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         .parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] =
  64                 DEFAULT_MIGRATE_COMPRESS_LEVEL,
  65         .parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
  66                 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
  67         .parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
  68                 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
  69     };
  70
  71     return &current_migration;
  72 }
  73
  74 /*
  75  * Called on -incoming with a defer: uri.
  76  * The migration can be started later after any parameters have been
  77  * changed.
  78  */
  79 static void deferred_incoming_migration(Error **errp)
  80 {
  81     if (deferred_incoming) {
  82         error_setg(errp, "Incoming migration already deferred");
  83     }
  84     deferred_incoming = true;
  85 }
  86
  87 void qemu_start_incoming_migration(const char *uri, Error **errp)
  88 {
  89     const char *p;
  90
  91     if (!strcmp(uri, "defer")) {
  92         deferred_incoming_migration(errp);
  93     } else if (strstart(uri, "tcp:", &p)) {
  94         tcp_start_incoming_migration(p, errp);
  95 #ifdef CONFIG_RDMA
  96     } else if (strstart(uri, "rdma:", &p)) {
  97         rdma_start_incoming_migration(p, errp);
  98 #endif
  99 #if !defined(WIN32)
 100     } else if (strstart(uri, "exec:", &p)) {
 101         exec_start_incoming_migration(p, errp);
 102     } else if (strstart(uri, "unix:", &p)) {
 103         unix_start_incoming_migration(p, errp);
 104     } else if (strstart(uri, "fd:", &p)) {
 105         fd_start_incoming_migration(p, errp);
 106 #endif
 107     } else {
 108         error_setg(errp, "unknown migration protocol: %s", uri);
 109     }
 110 }
 111
 112 static void process_incoming_migration_co(void *opaque)
 113 {
 114     QEMUFile *f = opaque;
 115     Error *local_err = NULL;
 116     int ret;
 117
 118     ret = qemu_loadvm_state(f);
 119     qemu_fclose(f);
 120     free_xbzrle_decoded_buf();
 121     if (ret < 0) {
 122         error_report("load of migration failed: %s", strerror(-ret));
 123         migrate_decompress_threads_join();
 124         exit(EXIT_FAILURE);
 125     }
 126     qemu_announce_self();
 127
 128     /* Make sure all file formats flush their mutable metadata */
 129     bdrv_invalidate_cache_all(&local_err);
 130     if (local_err) {
 131         error_report_err(local_err);
 132         migrate_decompress_threads_join();
 133         exit(EXIT_FAILURE);
 134     }
 135
 136     if (autostart) {
 137         vm_start();
 138     } else {
 139         runstate_set(RUN_STATE_PAUSED);
 140     }
 141     migrate_decompress_threads_join();
 142 }
 143
 144 void process_incoming_migration(QEMUFile *f)
 145 {
 146     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
 147     int fd = qemu_get_fd(f);
 148
 149     assert(fd != -1);
 150     migrate_decompress_threads_create();
 151     qemu_set_nonblock(fd);
 152     qemu_coroutine_enter(co, f);
 153 }
 154
 155 /* amount of nanoseconds we are willing to wait for migration to be down.
