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     bitmap = bdrv_dirty_bitmap_first(bs);
 278     if (!bitmap) {
 279         return 0;
 280     }
 281
 282     if (!bs_name || strcmp(bs_name, "") == 0) {
 283         error_report("Bitmap '%s' in unnamed node can't be migrated",
 284                      bdrv_dirty_bitmap_name(bitmap));
 285         return -1;
 286     }
 287
 288     FOR_EACH_DIRTY_BITMAP(bs, bitmap) {
 289         if (!bdrv_dirty_bitmap_name(bitmap)) {
 290             continue;
 291         }
 292
 293         if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_DEFAULT, &local_err)) {
 294             error_report_err(local_err);
 295             return -1;
 296         }
 297
 298         bdrv_ref(bs);
 299         bdrv_dirty_bitmap_set_busy(bitmap, true);
 300
 301         dbms = g_new0(DirtyBitmapMigBitmapState, 1);
 302         dbms->bs = bs;
 303         dbms->node_name = bs_name;
 304         dbms->bitmap = bitmap;
 305         dbms->total_sectors = bdrv_nb_sectors(bs);
 306         dbms->sectors_per_chunk = CHUNK_SIZE * 8 *
 307             bdrv_dirty_bitmap_granularity(bitmap) >> BDRV_SECTOR_BITS;
 308         if (bdrv_dirty_bitmap_enabled(bitmap)) {
 309             dbms->flags |= DIRTY_BITMAP_MIG_START_FLAG_ENABLED;
 310         }
 311         if (bdrv_dirty_bitmap_get_persistence(bitmap)) {
 312             dbms->flags |= DIRTY_BITMAP_MIG_START_FLAG_PERSISTENT;
 313         }
 314
 315         QSIMPLEQ_INSERT_TAIL(&dirty_bitmap_mig_state.dbms_list,
 316                              dbms, entry);
 317     }
 318
 319     return 0;
 320 }
 321
 322 /* Called with iothread lock taken. */
 323 static int init_dirty_bitmap_migration(void)
 324 {
 325     BlockDriverState *bs;
 326     DirtyBitmapMigBitmapState *dbms;
 327     GHashTable *handled_by_blk = g_hash_table_new(NULL, NULL);
 328     BlockBackend *blk;
 329
 330     dirty_bitmap_mig_state.bulk_completed = false;
 331     dirty_bitmap_mig_state.prev_bs = NULL;
 332     dirty_bitmap_mig_state.prev_bitmap = NULL;
 333     dirty_bitmap_mig_state.no_bitmaps = false;
 334
 335     /*
 336      * Use blockdevice name for direct (or filtered) children of named block
 337      * backends.
 338      */
 339     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
 340         const char *name = blk_name(blk);
 341
 342         if (!name || strcmp(name, "") == 0) {
 343             continue;
 344         }
 345
 346         bs = blk_bs(blk);
 347
 348         /* Skip filters without bitmaps */
 349         while (bs && bs->drv && bs->drv->is_filter &&
 350                !bdrv_has_named_bitmaps(bs))
 351         {
 352             if (bs->backing) {
 353                 bs = bs->backing->bs;
 354             } else if (bs->file) {
 355                 bs = bs->file->bs;
 356             } else {
 357                 bs = NULL;
 358             }
 359         }
 360
 361         if (bs && bs->drv && !bs->drv->is_filter) {
 362             if (add_bitmaps_to_list(bs, name)) {
 363                 goto fail;
 364             }
 365             g_hash_table_add(handled_by_blk, bs);
 366         }
 367     }
 368
 369     for (bs = bdrv_next_all_states(NULL); bs; bs = bdrv_next_all_states(bs)) {
 370         if (g_hash_table_contains(handled_by_blk, bs)) {
 371             continue;
 372         }
 373
 374         if (add_bitmaps_to_list(bs, bdrv_get_node_name(bs))) {
 375             goto fail;
 376         }
 377     }
 378
 379     /* unset migration flags here, to not roll back it */
 380     QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {
