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     /* Written during setup phase.  Can be read without a lock.  */
  47     BlockDriverState *bs;
  48     int shared_base;
  49     int64_t total_sectors;
  50     QSIMPLEQ_ENTRY(BlkMigDevState) entry;
  51
  52     /* Only used by migration thread.  Does not need a lock.  */
  53     int bulk_completed;
  54     int64_t cur_sector;
  55     int64_t cur_dirty;
  56
  57     /* Protected by block migration lock.  */
  58     unsigned long *aio_bitmap;
  59     int64_t completed_sectors;
  60 } BlkMigDevState;
  61
  62 typedef struct BlkMigBlock {
  63     /* Only used by migration thread.  */
  64     uint8_t *buf;
  65     BlkMigDevState *bmds;
  66     int64_t sector;
  67     int nr_sectors;
  68     struct iovec iov;
  69     QEMUIOVector qiov;
  70     BlockDriverAIOCB *aiocb;
  71
  72     /* Protected by block migration lock.  */
  73     int ret;
  74     QSIMPLEQ_ENTRY(BlkMigBlock) entry;
  75 } BlkMigBlock;
  76
  77 typedef struct BlkMigState {
  78     /* Written during setup phase.  Can be read without a lock.  */
  79     int blk_enable;
  80     int shared_base;
  81     QSIMPLEQ_HEAD(bmds_list, BlkMigDevState) bmds_list;
  82     int64_t total_sector_sum;
  83
  84     /* Protected by lock.  */
  85     QSIMPLEQ_HEAD(blk_list, BlkMigBlock) blk_list;
  86     int submitted;
  87     int read_done;
  88
  89     /* Only used by migration thread.  Does not need a lock.  */
  90     int transferred;
  91     int prev_progress;
  92     int bulk_completed;
  93
  94     /* Lock must be taken _inside_ the iothread lock.  */
  95     QemuMutex lock;
  96 } BlkMigState;
  97
  98 static BlkMigState block_mig_state;
  99
 100 static void blk_mig_lock(void)
 101 {
 102     qemu_mutex_lock(&block_mig_state.lock);
 103 }
 104
 105 static void blk_mig_unlock(void)
 106 {
 107     qemu_mutex_unlock(&block_mig_state.lock);
 108 }
 109
 110 static void blk_send(QEMUFile *f, BlkMigBlock * blk)
 111 {
 112     int len;
 113
 114     /* sector number and flags */
 115     qemu_put_be64(f, (blk->sector << BDRV_SECTOR_BITS)
 116                      | BLK_MIG_FLAG_DEVICE_BLOCK);
 117
 118     /* device name */
 119     len = strlen(blk->bmds->bs->device_name);
 120     qemu_put_byte(f, len);
 121     qemu_put_buffer(f, (uint8_t *)blk->bmds->bs->device_name, len);
 122
 123     qemu_put_buffer(f, blk->buf, BLOCK_SIZE);
 124 }
 125
 126 int blk_mig_active(void)
 127 {
 128     return !QSIMPLEQ_EMPTY(&block_mig_state.bmds_list);
 129 }
 130
 131 uint64_t blk_mig_bytes_transferred(void)
 132 {
 133     BlkMigDevState *bmds;
 134     uint64_t sum = 0;
 135
 136     blk_mig_lock();
 137     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 138         sum += bmds->completed_sectors;
 139     }
 140     blk_mig_unlock();
 141     return sum << BDRV_SECTOR_BITS;
 142 }
 143
 144 uint64_t blk_mig_bytes_remaining(void)
 145 {
 146     return blk_mig_bytes_total() - blk_mig_bytes_transferred();
 147 }
 148
 149 uint64_t blk_mig_bytes_total(void)
 150 {
 151     BlkMigDevState *bmds;
 152     uint64_t sum = 0;
 153
 154     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 155         sum += bmds->total_sectors;
 156     }
 157     return sum << BDRV_SECTOR_BITS;
 158 }
 159
 160
 161 /* Called with migration lock held.  */
 162
 163 static int bmds_aio_inflight(BlkMigDevState *bmds, int64_t sector)
 164 {
 165     int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
 166
 167     if ((sector << BDRV_SECTOR_BITS) < bdrv_getlength(bmds->bs)) {
 168         return !!(bmds->aio_bitmap[chunk / (sizeof(unsigned long) * 8)] &
 169             (1UL << (chunk % (sizeof(unsigned long) * 8))));
 170     } else {
 171         return 0;
 172     }
 173 }
 174
 175 /* Called with migration lock held.  */
 176
 177 static void bmds_set_aio_inflight(BlkMigDevState *bmds, int64_t sector_num,
 178                              int nb_sectors, int set)
 179 {
 180     int64_t start, end;