 156  * the choice of nanoseconds is because it is the maximum resolution that
 157  * get_clock() can achieve. It is an internal measure. All user-visible
 158  * units must be in seconds */
 159 static uint64_t max_downtime = 300000000;
 160
 161 uint64_t migrate_max_downtime(void)
 162 {
 163     return max_downtime;
 164 }
 165
 166 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 167 {
 168     MigrationCapabilityStatusList *head = NULL;
 169     MigrationCapabilityStatusList *caps;
 170     MigrationState *s = migrate_get_current();
 171     int i;
 172
 173     caps = NULL; /* silence compiler warning */
 174     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 175         if (head == NULL) {
 176             head = g_malloc0(sizeof(*caps));
 177             caps = head;
 178         } else {
 179             caps->next = g_malloc0(sizeof(*caps));
 180             caps = caps->next;
 181         }
 182         caps->value =
 183             g_malloc(sizeof(*caps->value));
 184         caps->value->capability = i;
 185         caps->value->state = s->enabled_capabilities[i];
 186     }
 187
 188     return head;
 189 }
 190
 191 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 192 {
 193     MigrationParameters *params;
 194     MigrationState *s = migrate_get_current();
 195
 196     params = g_malloc0(sizeof(*params));
 197     params->compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 198     params->compress_threads =
 199             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 200     params->decompress_threads =
 201             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 202
 203     return params;
 204 }
 205
 206 static void get_xbzrle_cache_stats(MigrationInfo *info)
 207 {
 208     if (migrate_use_xbzrle()) {
 209         info->has_xbzrle_cache = true;
 210         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 211         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 212         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 213         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 214         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 215         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
 216         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 217     }
 218 }
 219
 220 MigrationInfo *qmp_query_migrate(Error **errp)
 221 {
 222     MigrationInfo *info = g_malloc0(sizeof(*info));
 223     MigrationState *s = migrate_get_current();
 224
 225     switch (s->state) {
 226     case MIGRATION_STATUS_NONE:
 227         /* no migration has happened ever */
 228         break;
 229     case MIGRATION_STATUS_SETUP:
 230         info->has_status = true;
 231         info->has_total_time = false;
 232         break;
 233     case MIGRATION_STATUS_ACTIVE:
 234     case MIGRATION_STATUS_CANCELLING:
 235         info->has_status = true;
 236         info->has_total_time = true;
 237         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 238             - s->total_time;
 239         info->has_expected_downtime = true;
 240         info->expected_downtime = s->expected_downtime;
 241         info->has_setup_time = true;
 242         info->setup_time = s->setup_time;
 243
 244         info->has_ram = true;
 245         info->ram = g_malloc0(sizeof(*info->ram));
 246         info->ram->transferred = ram_bytes_transferred();
 247         info->ram->remaining = ram_bytes_remaining();
 248         info->ram->total = ram_bytes_total();
 249         info->ram->duplicate = dup_mig_pages_transferred();
 250         info->ram->skipped = skipped_mig_pages_transferred();
 251         info->ram->normal = norm_mig_pages_transferred();
 252         info->ram->normal_bytes = norm_mig_bytes_transferred();
 253         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 254         info->ram->mbps = s->mbps;
 255         info->ram->dirty_sync_count = s->dirty_sync_count;
 256
 257         if (blk_mig_active()) {
 258             info->has_disk = true;
 259             info->disk = g_malloc0(sizeof(*info->disk));
 260             info->disk->transferred = blk_mig_bytes_transferred();
 261             info->disk->remaining = blk_mig_bytes_remaining();
 262             info->disk->total = blk_mig_bytes_total();
 263         }
 264
 265         get_xbzrle_cache_stats(info);
 266         break;
 267     case MIGRATION_STATUS_COMPLETED:
 268         get_xbzrle_cache_stats(info);
 269
 270         info->has_status = true;
 271         info->has_total_time = true;
 272         info->total_time = s->total_time;
 273         info->has_downtime = true;
 274         info->downtime = s->downtime;
 275         info->has_setup_time = true;
 276         info->setup_time = s->setup_time;
 277
 278         info->has_ram = true;
 279         info->ram = g_malloc0(sizeof(*info->ram));
 280         info->ram->transferred = ram_bytes_transferred();
 281         info->ram->remaining = 0;
 282         info->ram->total = ram_bytes_total();
 283         info->ram->duplicate = dup_mig_pages_transferred();
 284         info->ram->skipped = skipped_mig_pages_transferred();
 285         info->ram->normal = norm_mig_pages_transferred();
 286         info->ram->normal_bytes = norm_mig_bytes_transferred();
