migration/block-dirty-bitmap.c

   1 /*
   2  * Block dirty bitmap postcopy migration
   3  *
   4  * Copyright IBM, Corp. 2009
   5  * Copyright (c) 2016-2017 Virtuozzo International GmbH. All rights reserved.
   6  *
   7  * Authors:
   8  *  Liran Schour   <lirans@il.ibm.com>
   9  *  Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2.  See
  12  * the COPYING file in the top-level directory.
  13  * This file is derived from migration/block.c, so it's author and IBM copyright
  14  * are here, although content is quite different.
  15  *
  16  * Contributions after 2012-01-13 are licensed under the terms of the
  17  * GNU GPL, version 2 or (at your option) any later version.
  18  *
  19  *                                ***
  20  *
  21  * Here postcopy migration of dirty bitmaps is realized. Only QMP-addressable
  22  * bitmaps are migrated.
  23  *
  24  * Bitmap migration implies creating bitmap with the same name and granularity
  25  * in destination QEMU. If the bitmap with the same name (for the same node)
  26  * already exists on destination an error will be generated.
  27  *
  28  * format of migration:
  29  *
  30  * # Header (shared for different chunk types)
  31  * 1, 2 or 4 bytes: flags (see qemu_{put,put}_flags)
  32  * [ 1 byte: node name size ] \  flags & DEVICE_NAME
  33  * [ n bytes: node name     ] /
  34  * [ 1 byte: bitmap name size ] \  flags & BITMAP_NAME
  35  * [ n bytes: bitmap name     ] /
  36  *
  37  * # Start of bitmap migration (flags & START)
  38  * header
  39  * be64: granularity
  40  * 1 byte: bitmap flags (corresponds to BdrvDirtyBitmap)
  41  *   bit 0    -  bitmap is enabled
  42  *   bit 1    -  bitmap is persistent
  43  *   bit 2    -  bitmap is autoloading
  44  *   bits 3-7 - reserved, must be zero
  45  *
  46  * # Complete of bitmap migration (flags & COMPLETE)
  47  * header
  48  *
  49  * # Data chunk of bitmap migration
  50  * header
  51  * be64: start sector
  52  * be32: number of sectors
  53  * [ be64: buffer size  ] \ ! (flags & ZEROES)
  54  * [ n bytes: buffer    ] /
  55  *
  56  * The last chunk in stream should contain flags & EOS. The chunk may skip
  57  * device and/or bitmap names, assuming them to be the same with the previous
  58  * chunk.
  59  */
  60
  61 #include "qemu/osdep.h"
  62 #include "block/block.h"
  63 #include "block/block_int.h"
  64 #include "sysemu/block-backend.h"
  65 #include "sysemu/runstate.h"
  66 #include "qemu/main-loop.h"
  67 #include "qemu/error-report.h"
  68 #include "migration/misc.h"
  69 #include "migration/migration.h"
  70 #include "qemu-file.h"
  71 #include "migration/vmstate.h"
  72 #include "migration/register.h"
  73 #include "qemu/hbitmap.h"
  74 #include "qemu/cutils.h"
  75 #include "qapi/error.h"
  76 #include "trace.h"
  77
  78 #define CHUNK_SIZE     (1 << 10)
  79
  80 /* Flags occupy one, two or four bytes (Big Endian). The size is determined as
  81  * follows:
  82  * in first (most significant) byte bit 8 is clear  -->  one byte