 181     unsigned long val, idx, bit;
 182
 183     start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
 184     end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
 185
 186     for (; start <= end; start++) {
 187         idx = start / (sizeof(unsigned long) * 8);
 188         bit = start % (sizeof(unsigned long) * 8);
 189         val = bmds->aio_bitmap[idx];
 190         if (set) {
 191             val |= 1UL << bit;
 192         } else {
 193             val &= ~(1UL << bit);
 194         }
 195         bmds->aio_bitmap[idx] = val;
 196     }
 197 }
 198
 199 static void alloc_aio_bitmap(BlkMigDevState *bmds)
 200 {
 201     BlockDriverState *bs = bmds->bs;
 202     int64_t bitmap_size;
 203
 204     bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
 205             BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
 206     bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
 207
 208     bmds->aio_bitmap = g_malloc0(bitmap_size);
 209 }
 210
 211 /* Never hold migration lock when yielding to the main loop!  */
 212
 213 static void blk_mig_read_cb(void *opaque, int ret)
 214 {
 215     BlkMigBlock *blk = opaque;
 216
 217     blk_mig_lock();
 218     blk->ret = ret;
 219
 220     QSIMPLEQ_INSERT_TAIL(&block_mig_state.blk_list, blk, entry);
 221     bmds_set_aio_inflight(blk->bmds, blk->sector, blk->nr_sectors, 0);
 222
 223     block_mig_state.submitted--;
 224     block_mig_state.read_done++;
 225     assert(block_mig_state.submitted >= 0);
 226     blk_mig_unlock();
 227 }
 228
 229 static int mig_save_device_bulk(QEMUFile *f, BlkMigDevState *bmds)
 230 {
 231     int64_t total_sectors = bmds->total_sectors;
 232     int64_t cur_sector = bmds->cur_sector;
 233     BlockDriverState *bs = bmds->bs;
 234     BlkMigBlock *blk;
 235     int nr_sectors;
 236
 237     if (bmds->shared_base) {
 238         while (cur_sector < total_sectors &&
 239                !bdrv_is_allocated(bs, cur_sector, MAX_IS_ALLOCATED_SEARCH,
 240                                   &nr_sectors)) {
 241             cur_sector += nr_sectors;
 242         }
 243     }
 244
 245     if (cur_sector >= total_sectors) {
 246         bmds->cur_sector = bmds->completed_sectors = total_sectors;
 247         return 1;
 248     }
 249
 250     bmds->completed_sectors = cur_sector;
 251
 252     cur_sector &= ~((int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK - 1);
 253
 254     /* we are going to transfer a full block even if it is not allocated */
 255     nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
 256
 257     if (total_sectors - cur_sector < BDRV_SECTORS_PER_DIRTY_CHUNK) {
 258         nr_sectors = total_sectors - cur_sector;
 259     }
 260
 261     blk = g_malloc(sizeof(BlkMigBlock));
 262     blk->buf = g_malloc(BLOCK_SIZE);
 263     blk->bmds = bmds;
 264     blk->sector = cur_sector;
 265     blk->nr_sectors = nr_sectors;
 266
 267     blk->iov.iov_base = blk->buf;
 268     blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE;
 269     qemu_iovec_init_external(&blk->qiov, &blk->iov, 1);
 270
 271     blk_mig_lock();
 272     block_mig_state.submitted++;
 273     blk_mig_unlock();
 274
 275     blk->aiocb = bdrv_aio_readv(bs, cur_sector, &blk->qiov,
 276                                 nr_sectors, blk_mig_read_cb, blk);
 277
 278     bdrv_reset_dirty(bs, cur_sector, nr_sectors);
 279     bmds->cur_sector = cur_sector + nr_sectors;
 280
 281     return (bmds->cur_sector >= total_sectors);
 282 }
 283
 284 static void set_dirty_tracking(int enable)
 285 {
 286     BlkMigDevState *bmds;
 287
 288     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 289         bdrv_set_dirty_tracking(bmds->bs, enable ? BLOCK_SIZE : 0);
 290     }
 291 }
 292
 293 static void init_blk_migration_it(void *opaque, BlockDriverState *bs)
 294 {
 295     BlkMigDevState *bmds;
 296     int64_t sectors;
 297
 298     if (!bdrv_is_read_only(bs)) {
 299         sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