 381         bdrv_dirty_bitmap_skip_store(dbms->bitmap, true);
 382     }
 383
 384     if (QSIMPLEQ_EMPTY(&dirty_bitmap_mig_state.dbms_list)) {
 385         dirty_bitmap_mig_state.no_bitmaps = true;
 386     }
 387
 388     g_hash_table_destroy(handled_by_blk);
 389
 390     return 0;
 391
 392 fail:
 393     g_hash_table_destroy(handled_by_blk);
 394     dirty_bitmap_mig_cleanup();
 395
 396     return -1;
 397 }
 398
 399 /* Called with no lock taken.  */
 400 static void bulk_phase_send_chunk(QEMUFile *f, DirtyBitmapMigBitmapState *dbms)
 401 {
 402     uint32_t nr_sectors = MIN(dbms->total_sectors - dbms->cur_sector,
 403                              dbms->sectors_per_chunk);
 404
 405     send_bitmap_bits(f, dbms, dbms->cur_sector, nr_sectors);
 406
 407     dbms->cur_sector += nr_sectors;
 408     if (dbms->cur_sector >= dbms->total_sectors) {
 409         dbms->bulk_completed = true;
 410     }
 411 }
 412
 413 /* Called with no lock taken.  */
 414 static void bulk_phase(QEMUFile *f, bool limit)
 415 {
 416     DirtyBitmapMigBitmapState *dbms;
 417
 418     QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {
 419         while (!dbms->bulk_completed) {
 420             bulk_phase_send_chunk(f, dbms);
 421             if (limit && qemu_file_rate_limit(f)) {
 422                 return;
 423             }
 424         }
 425     }
 426
 427     dirty_bitmap_mig_state.bulk_completed = true;
 428 }
 429
 430 /* for SaveVMHandlers */
 431 static void dirty_bitmap_save_cleanup(void *opaque)
 432 {
 433     dirty_bitmap_mig_cleanup();
 434 }
 435
 436 static int dirty_bitmap_save_iterate(QEMUFile *f, void *opaque)
 437 {
 438     trace_dirty_bitmap_save_iterate(migration_in_postcopy());
 439
 440     if (migration_in_postcopy() && !dirty_bitmap_mig_state.bulk_completed) {
 441         bulk_phase(f, true);
 442     }
 443
 444     qemu_put_bitmap_flags(f, DIRTY_BITMAP_MIG_FLAG_EOS);
 445
 446     return dirty_bitmap_mig_state.bulk_completed;
 447 }
 448
 449 /* Called with iothread lock taken.  */
 450
 451 static int dirty_bitmap_save_complete(QEMUFile *f, void *opaque)
 452 {
 453     DirtyBitmapMigBitmapState *dbms;
 454     trace_dirty_bitmap_save_complete_enter();
 455
 456     if (!dirty_bitmap_mig_state.bulk_completed) {
 457         bulk_phase(f, false);
 458     }
 459
 460     QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {
 461         send_bitmap_complete(f, dbms);
 462     }
 463
 464     qemu_put_bitmap_flags(f, DIRTY_BITMAP_MIG_FLAG_EOS);
 465
 466     trace_dirty_bitmap_save_complete_finish();
 467
 468     dirty_bitmap_mig_cleanup();
 469     return 0;
 470 }
 471
 472 static void dirty_bitmap_save_pending(QEMUFile *f, void *opaque,
 473                                       uint64_t max_size,
 474                                       uint64_t *res_precopy_only,
 475                                       uint64_t *res_compatible,
 476                                       uint64_t *res_postcopy_only)
 477 {
 478     DirtyBitmapMigBitmapState *dbms;
 479     uint64_t pending = 0;
 480
 481     qemu_mutex_lock_iothread();
 482
 483     QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {
 484         uint64_t gran = bdrv_dirty_bitmap_granularity(dbms->bitmap);
 485         uint64_t sectors = dbms->bulk_completed ? 0 :
 486                            dbms->total_sectors - dbms->cur_sector;
 487
 488         pending += DIV_ROUND_UP(sectors * BDRV_SECTOR_SIZE, gran);
 489     }
 490
 491     qemu_mutex_unlock_iothread();
 492
 493     trace_dirty_bitmap_save_pending(pending, max_size);