  83  * in first byte bit 8 is set    -->  two or four bytes, depending on second
  84  *                                    byte:
  85  *    | in second byte bit 8 is clear  -->  two bytes
  86  *    | in second byte bit 8 is set    -->  four bytes
  87  */
  88 #define DIRTY_BITMAP_MIG_FLAG_EOS           0x01
  89 #define DIRTY_BITMAP_MIG_FLAG_ZEROES        0x02
  90 #define DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME   0x04
  91 #define DIRTY_BITMAP_MIG_FLAG_DEVICE_NAME   0x08
  92 #define DIRTY_BITMAP_MIG_FLAG_START         0x10
  93 #define DIRTY_BITMAP_MIG_FLAG_COMPLETE      0x20
  94 #define DIRTY_BITMAP_MIG_FLAG_BITS          0x40
  95
  96 #define DIRTY_BITMAP_MIG_EXTRA_FLAGS        0x80
  97
  98 #define DIRTY_BITMAP_MIG_START_FLAG_ENABLED          0x01
  99 #define DIRTY_BITMAP_MIG_START_FLAG_PERSISTENT       0x02
 100 /* 0x04 was "AUTOLOAD" flags on elder versions, no it is ignored */
 101 #define DIRTY_BITMAP_MIG_START_FLAG_RESERVED_MASK    0xf8
 102
 103 typedef struct DirtyBitmapMigBitmapState {
 104     /* Written during setup phase. */
 105     BlockDriverState *bs;
 106     const char *node_name;
 107     BdrvDirtyBitmap *bitmap;
 108     uint64_t total_sectors;
 109     uint64_t sectors_per_chunk;
 110     QSIMPLEQ_ENTRY(DirtyBitmapMigBitmapState) entry;
 111     uint8_t flags;
 112
 113     /* For bulk phase. */
 114     bool bulk_completed;
 115     uint64_t cur_sector;
 116 } DirtyBitmapMigBitmapState;
 117
 118 typedef struct DirtyBitmapMigState {
 119     QSIMPLEQ_HEAD(, DirtyBitmapMigBitmapState) dbms_list;
 120
 121     bool bulk_completed;
 122     bool no_bitmaps;
 123
 124     /* for send_bitmap_bits() */
 125     BlockDriverState *prev_bs;
 126     BdrvDirtyBitmap *prev_bitmap;
 127 } DirtyBitmapMigState;
 128
 129 typedef struct DirtyBitmapLoadState {
 130     uint32_t flags;
 131     char node_name[256];
 132     char bitmap_name[256];
 133     BlockDriverState *bs;
 134     BdrvDirtyBitmap *bitmap;
 135 } DirtyBitmapLoadState;
 136
 137 static DirtyBitmapMigState dirty_bitmap_mig_state;
 138
 139 typedef struct DirtyBitmapLoadBitmapState {
 140     BlockDriverState *bs;
 141     BdrvDirtyBitmap *bitmap;
 142     bool migrated;
 143 } DirtyBitmapLoadBitmapState;
 144 static GSList *enabled_bitmaps;
 145 QemuMutex finish_lock;
 146
 147 void init_dirty_bitmap_incoming_migration(void)
 148 {
 149     qemu_mutex_init(&finish_lock);
 150 }
 151
 152 static uint32_t qemu_get_bitmap_flags(QEMUFile *f)
 153 {
 154     uint8_t flags = qemu_get_byte(f);
 155     if (flags & DIRTY_BITMAP_MIG_EXTRA_FLAGS) {
 156         flags = flags << 8 | qemu_get_byte(f);
 157         if (flags & DIRTY_BITMAP_MIG_EXTRA_FLAGS) {
 158             flags = flags << 16 | qemu_get_be16(f);
 159         }
 160     }
 161
 162     return flags;
 163 }
 164
 165 static void qemu_put_bitmap_flags(QEMUFile *f, uint32_t flags)
 166 {
 167     /* The code currently do not send flags more than one byte */