 287         info->ram->mbps = s->mbps;
 288         info->ram->dirty_sync_count = s->dirty_sync_count;
 289         break;
 290     case MIGRATION_STATUS_FAILED:
 291         info->has_status = true;
 292         break;
 293     case MIGRATION_STATUS_CANCELLED:
 294         info->has_status = true;
 295         break;
 296     }
 297     info->status = s->state;
 298
 299     return info;
 300 }
 301
 302 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 303                                   Error **errp)
 304 {
 305     MigrationState *s = migrate_get_current();
 306     MigrationCapabilityStatusList *cap;
 307
 308     if (s->state == MIGRATION_STATUS_ACTIVE ||
 309         s->state == MIGRATION_STATUS_SETUP) {
 310         error_set(errp, QERR_MIGRATION_ACTIVE);
 311         return;
 312     }
 313
 314     for (cap = params; cap; cap = cap->next) {
 315         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 316     }
 317 }
 318
 319 void qmp_migrate_set_parameters(bool has_compress_level,
 320                                 int64_t compress_level,
 321                                 bool has_compress_threads,
 322                                 int64_t compress_threads,
 323                                 bool has_decompress_threads,
 324                                 int64_t decompress_threads, Error **errp)
 325 {
 326     MigrationState *s = migrate_get_current();
 327
 328     if (has_compress_level && (compress_level < 0 || compress_level > 9)) {
 329         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
 330                   "is invalid, it should be in the range of 0 to 9");
 331         return;
 332     }
 333     if (has_compress_threads &&
 334             (compress_threads < 1 || compress_threads > 255)) {
 335         error_set(errp, QERR_INVALID_PARAMETER_VALUE,
 336                   "compress_threads",
 337                   "is invalid, it should be in the range of 1 to 255");
 338         return;
 339     }
 340     if (has_decompress_threads &&
 341             (decompress_threads < 1 || decompress_threads > 255)) {
 342         error_set(errp, QERR_INVALID_PARAMETER_VALUE,
 343                   "decompress_threads",
 344                   "is invalid, it should be in the range of 1 to 255");
 345         return;
 346     }
 347
 348     if (has_compress_level) {
 349         s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 350     }
 351     if (has_compress_threads) {
 352         s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] = compress_threads;
 353     }
 354     if (has_decompress_threads) {
 355         s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 356                                                     decompress_threads;
 357     }
 358 }
 359
 360 /* shared migration helpers */
 361
 362 static void migrate_set_state(MigrationState *s, int old_state, int new_state)
 363 {
 364     if (atomic_cmpxchg(&s->state, old_state, new_state) == new_state) {
 365         trace_migrate_set_state(new_state);
 366     }
 367 }
 368
 369 static void migrate_fd_cleanup(void *opaque)
 370 {
 371     MigrationState *s = opaque;
 372
 373     qemu_bh_delete(s->cleanup_bh);
 374     s->cleanup_bh = NULL;
 375
 376     if (s->file) {
 377         trace_migrate_fd_cleanup();
 378         qemu_mutex_unlock_iothread();
 379         qemu_thread_join(&s->thread);
 380         qemu_mutex_lock_iothread();
 381
 382         migrate_compress_threads_join();
 383         qemu_fclose(s->file);
 384         s->file = NULL;
 385     }
 386
 387     assert(s->state != MIGRATION_STATUS_ACTIVE);
 388
 389     if (s->state != MIGRATION_STATUS_COMPLETED) {
 390         qemu_savevm_state_cancel();
 391         if (s->state == MIGRATION_STATUS_CANCELLING) {
 392             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
 393                               MIGRATION_STATUS_CANCELLED);
 394         }
 395     }
 396
 397     notifier_list_notify(&migration_state_notifiers, s);
 398 }
 399
 400 void migrate_fd_error(MigrationState *s)
 401 {
 402     trace_migrate_fd_error();
 403     assert(s->file == NULL);
 404     s->state = MIGRATION_STATUS_FAILED;
 405     trace_migrate_set_state(MIGRATION_STATUS_FAILED);
 406     notifier_list_notify(&migration_state_notifiers, s);
 407 }
 408
 409 static void migrate_fd_cancel(MigrationState *s)
 410 {
 411     int old_state ;
 412     QEMUFile *f = migrate_get_current()->file;
 413     trace_migrate_fd_cancel();
 414
 415     do {
 416         old_state = s->state;
 417         if (old_state != MIGRATION_STATUS_SETUP &&
 418             old_state != MIGRATION_STATUS_ACTIVE) {
 419             break;
 420         }
 421         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
 422     } while (s->state != MIGRATION_STATUS_CANCELLING);
 423
 424     /*
 425      * If we're unlucky the migration code might be stuck somewhere in a
 426      * send/write while the network has failed and is waiting to timeout;
 427      * if we've got shutdown(2) available then we can force it to quit.
 428      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
 429      * called in a bh, so there is no race against this cancel.