 494
 495     *res_postcopy_only += pending;
 496 }
 497
 498 /* First occurrence of this bitmap. It should be created if doesn't exist */
 499 static int dirty_bitmap_load_start(QEMUFile *f, DirtyBitmapLoadState *s)
 500 {
 501     Error *local_err = NULL;
 502     uint32_t granularity = qemu_get_be32(f);
 503     uint8_t flags = qemu_get_byte(f);
 504
 505     if (s->bitmap) {
 506         error_report("Bitmap with the same name ('%s') already exists on "
 507                      "destination", bdrv_dirty_bitmap_name(s->bitmap));
 508         return -EINVAL;
 509     } else {
 510         s->bitmap = bdrv_create_dirty_bitmap(s->bs, granularity,
 511                                              s->bitmap_name, &local_err);
 512         if (!s->bitmap) {
 513             error_report_err(local_err);
 514             return -EINVAL;
 515         }
 516     }
 517
 518     if (flags & DIRTY_BITMAP_MIG_START_FLAG_RESERVED_MASK) {
 519         error_report("Unknown flags in migrated dirty bitmap header: %x",
 520                      flags);
 521         return -EINVAL;
 522     }
 523
 524     if (flags & DIRTY_BITMAP_MIG_START_FLAG_PERSISTENT) {
 525         bdrv_dirty_bitmap_set_persistence(s->bitmap, true);
 526     }
 527
 528     bdrv_disable_dirty_bitmap(s->bitmap);
 529     if (flags & DIRTY_BITMAP_MIG_START_FLAG_ENABLED) {
 530         DirtyBitmapLoadBitmapState *b;
 531
 532         bdrv_dirty_bitmap_create_successor(s->bitmap, &local_err);
 533         if (local_err) {
 534             error_report_err(local_err);
 535             return -EINVAL;
 536         }
 537
 538         b = g_new(DirtyBitmapLoadBitmapState, 1);
 539         b->bs = s->bs;
 540         b->bitmap = s->bitmap;
 541         b->migrated = false;
 542         enabled_bitmaps = g_slist_prepend(enabled_bitmaps, b);
 543     }
 544
 545     return 0;
 546 }
 547
 548 void dirty_bitmap_mig_before_vm_start(void)
 549 {
 550     GSList *item;
 551
 552     qemu_mutex_lock(&finish_lock);
 553
 554     for (item = enabled_bitmaps; item; item = g_slist_next(item)) {
 555         DirtyBitmapLoadBitmapState *b = item->data;
 556
 557         if (b->migrated) {
 558             bdrv_enable_dirty_bitmap_locked(b->bitmap);
 559         } else {
 560             bdrv_dirty_bitmap_enable_successor(b->bitmap);
 561         }
 562
 563         g_free(b);
 564     }
 565
 566     g_slist_free(enabled_bitmaps);
 567     enabled_bitmaps = NULL;
 568
 569     qemu_mutex_unlock(&finish_lock);
 570 }
 571
 572 static void dirty_bitmap_load_complete(QEMUFile *f, DirtyBitmapLoadState *s)
 573 {
 574     GSList *item;
 575     trace_dirty_bitmap_load_complete();
 576     bdrv_dirty_bitmap_deserialize_finish(s->bitmap);
 577
 578     qemu_mutex_lock(&finish_lock);
 579
 580     for (item = enabled_bitmaps; item; item = g_slist_next(item)) {
 581         DirtyBitmapLoadBitmapState *b = item->data;
 582
 583         if (b->bitmap == s->bitmap) {
 584             b->migrated = true;
 585             break;
 586         }
 587     }
 588
 589     if (bdrv_dirty_bitmap_has_successor(s->bitmap)) {
 590         bdrv_dirty_bitmap_lock(s->bitmap);
 591         if (enabled_bitmaps == NULL) {
 592             /* in postcopy */
 593             bdrv_reclaim_dirty_bitmap_locked(s->bitmap, &error_abort);
 594             bdrv_enable_dirty_bitmap_locked(s->bitmap);
 595         } else {
 596             /* target not started, successor must be empty */
 597             int64_t count = bdrv_get_dirty_count(s->bitmap);
 598             BdrvDirtyBitmap *ret = bdrv_reclaim_dirty_bitmap_locked(s->bitmap,