 168     assert(!(flags & (0xffffff00 | DIRTY_BITMAP_MIG_EXTRA_FLAGS)));
 169
 170     qemu_put_byte(f, flags);
 171 }
 172
 173 static void send_bitmap_header(QEMUFile *f, DirtyBitmapMigBitmapState *dbms,
 174                                uint32_t additional_flags)
 175 {
 176     BlockDriverState *bs = dbms->bs;
 177     BdrvDirtyBitmap *bitmap = dbms->bitmap;
 178     uint32_t flags = additional_flags;
 179     trace_send_bitmap_header_enter();
 180
 181     if (bs != dirty_bitmap_mig_state.prev_bs) {
 182         dirty_bitmap_mig_state.prev_bs = bs;
 183         flags |= DIRTY_BITMAP_MIG_FLAG_DEVICE_NAME;
 184     }
 185
 186     if (bitmap != dirty_bitmap_mig_state.prev_bitmap) {
 187         dirty_bitmap_mig_state.prev_bitmap = bitmap;
 188         flags |= DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME;
 189     }
 190
 191     qemu_put_bitmap_flags(f, flags);
 192
 193     if (flags & DIRTY_BITMAP_MIG_FLAG_DEVICE_NAME) {
 194         qemu_put_counted_string(f, dbms->node_name);
 195     }
 196
 197     if (flags & DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME) {
 198         qemu_put_counted_string(f, bdrv_dirty_bitmap_name(bitmap));
 199     }
 200 }
 201
 202 static void send_bitmap_start(QEMUFile *f, DirtyBitmapMigBitmapState *dbms)
 203 {
 204     send_bitmap_header(f, dbms, DIRTY_BITMAP_MIG_FLAG_START);
 205     qemu_put_be32(f, bdrv_dirty_bitmap_granularity(dbms->bitmap));
 206     qemu_put_byte(f, dbms->flags);
 207 }
 208
 209 static void send_bitmap_complete(QEMUFile *f, DirtyBitmapMigBitmapState *dbms)
 210 {
 211     send_bitmap_header(f, dbms, DIRTY_BITMAP_MIG_FLAG_COMPLETE);
 212 }
 213
 214 static void send_bitmap_bits(QEMUFile *f, DirtyBitmapMigBitmapState *dbms,
 215                              uint64_t start_sector, uint32_t nr_sectors)
 216 {
 217     /* align for buffer_is_zero() */
 218     uint64_t align = 4 * sizeof(long);
 219     uint64_t unaligned_size =
 220         bdrv_dirty_bitmap_serialization_size(
 221             dbms->bitmap, start_sector << BDRV_SECTOR_BITS,
 222             (uint64_t)nr_sectors << BDRV_SECTOR_BITS);
 223     uint64_t buf_size = QEMU_ALIGN_UP(unaligned_size, align);
 224     uint8_t *buf = g_malloc0(buf_size);
 225     uint32_t flags = DIRTY_BITMAP_MIG_FLAG_BITS;
 226
 227     bdrv_dirty_bitmap_serialize_part(
 228         dbms->bitmap, buf, start_sector << BDRV_SECTOR_BITS,
 229         (uint64_t)nr_sectors << BDRV_SECTOR_BITS);
 230
 231     if (buffer_is_zero(buf, buf_size)) {
 232         g_free(buf);
 233         buf = NULL;
 234         flags |= DIRTY_BITMAP_MIG_FLAG_ZEROES;
 235     }
 236
 237     trace_send_bitmap_bits(flags, start_sector, nr_sectors, buf_size);
 238
 239     send_bitmap_header(f, dbms, flags);
 240
 241     qemu_put_be64(f, start_sector);
 242     qemu_put_be32(f, nr_sectors);
 243
 244     /* if a block is zero we need to flush here since the network
 245      * bandwidth is now a lot higher than the storage device bandwidth.
