migration/migration.c

   1 /*
   2  * QEMU live migration
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <[email protected]>
   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/error-report.h"
  18 #include "qemu/main-loop.h"
  19 #include "migration/migration.h"
  20 #include "migration/qemu-file.h"
  21 #include "sysemu/sysemu.h"
  22 #include "block/block.h"
  23 #include "qapi/qmp/qerror.h"
  24 #include "qemu/sockets.h"
  25 #include "migration/block.h"
  26 #include "qemu/thread.h"
  27 #include "qmp-commands.h"
  28 #include "trace.h"
  29 #include "qapi/util.h"
  30 #include "qapi-event.h"
  31
  32 #define MAX_THROTTLE  (32 << 20)      /* Migration speed throttling */
  33
  34 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
  35  * data. */
  36 #define BUFFER_DELAY     100
  37 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
  38
  39 /* Default compression thread count */
  40 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
  41 /* Default decompression thread count, usually decompression is at
  42  * least 4 times as fast as compression.*/
  43 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
  44 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
  45 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
  46
  47 /* Migration XBZRLE default cache size */
  48 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
  49
  50 static NotifierList migration_state_notifiers =
  51     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  52
  53 static bool deferred_incoming;
  54
  55 /* When we add fault tolerance, we could have several
  56    migrations at once.  For now we don't need to add
  57    dynamic creation of migration */
  58
  59 /* For outgoing */
  60 MigrationState *migrate_get_current(void)
  61 {
  62     static MigrationState current_migration = {
  63         .state = MIGRATION_STATUS_NONE,
  64         .bandwidth_limit = MAX_THROTTLE,
  65         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
  66         .mbps = -1,
  67         .parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] =
  68                 DEFAULT_MIGRATE_COMPRESS_LEVEL,
  69         .parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
  70                 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
  71         .parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
  72                 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
  73     };
  74
  75     return &current_migration;
  76 }
  77
  78 /* For incoming */
  79 static MigrationIncomingState *mis_current;
  80
  81 MigrationIncomingState *migration_incoming_get_current(void)
  82 {
  83     return mis_current;
  84 }
  85
  86 MigrationIncomingState *migration_incoming_state_new(QEMUFile* f)
  87 {
  88     mis_current = g_malloc0(sizeof(MigrationIncomingState));
  89     mis_current->file = f;
  90     QLIST_INIT(&mis_current->loadvm_handlers);
  91
  92     return mis_current;
  93 }
  94
  95 void migration_incoming_state_destroy(void)
  96 {
  97     loadvm_free_handlers(mis_current);
  98     g_free(mis_current);
  99     mis_current = NULL;
 100 }
 101
 102
 103 typedef struct {
 104     bool optional;
 105     uint32_t size;
 106     uint8_t runstate[100];
 107 } GlobalState;
 108
 109 static GlobalState global_state;
 110
 111 static int global_state_store(void)
 112 {
 113     if (!runstate_store((char *)global_state.runstate,
 114                         sizeof(global_state.runstate))) {
 115         error_report("runstate name too big: %s", global_state.runstate);
 116         trace_migrate_state_too_big();
 117         return -EINVAL;
 118     }
 119     return 0;
 120 }
 121
 122 static char *global_state_get_runstate(void)
 123 {
 124     return (char *)global_state.runstate;
 125 }
 126
 127 void global_state_set_optional(void)
 128 {
 129     global_state.optional = true;
 130 }
 131
 132 static bool global_state_needed(void *opaque)
 133 {
 134     GlobalState *s = opaque;
