migration/migration.c

   1 /*
   2  * QEMU live migration
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  * 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 "qemu/main-loop.h"
  18 #include "migration/migration.h"
  19 #include "monitor/monitor.h"
  20 #include "migration/qemu-file.h"
  21 #include "sysemu/sysemu.h"
  22 #include "block/block.h"
  23 #include "qemu/sockets.h"
  24 #include "migration/block.h"
  25 #include "qemu/thread.h"
  26 #include "qmp-commands.h"
  27 #include "trace.h"
  28
  29 #define MAX_THROTTLE  (32 << 20)      /* Migration speed throttling */
  30
  31 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
  32  * data. */
  33 #define BUFFER_DELAY     100
  34 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
  35
  36 /* Default compression thread count */
  37 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
  38 /* Default decompression thread count, usually decompression is at
  39  * least 4 times as fast as compression.*/
  40 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
  41 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
  42 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
  43
  44 /* Migration XBZRLE default cache size */
  45 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
  46
  47 static NotifierList migration_state_notifiers =
  48     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  49
  50 static bool deferred_incoming;
  51
  52 /* When we add fault tolerance, we could have several
  53    migrations at once.  For now we don't need to add
  54    dynamic creation of migration */
  55
  56 MigrationState *migrate_get_current(void)
  57 {
  58     static MigrationState current_migration = {
  59         .state = MIGRATION_STATUS_NONE,
  60         .bandwidth_limit = MAX_THROTTLE,
  61         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
  62         .mbps = -1,
  63         .compress_thread_count = DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
  64         .decompress_thread_count = DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
  65         .compress_level = DEFAULT_MIGRATE_COMPRESS_LEVEL,
  66     };
  67
  68     return &current_migration;
  69 }
  70
  71 /*
  72  * Called on -incoming with a defer: uri.
  73  * The migration can be started later after any parameters have been
  74  * changed.
  75  */
  76 static void deferred_incoming_migration(Error **errp)
  77 {
  78     if (deferred_incoming) {
  79         error_setg(errp, "Incoming migration already deferred");
  80     }
  81     deferred_incoming = true;
  82 }
  83
  84 void qemu_start_incoming_migration(const char *uri, Error **errp)
  85 {
  86     const char *p;
  87
  88     if (!strcmp(uri, "defer")) {
  89         deferred_incoming_migration(errp);
  90     } else if (strstart(uri, "tcp:", &p)) {
  91         tcp_start_incoming_migration(p, errp);
  92 #ifdef CONFIG_RDMA
  93     } else if (strstart(uri, "rdma:", &p)) {
  94         rdma_start_incoming_migration(p, errp);
  95 #endif
  96 #if !defined(WIN32)
  97     } else if (strstart(uri, "exec:", &p)) {
  98         exec_start_incoming_migration(p, errp);
  99     } else if (strstart(uri, "unix:", &p)) {
 100         unix_start_incoming_migration(p, errp);
 101     } else if (strstart(uri, "fd:", &p)) {
 102         fd_start_incoming_migration(p, errp);
 103 #endif
 104     } else {
 105         error_setg(errp, "unknown migration protocol: %s", uri);
 106     }
 107 }
 108
 109 static void process_incoming_migration_co(void *opaque)
 110 {
 111     QEMUFile *f = opaque;
 112     Error *local_err = NULL;
 113     int ret;
 114
 115     ret = qemu_loadvm_state(f);
 116     qemu_fclose(f);
 117     free_xbzrle_decoded_buf();
 118     if (ret < 0) {
 119         error_report("load of migration failed: %s", strerror(-ret));
 120         migrate_decompress_threads_join();
 121         exit(EXIT_FAILURE);
 122     }
 123     qemu_announce_self();
 124
 125     /* Make sure all file formats flush their mutable metadata */
 126     bdrv_invalidate_cache_all(&local_err);
 127     if (local_err) {
 128         error_report_err(local_err);
 129         migrate_decompress_threads_join();
 130         exit(EXIT_FAILURE);
 131     }
 132
 133     if (autostart) {
 134         vm_start();
 135     } else {
 136         runstate_set(RUN_STATE_PAUSED);
 137     }
 138     migrate_decompress_threads_join();
 139 }
 140
 141 void process_incoming_migration(QEMUFile *f)
 142 {
 143     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
 144     int fd = qemu_get_fd(f);
 145
 146     assert(fd != -1);
 147     migrate_decompress_threads_create();
 148     qemu_set_nonblock(fd);
 149     qemu_coroutine_enter(co, f);
 150 }
 151
 152 /* amount of nanoseconds we are willing to wait for migration to be down.
