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 /* For outgoing */
  57 MigrationState *migrate_get_current(void)
  58 {
  59     static MigrationState current_migration = {
  60         .state = MIGRATION_STATUS_NONE,
  61         .bandwidth_limit = MAX_THROTTLE,
  62         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
  63         .mbps = -1,
  64         .parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] =
  65                 DEFAULT_MIGRATE_COMPRESS_LEVEL,
  66         .parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
  67                 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
  68         .parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
  69                 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
  70     };
  71
  72     return &current_migration;
  73 }
  74
  75 /* For incoming */
  76 static MigrationIncomingState *mis_current;
  77
  78 MigrationIncomingState *migration_incoming_get_current(void)
  79 {
  80     return mis_current;
  81 }
  82
  83 MigrationIncomingState *migration_incoming_state_new(QEMUFile* f)
  84 {
  85     mis_current = g_malloc0(sizeof(MigrationIncomingState));
  86     mis_current->file = f;
  87     QLIST_INIT(&mis_current->loadvm_handlers);
  88
  89     return mis_current;
  90 }
  91
  92 void migration_incoming_state_destroy(void)
  93 {
  94     loadvm_free_handlers(mis_current);
  95     g_free(mis_current);
  96     mis_current = NULL;
  97 }
  98
  99 /*
 100  * Called on -incoming with a defer: uri.
 101  * The migration can be started later after any parameters have been
 102  * changed.
 103  */
 104 static void deferred_incoming_migration(Error **errp)
 105 {
 106     if (deferred_incoming) {
 107         error_setg(errp, "Incoming migration already deferred");
 108     }
 109     deferred_incoming = true;
 110 }
 111
 112 void qemu_start_incoming_migration(const char *uri, Error **errp)
 113 {
 114     const char *p;
 115
 116     if (!strcmp(uri, "defer")) {
 117         deferred_incoming_migration(errp);
 118     } else if (strstart(uri, "tcp:", &p)) {
 119         tcp_start_incoming_migration(p, errp);
 120 #ifdef CONFIG_RDMA
 121     } else if (strstart(uri, "rdma:", &p)) {
 122         rdma_start_incoming_migration(p, errp);
 123 #endif
 124 #if !defined(WIN32)
 125     } else if (strstart(uri, "exec:", &p)) {
 126         exec_start_incoming_migration(p, errp);
 127     } else if (strstart(uri, "unix:", &p)) {
 128         unix_start_incoming_migration(p, errp);
 129     } else if (strstart(uri, "fd:", &p)) {
 130         fd_start_incoming_migration(p, errp);
 131 #endif
 132     } else {
 133         error_setg(errp, "unknown migration protocol: %s", uri);
 134     }
 135 }
 136
 137 static void process_incoming_migration_co(void *opaque)
 138 {
 139     QEMUFile *f = opaque;
 140     Error *local_err = NULL;
 141     int ret;
 142
 143     migration_incoming_state_new(f);
 144
 145     ret = qemu_loadvm_state(f);
 146
 147     qemu_fclose(f);
 148     free_xbzrle_decoded_buf();
 149     migration_incoming_state_destroy();
 150
 151     if (ret < 0) {
 152         error_report("load of migration failed: %s", strerror(-ret));
 153         migrate_decompress_threads_join();
 154         exit(EXIT_FAILURE);
 155     }
 156     qemu_announce_self();
 157
 158     /* Make sure all file formats flush their mutable metadata */
 159     bdrv_invalidate_cache_all(&local_err);
 160     if (local_err) {
 161         error_report_err(local_err);
 162         migrate_decompress_threads_join();
 163         exit(EXIT_FAILURE);
 164     }
 165
 166     if (autostart) {
 167         vm_start();
 168     } else {
 169         runstate_set(RUN_STATE_PAUSED);
 170     }
 171     migrate_decompress_threads_join();
 172 }
 173
 174 void process_incoming_migration(QEMUFile *f)
 175 {
 176     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
 177     int fd = qemu_get_fd(f);
 178
 179     assert(fd != -1);
 180     migrate_decompress_threads_create();
 181     qemu_set_nonblock(fd);
 182     qemu_coroutine_enter(co, f);
 183 }
 184
 185 /* amount of nanoseconds we are willing to wait for migration to be down.