 430      */
 431     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
 432         qemu_file_shutdown(f);
 433     }
 434 }
 435
 436 void add_migration_state_change_notifier(Notifier *notify)
 437 {
 438     notifier_list_add(&migration_state_notifiers, notify);
 439 }
 440
 441 void remove_migration_state_change_notifier(Notifier *notify)
 442 {
 443     notifier_remove(notify);
 444 }
 445
 446 bool migration_in_setup(MigrationState *s)
 447 {
 448     return s->state == MIGRATION_STATUS_SETUP;
 449 }
 450
 451 bool migration_has_finished(MigrationState *s)
 452 {
 453     return s->state == MIGRATION_STATUS_COMPLETED;
 454 }
 455
 456 bool migration_has_failed(MigrationState *s)
 457 {
 458     return (s->state == MIGRATION_STATUS_CANCELLED ||
 459             s->state == MIGRATION_STATUS_FAILED);
 460 }
 461
 462 static MigrationState *migrate_init(const MigrationParams *params)
 463 {
 464     MigrationState *s = migrate_get_current();
 465     int64_t bandwidth_limit = s->bandwidth_limit;
 466     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 467     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 468     int compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 469     int compress_thread_count =
 470             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 471     int decompress_thread_count =
 472             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 473
 474     memcpy(enabled_capabilities, s->enabled_capabilities,
 475            sizeof(enabled_capabilities));
 476
 477     memset(s, 0, sizeof(*s));
 478     s->params = *params;
 479     memcpy(s->enabled_capabilities, enabled_capabilities,
 480            sizeof(enabled_capabilities));
 481     s->xbzrle_cache_size = xbzrle_cache_size;
 482
 483     s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 484     s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
 485                compress_thread_count;
 486     s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 487                decompress_thread_count;
 488     s->bandwidth_limit = bandwidth_limit;
 489     s->state = MIGRATION_STATUS_SETUP;
 490     trace_migrate_set_state(MIGRATION_STATUS_SETUP);
 491
 492     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 493     return s;
 494 }
 495
 496 static GSList *migration_blockers;
 497
 498 void migrate_add_blocker(Error *reason)
 499 {
 500     migration_blockers = g_slist_prepend(migration_blockers, reason);
 501 }
 502
 503 void migrate_del_blocker(Error *reason)
 504 {
 505     migration_blockers = g_slist_remove(migration_blockers, reason);
 506 }
 507
 508 void qmp_migrate_incoming(const char *uri, Error **errp)
 509 {
 510     Error *local_err = NULL;
 511     static bool once = true;
 512
 513     if (!deferred_incoming) {
 514         error_setg(errp, "For use with '-incoming defer'");
 515         return;
 516     }
 517     if (!once) {
 518         error_setg(errp, "The incoming migration has already been started");
 519     }
 520
 521     qemu_start_incoming_migration(uri, &local_err);
 522
 523     if (local_err) {
 524         error_propagate(errp, local_err);
 525         return;
 526     }
 527
 528     once = false;
 529 }
 530
 531 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 532                  bool has_inc, bool inc, bool has_detach, bool detach,
 533                  Error **errp)
 534 {
 535     Error *local_err = NULL;
 536     MigrationState *s = migrate_get_current();
 537     MigrationParams params;
 538     const char *p;
 539
 540     params.blk = has_blk && blk;
 541     params.shared = has_inc && inc;
 542
 543     if (s->state == MIGRATION_STATUS_ACTIVE ||
 544         s->state == MIGRATION_STATUS_SETUP ||
 545         s->state == MIGRATION_STATUS_CANCELLING) {
 546         error_set(errp, QERR_MIGRATION_ACTIVE);
 547         return;
 548     }
 549
 550     if (runstate_check(RUN_STATE_INMIGRATE)) {
 551         error_setg(errp, "Guest is waiting for an incoming migration");
 552         return;
 553     }
 554
 555     if (qemu_savevm_state_blocked(errp)) {
 556         return;
 557     }
 558
 559     if (migration_blockers) {
 560         *errp = error_copy(migration_blockers->data);
 561         return;
 562     }
 563
 564     s = migrate_init(&params);
 565
 566     if (strstart(uri, "tcp:", &p)) {
 567         tcp_start_outgoing_migration(s, p, &local_err);
 568 #ifdef CONFIG_RDMA
 569     } else if (strstart(uri, "rdma:", &p)) {
 570         rdma_start_outgoing_migration(s, p, &local_err);
 571 #endif
 572 #if !defined(WIN32)
 573     } else if (strstart(uri, "exec:", &p)) {