 135     char *runstate = (char *)s->runstate;
 136
 137     /* If it is not optional, it is mandatory */
 138
 139     if (s->optional == false) {
 140         return true;
 141     }
 142
 143     /* If state is running or paused, it is not needed */
 144
 145     if (strcmp(runstate, "running") == 0 ||
 146         strcmp(runstate, "paused") == 0) {
 147         return false;
 148     }
 149
 150     /* for any other state it is needed */
 151     return true;
 152 }
 153
 154 static int global_state_post_load(void *opaque, int version_id)
 155 {
 156     GlobalState *s = opaque;
 157     int ret = 0;
 158     char *runstate = (char *)s->runstate;
 159
 160     trace_migrate_global_state_post_load(runstate);
 161
 162     if (strcmp(runstate, "running") != 0) {
 163         Error *local_err = NULL;
 164         int r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE_MAX,
 165                                 -1, &local_err);
 166
 167         if (r == -1) {
 168             if (local_err) {
 169                 error_report_err(local_err);
 170             }
 171             return -EINVAL;
 172         }
 173         ret = vm_stop_force_state(r);
 174     }
 175
 176    return ret;
 177 }
 178
 179 static void global_state_pre_save(void *opaque)
 180 {
 181     GlobalState *s = opaque;
 182
 183     trace_migrate_global_state_pre_save((char *)s->runstate);
 184     s->size = strlen((char *)s->runstate) + 1;
 185 }
 186
 187 static const VMStateDescription vmstate_globalstate = {
 188     .name = "globalstate",
 189     .version_id = 1,
 190     .minimum_version_id = 1,
 191     .post_load = global_state_post_load,
 192     .pre_save = global_state_pre_save,
 193     .needed = global_state_needed,
 194     .fields = (VMStateField[]) {
 195         VMSTATE_UINT32(size, GlobalState),
 196         VMSTATE_BUFFER(runstate, GlobalState),
 197         VMSTATE_END_OF_LIST()
 198     },
 199 };
 200
 201 void register_global_state(void)
 202 {
 203     /* We would use it independently that we receive it */
 204     strcpy((char *)&global_state.runstate, "");
 205     vmstate_register(NULL, 0, &vmstate_globalstate, &global_state);
 206 }
 207
 208 /*
 209  * Called on -incoming with a defer: uri.
 210  * The migration can be started later after any parameters have been
 211  * changed.
 212  */
 213 static void deferred_incoming_migration(Error **errp)
 214 {
 215     if (deferred_incoming) {
 216         error_setg(errp, "Incoming migration already deferred");
 217     }
 218     deferred_incoming = true;
 219 }
 220
 221 void qemu_start_incoming_migration(const char *uri, Error **errp)
 222 {
 223     const char *p;
 224
 225     if (!strcmp(uri, "defer")) {
 226         deferred_incoming_migration(errp);
 227     } else if (strstart(uri, "tcp:", &p)) {
 228         tcp_start_incoming_migration(p, errp);
 229 #ifdef CONFIG_RDMA
 230     } else if (strstart(uri, "rdma:", &p)) {
 231         rdma_start_incoming_migration(p, errp);
 232 #endif
 233 #if !defined(WIN32)
 234     } else if (strstart(uri, "exec:", &p)) {
 235         exec_start_incoming_migration(p, errp);
 236     } else if (strstart(uri, "unix:", &p)) {
 237         unix_start_incoming_migration(p, errp);
 238     } else if (strstart(uri, "fd:", &p)) {
 239         fd_start_incoming_migration(p, errp);
 240 #endif
 241     } else {
 242         error_setg(errp, "unknown migration protocol: %s", uri);
 243     }
 244 }
 245
 246 static void process_incoming_migration_co(void *opaque)
 247 {
 248     QEMUFile *f = opaque;
 249     Error *local_err = NULL;
 250     int ret;
 251
 252     migration_incoming_state_new(f);
 253
 254     ret = qemu_loadvm_state(f);
 255
 256     qemu_fclose(f);
 257     free_xbzrle_decoded_buf();
 258     migration_incoming_state_destroy();
 259
 260     if (ret < 0) {
 261         error_report("load of migration failed: %s", strerror(-ret));
 262         migrate_decompress_threads_join();
 263         exit(EXIT_FAILURE);