 186  * the choice of nanoseconds is because it is the maximum resolution that
 187  * get_clock() can achieve. It is an internal measure. All user-visible
 188  * units must be in seconds */
 189 static uint64_t max_downtime = 300000000;
 190
 191 uint64_t migrate_max_downtime(void)
 192 {
 193     return max_downtime;
 194 }
 195
 196 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 197 {
 198     MigrationCapabilityStatusList *head = NULL;
 199     MigrationCapabilityStatusList *caps;
 200     MigrationState *s = migrate_get_current();
 201     int i;
 202
 203     caps = NULL; /* silence compiler warning */
 204     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 205         if (head == NULL) {
 206             head = g_malloc0(sizeof(*caps));
 207             caps = head;
 208         } else {
 209             caps->next = g_malloc0(sizeof(*caps));
 210             caps = caps->next;
 211         }
 212         caps->value =
 213             g_malloc(sizeof(*caps->value));
 214         caps->value->capability = i;
 215         caps->value->state = s->enabled_capabilities[i];
 216     }
 217
 218     return head;
 219 }
 220
 221 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 222 {
 223     MigrationParameters *params;
 224     MigrationState *s = migrate_get_current();
 225
 226     params = g_malloc0(sizeof(*params));
 227     params->compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 228     params->compress_threads =
 229             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 230     params->decompress_threads =
 231             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 232
 233     return params;
 234 }
 235
 236 static void get_xbzrle_cache_stats(MigrationInfo *info)
 237 {
 238     if (migrate_use_xbzrle()) {
 239         info->has_xbzrle_cache = true;
 240         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 241         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 242         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 243         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 244         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 245         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
 246         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 247     }
 248 }
 249
 250 MigrationInfo *qmp_query_migrate(Error **errp)
 251 {
 252     MigrationInfo *info = g_malloc0(sizeof(*info));
 253     MigrationState *s = migrate_get_current();
 254
 255     switch (s->state) {
 256     case MIGRATION_STATUS_NONE:
 257         /* no migration has happened ever */
 258         break;
 259     case MIGRATION_STATUS_SETUP:
 260         info->has_status = true;
 261         info->has_total_time = false;
 262         break;
 263     case MIGRATION_STATUS_ACTIVE:
 264     case MIGRATION_STATUS_CANCELLING:
 265         info->has_status = true;
 266         info->has_total_time = true;
 267         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 268             - s->total_time;
 269         info->has_expected_downtime = true;
 270         info->expected_downtime = s->expected_downtime;
 271         info->has_setup_time = true;
 272         info->setup_time = s->setup_time;
 273
 274         info->has_ram = true;
 275         info->ram = g_malloc0(sizeof(*info->ram));
 276         info->ram->transferred = ram_bytes_transferred();
 277         info->ram->remaining = ram_bytes_remaining();
 278         info->ram->total = ram_bytes_total();
 279         info->ram->duplicate = dup_mig_pages_transferred();
 280         info->ram->skipped = skipped_mig_pages_transferred();
 281         info->ram->normal = norm_mig_pages_transferred();
 282         info->ram->normal_bytes = norm_mig_bytes_transferred();
 283         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 284         info->ram->mbps = s->mbps;
 285         info->ram->dirty_sync_count = s->dirty_sync_count;
 286
 287         if (blk_mig_active()) {
 288             info->has_disk = true;
 289             info->disk = g_malloc0(sizeof(*info->disk));
 290             info->disk->transferred = blk_mig_bytes_transferred();
 291             info->disk->remaining = blk_mig_bytes_remaining();
 292             info->disk->total = blk_mig_bytes_total();
 293         }
 294
 295         get_xbzrle_cache_stats(info);
 296         break;
 297     case MIGRATION_STATUS_COMPLETED:
 298         get_xbzrle_cache_stats(info);
 299
 300         info->has_status = true;
 301         info->has_total_time = true;
 302         info->total_time = s->total_time;
 303         info->has_downtime = true;
 304         info->downtime = s->downtime;
 305         info->has_setup_time = true;
 306         info->setup_time = s->setup_time;
 307
 308         info->has_ram = true;
 309         info->ram = g_malloc0(sizeof(*info->ram));
 310         info->ram->transferred = ram_bytes_transferred();
 311         info->ram->remaining = 0;
 312         info->ram->total = ram_bytes_total();
 313         info->ram->duplicate = dup_mig_pages_transferred();
 314         info->ram->skipped = skipped_mig_pages_transferred();