 246      * thus if we queue zero blocks we slow down the migration. */
 247     if (flags & DIRTY_BITMAP_MIG_FLAG_ZEROES) {
 248         qemu_fflush(f);
 249     } else {
 250         qemu_put_be64(f, buf_size);
 251         qemu_put_buffer(f, buf, buf_size);
 252     }
 253
 254     g_free(buf);
 255 }
 256
 257 /* Called with iothread lock taken.  */
 258 static void dirty_bitmap_mig_cleanup(void)
 259 {
 260     DirtyBitmapMigBitmapState *dbms;
 261
 262     while ((dbms = QSIMPLEQ_FIRST(&dirty_bitmap_mig_state.dbms_list)) != NULL) {
 263         QSIMPLEQ_REMOVE_HEAD(&dirty_bitmap_mig_state.dbms_list, entry);
 264         bdrv_dirty_bitmap_set_busy(dbms->bitmap, false);
 265         bdrv_unref(dbms->bs);
 266         g_free(dbms);
 267     }
 268 }
 269
 270 /* Called with iothread lock taken. */
 271 static int add_bitmaps_to_list(BlockDriverState *bs, const char *bs_name)
 272 {
 273     BdrvDirtyBitmap *bitmap;
 274     DirtyBitmapMigBitmapState *dbms;
 275     Error *local_err = NULL;
 276
 277     FOR_EACH_DIRTY_BITMAP(bs, bitmap) {
 278         if (!bdrv_dirty_bitmap_name(bitmap)) {
 279             continue;
 280         }
 281
 282         if (!bs_name || strcmp(bs_name, "") == 0) {
 283             error_report("Found bitmap '%s' in unnamed node %p. It can't "
 284                          "be migrated", bdrv_dirty_bitmap_name(bitmap), bs);
 285             return -1;
 286         }
 287
 288         if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_DEFAULT, &local_err)) {
 289             error_report_err(local_err);
 290             return -1;
 291         }
 292
 293         bdrv_ref(bs);
 294         bdrv_dirty_bitmap_set_busy(bitmap, true);
 295
 296         dbms = g_new0(DirtyBitmapMigBitmapState, 1);
 297         dbms->bs = bs;
 298         dbms->node_name = bs_name;
 299         dbms->bitmap = bitmap;
 300         dbms->total_sectors = bdrv_nb_sectors(bs);
 301         dbms->sectors_per_chunk = CHUNK_SIZE * 8 *
 302             bdrv_dirty_bitmap_granularity(bitmap) >> BDRV_SECTOR_BITS;
 303         if (bdrv_dirty_bitmap_enabled(bitmap)) {
 304             dbms->flags |= DIRTY_BITMAP_MIG_START_FLAG_ENABLED;
 305         }
 306         if (bdrv_dirty_bitmap_get_persistence(bitmap)) {
 307             dbms->flags |= DIRTY_BITMAP_MIG_START_FLAG_PERSISTENT;
 308         }
 309
 310         QSIMPLEQ_INSERT_TAIL(&dirty_bitmap_mig_state.dbms_list,
 311                              dbms, entry);
 312     }
 313
 314     return 0;
 315 }
 316
 317 /* Called with iothread lock taken. */
 318 static int init_dirty_bitmap_migration(void)
 319 {
 320     BlockDriverState *bs;
 321     DirtyBitmapMigBitmapState *dbms;
 322     GHashTable *handled_by_blk = g_hash_table_new(NULL, NULL);
 323     BlockBackend *blk;
 324
 325     dirty_bitmap_mig_state.bulk_completed = false;
 326     dirty_bitmap_mig_state.prev_bs = NULL;
 327     dirty_bitmap_mig_state.prev_bitmap = NULL;
 328     dirty_bitmap_mig_state.no_bitmaps = false;
 329
 330     /*
 331      * Use blockdevice name for direct (or filtered) children of named block
 332      * backends.
 333      */
 334     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 335         const char *name = blk_name(blk);
 336
 337         if (!name || strcmp(name, "") == 0) {
 338             continue;
 339         }
 340
 341         bs = blk_bs(blk);
 342
 343         /* Skip filters without bitmaps */
 344         while (bs && bs->drv && bs->drv->is_filter &&
 345                !bdrv_has_named_bitmaps(bs))
 346         {
 347             if (bs->backing) {
 348                 bs = bs->backing->bs;
 349             } else if (bs->file) {
 350                 bs = bs->file->bs;
 351             } else {
 352                 bs = NULL;