 264     }
 265     qemu_announce_self();
 266
 267     /* Make sure all file formats flush their mutable metadata */
 268     bdrv_invalidate_cache_all(&local_err);
 269     if (local_err) {
 270         error_report_err(local_err);
 271         migrate_decompress_threads_join();
 272         exit(EXIT_FAILURE);
 273     }
 274
 275     /* runstate == "" means that we haven't received it through the
 276      * wire, so we obey autostart.  runstate == runing means that we
 277      * need to run it, we need to make sure that we do it after
 278      * everything else has finished.  Every other state change is done
 279      * at the post_load function */
 280
 281     if (strcmp(global_state_get_runstate(), "running") == 0) {
 282         vm_start();
 283     } else if (strcmp(global_state_get_runstate(), "") == 0) {
 284         if (autostart) {
 285             vm_start();
 286         } else {
 287             runstate_set(RUN_STATE_PAUSED);
 288         }
 289     }
 290     migrate_decompress_threads_join();
 291 }
 292
 293 void process_incoming_migration(QEMUFile *f)
 294 {
 295     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
 296     int fd = qemu_get_fd(f);
 297
 298     assert(fd != -1);
 299     migrate_decompress_threads_create();
 300     qemu_set_nonblock(fd);
 301     qemu_coroutine_enter(co, f);
 302 }
 303
 304 /* amount of nanoseconds we are willing to wait for migration to be down.
 305  * the choice of nanoseconds is because it is the maximum resolution that
 306  * get_clock() can achieve. It is an internal measure. All user-visible
 307  * units must be in seconds */
 308 static uint64_t max_downtime = 300000000;
 309
 310 uint64_t migrate_max_downtime(void)
 311 {
 312     return max_downtime;
 313 }
 314
 315 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 316 {
 317     MigrationCapabilityStatusList *head = NULL;
 318     MigrationCapabilityStatusList *caps;
 319     MigrationState *s = migrate_get_current();
 320     int i;
 321
 322     caps = NULL; /* silence compiler warning */
 323     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 324         if (head == NULL) {
 325             head = g_malloc0(sizeof(*caps));
 326             caps = head;
 327         } else {
 328             caps->next = g_malloc0(sizeof(*caps));
 329             caps = caps->next;
 330         }
 331         caps->value =
 332             g_malloc(sizeof(*caps->value));
 333         caps->value->capability = i;
 334         caps->value->state = s->enabled_capabilities[i];
 335     }
 336
 337     return head;
 338 }
 339
 340 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 341 {
 342     MigrationParameters *params;
 343     MigrationState *s = migrate_get_current();
 344
 345     params = g_malloc0(sizeof(*params));
 346     params->compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 347     params->compress_threads =
 348             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 349     params->decompress_threads =
 350             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 351
 352     return params;
 353 }
 354
 355 static void get_xbzrle_cache_stats(MigrationInfo *info)
 356 {
 357     if (migrate_use_xbzrle()) {
 358         info->has_xbzrle_cache = true;
 359         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 360         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 361         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 362         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 363         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 364         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
 365         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 366     }
 367 }
 368
 369 MigrationInfo *qmp_query_migrate(Error **errp)
 370 {
 371     MigrationInfo *info = g_malloc0(sizeof(*info));
 372     MigrationState *s = migrate_get_current();
 373
 374     switch (s->state) {