 153  * the choice of nanoseconds is because it is the maximum resolution that
 154  * get_clock() can achieve. It is an internal measure. All user-visible
 155  * units must be in seconds */
 156 static uint64_t max_downtime = 300000000;
 157
 158 uint64_t migrate_max_downtime(void)
 159 {
 160     return max_downtime;
 161 }
 162
 163 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 164 {
 165     MigrationCapabilityStatusList *head = NULL;
 166     MigrationCapabilityStatusList *caps;
 167     MigrationState *s = migrate_get_current();
 168     int i;
 169
 170     caps = NULL; /* silence compiler warning */
 171     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 172         if (head == NULL) {
 173             head = g_malloc0(sizeof(*caps));
 174             caps = head;
 175         } else {
 176             caps->next = g_malloc0(sizeof(*caps));
 177             caps = caps->next;
 178         }
 179         caps->value =
 180             g_malloc(sizeof(*caps->value));
 181         caps->value->capability = i;
 182         caps->value->state = s->enabled_capabilities[i];
 183     }
 184
 185     return head;
 186 }
 187
 188 static void get_xbzrle_cache_stats(MigrationInfo *info)
 189 {
 190     if (migrate_use_xbzrle()) {
 191         info->has_xbzrle_cache = true;
 192         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 193         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 194         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 195         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 196         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 197         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
 198         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 199     }
 200 }
 201
 202 MigrationInfo *qmp_query_migrate(Error **errp)
 203 {
 204     MigrationInfo *info = g_malloc0(sizeof(*info));
 205     MigrationState *s = migrate_get_current();
 206
 207     switch (s->state) {
 208     case MIGRATION_STATUS_NONE:
 209         /* no migration has happened ever */
 210         break;
 211     case MIGRATION_STATUS_SETUP:
 212         info->has_status = true;
 213         info->has_total_time = false;
 214         break;
 215     case MIGRATION_STATUS_ACTIVE:
 216     case MIGRATION_STATUS_CANCELLING:
 217         info->has_status = true;
 218         info->has_total_time = true;
 219         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 220             - s->total_time;
 221         info->has_expected_downtime = true;
 222         info->expected_downtime = s->expected_downtime;
 223         info->has_setup_time = true;
 224         info->setup_time = s->setup_time;
 225
 226         info->has_ram = true;
 227         info->ram = g_malloc0(sizeof(*info->ram));
 228         info->ram->transferred = ram_bytes_transferred();
 229         info->ram->remaining = ram_bytes_remaining();
 230         info->ram->total = ram_bytes_total();
 231         info->ram->duplicate = dup_mig_pages_transferred();
 232         info->ram->skipped = skipped_mig_pages_transferred();
 233         info->ram->normal = norm_mig_pages_transferred();
 234         info->ram->normal_bytes = norm_mig_bytes_transferred();
 235         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 236         info->ram->mbps = s->mbps;
 237         info->ram->dirty_sync_count = s->dirty_sync_count;
 238
 239         if (blk_mig_active()) {
 240             info->has_disk = true;
 241             info->disk = g_malloc0(sizeof(*info->disk));
 242             info->disk->transferred = blk_mig_bytes_transferred();
 243             info->disk->remaining = blk_mig_bytes_remaining();
 244             info->disk->total = blk_mig_bytes_total();
 245         }
 246
 247         get_xbzrle_cache_stats(info);
 248         break;
 249     case MIGRATION_STATUS_COMPLETED:
 250         get_xbzrle_cache_stats(info);
 251
 252         info->has_status = true;
 253         info->has_total_time = true;
 254         info->total_time = s->total_time;