 353             }
 354         }
 355
 356         if (bs && bs->drv && !bs->drv->is_filter) {
 357             if (add_bitmaps_to_list(bs, name)) {
 358                 goto fail;
 359             }
 360             g_hash_table_add(handled_by_blk, bs);
 361         }
 362     }
 363
 364     for (bs = bdrv_next_all_states(NULL); bs; bs = bdrv_next_all_states(bs)) {
 365         if (g_hash_table_contains(handled_by_blk, bs)) {
 366             continue;
 367         }
 368
 369         if (add_bitmaps_to_list(bs, bdrv_get_node_name(bs))) {
 370             goto fail;
 371         }
 372     }
 373
 374     /* unset migration flags here, to not roll back it */
 375     QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {
 376         bdrv_dirty_bitmap_skip_store(dbms->bitmap, true);
 377     }
 378
 379     if (QSIMPLEQ_EMPTY(&dirty_bitmap_mig_state.dbms_list)) {
 380         dirty_bitmap_mig_state.no_bitmaps = true;
 381     }
 382
 383     g_hash_table_destroy(handled_by_blk);
 384
 385     return 0;
 386
 387 fail:
 388     g_hash_table_destroy(handled_by_blk);
 389     dirty_bitmap_mig_cleanup();
 390
 391     return -1;
 392 }
 393
 394 /* Called with no lock taken.  */
 395 static void bulk_phase_send_chunk(QEMUFile *f, DirtyBitmapMigBitmapState *dbms)
 396 {
 397     uint32_t nr_sectors = MIN(dbms->total_sectors - dbms->cur_sector,
 398                              dbms->sectors_per_chunk);
 399
 400     send_bitmap_bits(f, dbms, dbms->cur_sector, nr_sectors);
 401
 402     dbms->cur_sector += nr_sectors;
 403     if (dbms->cur_sector >= dbms->total_sectors) {
 404         dbms->bulk_completed = true;
 405     }
 406 }
 407
 408 /* Called with no lock taken.  */
 409 static void bulk_phase(QEMUFile *f, bool limit)
 410 {
 411     DirtyBitmapMigBitmapState *dbms;
 412
 413     QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {
 414         while (!dbms->bulk_completed) {
 415             bulk_phase_send_chunk(f, dbms);
 416             if (limit && qemu_file_rate_limit(f)) {
 417                 return;
 418             }
 419         }
 420     }
 421
 422     dirty_bitmap_mig_state.bulk_completed = true;
 423 }
 424
 425 /* for SaveVMHandlers */
 426 static void dirty_bitmap_save_cleanup(void *opaque)
 427 {
 428     dirty_bitmap_mig_cleanup();
 429 }
 430
 431 static int dirty_bitmap_save_iterate(QEMUFile *f, void *opaque)
 432 {
 433     trace_dirty_bitmap_save_iterate(migration_in_postcopy());
 434
 435     if (migration_in_postcopy() && !dirty_bitmap_mig_state.bulk_completed) {
 436         bulk_phase(f, true);
 437     }
 438
 439     qemu_put_bitmap_flags(f, DIRTY_BITMAP_MIG_FLAG_EOS);
 440
 441     return dirty_bitmap_mig_state.bulk_completed;
 442 }
 443
 444 /* Called with iothread lock taken.  */
 445
 446 static int dirty_bitmap_save_complete(QEMUFile *f, void *opaque)
 447 {
 448     DirtyBitmapMigBitmapState *dbms;
 449     trace_dirty_bitmap_save_complete_enter();
 450
 451     if (!dirty_bitmap_mig_state.bulk_completed) {
 452         bulk_phase(f, false);
 453     }
 454
 455     QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {
 456         send_bitmap_complete(f, dbms);
 457     }
 458
 459     qemu_put_bitmap_flags(f, DIRTY_BITMAP_MIG_FLAG_EOS);
 460
 461     trace_dirty_bitmap_save_complete_finish();
 462
 463     dirty_bitmap_mig_cleanup();
 464     return 0;
 465 }
 466
 467 static void dirty_bitmap_save_pending(QEMUFile *f, void *opaque,
 468                                       uint64_t max_size,
 469                                       uint64_t *res_precopy_only,
 470                                       uint64_t *res_compatible,
 471                                       uint64_t *res_postcopy_only)
 472 {
 473     DirtyBitmapMigBitmapState *dbms;