 375     case MIGRATION_STATUS_NONE:
 376         /* no migration has happened ever */
 377         break;
 378     case MIGRATION_STATUS_SETUP:
 379         info->has_status = true;
 380         info->has_total_time = false;
 381         break;
 382     case MIGRATION_STATUS_ACTIVE:
 383     case MIGRATION_STATUS_CANCELLING:
 384         info->has_status = true;
 385         info->has_total_time = true;
 386         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 387             - s->total_time;
 388         info->has_expected_downtime = true;
 389         info->expected_downtime = s->expected_downtime;
 390         info->has_setup_time = true;
 391         info->setup_time = s->setup_time;
 392
 393         info->has_ram = true;
 394         info->ram = g_malloc0(sizeof(*info->ram));
 395         info->ram->transferred = ram_bytes_transferred();
 396         info->ram->remaining = ram_bytes_remaining();
 397         info->ram->total = ram_bytes_total();
 398         info->ram->duplicate = dup_mig_pages_transferred();
 399         info->ram->skipped = skipped_mig_pages_transferred();
 400         info->ram->normal = norm_mig_pages_transferred();
 401         info->ram->normal_bytes = norm_mig_bytes_transferred();
 402         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 403         info->ram->mbps = s->mbps;
 404         info->ram->dirty_sync_count = s->dirty_sync_count;
 405
 406         if (blk_mig_active()) {
 407             info->has_disk = true;
 408             info->disk = g_malloc0(sizeof(*info->disk));
 409             info->disk->transferred = blk_mig_bytes_transferred();
 410             info->disk->remaining = blk_mig_bytes_remaining();
 411             info->disk->total = blk_mig_bytes_total();
 412         }
 413
 414         get_xbzrle_cache_stats(info);
 415         break;
 416     case MIGRATION_STATUS_COMPLETED:
 417         get_xbzrle_cache_stats(info);
 418
 419         info->has_status = true;
 420         info->has_total_time = true;
 421         info->total_time = s->total_time;
 422         info->has_downtime = true;
 423         info->downtime = s->downtime;
 424         info->has_setup_time = true;
 425         info->setup_time = s->setup_time;
 426
 427         info->has_ram = true;
 428         info->ram = g_malloc0(sizeof(*info->ram));
 429         info->ram->transferred = ram_bytes_transferred();
 430         info->ram->remaining = 0;
 431         info->ram->total = ram_bytes_total();
 432         info->ram->duplicate = dup_mig_pages_transferred();
 433         info->ram->skipped = skipped_mig_pages_transferred();
 434         info->ram->normal = norm_mig_pages_transferred();
 435         info->ram->normal_bytes = norm_mig_bytes_transferred();
 436         info->ram->mbps = s->mbps;
 437         info->ram->dirty_sync_count = s->dirty_sync_count;
 438         break;
 439     case MIGRATION_STATUS_FAILED:
 440         info->has_status = true;
 441         break;
 442     case MIGRATION_STATUS_CANCELLED:
 443         info->has_status = true;
 444         break;
 445     }
 446     info->status = s->state;
 447
 448     return info;
 449 }
 450
 451 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 452                                   Error **errp)
 453 {
 454     MigrationState *s = migrate_get_current();
 455     MigrationCapabilityStatusList *cap;
 456
 457     if (s->state == MIGRATION_STATUS_ACTIVE ||
 458         s->state == MIGRATION_STATUS_SETUP) {
 459         error_setg(errp, QERR_MIGRATION_ACTIVE);
 460         return;
 461     }
 462
 463     for (cap = params; cap; cap = cap->next) {
 464         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 465     }
 466 }
 467
 468 void qmp_migrate_set_parameters(bool has_compress_level,
 469                                 int64_t compress_level,
 470                                 bool has_compress_threads,
 471                                 int64_t compress_threads,
 472                                 bool has_decompress_threads,
