block-migration.c

   1 /*
   2  * QEMU live block migration
   3  *
   4  * Copyright IBM, Corp. 2009
   5  *
   6  * Authors:
   7  *  Liran Schour   <lirans@il.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  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu-common.h"
  17 #include "block/block_int.h"
  18 #include "hw/hw.h"
  19 #include "qemu/queue.h"
  20 #include "qemu/timer.h"
  21 #include "migration/block.h"
  22 #include "migration/migration.h"
  23 #include "sysemu/blockdev.h"
  24 #include <assert.h>
  25
  26 #define BLOCK_SIZE                       (1 << 20)
  27 #define BDRV_SECTORS_PER_DIRTY_CHUNK     (BLOCK_SIZE >> BDRV_SECTOR_BITS)
  28
  29 #define BLK_MIG_FLAG_DEVICE_BLOCK       0x01
  30 #define BLK_MIG_FLAG_EOS                0x02
  31 #define BLK_MIG_FLAG_PROGRESS           0x04
  32
  33 #define MAX_IS_ALLOCATED_SEARCH 65536
  34
  35 //#define DEBUG_BLK_MIGRATION
  36
  37 #ifdef DEBUG_BLK_MIGRATION
  38 #define DPRINTF(fmt, ...) \
  39     do { printf("blk_migration: " fmt, ## __VA_ARGS__); } while (0)
  40 #else
  41 #define DPRINTF(fmt, ...) \
  42     do { } while (0)
  43 #endif
  44
  45 typedef struct BlkMigDevState {
  46     BlockDriverState *bs;
  47     int bulk_completed;
  48     int shared_base;
  49     int64_t cur_sector;
  50     int64_t cur_dirty;
  51     int64_t completed_sectors;
  52     int64_t total_sectors;
  53     int64_t dirty;
  54     QSIMPLEQ_ENTRY(BlkMigDevState) entry;
  55     unsigned long *aio_bitmap;
  56 } BlkMigDevState;
  57
  58 typedef struct BlkMigBlock {
  59     uint8_t *buf;
  60     BlkMigDevState *bmds;
  61     int64_t sector;
  62     int nr_sectors;
  63     struct iovec iov;
  64     QEMUIOVector qiov;
  65     BlockDriverAIOCB *aiocb;
  66     int ret;
  67     QSIMPLEQ_ENTRY(BlkMigBlock) entry;
  68 } BlkMigBlock;
  69
  70 typedef struct BlkMigState {
  71     int blk_enable;
  72     int shared_base;
  73     QSIMPLEQ_HEAD(bmds_list, BlkMigDevState) bmds_list;
  74     QSIMPLEQ_HEAD(blk_list, BlkMigBlock) blk_list;
  75     int submitted;
  76     int read_done;
  77     int transferred;
  78     int64_t total_sector_sum;
  79     int prev_progress;
  80     int bulk_completed;
  81     long double prev_time_offset;
  82 } BlkMigState;
  83
  84 static BlkMigState block_mig_state;
  85
  86 static void blk_send(QEMUFile *f, BlkMigBlock * blk)
  87 {
  88     int len;
  89
  90     /* sector number and flags */
  91     qemu_put_be64(f, (blk->sector << BDRV_SECTOR_BITS)
  92                      | BLK_MIG_FLAG_DEVICE_BLOCK);
  93
  94     /* device name */
  95     len = strlen(blk->bmds->bs->device_name);
  96     qemu_put_byte(f, len);
  97     qemu_put_buffer(f, (uint8_t *)blk->bmds->bs->device_name, len);
  98
  99     qemu_put_buffer(f, blk->buf, BLOCK_SIZE);
 100 }
 101
 102 int blk_mig_active(void)
 103 {
 104     return !QSIMPLEQ_EMPTY(&block_mig_state.bmds_list);
 105 }
 106
 107 uint64_t blk_mig_bytes_transferred(void)
 108 {
 109     BlkMigDevState *bmds;
 110     uint64_t sum = 0;
 111
 112     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 113         sum += bmds->completed_sectors;
 114     }
 115     return sum << BDRV_SECTOR_BITS;
 116 }
 117
 118 uint64_t blk_mig_bytes_remaining(void)
 119 {
 120     return blk_mig_bytes_total() - blk_mig_bytes_transferred();
 121 }
 122
 123 uint64_t blk_mig_bytes_total(void)
 124 {
 125     BlkMigDevState *bmds;
 126     uint64_t sum = 0;
 127
 128     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 129         sum += bmds->total_sectors;
 130     }
 131     return sum << BDRV_SECTOR_BITS;
 132 }
 133
 134 static int bmds_aio_inflight(BlkMigDevState *bmds, int64_t sector)
 135 {
 136     int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
 137
 138     if ((sector << BDRV_SECTOR_BITS) < bdrv_getlength(bmds->bs)) {
 139         return !!(bmds->aio_bitmap[chunk / (sizeof(unsigned long) * 8)] &
 140             (1UL << (chunk % (sizeof(unsigned long) * 8))));