 599                                                                     NULL);
 600             /* bdrv_reclaim_dirty_bitmap can fail only on no successor (it
 601              * must be) or on merge fail, but merge can't fail when second
 602              * bitmap is empty
 603              */
 604             assert(ret == s->bitmap &&
 605                    count == bdrv_get_dirty_count(s->bitmap));
 606         }
 607         bdrv_dirty_bitmap_unlock(s->bitmap);
 608     }
 609
 610     qemu_mutex_unlock(&finish_lock);
 611 }
 612
 613 static int dirty_bitmap_load_bits(QEMUFile *f, DirtyBitmapLoadState *s)
 614 {
 615     uint64_t first_byte = qemu_get_be64(f) << BDRV_SECTOR_BITS;
 616     uint64_t nr_bytes = (uint64_t)qemu_get_be32(f) << BDRV_SECTOR_BITS;
 617     trace_dirty_bitmap_load_bits_enter(first_byte >> BDRV_SECTOR_BITS,
 618                                        nr_bytes >> BDRV_SECTOR_BITS);
 619
 620     if (s->flags & DIRTY_BITMAP_MIG_FLAG_ZEROES) {
 621         trace_dirty_bitmap_load_bits_zeroes();
 622         bdrv_dirty_bitmap_deserialize_zeroes(s->bitmap, first_byte, nr_bytes,
 623                                              false);
 624     } else {
 625         size_t ret;
 626         uint8_t *buf;
 627         uint64_t buf_size = qemu_get_be64(f);
 628         uint64_t needed_size =
 629             bdrv_dirty_bitmap_serialization_size(s->bitmap,
 630                                                  first_byte, nr_bytes);
 631
 632         if (needed_size > buf_size ||
 633             buf_size > QEMU_ALIGN_UP(needed_size, 4 * sizeof(long))
 634              /* Here used same alignment as in send_bitmap_bits */
 635         ) {
 636             error_report("Migrated bitmap granularity doesn't "
 637                          "match the destination bitmap '%s' granularity",
 638                          bdrv_dirty_bitmap_name(s->bitmap));
 639             return -EINVAL;
 640         }
 641
 642         buf = g_malloc(buf_size);
 643         ret = qemu_get_buffer(f, buf, buf_size);
 644         if (ret != buf_size) {
 645             error_report("Failed to read bitmap bits");
 646             g_free(buf);
 647             return -EIO;
 648         }
 649
 650         bdrv_dirty_bitmap_deserialize_part(s->bitmap, buf, first_byte, nr_bytes,
 651                                            false);
 652         g_free(buf);
 653     }
 654
 655     return 0;
 656 }
 657
 658 static int dirty_bitmap_load_header(QEMUFile *f, DirtyBitmapLoadState *s)
 659 {
 660     Error *local_err = NULL;
 661     bool nothing;
 662     s->flags = qemu_get_bitmap_flags(f);
 663     trace_dirty_bitmap_load_header(s->flags);
 664
 665     nothing = s->flags == (s->flags & DIRTY_BITMAP_MIG_FLAG_EOS);
 666
 667     if (s->flags & DIRTY_BITMAP_MIG_FLAG_DEVICE_NAME) {
 668         if (!qemu_get_counted_string(f, s->node_name)) {
 669             error_report("Unable to read node name string");
 670             return -EINVAL;
 671         }
 672         s->bs = bdrv_lookup_bs(s->node_name, s->node_name, &local_err);
 673         if (!s->bs) {
 674             error_report_err(local_err);
 675             return -EINVAL;
 676         }
 677     } else if (!s->bs && !nothing) {
 678         error_report("Error: block device name is not set");
 679         return -EINVAL;
 680     }
 681
 682     if (s->flags & DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME) {
 683         if (!qemu_get_counted_string(f, s->bitmap_name)) {
 684             error_report("Unable to read bitmap name string");