 473                                 int64_t decompress_threads, Error **errp)
 474 {
 475     MigrationState *s = migrate_get_current();
 476
 477     if (has_compress_level && (compress_level < 0 || compress_level > 9)) {
 478         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
 479                    "is invalid, it should be in the range of 0 to 9");
 480         return;
 481     }
 482     if (has_compress_threads &&
 483             (compress_threads < 1 || compress_threads > 255)) {
 484         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 485                    "compress_threads",
 486                    "is invalid, it should be in the range of 1 to 255");
 487         return;
 488     }
 489     if (has_decompress_threads &&
 490             (decompress_threads < 1 || decompress_threads > 255)) {
 491         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 492                    "decompress_threads",
 493                    "is invalid, it should be in the range of 1 to 255");
 494         return;
 495     }
 496
 497     if (has_compress_level) {
 498         s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 499     }
 500     if (has_compress_threads) {
 501         s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] = compress_threads;
 502     }
 503     if (has_decompress_threads) {
 504         s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 505                                                     decompress_threads;
 506     }
 507 }
 508
 509 /* shared migration helpers */
 510
 511 static void migrate_set_state(MigrationState *s, int old_state, int new_state)
 512 {
 513     if (atomic_cmpxchg(&s->state, old_state, new_state) == old_state) {
 514         qapi_event_send_migration(new_state, &error_abort);
 515         trace_migrate_set_state(new_state);
 516     }
 517 }
 518
 519 static void migrate_fd_cleanup(void *opaque)
 520 {
 521     MigrationState *s = opaque;
 522
 523     qemu_bh_delete(s->cleanup_bh);
 524     s->cleanup_bh = NULL;
 525
 526     if (s->file) {
 527         trace_migrate_fd_cleanup();
 528         qemu_mutex_unlock_iothread();
 529         qemu_thread_join(&s->thread);
 530         qemu_mutex_lock_iothread();
 531
 532         migrate_compress_threads_join();
 533         qemu_fclose(s->file);
 534         s->file = NULL;
 535     }
 536
 537     assert(s->state != MIGRATION_STATUS_ACTIVE);
 538
 539     if (s->state != MIGRATION_STATUS_COMPLETED) {
 540         qemu_savevm_state_cancel();
 541         if (s->state == MIGRATION_STATUS_CANCELLING) {
 542             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
 543                               MIGRATION_STATUS_CANCELLED);
 544         }
 545     }
 546
 547     notifier_list_notify(&migration_state_notifiers, s);
 548 }
 549
 550 void migrate_fd_error(MigrationState *s)
 551 {
 552     trace_migrate_fd_error();
 553     assert(s->file == NULL);
 554     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 555     notifier_list_notify(&migration_state_notifiers, s);
 556 }
 557
 558 static void migrate_fd_cancel(MigrationState *s)
 559 {
 560     int old_state ;
 561     QEMUFile *f = migrate_get_current()->file;
 562     trace_migrate_fd_cancel();
 563
 564     do {
 565         old_state = s->state;
 566         if (old_state != MIGRATION_STATUS_SETUP &&
 567             old_state != MIGRATION_STATUS_ACTIVE) {
 568             break;
 569         }
 570         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
 571     } while (s->state != MIGRATION_STATUS_CANCELLING);
 572
 573     /*
 574      * If we're unlucky the migration code might be stuck somewhere in a
 575      * send/write while the network has failed and is waiting to timeout;
 576      * if we've got shutdown(2) available then we can force it to quit.
 577      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
 578      * called in a bh, so there is no race against this cancel.