 141     } else {
 142         return 0;
 143     }
 144 }
 145
 146 static void bmds_set_aio_inflight(BlkMigDevState *bmds, int64_t sector_num,
 147                              int nb_sectors, int set)
 148 {
 149     int64_t start, end;
 150     unsigned long val, idx, bit;
 151
 152     start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
 153     end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
 154
 155     for (; start <= end; start++) {
 156         idx = start / (sizeof(unsigned long) * 8);
 157         bit = start % (sizeof(unsigned long) * 8);
 158         val = bmds->aio_bitmap[idx];
 159         if (set) {
 160             val |= 1UL << bit;
 161         } else {
 162             val &= ~(1UL << bit);
 163         }
 164         bmds->aio_bitmap[idx] = val;
 165     }
 166 }
 167
 168 static void alloc_aio_bitmap(BlkMigDevState *bmds)
 169 {
 170     BlockDriverState *bs = bmds->bs;
 171     int64_t bitmap_size;
 172
 173     bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
 174             BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
 175     bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
 176
 177     bmds->aio_bitmap = g_malloc0(bitmap_size);
 178 }
 179
 180 static void blk_mig_read_cb(void *opaque, int ret)
 181 {
 182     long double curr_time = qemu_get_clock_ns(rt_clock);
 183     BlkMigBlock *blk = opaque;
 184
 185     blk->ret = ret;
 186
 187     block_mig_state.prev_time_offset = curr_time;
 188
 189     QSIMPLEQ_INSERT_TAIL(&block_mig_state.blk_list, blk, entry);
 190     bmds_set_aio_inflight(blk->bmds, blk->sector, blk->nr_sectors, 0);
 191
 192     block_mig_state.submitted--;
 193     block_mig_state.read_done++;
 194     assert(block_mig_state.submitted >= 0);
 195 }
 196
 197 static int mig_save_device_bulk(QEMUFile *f, BlkMigDevState *bmds)
 198 {
 199     int64_t total_sectors = bmds->total_sectors;
 200     int64_t cur_sector = bmds->cur_sector;
 201     BlockDriverState *bs = bmds->bs;
 202     BlkMigBlock *blk;
 203     int nr_sectors;
 204
 205     if (bmds->shared_base) {
 206         while (cur_sector < total_sectors &&
 207                !bdrv_is_allocated(bs, cur_sector, MAX_IS_ALLOCATED_SEARCH,
 208                                   &nr_sectors)) {
 209             cur_sector += nr_sectors;
 210         }
 211     }
 212
 213     if (cur_sector >= total_sectors) {
 214         bmds->cur_sector = bmds->completed_sectors = total_sectors;
 215         return 1;
 216     }
 217
 218     bmds->completed_sectors = cur_sector;
 219
 220     cur_sector &= ~((int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK - 1);
 221
 222     /* we are going to transfer a full block even if it is not allocated */
 223     nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
 224
 225     if (total_sectors - cur_sector < BDRV_SECTORS_PER_DIRTY_CHUNK) {
 226         nr_sectors = total_sectors - cur_sector;
 227     }
 228
 229     blk = g_malloc(sizeof(BlkMigBlock));
 230     blk->buf = g_malloc(BLOCK_SIZE);
 231     blk->bmds = bmds;
 232     blk->sector = cur_sector;
 233     blk->nr_sectors = nr_sectors;
 234
 235     blk->iov.iov_base = blk->buf;
 236     blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE;
 237     qemu_iovec_init_external(&blk->qiov, &blk->iov, 1);
 238
 239     if (block_mig_state.submitted == 0) {
 240         block_mig_state.prev_time_offset = qemu_get_clock_ns(rt_clock);
 241     }
 242
 243     blk->aiocb = bdrv_aio_readv(bs, cur_sector, &blk->qiov,
 244                                 nr_sectors, blk_mig_read_cb, blk);
 245     block_mig_state.submitted++;
 246
 247     bdrv_reset_dirty(bs, cur_sector, nr_sectors);
 248     bmds->cur_sector = cur_sector + nr_sectors;
 249
 250     return (bmds->cur_sector >= total_sectors);
 251 }
 252
 253 static void set_dirty_tracking(int enable)
 254 {
 255     BlkMigDevState *bmds;
 256
 257     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 258         bdrv_set_dirty_tracking(bmds->bs, enable ? BLOCK_SIZE : 0);