 255         info->has_downtime = true;
 256         info->downtime = s->downtime;
 257         info->has_setup_time = true;
 258         info->setup_time = s->setup_time;
 259
 260         info->has_ram = true;
 261         info->ram = g_malloc0(sizeof(*info->ram));
 262         info->ram->transferred = ram_bytes_transferred();
 263         info->ram->remaining = 0;
 264         info->ram->total = ram_bytes_total();
 265         info->ram->duplicate = dup_mig_pages_transferred();
 266         info->ram->skipped = skipped_mig_pages_transferred();
 267         info->ram->normal = norm_mig_pages_transferred();
 268         info->ram->normal_bytes = norm_mig_bytes_transferred();
 269         info->ram->mbps = s->mbps;
 270         info->ram->dirty_sync_count = s->dirty_sync_count;
 271         break;
 272     case MIGRATION_STATUS_FAILED:
 273         info->has_status = true;
 274         break;
 275     case MIGRATION_STATUS_CANCELLED:
 276         info->has_status = true;
 277         break;
 278     }
 279     info->status = s->state;
 280
 281     return info;
 282 }
 283
 284 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 285                                   Error **errp)
 286 {
 287     MigrationState *s = migrate_get_current();
 288     MigrationCapabilityStatusList *cap;
 289
 290     if (s->state == MIGRATION_STATUS_ACTIVE ||
 291         s->state == MIGRATION_STATUS_SETUP) {
 292         error_set(errp, QERR_MIGRATION_ACTIVE);
 293         return;
 294     }
 295
 296     for (cap = params; cap; cap = cap->next) {
 297         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 298     }
 299 }
 300
 301 /* shared migration helpers */
 302
 303 static void migrate_set_state(MigrationState *s, int old_state, int new_state)
 304 {
 305     if (atomic_cmpxchg(&s->state, old_state, new_state) == new_state) {
 306         trace_migrate_set_state(new_state);
 307     }
 308 }
 309
 310 static void migrate_fd_cleanup(void *opaque)
 311 {
 312     MigrationState *s = opaque;
 313
 314     qemu_bh_delete(s->cleanup_bh);
 315     s->cleanup_bh = NULL;
 316
 317     if (s->file) {
 318         trace_migrate_fd_cleanup();
 319         qemu_mutex_unlock_iothread();
 320         qemu_thread_join(&s->thread);
 321         qemu_mutex_lock_iothread();
 322
 323         migrate_compress_threads_join();
 324         qemu_fclose(s->file);
 325         s->file = NULL;
 326     }
 327
 328     assert(s->state != MIGRATION_STATUS_ACTIVE);
 329
 330     if (s->state != MIGRATION_STATUS_COMPLETED) {
 331         qemu_savevm_state_cancel();
 332         if (s->state == MIGRATION_STATUS_CANCELLING) {
 333             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
 334                               MIGRATION_STATUS_CANCELLED);
 335         }
 336     }
 337
 338     notifier_list_notify(&migration_state_notifiers, s);
 339 }
 340
 341 void migrate_fd_error(MigrationState *s)
 342 {
 343     trace_migrate_fd_error();
 344     assert(s->file == NULL);
 345     s->state = MIGRATION_STATUS_FAILED;
 346     trace_migrate_set_state(MIGRATION_STATUS_FAILED);
 347     notifier_list_notify(&migration_state_notifiers, s);
 348 }
 349
 350 static void migrate_fd_cancel(MigrationState *s)
 351 {
 352     int old_state ;
 353     QEMUFile *f = migrate_get_current()->file;
 354     trace_migrate_fd_cancel();
 355
 356     do {
 357         old_state = s->state;
 358         if (old_state != MIGRATION_STATUS_SETUP &&
 359             old_state != MIGRATION_STATUS_ACTIVE) {
 360             break;
 361         }
 362         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
 363     } while (s->state != MIGRATION_STATUS_CANCELLING);
 364
 365     /*
 366      * If we're unlucky the migration code might be stuck somewhere in a
 367      * send/write while the network has failed and is waiting to timeout;
 368      * if we've got shutdown(2) available then we can force it to quit.
 369      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
 370      * called in a bh, so there is no race against this cancel.