 579      */
 580     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
 581         qemu_file_shutdown(f);
 582     }
 583 }
 584
 585 void add_migration_state_change_notifier(Notifier *notify)
 586 {
 587     notifier_list_add(&migration_state_notifiers, notify);
 588 }
 589
 590 void remove_migration_state_change_notifier(Notifier *notify)
 591 {
 592     notifier_remove(notify);
 593 }
 594
 595 bool migration_in_setup(MigrationState *s)
 596 {
 597     return s->state == MIGRATION_STATUS_SETUP;
 598 }
 599
 600 bool migration_has_finished(MigrationState *s)
 601 {
 602     return s->state == MIGRATION_STATUS_COMPLETED;
 603 }
 604
 605 bool migration_has_failed(MigrationState *s)
 606 {
 607     return (s->state == MIGRATION_STATUS_CANCELLED ||
 608             s->state == MIGRATION_STATUS_FAILED);
 609 }
 610
 611 static MigrationState *migrate_init(const MigrationParams *params)
 612 {
 613     MigrationState *s = migrate_get_current();
 614     int64_t bandwidth_limit = s->bandwidth_limit;
 615     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 616     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 617     int compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 618     int compress_thread_count =
 619             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 620     int decompress_thread_count =
 621             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 622
 623     memcpy(enabled_capabilities, s->enabled_capabilities,
 624            sizeof(enabled_capabilities));
 625
 626     memset(s, 0, sizeof(*s));
 627     s->params = *params;
 628     memcpy(s->enabled_capabilities, enabled_capabilities,
 629            sizeof(enabled_capabilities));
 630     s->xbzrle_cache_size = xbzrle_cache_size;
 631
 632     s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 633     s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
 634                compress_thread_count;
 635     s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 636                decompress_thread_count;
 637     s->bandwidth_limit = bandwidth_limit;
 638     migrate_set_state(s, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
 639
 640     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 641     return s;
 642 }
 643
 644 static GSList *migration_blockers;
 645
 646 void migrate_add_blocker(Error *reason)
 647 {
 648     migration_blockers = g_slist_prepend(migration_blockers, reason);
 649 }
 650
 651 void migrate_del_blocker(Error *reason)
 652 {
 653     migration_blockers = g_slist_remove(migration_blockers, reason);
 654 }
 655
 656 void qmp_migrate_incoming(const char *uri, Error **errp)
 657 {
 658     Error *local_err = NULL;
 659     static bool once = true;
 660
 661     if (!deferred_incoming) {
 662         error_setg(errp, "For use with '-incoming defer'");
 663         return;
 664     }
 665     if (!once) {
 666         error_setg(errp, "The incoming migration has already been started");
 667     }
 668
 669     qemu_start_incoming_migration(uri, &local_err);
 670
 671     if (local_err) {
 672         error_propagate(errp, local_err);
 673         return;
 674     }
 675
 676     once = false;
 677 }
 678
 679 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 680                  bool has_inc, bool inc, bool has_detach, bool detach,
 681                  Error **errp)
 682 {
 683     Error *local_err = NULL;
 684     MigrationState *s = migrate_get_current();
 685     MigrationParams params;
 686     const char *p;
 687
 688     params.blk = has_blk && blk;
 689     params.shared = has_inc && inc;
 690
 691     if (s->state == MIGRATION_STATUS_ACTIVE ||
 692         s->state == MIGRATION_STATUS_SETUP ||
 693         s->state == MIGRATION_STATUS_CANCELLING) {
 694         error_setg(errp, QERR_MIGRATION_ACTIVE);
 695         return;
 696     }
 697     if (runstate_check(RUN_STATE_INMIGRATE)) {
 698         error_setg(errp, "Guest is waiting for an incoming migration");
 699         return;
 700     }
 701
 702     if (qemu_savevm_state_blocked(errp)) {
 703         return;
 704     }
 705
 706     if (migration_blockers) {
 707         *errp = error_copy(migration_blockers->data);
 708         return;
 709     }
 710
 711     /* We are starting a new migration, so we want to start in a clean
 712        state.  This change is only needed if previous migration
 713        failed/was cancelled.  We don't use migrate_set_state() because
 714        we are setting the initial state, not changing it. */
 715     s->state = MIGRATION_STATUS_NONE;
 716
 717     s = migrate_init(&params);
 718
 719     if (strstart(uri, "tcp:", &p)) {
 720         tcp_start_outgoing_migration(s, p, &local_err);
 721 #ifdef CONFIG_RDMA
 722     } else if (strstart(uri, "rdma:", &p)) {