 259     }
 260 }
 261
 262 static void init_blk_migration_it(void *opaque, BlockDriverState *bs)
 263 {
 264     BlkMigDevState *bmds;
 265     int64_t sectors;
 266
 267     if (!bdrv_is_read_only(bs)) {
 268         sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
 269         if (sectors <= 0) {
 270             return;
 271         }
 272
 273         bmds = g_malloc0(sizeof(BlkMigDevState));
 274         bmds->bs = bs;
 275         bmds->bulk_completed = 0;
 276         bmds->total_sectors = sectors;
 277         bmds->completed_sectors = 0;
 278         bmds->shared_base = block_mig_state.shared_base;
 279         alloc_aio_bitmap(bmds);
 280         drive_get_ref(drive_get_by_blockdev(bs));
 281         bdrv_set_in_use(bs, 1);
 282
 283         block_mig_state.total_sector_sum += sectors;
 284
 285         if (bmds->shared_base) {
 286             DPRINTF("Start migration for %s with shared base image\n",
 287                     bs->device_name);
 288         } else {
 289             DPRINTF("Start full migration for %s\n", bs->device_name);
 290         }
 291
 292         QSIMPLEQ_INSERT_TAIL(&block_mig_state.bmds_list, bmds, entry);
 293     }
 294 }
 295
 296 static void init_blk_migration(QEMUFile *f)
 297 {
 298     block_mig_state.submitted = 0;
 299     block_mig_state.read_done = 0;
 300     block_mig_state.transferred = 0;
 301     block_mig_state.total_sector_sum = 0;
 302     block_mig_state.prev_progress = -1;
 303     block_mig_state.bulk_completed = 0;
 304
 305     bdrv_iterate(init_blk_migration_it, NULL);
 306 }
 307
 308 static int blk_mig_save_bulked_block(QEMUFile *f)
 309 {
 310     int64_t completed_sector_sum = 0;
 311     BlkMigDevState *bmds;
 312     int progress;
 313     int ret = 0;
 314
 315     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 316         if (bmds->bulk_completed == 0) {
 317             if (mig_save_device_bulk(f, bmds) == 1) {
 318                 /* completed bulk section for this device */
 319                 bmds->bulk_completed = 1;
 320             }
 321             completed_sector_sum += bmds->completed_sectors;
 322             ret = 1;
 323             break;
 324         } else {
 325             completed_sector_sum += bmds->completed_sectors;
 326         }
 327     }
 328
 329     if (block_mig_state.total_sector_sum != 0) {
 330         progress = completed_sector_sum * 100 /
 331                    block_mig_state.total_sector_sum;
 332     } else {
 333         progress = 100;
 334     }
 335     if (progress != block_mig_state.prev_progress) {
 336         block_mig_state.prev_progress = progress;
 337         qemu_put_be64(f, (progress << BDRV_SECTOR_BITS)
 338                          | BLK_MIG_FLAG_PROGRESS);
 339         DPRINTF("Completed %d %%\r", progress);
 340     }
 341
 342     return ret;
 343 }
 344
 345 static void blk_mig_reset_dirty_cursor(void)
 346 {
 347     BlkMigDevState *bmds;
 348
 349     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 350         bmds->cur_dirty = 0;
 351     }
 352 }
 353
 354 static int mig_save_device_dirty(QEMUFile *f, BlkMigDevState *bmds,
 355                                  int is_async)
 356 {
 357     BlkMigBlock *blk;
 358     int64_t total_sectors = bmds->total_sectors;
 359     int64_t sector;
 360     int nr_sectors;
 361     int ret = -EIO;
 362
 363     for (sector = bmds->cur_dirty; sector < bmds->total_sectors;) {
 364         if (bmds_aio_inflight(bmds, sector)) {
 365             bdrv_drain_all();
 366         }
 367         if (bdrv_get_dirty(bmds->bs, sector)) {
 368
 369             if (total_sectors - sector < BDRV_SECTORS_PER_DIRTY_CHUNK) {
 370                 nr_sectors = total_sectors - sector;
 371             } else {
 372                 nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
 373             }
 374             blk = g_malloc(sizeof(BlkMigBlock));
 375             blk->buf = g_malloc(BLOCK_SIZE);