 474     uint64_t pending = 0;
 475
 476     qemu_mutex_lock_iothread();
 477
 478     QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {
 479         uint64_t gran = bdrv_dirty_bitmap_granularity(dbms->bitmap);
 480         uint64_t sectors = dbms->bulk_completed ? 0 :
 481                            dbms->total_sectors - dbms->cur_sector;
 482
 483         pending += DIV_ROUND_UP(sectors * BDRV_SECTOR_SIZE, gran);
 484     }
 485
 486     qemu_mutex_unlock_iothread();
 487
 488     trace_dirty_bitmap_save_pending(pending, max_size);
 489
 490     *res_postcopy_only += pending;
 491 }
 492
 493 /* First occurrence of this bitmap. It should be created if doesn't exist */
 494 static int dirty_bitmap_load_start(QEMUFile *f, DirtyBitmapLoadState *s)
 495 {
 496     Error *local_err = NULL;
 497     uint32_t granularity = qemu_get_be32(f);
 498     uint8_t flags = qemu_get_byte(f);
 499
 500     if (s->bitmap) {
 501         error_report("Bitmap with the same name ('%s') already exists on "
 502                      "destination", bdrv_dirty_bitmap_name(s->bitmap));
 503         return -EINVAL;
 504     } else {
 505         s->bitmap = bdrv_create_dirty_bitmap(s->bs, granularity,
 506                                              s->bitmap_name, &local_err);
 507         if (!s->bitmap) {
 508             error_report_err(local_err);
 509             return -EINVAL;
 510         }
 511     }
 512
 513     if (flags & DIRTY_BITMAP_MIG_START_FLAG_RESERVED_MASK) {
 514         error_report("Unknown flags in migrated dirty bitmap header: %x",
 515                      flags);
 516         return -EINVAL;
 517     }
 518
 519     if (flags & DIRTY_BITMAP_MIG_START_FLAG_PERSISTENT) {
 520         bdrv_dirty_bitmap_set_persistence(s->bitmap, true);
 521     }
 522
 523     bdrv_disable_dirty_bitmap(s->bitmap);
 524     if (flags & DIRTY_BITMAP_MIG_START_FLAG_ENABLED) {
 525         DirtyBitmapLoadBitmapState *b;
 526
 527         bdrv_dirty_bitmap_create_successor(s->bitmap, &local_err);
 528         if (local_err) {
 529             error_report_err(local_err);
 530             return -EINVAL;
 531         }
 532
 533         b = g_new(DirtyBitmapLoadBitmapState, 1);
 534         b->bs = s->bs;
 535         b->bitmap = s->bitmap;
 536         b->migrated = false;
 537         enabled_bitmaps = g_slist_prepend(enabled_bitmaps, b);
 538     }
 539
 540     return 0;
 541 }
 542
 543 void dirty_bitmap_mig_before_vm_start(void)
 544 {
 545     GSList *item;
 546
 547     qemu_mutex_lock(&finish_lock);
 548
 549     for (item = enabled_bitmaps; item; item = g_slist_next(item)) {
 550         DirtyBitmapLoadBitmapState *b = item->data;
 551
 552         if (b->migrated) {
 553             bdrv_enable_dirty_bitmap_locked(b->bitmap);
 554         } else {
 555             bdrv_dirty_bitmap_enable_successor(b->bitmap);
 556         }
 557
 558         g_free(b);
 559     }
 560
 561     g_slist_free(enabled_bitmaps);
 562     enabled_bitmaps = NULL;
 563
 564     qemu_mutex_unlock(&finish_lock);
 565 }
 566
 567 static void dirty_bitmap_load_complete(QEMUFile *f, DirtyBitmapLoadState *s)
 568 {
 569     GSList *item;
 570     trace_dirty_bitmap_load_complete();
 571     bdrv_dirty_bitmap_deserialize_finish(s->bitmap);
 572
 573     qemu_mutex_lock(&finish_lock);
 574
 575     for (item = enabled_bitmaps; item; item = g_slist_next(item)) {
 576         DirtyBitmapLoadBitmapState *b = item->data;
 577
 578         if (b->bitmap == s->bitmap) {
 579             b->migrated = true;
 580             break;
 581         }
 582     }
 583
 584     if (bdrv_dirty_bitmap_has_successor(s->bitmap)) {
 585         bdrv_dirty_bitmap_lock(s->bitmap);
 586         if (enabled_bitmaps == NULL) {
 587             /* in postcopy */