 371      */
 372     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
 373         qemu_file_shutdown(f);
 374     }
 375 }
 376
 377 void add_migration_state_change_notifier(Notifier *notify)
 378 {
 379     notifier_list_add(&migration_state_notifiers, notify);
 380 }
 381
 382 void remove_migration_state_change_notifier(Notifier *notify)
 383 {
 384     notifier_remove(notify);
 385 }
 386
 387 bool migration_in_setup(MigrationState *s)
 388 {
 389     return s->state == MIGRATION_STATUS_SETUP;
 390 }
 391
 392 bool migration_has_finished(MigrationState *s)
 393 {
 394     return s->state == MIGRATION_STATUS_COMPLETED;
 395 }
 396
 397 bool migration_has_failed(MigrationState *s)
 398 {
 399     return (s->state == MIGRATION_STATUS_CANCELLED ||
 400             s->state == MIGRATION_STATUS_FAILED);
 401 }
 402
 403 static MigrationState *migrate_init(const MigrationParams *params)
 404 {
 405     MigrationState *s = migrate_get_current();
 406     int64_t bandwidth_limit = s->bandwidth_limit;
 407     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 408     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 409     int compress_level = s->compress_level;
 410     int compress_thread_count = s->compress_thread_count;
 411     int decompress_thread_count = s->decompress_thread_count;
 412
 413     memcpy(enabled_capabilities, s->enabled_capabilities,
 414            sizeof(enabled_capabilities));
 415
 416     memset(s, 0, sizeof(*s));
 417     s->params = *params;
 418     memcpy(s->enabled_capabilities, enabled_capabilities,
 419            sizeof(enabled_capabilities));
 420     s->xbzrle_cache_size = xbzrle_cache_size;
 421
 422     s->compress_level = compress_level;
 423     s->compress_thread_count = compress_thread_count;
 424     s->decompress_thread_count = decompress_thread_count;
 425     s->bandwidth_limit = bandwidth_limit;
 426     s->state = MIGRATION_STATUS_SETUP;
 427     trace_migrate_set_state(MIGRATION_STATUS_SETUP);
 428
 429     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 430     return s;
 431 }
 432
 433 static GSList *migration_blockers;
 434
 435 void migrate_add_blocker(Error *reason)
 436 {
 437     migration_blockers = g_slist_prepend(migration_blockers, reason);
 438 }
 439
 440 void migrate_del_blocker(Error *reason)
 441 {
 442     migration_blockers = g_slist_remove(migration_blockers, reason);
 443 }
 444
 445 void qmp_migrate_incoming(const char *uri, Error **errp)
 446 {
 447     Error *local_err = NULL;
 448     static bool once = true;
 449
 450     if (!deferred_incoming) {
 451         error_setg(errp, "For use with '-incoming defer'");
 452         return;
 453     }
 454     if (!once) {
 455         error_setg(errp, "The incoming migration has already been started");
 456     }
 457
 458     qemu_start_incoming_migration(uri, &local_err);
 459
 460     if (local_err) {
 461         error_propagate(errp, local_err);
 462         return;
 463     }
 464
 465     once = false;
 466 }
 467
 468 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 469                  bool has_inc, bool inc, bool has_detach, bool detach,
 470                  Error **errp)
 471 {
 472     Error *local_err = NULL;
 473     MigrationState *s = migrate_get_current();
 474     MigrationParams params;
 475     const char *p;
 476
 477     params.blk = has_blk && blk;
 478     params.shared = has_inc && inc;
 479
 480     if (s->state == MIGRATION_STATUS_ACTIVE ||
 481         s->state == MIGRATION_STATUS_SETUP ||
 482         s->state == MIGRATION_STATUS_CANCELLING) {
 483         error_set(errp, QERR_MIGRATION_ACTIVE);
 484         return;
 485     }
 486
 487     if (runstate_check(RUN_STATE_INMIGRATE)) {
 488         error_setg(errp, "Guest is waiting for an incoming migration");
 489         return;
 490     }
 491
 492     if (qemu_savevm_state_blocked(errp)) {
 493         return;
 494     }
 495
 496     if (migration_blockers) {
 497         *errp = error_copy(migration_blockers->data);
 498         return;
 499     }
 500
 501     s = migrate_init(&params);
 502
 503     if (strstart(uri, "tcp:", &p)) {
 504         tcp_start_outgoing_migration(s, p, &local_err);
 505 #ifdef CONFIG_RDMA
 506     } else if (strstart(uri, "rdma:", &p)) {
 507         rdma_start_outgoing_migration(s, p, &local_err);
 508 #endif
 509 #if !defined(WIN32)
 510     } else if (strstart(uri, "exec:", &p)) {
 511         exec_start_outgoing_migration(s, p, &local_err);
 512     } else if (strstart(uri, "unix:", &p)) {