 574         exec_start_outgoing_migration(s, p, &local_err);
 575     } else if (strstart(uri, "unix:", &p)) {
 576         unix_start_outgoing_migration(s, p, &local_err);
 577     } else if (strstart(uri, "fd:", &p)) {
 578         fd_start_outgoing_migration(s, p, &local_err);
 579 #endif
 580     } else {
 581         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 582         s->state = MIGRATION_STATUS_FAILED;
 583         return;
 584     }
 585
 586     if (local_err) {
 587         migrate_fd_error(s);
 588         error_propagate(errp, local_err);
 589         return;
 590     }
 591 }
 592
 593 void qmp_migrate_cancel(Error **errp)
 594 {
 595     migrate_fd_cancel(migrate_get_current());
 596 }
 597
 598 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 599 {
 600     MigrationState *s = migrate_get_current();
 601     int64_t new_size;
 602
 603     /* Check for truncation */
 604     if (value != (size_t)value) {
 605         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 606                   "exceeding address space");
 607         return;
 608     }
 609
 610     /* Cache should not be larger than guest ram size */
 611     if (value > ram_bytes_total()) {
 612         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 613                   "exceeds guest ram size ");
 614         return;
 615     }
 616
 617     new_size = xbzrle_cache_resize(value);
 618     if (new_size < 0) {
 619         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 620                   "is smaller than page size");
 621         return;
 622     }
 623
 624     s->xbzrle_cache_size = new_size;
 625 }
 626
 627 int64_t qmp_query_migrate_cache_size(Error **errp)
 628 {
 629     return migrate_xbzrle_cache_size();
 630 }
 631
 632 void qmp_migrate_set_speed(int64_t value, Error **errp)
 633 {
 634     MigrationState *s;
 635
 636     if (value < 0) {
 637         value = 0;
 638     }
 639     if (value > SIZE_MAX) {
 640         value = SIZE_MAX;
 641     }
 642
 643     s = migrate_get_current();
 644     s->bandwidth_limit = value;
 645     if (s->file) {
 646         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
 647     }
 648 }
 649
 650 void qmp_migrate_set_downtime(double value, Error **errp)
 651 {
 652     value *= 1e9;
 653     value = MAX(0, MIN(UINT64_MAX, value));
 654     max_downtime = (uint64_t)value;
 655 }
 656
 657 bool migrate_auto_converge(void)
 658 {
 659     MigrationState *s;
 660
 661     s = migrate_get_current();
 662
 663     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
 664 }
 665
 666 bool migrate_zero_blocks(void)
 667 {
 668     MigrationState *s;
 669
 670     s = migrate_get_current();
 671
 672     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
 673 }
 674
 675 bool migrate_use_compression(void)
 676 {
 677     MigrationState *s;
 678
 679     s = migrate_get_current();
 680
 681     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
 682 }
 683
 684 int migrate_compress_level(void)
 685 {
 686     MigrationState *s;
 687
 688     s = migrate_get_current();
 689
 690     return s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 691 }
 692
 693 int migrate_compress_threads(void)
 694 {
 695     MigrationState *s;
 696
 697     s = migrate_get_current();
 698
 699     return s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 700 }
 701
 702 int migrate_decompress_threads(void)
 703 {
 704     MigrationState *s;
 705
 706     s = migrate_get_current();
 707
 708     return s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 709 }
 710
 711 int migrate_use_xbzrle(void)
 712 {
 713     MigrationState *s;
 714
 715     s = migrate_get_current();
 716
 717     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 718 }
 719
 720 int64_t migrate_xbzrle_cache_size(void)
 721 {
 722     MigrationState *s;
 723
 724     s = migrate_get_current();
 725
 726     return s->xbzrle_cache_size;
 727 }
 728
 729 /* migration thread support */
 730
 731 static void *migration_thread(void *opaque)
 732 {
 733     MigrationState *s = opaque;
 734     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 735     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 736     int64_t initial_bytes = 0;
 737     int64_t max_size = 0;
 738     int64_t start_time = initial_time;
 739     bool old_vm_running = false;
 740
 741     qemu_savevm_state_begin(s->file, &s->params);
 742
 743     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
 744     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
 745
 746     while (s->state == MIGRATION_STATUS_ACTIVE) {
 747         int64_t current_time;
 748         uint64_t pending_size;