 376             blk->bmds = bmds;
 377             blk->sector = sector;
 378             blk->nr_sectors = nr_sectors;
 379
 380             if (is_async) {
 381                 blk->iov.iov_base = blk->buf;
 382                 blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE;
 383                 qemu_iovec_init_external(&blk->qiov, &blk->iov, 1);
 384
 385                 if (block_mig_state.submitted == 0) {
 386                     block_mig_state.prev_time_offset = qemu_get_clock_ns(rt_clock);
 387                 }
 388
 389                 blk->aiocb = bdrv_aio_readv(bmds->bs, sector, &blk->qiov,
 390                                             nr_sectors, blk_mig_read_cb, blk);
 391                 block_mig_state.submitted++;
 392                 bmds_set_aio_inflight(bmds, sector, nr_sectors, 1);
 393             } else {
 394                 ret = bdrv_read(bmds->bs, sector, blk->buf, nr_sectors);
 395                 if (ret < 0) {
 396                     goto error;
 397                 }
 398                 blk_send(f, blk);
 399
 400                 g_free(blk->buf);
 401                 g_free(blk);
 402             }
 403
 404             bdrv_reset_dirty(bmds->bs, sector, nr_sectors);
 405             break;
 406         }
 407         sector += BDRV_SECTORS_PER_DIRTY_CHUNK;
 408         bmds->cur_dirty = sector;
 409     }
 410
 411     return (bmds->cur_dirty >= bmds->total_sectors);
 412
 413 error:
 414     DPRINTF("Error reading sector %" PRId64 "\n", sector);
 415     g_free(blk->buf);
 416     g_free(blk);
 417     return ret;
 418 }
 419
 420 /* return value:
 421  * 0: too much data for max_downtime
 422  * 1: few enough data for max_downtime
 423 */
 424 static int blk_mig_save_dirty_block(QEMUFile *f, int is_async)
 425 {
 426     BlkMigDevState *bmds;
 427     int ret = 1;
 428
 429     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 430         ret = mig_save_device_dirty(f, bmds, is_async);
 431         if (ret <= 0) {
 432             break;
 433         }
 434     }
 435
 436     return ret;
 437 }
 438
 439 static int flush_blks(QEMUFile *f)
 440 {
 441     BlkMigBlock *blk;
 442     int ret = 0;
 443
 444     DPRINTF("%s Enter submitted %d read_done %d transferred %d\n",
 445             __FUNCTION__, block_mig_state.submitted, block_mig_state.read_done,
 446             block_mig_state.transferred);
 447
 448     while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) {
 449         if (qemu_file_rate_limit(f)) {
 450             break;
 451         }
 452         if (blk->ret < 0) {
 453             ret = blk->ret;
 454             break;
 455         }
 456         blk_send(f, blk);
 457
 458         QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry);
 459         g_free(blk->buf);
 460         g_free(blk);
 461
 462         block_mig_state.read_done--;
 463         block_mig_state.transferred++;
 464         assert(block_mig_state.read_done >= 0);
 465     }
 466
 467     DPRINTF("%s Exit submitted %d read_done %d transferred %d\n", __FUNCTION__,
 468             block_mig_state.submitted, block_mig_state.read_done,
 469             block_mig_state.transferred);
 470     return ret;
 471 }
 472
 473 static int64_t get_remaining_dirty(void)
 474 {
 475     BlkMigDevState *bmds;
 476     int64_t dirty = 0;
 477
 478     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 479         dirty += bdrv_get_dirty_count(bmds->bs);
 480     }
 481
 482     return dirty << BDRV_SECTOR_BITS;
 483 }
 484
 485 static void blk_mig_cleanup(void)
 486 {
 487     BlkMigDevState *bmds;
 488     BlkMigBlock *blk;
 489
 490     bdrv_drain_all();
 491
 492     set_dirty_tracking(0);
 493
 494     while ((bmds = QSIMPLEQ_FIRST(&block_mig_state.bmds_list)) != NULL) {
 495         QSIMPLEQ_REMOVE_HEAD(&block_mig_state.bmds_list, entry);
 496         bdrv_set_in_use(bmds->bs, 0);
 497         drive_put_ref(drive_get_by_blockdev(bmds->bs));