 300         if (sectors <= 0) {
 301             return;
 302         }
 303
 304         bmds = g_malloc0(sizeof(BlkMigDevState));
 305         bmds->bs = bs;
 306         bmds->bulk_completed = 0;
 307         bmds->total_sectors = sectors;
 308         bmds->completed_sectors = 0;
 309         bmds->shared_base = block_mig_state.shared_base;
 310         alloc_aio_bitmap(bmds);
 311         drive_get_ref(drive_get_by_blockdev(bs));
 312         bdrv_set_in_use(bs, 1);
 313
 314         block_mig_state.total_sector_sum += sectors;
 315
 316         if (bmds->shared_base) {
 317             DPRINTF("Start migration for %s with shared base image\n",
 318                     bs->device_name);
 319         } else {
 320             DPRINTF("Start full migration for %s\n", bs->device_name);
 321         }
 322
 323         QSIMPLEQ_INSERT_TAIL(&block_mig_state.bmds_list, bmds, entry);
 324     }
 325 }
 326
 327 static void init_blk_migration(QEMUFile *f)
 328 {
 329     block_mig_state.submitted = 0;
 330     block_mig_state.read_done = 0;
 331     block_mig_state.transferred = 0;
 332     block_mig_state.total_sector_sum = 0;
 333     block_mig_state.prev_progress = -1;
 334     block_mig_state.bulk_completed = 0;
 335
 336     bdrv_iterate(init_blk_migration_it, NULL);
 337 }
 338
 339 static int blk_mig_save_bulked_block(QEMUFile *f)
 340 {
 341     int64_t completed_sector_sum = 0;
 342     BlkMigDevState *bmds;
 343     int progress;
 344     int ret = 0;
 345
 346     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 347         if (bmds->bulk_completed == 0) {
 348             if (mig_save_device_bulk(f, bmds) == 1) {
 349                 /* completed bulk section for this device */
 350                 bmds->bulk_completed = 1;
 351             }
 352             completed_sector_sum += bmds->completed_sectors;
 353             ret = 1;
 354             break;
 355         } else {
 356             completed_sector_sum += bmds->completed_sectors;
 357         }
 358     }
 359
 360     if (block_mig_state.total_sector_sum != 0) {
 361         progress = completed_sector_sum * 100 /
 362                    block_mig_state.total_sector_sum;
 363     } else {
 364         progress = 100;
 365     }
 366     if (progress != block_mig_state.prev_progress) {
 367         block_mig_state.prev_progress = progress;
 368         qemu_put_be64(f, (progress << BDRV_SECTOR_BITS)
 369                          | BLK_MIG_FLAG_PROGRESS);
 370         DPRINTF("Completed %d %%\r", progress);
 371     }
 372
 373     return ret;
 374 }
 375
 376 static void blk_mig_reset_dirty_cursor(void)
 377 {
 378     BlkMigDevState *bmds;
 379
 380     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 381         bmds->cur_dirty = 0;
 382     }
 383 }
 384
 385 static int mig_save_device_dirty(QEMUFile *f, BlkMigDevState *bmds,
 386                                  int is_async)
 387 {
 388     BlkMigBlock *blk;
 389     int64_t total_sectors = bmds->total_sectors;
 390     int64_t sector;
 391     int nr_sectors;
 392     int ret = -EIO;
 393
 394     for (sector = bmds->cur_dirty; sector < bmds->total_sectors;) {
 395         blk_mig_lock();
 396         if (bmds_aio_inflight(bmds, sector)) {
 397             blk_mig_unlock();
 398             bdrv_drain_all();
 399         } else {
 400             blk_mig_unlock();
 401         }
 402         if (bdrv_get_dirty(bmds->bs, sector)) {
 403
 404             if (total_sectors - sector < BDRV_SECTORS_PER_DIRTY_CHUNK) {
 405                 nr_sectors = total_sectors - sector;
 406             } else {
 407                 nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
 408             }
 409             blk = g_malloc(sizeof(BlkMigBlock));
 410             blk->buf = g_malloc(BLOCK_SIZE);
 411             blk->bmds = bmds;
 412             blk->sector = sector;
 413             blk->nr_sectors = nr_sectors;
 414
 415             if (is_async) {