 723         rdma_start_outgoing_migration(s, p, &local_err);
 724 #endif
 725 #if !defined(WIN32)
 726     } else if (strstart(uri, "exec:", &p)) {
 727         exec_start_outgoing_migration(s, p, &local_err);
 728     } else if (strstart(uri, "unix:", &p)) {
 729         unix_start_outgoing_migration(s, p, &local_err);
 730     } else if (strstart(uri, "fd:", &p)) {
 731         fd_start_outgoing_migration(s, p, &local_err);
 732 #endif
 733     } else {
 734         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
 735                    "a valid migration protocol");
 736         migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 737         return;
 738     }
 739
 740     if (local_err) {
 741         migrate_fd_error(s);
 742         error_propagate(errp, local_err);
 743         return;
 744     }
 745 }
 746
 747 void qmp_migrate_cancel(Error **errp)
 748 {
 749     migrate_fd_cancel(migrate_get_current());
 750 }
 751
 752 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 753 {
 754     MigrationState *s = migrate_get_current();
 755     int64_t new_size;
 756
 757     /* Check for truncation */
 758     if (value != (size_t)value) {
 759         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 760                    "exceeding address space");
 761         return;
 762     }
 763
 764     /* Cache should not be larger than guest ram size */
 765     if (value > ram_bytes_total()) {
 766         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 767                    "exceeds guest ram size ");
 768         return;
 769     }
 770
 771     new_size = xbzrle_cache_resize(value);
 772     if (new_size < 0) {
 773         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 774                    "is smaller than page size");
 775         return;
 776     }
 777
 778     s->xbzrle_cache_size = new_size;
 779 }
 780
 781 int64_t qmp_query_migrate_cache_size(Error **errp)
 782 {
 783     return migrate_xbzrle_cache_size();
 784 }
 785
 786 void qmp_migrate_set_speed(int64_t value, Error **errp)
 787 {
 788     MigrationState *s;
 789
 790     if (value < 0) {
 791         value = 0;
 792     }
 793     if (value > SIZE_MAX) {
 794         value = SIZE_MAX;
 795     }
 796
 797     s = migrate_get_current();
 798     s->bandwidth_limit = value;
 799     if (s->file) {
 800         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
 801     }
 802 }
 803
 804 void qmp_migrate_set_downtime(double value, Error **errp)
 805 {
 806     value *= 1e9;
 807     value = MAX(0, MIN(UINT64_MAX, value));
 808     max_downtime = (uint64_t)value;
 809 }
 810
 811 bool migrate_auto_converge(void)
 812 {
 813     MigrationState *s;
 814
 815     s = migrate_get_current();
 816
 817     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
 818 }
 819
 820 bool migrate_zero_blocks(void)
 821 {
 822     MigrationState *s;
 823
 824     s = migrate_get_current();
 825
 826     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
 827 }
 828
 829 bool migrate_use_compression(void)
 830 {
 831     MigrationState *s;
 832
 833     s = migrate_get_current();
 834
 835     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
 836 }
 837
 838 int migrate_compress_level(void)
 839 {
 840     MigrationState *s;
 841
 842     s = migrate_get_current();
 843
 844     return s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 845 }
 846
 847 int migrate_compress_threads(void)
 848 {
 849     MigrationState *s;
 850
 851     s = migrate_get_current();
 852
 853     return s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 854 }
 855
 856 int migrate_decompress_threads(void)
 857 {
 858     MigrationState *s;
 859
 860     s = migrate_get_current();
 861
 862     return s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 863 }
 864
 865 int migrate_use_xbzrle(void)
 866 {
 867     MigrationState *s;
 868
 869     s = migrate_get_current();
 870
 871     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 872 }
 873
 874 int64_t migrate_xbzrle_cache_size(void)
 875 {
 876     MigrationState *s;
 877
 878     s = migrate_get_current();
 879
 880     return s->xbzrle_cache_size;
 881 }
 882
 883 /* migration thread support */
 884
 885 static void *migration_thread(void *opaque)
 886 {
 887     MigrationState *s = opaque;
 888     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 889     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 890     int64_t initial_bytes = 0;
 891     int64_t max_size = 0;
 892     int64_t start_time = initial_time;
 893     bool old_vm_running = false;
 894
 895     qemu_savevm_state_header(s->file);
 896     qemu_savevm_state_begin(s->file, &s->params);
 897
 898     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
 899     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