 315         info->ram->normal = norm_mig_pages_transferred();
 316         info->ram->normal_bytes = norm_mig_bytes_transferred();
 317         info->ram->mbps = s->mbps;
 318         info->ram->dirty_sync_count = s->dirty_sync_count;
 319         break;
 320     case MIGRATION_STATUS_FAILED:
 321         info->has_status = true;
 322         break;
 323     case MIGRATION_STATUS_CANCELLED:
 324         info->has_status = true;
 325         break;
 326     }
 327     info->status = s->state;
 328
 329     return info;
 330 }
 331
 332 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 333                                   Error **errp)
 334 {
 335     MigrationState *s = migrate_get_current();
 336     MigrationCapabilityStatusList *cap;
 337
 338     if (s->state == MIGRATION_STATUS_ACTIVE ||
 339         s->state == MIGRATION_STATUS_SETUP) {
 340         error_set(errp, QERR_MIGRATION_ACTIVE);
 341         return;
 342     }
 343
 344     for (cap = params; cap; cap = cap->next) {
 345         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 346     }
 347 }
 348
 349 void qmp_migrate_set_parameters(bool has_compress_level,
 350                                 int64_t compress_level,
 351                                 bool has_compress_threads,
 352                                 int64_t compress_threads,
 353                                 bool has_decompress_threads,
 354                                 int64_t decompress_threads, Error **errp)
 355 {
 356     MigrationState *s = migrate_get_current();
 357
 358     if (has_compress_level && (compress_level < 0 || compress_level > 9)) {
 359         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
 360                   "is invalid, it should be in the range of 0 to 9");
 361         return;
 362     }
 363     if (has_compress_threads &&
 364             (compress_threads < 1 || compress_threads > 255)) {
 365         error_set(errp, QERR_INVALID_PARAMETER_VALUE,
 366                   "compress_threads",
 367                   "is invalid, it should be in the range of 1 to 255");
 368         return;
 369     }
 370     if (has_decompress_threads &&
 371             (decompress_threads < 1 || decompress_threads > 255)) {
 372         error_set(errp, QERR_INVALID_PARAMETER_VALUE,
 373                   "decompress_threads",
 374                   "is invalid, it should be in the range of 1 to 255");
 375         return;
 376     }
 377
 378     if (has_compress_level) {
 379         s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 380     }
 381     if (has_compress_threads) {
 382         s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] = compress_threads;
 383     }
 384     if (has_decompress_threads) {
 385         s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 386                                                     decompress_threads;
 387     }
 388 }
 389
 390 /* shared migration helpers */
 391
 392 static void migrate_set_state(MigrationState *s, int old_state, int new_state)
 393 {
 394     if (atomic_cmpxchg(&s->state, old_state, new_state) == new_state) {
 395         trace_migrate_set_state(new_state);
 396     }
 397 }
 398
 399 static void migrate_fd_cleanup(void *opaque)
 400 {
 401     MigrationState *s = opaque;
 402
 403     qemu_bh_delete(s->cleanup_bh);
 404     s->cleanup_bh = NULL;
 405
 406     if (s->file) {
 407         trace_migrate_fd_cleanup();
 408         qemu_mutex_unlock_iothread();
 409         qemu_thread_join(&s->thread);
 410         qemu_mutex_lock_iothread();
 411
 412         migrate_compress_threads_join();
 413         qemu_fclose(s->file);
 414         s->file = NULL;
 415     }
 416
 417     assert(s->state != MIGRATION_STATUS_ACTIVE);
 418
 419     if (s->state != MIGRATION_STATUS_COMPLETED) {
 420         qemu_savevm_state_cancel();
 421         if (s->state == MIGRATION_STATUS_CANCELLING) {
 422             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
 423                               MIGRATION_STATUS_CANCELLED);
 424         }
 425     }
 426
 427     notifier_list_notify(&migration_state_notifiers, s);
 428 }
 429
 430 void migrate_fd_error(MigrationState *s)
 431 {
 432     trace_migrate_fd_error();
 433     assert(s->file == NULL);
 434     s->state = MIGRATION_STATUS_FAILED;
 435     trace_migrate_set_state(MIGRATION_STATUS_FAILED);
 436     notifier_list_notify(&migration_state_notifiers, s);
 437 }
 438
 439 static void migrate_fd_cancel(MigrationState *s)
 440 {
 441     int old_state ;
 442     QEMUFile *f = migrate_get_current()->file;
 443     trace_migrate_fd_cancel();
 444
 445     do {
 446         old_state = s->state;
 447         if (old_state != MIGRATION_STATUS_SETUP &&
 448             old_state != MIGRATION_STATUS_ACTIVE) {
 449             break;
 450         }
 451         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
 452     } while (s->state != MIGRATION_STATUS_CANCELLING);
 453
 454     /*
 455      * If we're unlucky the migration code might be stuck somewhere in a
 456      * send/write while the network has failed and is waiting to timeout;
 457      * if we've got shutdown(2) available then we can force it to quit.