 416                 blk->iov.iov_base = blk->buf;
 417                 blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE;
 418                 qemu_iovec_init_external(&blk->qiov, &blk->iov, 1);
 419
 420                 blk->aiocb = bdrv_aio_readv(bmds->bs, sector, &blk->qiov,
 421                                             nr_sectors, blk_mig_read_cb, blk);
 422
 423                 blk_mig_lock();
 424                 block_mig_state.submitted++;
 425                 bmds_set_aio_inflight(bmds, sector, nr_sectors, 1);
 426                 blk_mig_unlock();
 427             } else {
 428                 ret = bdrv_read(bmds->bs, sector, blk->buf, nr_sectors);
 429                 if (ret < 0) {
 430                     goto error;
 431                 }
 432                 blk_send(f, blk);
 433
 434                 g_free(blk->buf);
 435                 g_free(blk);
 436             }
 437
 438             bdrv_reset_dirty(bmds->bs, sector, nr_sectors);
 439             break;
 440         }
 441         sector += BDRV_SECTORS_PER_DIRTY_CHUNK;
 442         bmds->cur_dirty = sector;
 443     }
 444
 445     return (bmds->cur_dirty >= bmds->total_sectors);
 446
 447 error:
 448     DPRINTF("Error reading sector %" PRId64 "\n", sector);
 449     g_free(blk->buf);
 450     g_free(blk);
 451     return ret;
 452 }
 453
 454 /* return value:
 455  * 0: too much data for max_downtime
 456  * 1: few enough data for max_downtime
 457 */
 458 static int blk_mig_save_dirty_block(QEMUFile *f, int is_async)
 459 {
 460     BlkMigDevState *bmds;
 461     int ret = 1;
 462
 463     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 464         ret = mig_save_device_dirty(f, bmds, is_async);
 465         if (ret <= 0) {
 466             break;
 467         }
 468     }
 469
 470     return ret;
 471 }
 472
 473 static int flush_blks(QEMUFile *f)
 474 {
 475     BlkMigBlock *blk;
 476     int ret = 0;
 477
 478     DPRINTF("%s Enter submitted %d read_done %d transferred %d\n",
 479             __FUNCTION__, block_mig_state.submitted, block_mig_state.read_done,
 480             block_mig_state.transferred);
 481
 482     blk_mig_lock();
 483     while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) {
 484         if (qemu_file_rate_limit(f)) {
 485             break;
 486         }
 487         if (blk->ret < 0) {
 488             ret = blk->ret;
 489             break;
 490         }
 491
 492         QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry);
 493         blk_mig_unlock();
 494         blk_send(f, blk);
 495         blk_mig_lock();
 496
 497         g_free(blk->buf);
 498         g_free(blk);
 499
 500         block_mig_state.read_done--;
 501         block_mig_state.transferred++;
 502         assert(block_mig_state.read_done >= 0);
 503     }
 504     blk_mig_unlock();
 505
 506     DPRINTF("%s Exit submitted %d read_done %d transferred %d\n", __FUNCTION__,
 507             block_mig_state.submitted, block_mig_state.read_done,
 508             block_mig_state.transferred);
 509     return ret;
 510 }
 511
 512 static int64_t get_remaining_dirty(void)
 513 {
 514     BlkMigDevState *bmds;
 515     int64_t dirty = 0;
 516
 517     QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
 518         dirty += bdrv_get_dirty_count(bmds->bs);
 519     }
 520
 521     return dirty << BDRV_SECTOR_BITS;
 522 }
 523
 524 static void blk_mig_cleanup(void)
 525 {
 526     BlkMigDevState *bmds;
 527     BlkMigBlock *blk;
 528
 529     bdrv_drain_all();
 530
 531     set_dirty_tracking(0);
 532
 533     blk_mig_lock();
 534     while ((bmds = QSIMPLEQ_FIRST(&block_mig_state.bmds_list)) != NULL) {
 535         QSIMPLEQ_REMOVE_HEAD(&block_mig_state.bmds_list, entry);
 536         bdrv_set_in_use(bmds->bs, 0);
 537         drive_put_ref(drive_get_by_blockdev(bmds->bs));
 538         g_free(bmds->aio_bitmap);
 539         g_free(bmds);
 540     }
 541
 542     while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) {