 588             bdrv_reclaim_dirty_bitmap_locked(s->bitmap, &error_abort);
 589             bdrv_enable_dirty_bitmap_locked(s->bitmap);
 590         } else {
 591             /* target not started, successor must be empty */
 592             int64_t count = bdrv_get_dirty_count(s->bitmap);
 593             BdrvDirtyBitmap *ret = bdrv_reclaim_dirty_bitmap_locked(s->bitmap,
 594                                                                     NULL);
 595             /* bdrv_reclaim_dirty_bitmap can fail only on no successor (it
 596              * must be) or on merge fail, but merge can't fail when second
 597              * bitmap is empty
 598              */
 599             assert(ret == s->bitmap &&
 600                    count == bdrv_get_dirty_count(s->bitmap));
 601         }
 602         bdrv_dirty_bitmap_unlock(s->bitmap);
 603     }
 604
 605     qemu_mutex_unlock(&finish_lock);
 606 }
 607
 608 static int dirty_bitmap_load_bits(QEMUFile *f, DirtyBitmapLoadState *s)
 609 {
 610     uint64_t first_byte = qemu_get_be64(f) << BDRV_SECTOR_BITS;
 611     uint64_t nr_bytes = (uint64_t)qemu_get_be32(f) << BDRV_SECTOR_BITS;
 612     trace_dirty_bitmap_load_bits_enter(first_byte >> BDRV_SECTOR_BITS,
 613                                        nr_bytes >> BDRV_SECTOR_BITS);
 614
 615     if (s->flags & DIRTY_BITMAP_MIG_FLAG_ZEROES) {
 616         trace_dirty_bitmap_load_bits_zeroes();
 617         bdrv_dirty_bitmap_deserialize_zeroes(s->bitmap, first_byte, nr_bytes,
 618                                              false);
 619     } else {
 620         size_t ret;
 621         uint8_t *buf;
 622         uint64_t buf_size = qemu_get_be64(f);
 623         uint64_t needed_size =
 624             bdrv_dirty_bitmap_serialization_size(s->bitmap,
 625                                                  first_byte, nr_bytes);
 626
 627         if (needed_size > buf_size ||
 628             buf_size > QEMU_ALIGN_UP(needed_size, 4 * sizeof(long))
 629              /* Here used same alignment as in send_bitmap_bits */
 630         ) {
 631             error_report("Migrated bitmap granularity doesn't "
 632                          "match the destination bitmap '%s' granularity",
 633                          bdrv_dirty_bitmap_name(s->bitmap));
 634             return -EINVAL;
 635         }
 636
 637         buf = g_malloc(buf_size);
 638         ret = qemu_get_buffer(f, buf, buf_size);
 639         if (ret != buf_size) {
 640             error_report("Failed to read bitmap bits");
 641             g_free(buf);
 642             return -EIO;
 643         }
 644
 645         bdrv_dirty_bitmap_deserialize_part(s->bitmap, buf, first_byte, nr_bytes,
 646                                            false);
 647         g_free(buf);
 648     }
 649
 650     return 0;
 651 }
 652
 653 static int dirty_bitmap_load_header(QEMUFile *f, DirtyBitmapLoadState *s)
 654 {
 655     Error *local_err = NULL;
 656     bool nothing;
 657     s->flags = qemu_get_bitmap_flags(f);
 658     trace_dirty_bitmap_load_header(s->flags);
 659
 660     nothing = s->flags == (s->flags & DIRTY_BITMAP_MIG_FLAG_EOS);
 661
 662     if (s->flags & DIRTY_BITMAP_MIG_FLAG_DEVICE_NAME) {
 663         if (!qemu_get_counted_string(f, s->node_name)) {
 664             error_report("Unable to read node name string");
 665             return -EINVAL;
 666         }
 667         s->bs = bdrv_lookup_bs(s->node_name, s->node_name, &local_err);
 668         if (!s->bs) {
 669             error_report_err(local_err);
 670             return -EINVAL;
 671         }
 672     } else if (!s->bs && !nothing) {
 673         error_report("Error: block device name is not set");
 674         return -EINVAL;
 675     }
 676