 458      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
 459      * called in a bh, so there is no race against this cancel.
 460      */
 461     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
 462         qemu_file_shutdown(f);
 463     }
 464 }
 465
 466 void add_migration_state_change_notifier(Notifier *notify)
 467 {
 468     notifier_list_add(&migration_state_notifiers, notify);
 469 }
 470
 471 void remove_migration_state_change_notifier(Notifier *notify)
 472 {
 473     notifier_remove(notify);
 474 }
 475
 476 bool migration_in_setup(MigrationState *s)
 477 {
 478     return s->state == MIGRATION_STATUS_SETUP;
 479 }
 480
 481 bool migration_has_finished(MigrationState *s)
 482 {
 483     return s->state == MIGRATION_STATUS_COMPLETED;
 484 }
 485
 486 bool migration_has_failed(MigrationState *s)
 487 {
 488     return (s->state == MIGRATION_STATUS_CANCELLED ||
 489             s->state == MIGRATION_STATUS_FAILED);
 490 }
 491
 492 static MigrationState *migrate_init(const MigrationParams *params)
 493 {
 494     MigrationState *s = migrate_get_current();
 495     int64_t bandwidth_limit = s->bandwidth_limit;
 496     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 497     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 498     int compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 499     int compress_thread_count =
 500             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 501     int decompress_thread_count =
 502             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 503
 504     memcpy(enabled_capabilities, s->enabled_capabilities,
 505            sizeof(enabled_capabilities));
 506
 507     memset(s, 0, sizeof(*s));
 508     s->params = *params;
 509     memcpy(s->enabled_capabilities, enabled_capabilities,
 510            sizeof(enabled_capabilities));
 511     s->xbzrle_cache_size = xbzrle_cache_size;
 512
 513     s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 514     s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
 515                compress_thread_count;
 516     s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 517                decompress_thread_count;
 518     s->bandwidth_limit = bandwidth_limit;
 519     s->state = MIGRATION_STATUS_SETUP;
 520     trace_migrate_set_state(MIGRATION_STATUS_SETUP);
 521
 522     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 523     return s;
 524 }
 525
 526 static GSList *migration_blockers;
 527
 528 void migrate_add_blocker(Error *reason)
 529 {
 530     migration_blockers = g_slist_prepend(migration_blockers, reason);
 531 }
 532
 533 void migrate_del_blocker(Error *reason)
 534 {
 535     migration_blockers = g_slist_remove(migration_blockers, reason);
 536 }
 537
 538 void qmp_migrate_incoming(const char *uri, Error **errp)
 539 {
 540     Error *local_err = NULL;
 541     static bool once = true;
 542
 543     if (!deferred_incoming) {
 544         error_setg(errp, "For use with '-incoming defer'");
 545         return;
 546     }
 547     if (!once) {
 548         error_setg(errp, "The incoming migration has already been started");
 549     }
 550
 551     qemu_start_incoming_migration(uri, &local_err);
 552
 553     if (local_err) {
 554         error_propagate(errp, local_err);
 555         return;
 556     }
 557
 558     once = false;
 559 }
 560
 561 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 562                  bool has_inc, bool inc, bool has_detach, bool detach,
 563                  Error **errp)
 564 {
 565     Error *local_err = NULL;
 566     MigrationState *s = migrate_get_current();
 567     MigrationParams params;
 568     const char *p;
 569
 570     params.blk = has_blk && blk;
 571     params.shared = has_inc && inc;
 572
 573     if (s->state == MIGRATION_STATUS_ACTIVE ||
 574         s->state == MIGRATION_STATUS_SETUP ||
 575         s->state == MIGRATION_STATUS_CANCELLING) {
 576         error_set(errp, QERR_MIGRATION_ACTIVE);
 577         return;
 578     }
 579
 580     if (runstate_check(RUN_STATE_INMIGRATE)) {
 581         error_setg(errp, "Guest is waiting for an incoming migration");
 582         return;
 583     }
 584
 585     if (qemu_savevm_state_blocked(errp)) {
 586         return;
 587     }
 588
 589     if (migration_blockers) {
 590         *errp = error_copy(migration_blockers->data);
 591         return;
 592     }
 593
 594     s = migrate_init(&params);
 595
 596     if (strstart(uri, "tcp:", &p)) {
 597         tcp_start_outgoing_migration(s, p, &local_err);
 598 #ifdef CONFIG_RDMA
 599     } else if (strstart(uri, "rdma:", &p)) {
 600         rdma_start_outgoing_migration(s, p, &local_err);