 900
 901     while (s->state == MIGRATION_STATUS_ACTIVE) {
 902         int64_t current_time;
 903         uint64_t pending_size;
 904
 905         if (!qemu_file_rate_limit(s->file)) {
 906             pending_size = qemu_savevm_state_pending(s->file, max_size);
 907             trace_migrate_pending(pending_size, max_size);
 908             if (pending_size && pending_size >= max_size) {
 909                 qemu_savevm_state_iterate(s->file);
 910             } else {
 911                 int ret;
 912
 913                 qemu_mutex_lock_iothread();
 914                 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 915                 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 916                 old_vm_running = runstate_is_running();
 917
 918                 ret = global_state_store();
 919                 if (!ret) {
 920                     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 921                     if (ret >= 0) {
 922                         qemu_file_set_rate_limit(s->file, INT64_MAX);
 923                         qemu_savevm_state_complete(s->file);
 924                     }
 925                 }
 926                 qemu_mutex_unlock_iothread();
 927
 928                 if (ret < 0) {
 929                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 930                                       MIGRATION_STATUS_FAILED);
 931                     break;
 932                 }
 933
 934                 if (!qemu_file_get_error(s->file)) {
 935                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 936                                       MIGRATION_STATUS_COMPLETED);
 937                     break;
 938                 }
 939             }
 940         }
 941
 942         if (qemu_file_get_error(s->file)) {
 943             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 944                               MIGRATION_STATUS_FAILED);
 945             break;
 946         }
 947         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 948         if (current_time >= initial_time + BUFFER_DELAY) {
 949             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
 950             uint64_t time_spent = current_time - initial_time;
 951             double bandwidth = transferred_bytes / time_spent;
 952             max_size = bandwidth * migrate_max_downtime() / 1000000;
 953
 954             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
 955                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
 956
 957             trace_migrate_transferred(transferred_bytes, time_spent,
 958                                       bandwidth, max_size);
 959             /* if we haven't sent anything, we don't want to recalculate
 960                10000 is a small enough number for our purposes */
 961             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
 962                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
 963             }
 964
 965             qemu_file_reset_rate_limit(s->file);
 966             initial_time = current_time;
 967             initial_bytes = qemu_ftell(s->file);
 968         }
 969         if (qemu_file_rate_limit(s->file)) {
 970             /* usleep expects microseconds */
 971             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
 972         }
 973     }
 974
 975     qemu_mutex_lock_iothread();
 976     if (s->state == MIGRATION_STATUS_COMPLETED) {
 977         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 978         uint64_t transferred_bytes = qemu_ftell(s->file);
 979         s->total_time = end_time - s->total_time;
 980         s->downtime = end_time - start_time;
 981         if (s->total_time) {
 982             s->mbps = (((double) transferred_bytes * 8.0) /
 983                        ((double) s->total_time)) / 1000;
 984         }
 985         runstate_set(RUN_STATE_POSTMIGRATE);
 986     } else {
 987         if (old_vm_running) {
 988             vm_start();
 989         }
 990     }
 991     qemu_bh_schedule(s->cleanup_bh);
 992     qemu_mutex_unlock_iothread();
 993
 994     return NULL;
 995 }
 996
 997 void migrate_fd_connect(MigrationState *s)
 998 {
 999     /* This is a best 1st approximation. ns to ms */
1000     s->expected_downtime = max_downtime/1000000;
1001     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
1002
1003     qemu_file_set_rate_limit(s->file,
1004                              s->bandwidth_limit / XFER_LIMIT_RATIO);
1005
1006     /* Notify before starting migration thread */
1007     notifier_list_notify(&migration_state_notifiers, s);
1008
1009     migrate_compress_threads_create();
1010     qemu_thread_create(&s->thread, "migration", migration_thread, s,
1011                        QEMU_THREAD_JOINABLE);
1012 }