 543         QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry);
 544         g_free(blk->buf);
 545         g_free(blk);
 546     }
 547     blk_mig_unlock();
 548 }
 549
 550 static void block_migration_cancel(void *opaque)
 551 {
 552     blk_mig_cleanup();
 553 }
 554
 555 static int block_save_setup(QEMUFile *f, void *opaque)
 556 {
 557     int ret;
 558
 559     DPRINTF("Enter save live setup submitted %d transferred %d\n",
 560             block_mig_state.submitted, block_mig_state.transferred);
 561
 562     init_blk_migration(f);
 563
 564     /* start track dirty blocks */
 565     set_dirty_tracking(1);
 566
 567     ret = flush_blks(f);
 568     blk_mig_reset_dirty_cursor();
 569     qemu_put_be64(f, BLK_MIG_FLAG_EOS);
 570
 571     return ret;
 572 }
 573
 574 static int block_save_iterate(QEMUFile *f, void *opaque)
 575 {
 576     int ret;
 577     int64_t last_ftell = qemu_ftell(f);
 578
 579     DPRINTF("Enter save live iterate submitted %d transferred %d\n",
 580             block_mig_state.submitted, block_mig_state.transferred);
 581
 582     ret = flush_blks(f);
 583     if (ret) {
 584         return ret;
 585     }
 586
 587     blk_mig_reset_dirty_cursor();
 588
 589     /* control the rate of transfer */
 590     blk_mig_lock();
 591     while ((block_mig_state.submitted +
 592             block_mig_state.read_done) * BLOCK_SIZE <
 593            qemu_file_get_rate_limit(f)) {
 594         blk_mig_unlock();
 595         if (block_mig_state.bulk_completed == 0) {
 596             /* first finish the bulk phase */
 597             if (blk_mig_save_bulked_block(f) == 0) {
 598                 /* finished saving bulk on all devices */
 599                 block_mig_state.bulk_completed = 1;
 600             }
 601             ret = 0;
 602         } else {
 603             ret = blk_mig_save_dirty_block(f, 1);
 604         }
 605         if (ret < 0) {
 606             return ret;
 607         }
 608         blk_mig_lock();
 609         if (ret != 0) {
 610             /* no more dirty blocks */
 611             break;
 612         }
 613     }
 614     blk_mig_unlock();
 615
 616     ret = flush_blks(f);
 617     if (ret) {
 618         return ret;
 619     }
 620
 621     qemu_put_be64(f, BLK_MIG_FLAG_EOS);
 622     return qemu_ftell(f) - last_ftell;
 623 }
 624
 625 static int block_save_complete(QEMUFile *f, void *opaque)
 626 {
 627     int ret;
 628
 629     DPRINTF("Enter save live complete submitted %d transferred %d\n",
 630             block_mig_state.submitted, block_mig_state.transferred);
 631
 632     ret = flush_blks(f);
 633     if (ret) {
 634         return ret;
 635     }
 636
 637     blk_mig_reset_dirty_cursor();
 638
 639     /* we know for sure that save bulk is completed and
 640        all async read completed */
 641     blk_mig_lock();
 642     assert(block_mig_state.submitted == 0);
 643     blk_mig_unlock();
 644
 645     do {
 646         ret = blk_mig_save_dirty_block(f, 0);
 647         if (ret < 0) {
 648             return ret;
 649         }
 650     } while (ret == 0);
 651
 652     /* report completion */
 653     qemu_put_be64(f, (100 << BDRV_SECTOR_BITS) | BLK_MIG_FLAG_PROGRESS);
 654
 655     DPRINTF("Block migration completed\n");
 656
 657     qemu_put_be64(f, BLK_MIG_FLAG_EOS);
 658
 659     blk_mig_cleanup();
 660     return 0;
 661 }
 662
 663 static uint64_t block_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
 664 {
 665     /* Estimate pending number of bytes to send */
 666     uint64_t pending;
 667
 668     blk_mig_lock();
 669     pending = get_remaining_dirty() +
 670                        block_mig_state.submitted * BLOCK_SIZE +
 671                        block_mig_state.read_done * BLOCK_SIZE;
 672
 673     /* Report at least one block pending during bulk phase */
 674     if (pending == 0 && !block_mig_state.bulk_completed) {
 675         pending = BLOCK_SIZE;
 676     }
 677     blk_mig_unlock();