 513         unix_start_outgoing_migration(s, p, &local_err);
 514     } else if (strstart(uri, "fd:", &p)) {
 515         fd_start_outgoing_migration(s, p, &local_err);
 516 #endif
 517     } else {
 518         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 519         s->state = MIGRATION_STATUS_FAILED;
 520         return;
 521     }
 522
 523     if (local_err) {
 524         migrate_fd_error(s);
 525         error_propagate(errp, local_err);
 526         return;
 527     }
 528 }
 529
 530 void qmp_migrate_cancel(Error **errp)
 531 {
 532     migrate_fd_cancel(migrate_get_current());
 533 }
 534
 535 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 536 {
 537     MigrationState *s = migrate_get_current();
 538     int64_t new_size;
 539
 540     /* Check for truncation */
 541     if (value != (size_t)value) {
 542         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 543                   "exceeding address space");
 544         return;
 545     }
 546
 547     /* Cache should not be larger than guest ram size */
 548     if (value > ram_bytes_total()) {
 549         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 550                   "exceeds guest ram size ");
 551         return;
 552     }
 553
 554     new_size = xbzrle_cache_resize(value);
 555     if (new_size < 0) {
 556         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 557                   "is smaller than page size");
 558         return;
 559     }
 560
 561     s->xbzrle_cache_size = new_size;
 562 }
 563
 564 int64_t qmp_query_migrate_cache_size(Error **errp)
 565 {
 566     return migrate_xbzrle_cache_size();
 567 }
 568
 569 void qmp_migrate_set_speed(int64_t value, Error **errp)
 570 {
 571     MigrationState *s;
 572
 573     if (value < 0) {
 574         value = 0;
 575     }
 576     if (value > SIZE_MAX) {
 577         value = SIZE_MAX;
 578     }
 579
 580     s = migrate_get_current();
 581     s->bandwidth_limit = value;
 582     if (s->file) {
 583         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
 584     }
 585 }
 586
 587 void qmp_migrate_set_downtime(double value, Error **errp)
 588 {
 589     value *= 1e9;
 590     value = MAX(0, MIN(UINT64_MAX, value));
 591     max_downtime = (uint64_t)value;
 592 }
 593
 594 bool migrate_auto_converge(void)
 595 {
 596     MigrationState *s;
 597
 598     s = migrate_get_current();
 599
 600     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
 601 }
 602
 603 bool migrate_zero_blocks(void)
 604 {
 605     MigrationState *s;
 606
 607     s = migrate_get_current();
 608
 609     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
 610 }
 611
 612 bool migrate_use_compression(void)
 613 {
 614     /* Disable compression before the patch series are applied */
 615     return false;
 616 }
 617
 618 int migrate_compress_level(void)
 619 {
 620     MigrationState *s;
 621
 622     s = migrate_get_current();
 623
 624     return s->compress_level;
 625 }
 626
 627 int migrate_compress_threads(void)
 628 {
 629     MigrationState *s;
 630
 631     s = migrate_get_current();
 632
 633     return s->compress_thread_count;
 634 }
 635
 636 int migrate_decompress_threads(void)
 637 {
 638     MigrationState *s;
 639
 640     s = migrate_get_current();
 641
 642     return s->decompress_thread_count;
 643 }
 644
 645 int migrate_use_xbzrle(void)
 646 {
 647     MigrationState *s;
 648
 649     s = migrate_get_current();
 650
 651     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 652 }
 653
 654 int64_t migrate_xbzrle_cache_size(void)
 655 {
 656     MigrationState *s;
 657
 658     s = migrate_get_current();
 659
 660     return s->xbzrle_cache_size;
 661 }
 662
 663 /* migration thread support */
 664
 665 static void *migration_thread(void *opaque)
 666 {
 667     MigrationState *s = opaque;
 668     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 669     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 670     int64_t initial_bytes = 0;
 671     int64_t max_size = 0;
 672     int64_t start_time = initial_time;
 673     bool old_vm_running = false;
 674
 675     qemu_savevm_state_begin(s->file, &s->params);
 676
 677     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
 678     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
 679
 680     while (s->state == MIGRATION_STATUS_ACTIVE) {
 681         int64_t current_time;
 682         uint64_t pending_size;
 683
 684         if (!qemu_file_rate_limit(s->file)) {
 685             pending_size = qemu_savevm_state_pending(s->file, max_size);