 498         g_free(bmds->aio_bitmap);
 499         g_free(bmds);
 500     }
 501
 502     while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) {
 503         QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry);
 504         g_free(blk->buf);
 505         g_free(blk);
 506     }
 507 }
 508
 509 static void block_migration_cancel(void *opaque)
 510 {
 511     blk_mig_cleanup();
 512 }
 513
 514 static int block_save_setup(QEMUFile *f, void *opaque)
 515 {
 516     int ret;
 517
 518     DPRINTF("Enter save live setup submitted %d transferred %d\n",
 519             block_mig_state.submitted, block_mig_state.transferred);
 520
 521     init_blk_migration(f);
 522
 523     /* start track dirty blocks */
 524     set_dirty_tracking(1);
 525
 526     ret = flush_blks(f);
 527     if (ret) {
 528         blk_mig_cleanup();
 529         return ret;
 530     }
 531
 532     blk_mig_reset_dirty_cursor();
 533
 534     qemu_put_be64(f, BLK_MIG_FLAG_EOS);
 535
 536     return 0;
 537 }
 538
 539 static int block_save_iterate(QEMUFile *f, void *opaque)
 540 {
 541     int ret;
 542     int64_t last_ftell = qemu_ftell(f);
 543
 544     DPRINTF("Enter save live iterate submitted %d transferred %d\n",
 545             block_mig_state.submitted, block_mig_state.transferred);
 546
 547     ret = flush_blks(f);
 548     if (ret) {
 549         blk_mig_cleanup();
 550         return ret;
 551     }
 552
 553     blk_mig_reset_dirty_cursor();
 554
 555     /* control the rate of transfer */
 556     while ((block_mig_state.submitted +
 557             block_mig_state.read_done) * BLOCK_SIZE <
 558            qemu_file_get_rate_limit(f)) {
 559         if (block_mig_state.bulk_completed == 0) {
 560             /* first finish the bulk phase */
 561             if (blk_mig_save_bulked_block(f) == 0) {
 562                 /* finished saving bulk on all devices */
 563                 block_mig_state.bulk_completed = 1;
 564             }
 565         } else {
 566             ret = blk_mig_save_dirty_block(f, 1);
 567             if (ret != 0) {
 568                 /* no more dirty blocks */
 569                 break;
 570             }
 571         }
 572     }
 573     if (ret < 0) {
 574         blk_mig_cleanup();
 575         return ret;
 576     }
 577
 578     ret = flush_blks(f);
 579     if (ret) {
 580         blk_mig_cleanup();
 581         return ret;
 582     }
 583
 584     qemu_put_be64(f, BLK_MIG_FLAG_EOS);
 585
 586     return qemu_ftell(f) - last_ftell;
 587 }
 588
 589 static int block_save_complete(QEMUFile *f, void *opaque)
 590 {
 591     int ret;
 592
 593     DPRINTF("Enter save live complete submitted %d transferred %d\n",
 594             block_mig_state.submitted, block_mig_state.transferred);
 595
 596     ret = flush_blks(f);
 597     if (ret) {
 598         blk_mig_cleanup();
 599         return ret;
 600     }
 601
 602     blk_mig_reset_dirty_cursor();
 603
 604     /* we know for sure that save bulk is completed and
 605        all async read completed */
 606     assert(block_mig_state.submitted == 0);
 607
 608     do {
 609         ret = blk_mig_save_dirty_block(f, 0);
 610     } while (ret == 0);
 611
 612     blk_mig_cleanup();
 613     if (ret < 0) {
 614         return ret;
 615     }
 616     /* report completion */
 617     qemu_put_be64(f, (100 << BDRV_SECTOR_BITS) | BLK_MIG_FLAG_PROGRESS);
 618
 619     DPRINTF("Block migration completed\n");
 620
 621     qemu_put_be64(f, BLK_MIG_FLAG_EOS);
 622
 623     return 0;
 624 }
 625
 626 static uint64_t block_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
 627 {
 628     /* Estimate pending number of bytes to send */
 629     uint64_t pending = get_remaining_dirty() +
 630                        block_mig_state.submitted * BLOCK_SIZE +
 631                        block_mig_state.read_done * BLOCK_SIZE;
 632
 633     /* Report at least one block pending during bulk phase */
 634     if (pending == 0 && !block_mig_state.bulk_completed) {