 601 #endif
 602 #if !defined(WIN32)
 603     } else if (strstart(uri, "exec:", &p)) {
 604         exec_start_outgoing_migration(s, p, &local_err);
 605     } else if (strstart(uri, "unix:", &p)) {
 606         unix_start_outgoing_migration(s, p, &local_err);
 607     } else if (strstart(uri, "fd:", &p)) {
 608         fd_start_outgoing_migration(s, p, &local_err);
 609 #endif
 610     } else {
 611         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 612         s->state = MIGRATION_STATUS_FAILED;
 613         return;
 614     }
 615
 616     if (local_err) {
 617         migrate_fd_error(s);
 618         error_propagate(errp, local_err);
 619         return;
 620     }
 621 }
 622
 623 void qmp_migrate_cancel(Error **errp)
 624 {
 625     migrate_fd_cancel(migrate_get_current());
 626 }
 627
 628 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 629 {
 630     MigrationState *s = migrate_get_current();
 631     int64_t new_size;
 632
 633     /* Check for truncation */
 634     if (value != (size_t)value) {
 635         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 636                   "exceeding address space");
 637         return;
 638     }
 639
 640     /* Cache should not be larger than guest ram size */
 641     if (value > ram_bytes_total()) {
 642         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 643                   "exceeds guest ram size ");
 644         return;
 645     }
 646
 647     new_size = xbzrle_cache_resize(value);
 648     if (new_size < 0) {
 649         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 650                   "is smaller than page size");
 651         return;
 652     }
 653
 654     s->xbzrle_cache_size = new_size;
 655 }
 656
 657 int64_t qmp_query_migrate_cache_size(Error **errp)
 658 {
 659     return migrate_xbzrle_cache_size();
 660 }
 661
 662 void qmp_migrate_set_speed(int64_t value, Error **errp)
 663 {
 664     MigrationState *s;
 665
 666     if (value < 0) {
 667         value = 0;
 668     }
 669     if (value > SIZE_MAX) {
 670         value = SIZE_MAX;
 671     }
 672
 673     s = migrate_get_current();
 674     s->bandwidth_limit = value;
 675     if (s->file) {
 676         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
 677     }
 678 }
 679
 680 void qmp_migrate_set_downtime(double value, Error **errp)
 681 {
 682     value *= 1e9;
 683     value = MAX(0, MIN(UINT64_MAX, value));
 684     max_downtime = (uint64_t)value;
 685 }
 686
 687 bool migrate_auto_converge(void)
 688 {
 689     MigrationState *s;
 690
 691     s = migrate_get_current();
 692
 693     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
 694 }
 695
 696 bool migrate_zero_blocks(void)
 697 {
 698     MigrationState *s;
 699
 700     s = migrate_get_current();
 701
 702     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
 703 }
 704
 705 bool migrate_use_compression(void)
 706 {
 707     MigrationState *s;
 708
 709     s = migrate_get_current();
 710
 711     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
 712 }
 713
 714 int migrate_compress_level(void)
 715 {
 716     MigrationState *s;
 717
 718     s = migrate_get_current();
 719
 720     return s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 721 }
 722
 723 int migrate_compress_threads(void)
 724 {
 725     MigrationState *s;
 726
 727     s = migrate_get_current();
 728
 729     return s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 730 }
 731
 732 int migrate_decompress_threads(void)
 733 {
 734     MigrationState *s;
 735
 736     s = migrate_get_current();
 737
 738     return s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 739 }
 740
 741 int migrate_use_xbzrle(void)
 742 {
 743     MigrationState *s;
 744
 745     s = migrate_get_current();
 746
 747     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 748 }
 749
 750 int64_t migrate_xbzrle_cache_size(void)
 751 {
 752     MigrationState *s;
 753
 754     s = migrate_get_current();
 755
 756     return s->xbzrle_cache_size;
 757 }
 758
 759 /* migration thread support */
 760
 761 static void *migration_thread(void *opaque)
 762 {
 763     MigrationState *s = opaque;
 764     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 765     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 766     int64_t initial_bytes = 0;
 767     int64_t max_size = 0;
 768     int64_t start_time = initial_time;
 769     bool old_vm_running = false;
 770
 771     qemu_savevm_state_header(s->file);
 772     qemu_savevm_state_begin(s->file, &s->params);
 773
 774     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
 775     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