 685             return -EINVAL;
 686         }
 687         s->bitmap = bdrv_find_dirty_bitmap(s->bs, s->bitmap_name);
 688
 689         /* bitmap may be NULL here, it wouldn't be an error if it is the
 690          * first occurrence of the bitmap */
 691         if (!s->bitmap && !(s->flags & DIRTY_BITMAP_MIG_FLAG_START)) {
 692             error_report("Error: unknown dirty bitmap "
 693                          "'%s' for block device '%s'",
 694                          s->bitmap_name, s->node_name);
 695             return -EINVAL;
 696         }
 697     } else if (!s->bitmap && !nothing) {
 698         error_report("Error: block device name is not set");
 699         return -EINVAL;
 700     }
 701
 702     return 0;
 703 }
 704
 705 static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id)
 706 {
 707     static DirtyBitmapLoadState s;
 708     int ret = 0;
 709
 710     trace_dirty_bitmap_load_enter();
 711
 712     if (version_id != 1) {
 713         return -EINVAL;
 714     }
 715
 716     do {
 717         ret = dirty_bitmap_load_header(f, &s);
 718         if (ret < 0) {
 719             return ret;
 720         }
 721
 722         if (s.flags & DIRTY_BITMAP_MIG_FLAG_START) {
 723             ret = dirty_bitmap_load_start(f, &s);
 724         } else if (s.flags & DIRTY_BITMAP_MIG_FLAG_COMPLETE) {
 725             dirty_bitmap_load_complete(f, &s);
 726         } else if (s.flags & DIRTY_BITMAP_MIG_FLAG_BITS) {
 727             ret = dirty_bitmap_load_bits(f, &s);
 728         }
 729
 730         if (!ret) {
 731             ret = qemu_file_get_error(f);
 732         }
 733
 734         if (ret) {
 735             return ret;
 736         }
 737     } while (!(s.flags & DIRTY_BITMAP_MIG_FLAG_EOS));
 738
 739     trace_dirty_bitmap_load_success();
 740     return 0;
 741 }
 742
 743 static int dirty_bitmap_save_setup(QEMUFile *f, void *opaque)
 744 {
 745     DirtyBitmapMigBitmapState *dbms = NULL;
 746     if (init_dirty_bitmap_migration() < 0) {
 747         return -1;
 748     }
 749
 750     QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {
 751         send_bitmap_start(f, dbms);
 752     }
 753     qemu_put_bitmap_flags(f, DIRTY_BITMAP_MIG_FLAG_EOS);
 754
 755     return 0;
 756 }
 757
 758 static bool dirty_bitmap_is_active(void *opaque)
 759 {
 760     return migrate_dirty_bitmaps() && !dirty_bitmap_mig_state.no_bitmaps;
 761 }
 762
 763 static bool dirty_bitmap_is_active_iterate(void *opaque)
 764 {
 765     return dirty_bitmap_is_active(opaque) && !runstate_is_running();
 766 }
 767
 768 static bool dirty_bitmap_has_postcopy(void *opaque)
 769 {
 770     return true;
 771 }
 772
 773 static SaveVMHandlers savevm_dirty_bitmap_handlers = {
 774     .save_setup = dirty_bitmap_save_setup,
 775     .save_live_complete_postcopy = dirty_bitmap_save_complete,
 776     .save_live_complete_precopy = dirty_bitmap_save_complete,
 777     .has_postcopy = dirty_bitmap_has_postcopy,
 778     .save_live_pending = dirty_bitmap_save_pending,
 779     .save_live_iterate = dirty_bitmap_save_iterate,
 780     .is_active_iterate = dirty_bitmap_is_active_iterate,
 781     .load_state = dirty_bitmap_load,
 782     .save_cleanup = dirty_bitmap_save_cleanup,
 783     .is_active = dirty_bitmap_is_active,
 784 };
 785
 786 void dirty_bitmap_mig_init(void)
 787 {
 788     QSIMPLEQ_INIT(&dirty_bitmap_mig_state.dbms_list);
 789
 790     register_savevm_live("dirty-bitmap", 0, 1,
 791                          &savevm_dirty_bitmap_handlers,
 792                          &dirty_bitmap_mig_state);
 793 }