 635         pending = BLOCK_SIZE;
 636     }
 637
 638     DPRINTF("Enter save live pending  %" PRIu64 "\n", pending);
 639     return pending;
 640 }
 641
 642 static int block_load(QEMUFile *f, void *opaque, int version_id)
 643 {
 644     static int banner_printed;
 645     int len, flags;
 646     char device_name[256];
 647     int64_t addr;
 648     BlockDriverState *bs, *bs_prev = NULL;
 649     uint8_t *buf;
 650     int64_t total_sectors = 0;
 651     int nr_sectors;
 652     int ret;
 653
 654     do {
 655         addr = qemu_get_be64(f);
 656
 657         flags = addr & ~BDRV_SECTOR_MASK;
 658         addr >>= BDRV_SECTOR_BITS;
 659
 660         if (flags & BLK_MIG_FLAG_DEVICE_BLOCK) {
 661             /* get device name */
 662             len = qemu_get_byte(f);
 663             qemu_get_buffer(f, (uint8_t *)device_name, len);
 664             device_name[len] = '\0';
 665
 666             bs = bdrv_find(device_name);
 667             if (!bs) {
 668                 fprintf(stderr, "Error unknown block device %s\n",
 669                         device_name);
 670                 return -EINVAL;
 671             }
 672
 673             if (bs != bs_prev) {
 674                 bs_prev = bs;
 675                 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
 676                 if (total_sectors <= 0) {
 677                     error_report("Error getting length of block device %s",
 678                                  device_name);
 679                     return -EINVAL;
 680                 }
 681             }
 682
 683             if (total_sectors - addr < BDRV_SECTORS_PER_DIRTY_CHUNK) {
 684                 nr_sectors = total_sectors - addr;
 685             } else {
 686                 nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
 687             }
 688
 689             buf = g_malloc(BLOCK_SIZE);
 690
 691             qemu_get_buffer(f, buf, BLOCK_SIZE);
 692             ret = bdrv_write(bs, addr, buf, nr_sectors);
 693
 694             g_free(buf);
 695             if (ret < 0) {
 696                 return ret;
 697             }
 698         } else if (flags & BLK_MIG_FLAG_PROGRESS) {
 699             if (!banner_printed) {
 700                 printf("Receiving block device images\n");
 701                 banner_printed = 1;
 702             }
 703             printf("Completed %d %%%c", (int)addr,
 704                    (addr == 100) ? '\n' : '\r');
 705             fflush(stdout);
 706         } else if (!(flags & BLK_MIG_FLAG_EOS)) {
 707             fprintf(stderr, "Unknown block migration flags: %#x\n", flags);
 708             return -EINVAL;
 709         }
 710         ret = qemu_file_get_error(f);
 711         if (ret != 0) {
 712             return ret;
 713         }
 714     } while (!(flags & BLK_MIG_FLAG_EOS));
 715
 716     return 0;
 717 }
 718
 719 static void block_set_params(const MigrationParams *params, void *opaque)
 720 {
 721     block_mig_state.blk_enable = params->blk;
 722     block_mig_state.shared_base = params->shared;
 723
 724     /* shared base means that blk_enable = 1 */
 725     block_mig_state.blk_enable |= params->shared;
 726 }
 727
 728 static bool block_is_active(void *opaque)
 729 {
 730     return block_mig_state.blk_enable == 1;
 731 }
 732
 733 SaveVMHandlers savevm_block_handlers = {
 734     .set_params = block_set_params,
 735     .save_live_setup = block_save_setup,
 736     .save_live_iterate = block_save_iterate,
 737     .save_live_complete = block_save_complete,
 738     .save_live_pending = block_save_pending,
 739     .load_state = block_load,
 740     .cancel = block_migration_cancel,
 741     .is_active = block_is_active,
 742 };
 743
 744 void blk_mig_init(void)
 745 {
 746     QSIMPLEQ_INIT(&block_mig_state.bmds_list);
 747     QSIMPLEQ_INIT(&block_mig_state.blk_list);
 748
 749     register_savevm_live(NULL, "block", 0, 1, &savevm_block_handlers,
 750                          &block_mig_state);
 751 }