 686             trace_migrate_pending(pending_size, max_size);
 687             if (pending_size && pending_size >= max_size) {
 688                 qemu_savevm_state_iterate(s->file);
 689             } else {
 690                 int ret;
 691
 692                 qemu_mutex_lock_iothread();
 693                 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 694                 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 695                 old_vm_running = runstate_is_running();
 696
 697                 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 698                 if (ret >= 0) {
 699                     qemu_file_set_rate_limit(s->file, INT64_MAX);
 700                     qemu_savevm_state_complete(s->file);
 701                 }
 702                 qemu_mutex_unlock_iothread();
 703
 704                 if (ret < 0) {
 705                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 706                                       MIGRATION_STATUS_FAILED);
 707                     break;
 708                 }
 709
 710                 if (!qemu_file_get_error(s->file)) {
 711                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 712                                       MIGRATION_STATUS_COMPLETED);
 713                     break;
 714                 }
 715             }
 716         }
 717
 718         if (qemu_file_get_error(s->file)) {
 719             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 720                               MIGRATION_STATUS_FAILED);
 721             break;
 722         }
 723         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 724         if (current_time >= initial_time + BUFFER_DELAY) {
 725             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
 726             uint64_t time_spent = current_time - initial_time;
 727             double bandwidth = transferred_bytes / time_spent;
 728             max_size = bandwidth * migrate_max_downtime() / 1000000;
 729
 730             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
 731                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
 732
 733             trace_migrate_transferred(transferred_bytes, time_spent,
 734                                       bandwidth, max_size);
 735             /* if we haven't sent anything, we don't want to recalculate
 736                10000 is a small enough number for our purposes */
 737             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
 738                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
 739             }
 740
 741             qemu_file_reset_rate_limit(s->file);
 742             initial_time = current_time;
 743             initial_bytes = qemu_ftell(s->file);
 744         }
 745         if (qemu_file_rate_limit(s->file)) {
 746             /* usleep expects microseconds */
 747             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
 748         }
 749     }
 750
 751     qemu_mutex_lock_iothread();
 752     if (s->state == MIGRATION_STATUS_COMPLETED) {
 753         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 754         uint64_t transferred_bytes = qemu_ftell(s->file);
 755         s->total_time = end_time - s->total_time;
 756         s->downtime = end_time - start_time;
 757         if (s->total_time) {
 758             s->mbps = (((double) transferred_bytes * 8.0) /
 759                        ((double) s->total_time)) / 1000;
 760         }
 761         runstate_set(RUN_STATE_POSTMIGRATE);
 762     } else {
 763         if (old_vm_running) {
 764             vm_start();
 765         }
 766     }
 767     qemu_bh_schedule(s->cleanup_bh);
 768     qemu_mutex_unlock_iothread();
 769
 770     return NULL;
 771 }
 772
 773 void migrate_fd_connect(MigrationState *s)
 774 {
 775     s->state = MIGRATION_STATUS_SETUP;
 776     trace_migrate_set_state(MIGRATION_STATUS_SETUP);
 777
 778     /* This is a best 1st approximation. ns to ms */
 779     s->expected_downtime = max_downtime/1000000;
 780     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
 781
 782     qemu_file_set_rate_limit(s->file,
 783                              s->bandwidth_limit / XFER_LIMIT_RATIO);
 784
 785     /* Notify before starting migration thread */
 786     notifier_list_notify(&migration_state_notifiers, s);
 787
 788     migrate_compress_threads_create();
 789     qemu_thread_create(&s->thread, "migration", migration_thread, s,
 790                        QEMU_THREAD_JOINABLE);
 791 }