793     remove_migration_state_change_notifier(&s->migration_state_notifier);
 794     virtio_blk_data_plane_destroy(s->dataplane);
 795     s->dataplane = NULL;
 796 #endif
 797     qemu_del_vm_change_state_handler(s->change);
 798     unregister_savevm(dev, "virtio-blk", s);
 799     blockdev_mark_auto_del(s->bs);
 800     virtio_cleanup(vdev);
 801 }
 802
 803 static void virtio_blk_instance_init(Object *obj)
 804 {
 805     VirtIOBlock *s = VIRTIO_BLK(obj);
 806
 807     object_property_add_link(obj, "iothread", TYPE_IOTHREAD,
 808                              (Object **)&s->blk.iothread,
 809                              qdev_prop_allow_set_link_before_realize,
 810                              OBJ_PROP_LINK_UNREF_ON_RELEASE, NULL);
 811 }
 812
 813 static Property virtio_blk_properties[] = {
 814     DEFINE_BLOCK_PROPERTIES(VirtIOBlock, blk.conf),
 815     DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, blk.conf),
 816     DEFINE_PROP_STRING("serial", VirtIOBlock, blk.serial),
 817     DEFINE_PROP_BIT("config-wce", VirtIOBlock, blk.config_wce, 0, true),
 818 #ifdef __linux__
 819     DEFINE_PROP_BIT("scsi", VirtIOBlock, blk.scsi, 0, true),
 820 #endif
 821 #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE
 822     DEFINE_PROP_BIT("x-data-plane", VirtIOBlock, blk.data_plane, 0, false),
 823 #endif
 824     DEFINE_PROP_END_OF_LIST(),
 825 };
 826
 827 static void virtio_blk_class_init(ObjectClass *klass, void *data)
 828 {
 829     DeviceClass *dc = DEVICE_CLASS(klass);
 830     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
 831
 832     dc->props = virtio_blk_properties;
 833     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
 834     vdc->realize = virtio_blk_device_realize;
 835     vdc->unrealize = virtio_blk_device_unrealize;
 836     vdc->get_config = virtio_blk_update_config;
 837     vdc->set_config = virtio_blk_set_config;
 838     vdc->get_features = virtio_blk_get_features;
 839     vdc->set_status = virtio_blk_set_status;
 840     vdc->reset = virtio_blk_reset;
 841     vdc->save = virtio_blk_save_device;
 842     vdc->load = virtio_blk_load_device;
 843 }
 844
 845 static const TypeInfo virtio_device_info = {
 846     .name = TYPE_VIRTIO_BLK,
 847     .parent = TYPE_VIRTIO_DEVICE,
 848     .instance_size = sizeof(VirtIOBlock),
 849     .instance_init = virtio_blk_instance_init,
 850     .class_init = virtio_blk_class_init,
 851 };
 852
 853 static void virtio_register_types(void)
 854 {
 855     type_register_static(&virtio_device_info);
 856 }
 857
 858 type_init(virtio_register_types)