 677     if (s->flags & DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME) {
 678         if (!qemu_get_counted_string(f, s->bitmap_name)) {
 679             error_report("Unable to read bitmap name string");
 680             return -EINVAL;
 681         }
 682         s->bitmap = bdrv_find_dirty_bitmap(s->bs, s->bitmap_name);
 683
 684         /* bitmap may be NULL here, it wouldn't be an error if it is the
 685          * first occurrence of the bitmap */
 686         if (!s->bitmap && !(s->flags & DIRTY_BITMAP_MIG_FLAG_START)) {
 687             error_report("Error: unknown dirty bitmap "
 688                          "'%s' for block device '%s'",
 689                          s->bitmap_name, s->node_name);
 690             return -EINVAL;
 691         }
 692     } else if (!s->bitmap && !nothing) {
 693         error_report("Error: block device name is not set");
 694         return -EINVAL;
 695     }
 696
 697     return 0;
 698 }
 699
 700 static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id)
 701 {
 702     static DirtyBitmapLoadState s;
 703     int ret = 0;
 704
 705     trace_dirty_bitmap_load_enter();
 706
 707     if (version_id != 1) {
 708         return -EINVAL;
 709     }
 710
 711     do {
 712         ret = dirty_bitmap_load_header(f, &s);
 713         if (ret < 0) {
 714             return ret;
 715         }
 716
 717         if (s.flags & DIRTY_BITMAP_MIG_FLAG_START) {
 718             ret = dirty_bitmap_load_start(f, &s);
 719         } else if (s.flags & DIRTY_BITMAP_MIG_FLAG_COMPLETE) {
 720             dirty_bitmap_load_complete(f, &s);
 721         } else if (s.flags & DIRTY_BITMAP_MIG_FLAG_BITS) {
 722             ret = dirty_bitmap_load_bits(f, &s);
 723         }
 724
 725         if (!ret) {
 726             ret = qemu_file_get_error(f);
 727         }
 728
 729         if (ret) {
 730             return ret;
 731         }
 732     } while (!(s.flags & DIRTY_BITMAP_MIG_FLAG_EOS));
 733
 734     trace_dirty_bitmap_load_success();
 735     return 0;
 736 }
 737
 738 static int dirty_bitmap_save_setup(QEMUFile *f, void *opaque)
 739 {
 740     DirtyBitmapMigBitmapState *dbms = NULL;
 741     if (init_dirty_bitmap_migration() < 0) {
 742         return -1;
 743     }
 744
 745     QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {
 746         send_bitmap_start(f, dbms);
 747     }
 748     qemu_put_bitmap_flags(f, DIRTY_BITMAP_MIG_FLAG_EOS);
 749
 750     return 0;
 751 }
 752
 753 static bool dirty_bitmap_is_active(void *opaque)
 754 {
 755     return migrate_dirty_bitmaps() && !dirty_bitmap_mig_state.no_bitmaps;
 756 }
 757
 758 static bool dirty_bitmap_is_active_iterate(void *opaque)
 759 {
 760     return dirty_bitmap_is_active(opaque) && !runstate_is_running();
 761 }
 762
 763 static bool dirty_bitmap_has_postcopy(void *opaque)
 764 {
 765     return true;
 766 }
 767
 768 static SaveVMHandlers savevm_dirty_bitmap_handlers = {
 769     .save_setup = dirty_bitmap_save_setup,
 770     .save_live_complete_postcopy = dirty_bitmap_save_complete,
 771     .save_live_complete_precopy = dirty_bitmap_save_complete,
 772     .has_postcopy = dirty_bitmap_has_postcopy,
 773     .save_live_pending = dirty_bitmap_save_pending,
 774     .save_live_iterate = dirty_bitmap_save_iterate,
 775     .is_active_iterate = dirty_bitmap_is_active_iterate,
 776     .load_state = dirty_bitmap_load,
 777     .save_cleanup = dirty_bitmap_save_cleanup,
 778     .is_active = dirty_bitmap_is_active,
 779 };
 780
 781 void dirty_bitmap_mig_init(void)
 782 {
 783     QSIMPLEQ_INIT(&dirty_bitmap_mig_state.dbms_list);
 784
 785     register_savevm_live("dirty-bitmap", 0, 1,
 786                          &savevm_dirty_bitmap_handlers,
 787                          &dirty_bitmap_mig_state);
 788 }