 678
 679     DPRINTF("Enter save live pending  %" PRIu64 "\n", pending);
 680     return pending;
 681 }
 682
 683 static int block_load(QEMUFile *f, void *opaque, int version_id)
 684 {
 685     static int banner_printed;
 686     int len, flags;
 687     char device_name[256];
 688     int64_t addr;
 689     BlockDriverState *bs, *bs_prev = NULL;
 690     uint8_t *buf;
 691     int64_t total_sectors = 0;
 692     int nr_sectors;
 693     int ret;
 694
 695     do {
 696         addr = qemu_get_be64(f);
 697
 698         flags = addr & ~BDRV_SECTOR_MASK;
 699         addr >>= BDRV_SECTOR_BITS;
 700
 701         if (flags & BLK_MIG_FLAG_DEVICE_BLOCK) {
 702             /* get device name */
 703             len = qemu_get_byte(f);
 704             qemu_get_buffer(f, (uint8_t *)device_name, len);
 705             device_name[len] = '\0';
 706
 707             bs = bdrv_find(device_name);
 708             if (!bs) {
 709                 fprintf(stderr, "Error unknown block device %s\n",
 710                         device_name);
 711                 return -EINVAL;
 712             }
 713
 714             if (bs != bs_prev) {
 715                 bs_prev = bs;
 716                 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
 717                 if (total_sectors <= 0) {
 718                     error_report("Error getting length of block device %s",
 719                                  device_name);
 720                     return -EINVAL;
 721                 }
 722             }
 723
 724             if (total_sectors - addr < BDRV_SECTORS_PER_DIRTY_CHUNK) {
 725                 nr_sectors = total_sectors - addr;
 726             } else {
 727                 nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
 728             }
 729
 730             buf = g_malloc(BLOCK_SIZE);
 731
 732             qemu_get_buffer(f, buf, BLOCK_SIZE);
 733             ret = bdrv_write(bs, addr, buf, nr_sectors);
 734
 735             g_free(buf);
 736             if (ret < 0) {
 737                 return ret;
 738             }
 739         } else if (flags & BLK_MIG_FLAG_PROGRESS) {
 740             if (!banner_printed) {
 741                 printf("Receiving block device images\n");
 742                 banner_printed = 1;
 743             }
 744             printf("Completed %d %%%c", (int)addr,
 745                    (addr == 100) ? '\n' : '\r');
 746             fflush(stdout);
 747         } else if (!(flags & BLK_MIG_FLAG_EOS)) {
 748             fprintf(stderr, "Unknown block migration flags: %#x\n", flags);
 749             return -EINVAL;
 750         }
 751         ret = qemu_file_get_error(f);
 752         if (ret != 0) {
 753             return ret;
 754         }
 755     } while (!(flags & BLK_MIG_FLAG_EOS));
 756
 757     return 0;
 758 }
 759
 760 static void block_set_params(const MigrationParams *params, void *opaque)
 761 {
 762     block_mig_state.blk_enable = params->blk;
 763     block_mig_state.shared_base = params->shared;
 764
 765     /* shared base means that blk_enable = 1 */
 766     block_mig_state.blk_enable |= params->shared;
 767 }
 768
 769 static bool block_is_active(void *opaque)
 770 {
 771     return block_mig_state.blk_enable == 1;
 772 }
 773
 774 SaveVMHandlers savevm_block_handlers = {
 775     .set_params = block_set_params,
 776     .save_live_setup = block_save_setup,
 777     .save_live_iterate = block_save_iterate,
 778     .save_live_complete = block_save_complete,
 779     .save_live_pending = block_save_pending,
 780     .load_state = block_load,
 781     .cancel = block_migration_cancel,
 782     .is_active = block_is_active,
 783 };
 784
 785 void blk_mig_init(void)
 786 {
 787     QSIMPLEQ_INIT(&block_mig_state.bmds_list);
 788     QSIMPLEQ_INIT(&block_mig_state.blk_list);
 789     qemu_mutex_init(&block_mig_state.lock);
 790
 791     register_savevm_live(NULL, "block", 0, 1, &savevm_block_handlers,
 792                          &block_mig_state);
 793 }