 776
 777     while (s->state == MIGRATION_STATUS_ACTIVE) {
 778         int64_t current_time;
 779         uint64_t pending_size;
 780
 781         if (!qemu_file_rate_limit(s->file)) {
 782             pending_size = qemu_savevm_state_pending(s->file, max_size);
 783             trace_migrate_pending(pending_size, max_size);
 784             if (pending_size && pending_size >= max_size) {
 785                 qemu_savevm_state_iterate(s->file);
 786             } else {
 787                 int ret;
 788
 789                 qemu_mutex_lock_iothread();
 790                 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 791                 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 792                 old_vm_running = runstate_is_running();
 793
 794                 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 795                 if (ret >= 0) {
 796                     qemu_file_set_rate_limit(s->file, INT64_MAX);
 797                     qemu_savevm_state_complete(s->file);
 798                 }
 799                 qemu_mutex_unlock_iothread();
 800
 801                 if (ret < 0) {
 802                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 803                                       MIGRATION_STATUS_FAILED);
 804                     break;
 805                 }
 806
 807                 if (!qemu_file_get_error(s->file)) {
 808                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 809                                       MIGRATION_STATUS_COMPLETED);
 810                     break;
 811                 }
 812             }
 813         }
 814
 815         if (qemu_file_get_error(s->file)) {
 816             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 817                               MIGRATION_STATUS_FAILED);
 818             break;
 819         }
 820         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 821         if (current_time >= initial_time + BUFFER_DELAY) {
 822             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
 823             uint64_t time_spent = current_time - initial_time;
 824             double bandwidth = transferred_bytes / time_spent;
 825             max_size = bandwidth * migrate_max_downtime() / 1000000;
 826
 827             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
 828                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
 829
 830             trace_migrate_transferred(transferred_bytes, time_spent,
 831                                       bandwidth, max_size);
 832             /* if we haven't sent anything, we don't want to recalculate
 833                10000 is a small enough number for our purposes */
 834             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
 835                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
 836             }
 837
 838             qemu_file_reset_rate_limit(s->file);
 839             initial_time = current_time;
 840             initial_bytes = qemu_ftell(s->file);
 841         }
 842         if (qemu_file_rate_limit(s->file)) {
 843             /* usleep expects microseconds */
 844             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
 845         }
 846     }
 847
 848     qemu_mutex_lock_iothread();
 849     if (s->state == MIGRATION_STATUS_COMPLETED) {
 850         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 851         uint64_t transferred_bytes = qemu_ftell(s->file);
 852         s->total_time = end_time - s->total_time;
 853         s->downtime = end_time - start_time;
 854         if (s->total_time) {
 855             s->mbps = (((double) transferred_bytes * 8.0) /
 856                        ((double) s->total_time)) / 1000;
 857         }
 858         runstate_set(RUN_STATE_POSTMIGRATE);
 859     } else {
 860         if (old_vm_running) {
 861             vm_start();
 862         }
 863     }
 864     qemu_bh_schedule(s->cleanup_bh);
 865     qemu_mutex_unlock_iothread();
 866
 867     return NULL;
 868 }
 869
 870 void migrate_fd_connect(MigrationState *s)
 871 {
 872     /* This is a best 1st approximation. ns to ms */
 873     s->expected_downtime = max_downtime/1000000;
 874     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
 875
 876     qemu_file_set_rate_limit(s->file,
 877                              s->bandwidth_limit / XFER_LIMIT_RATIO);
 878
 879     /* Notify before starting migration thread */
 880     notifier_list_notify(&migration_state_notifiers, s);
 881
 882     migrate_compress_threads_create();
 883     qemu_thread_create(&s->thread, "migration", migration_thread, s,
 884                        QEMU_THREAD_JOINABLE);
 885 }