 749
 750         if (!qemu_file_rate_limit(s->file)) {
 751             pending_size = qemu_savevm_state_pending(s->file, max_size);
 752             trace_migrate_pending(pending_size, max_size);
 753             if (pending_size && pending_size >= max_size) {
 754                 qemu_savevm_state_iterate(s->file);
 755             } else {
 756                 int ret;
 757
 758                 qemu_mutex_lock_iothread();
 759                 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 760                 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 761                 old_vm_running = runstate_is_running();
 762
 763                 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 764                 if (ret >= 0) {
 765                     qemu_file_set_rate_limit(s->file, INT64_MAX);
 766                     qemu_savevm_state_complete(s->file);
 767                 }
 768                 qemu_mutex_unlock_iothread();
 769
 770                 if (ret < 0) {
 771                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 772                                       MIGRATION_STATUS_FAILED);
 773                     break;
 774                 }
 775
 776                 if (!qemu_file_get_error(s->file)) {
 777                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 778                                       MIGRATION_STATUS_COMPLETED);
 779                     break;
 780                 }
 781             }
 782         }
 783
 784         if (qemu_file_get_error(s->file)) {
 785             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 786                               MIGRATION_STATUS_FAILED);
 787             break;
 788         }
 789         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 790         if (current_time >= initial_time + BUFFER_DELAY) {
 791             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
 792             uint64_t time_spent = current_time - initial_time;
 793             double bandwidth = transferred_bytes / time_spent;
 794             max_size = bandwidth * migrate_max_downtime() / 1000000;
 795
 796             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
 797                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
 798
 799             trace_migrate_transferred(transferred_bytes, time_spent,
 800                                       bandwidth, max_size);
 801             /* if we haven't sent anything, we don't want to recalculate
 802                10000 is a small enough number for our purposes */
 803             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
 804                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
 805             }
 806
 807             qemu_file_reset_rate_limit(s->file);
 808             initial_time = current_time;
 809             initial_bytes = qemu_ftell(s->file);
 810         }
 811         if (qemu_file_rate_limit(s->file)) {
 812             /* usleep expects microseconds */
 813             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
 814         }
 815     }
 816
 817     qemu_mutex_lock_iothread();
 818     if (s->state == MIGRATION_STATUS_COMPLETED) {
 819         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 820         uint64_t transferred_bytes = qemu_ftell(s->file);
 821         s->total_time = end_time - s->total_time;
 822         s->downtime = end_time - start_time;
 823         if (s->total_time) {
 824             s->mbps = (((double) transferred_bytes * 8.0) /
 825                        ((double) s->total_time)) / 1000;
 826         }
 827         runstate_set(RUN_STATE_POSTMIGRATE);
 828     } else {
 829         if (old_vm_running) {
 830             vm_start();
 831         }
 832     }
 833     qemu_bh_schedule(s->cleanup_bh);
 834     qemu_mutex_unlock_iothread();
 835
 836     return NULL;
 837 }
 838
 839 void migrate_fd_connect(MigrationState *s)
 840 {
 841     s->state = MIGRATION_STATUS_SETUP;
 842     trace_migrate_set_state(MIGRATION_STATUS_SETUP);
 843
 844     /* This is a best 1st approximation. ns to ms */
 845     s->expected_downtime = max_downtime/1000000;
 846     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
 847
 848     qemu_file_set_rate_limit(s->file,
 849                              s->bandwidth_limit / XFER_LIMIT_RATIO);
 850
 851     /* Notify before starting migration thread */
 852     notifier_list_notify(&migration_state_notifiers, s);
 853
 854     migrate_compress_threads_create();
 855     qemu_thread_create(&s->thread, "migration", migration_thread, s,
 856                        QEMU_THREAD_JOINABLE);
 857 }