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/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 "qemu/rcu.h"
  26 #include "migration/block.h"
  27 #include "qemu/thread.h"
  28 #include "qmp-commands.h"
  29 #include "trace.h"
  30 #include "qapi/util.h"
  31 #include "qapi-event.h"
  32
  33 #define MAX_THROTTLE  (32 << 20)      /* Migration speed throttling */
  34
  35 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
  36  * data. */
  37 #define BUFFER_DELAY     100
  38 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
  39
  40 /* Default compression thread count */
  41 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
  42 /* Default decompression thread count, usually decompression is at
  43  * least 4 times as fast as compression.*/
  44 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
  45 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
  46 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
  47
  48 /* Migration XBZRLE default cache size */
  49 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
  50
  51 static NotifierList migration_state_notifiers =
  52     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  53
  54 static bool deferred_incoming;
  55
  56 /* When we add fault tolerance, we could have several
  57    migrations at once.  For now we don't need to add
  58    dynamic creation of migration */
  59
  60 /* For outgoing */
  61 MigrationState *migrate_get_current(void)
  62 {
  63     static MigrationState current_migration = {
  64         .state = MIGRATION_STATUS_NONE,
  65         .bandwidth_limit = MAX_THROTTLE,
  66         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
  67         .mbps = -1,
  68         .parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] =
  69                 DEFAULT_MIGRATE_COMPRESS_LEVEL,
  70         .parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
  71                 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
  72         .parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
  73                 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
  74     };
  75
  76     return &current_migration;
  77 }
  78
  79 /* For incoming */
  80 static MigrationIncomingState *mis_current;
  81
  82 MigrationIncomingState *migration_incoming_get_current(void)
  83 {
  84     return mis_current;
  85 }
  86
  87 MigrationIncomingState *migration_incoming_state_new(QEMUFile* f)
  88 {
  89     mis_current = g_malloc0(sizeof(MigrationIncomingState));
  90     mis_current->file = f;
  91     QLIST_INIT(&mis_current->loadvm_handlers);
  92
  93     return mis_current;
  94 }
  95
  96 void migration_incoming_state_destroy(void)
  97 {
  98     loadvm_free_handlers(mis_current);
  99     g_free(mis_current);
 100     mis_current = NULL;
 101 }
 102
 103
 104 typedef struct {
 105     bool optional;
 106     uint32_t size;
 107     uint8_t runstate[100];
 108     RunState state;
 109     bool received;
 110 } GlobalState;
 111
 112 static GlobalState global_state;
 113
 114 int global_state_store(void)
 115 {
 116     if (!runstate_store((char *)global_state.runstate,
 117                         sizeof(global_state.runstate))) {
 118         error_report("runstate name too big: %s", global_state.runstate);
 119         trace_migrate_state_too_big();
 120         return -EINVAL;
 121     }
 122     return 0;
 123 }
 124
 125 static bool global_state_received(void)
 126 {
 127     return global_state.received;
 128 }
 129
 130 static RunState global_state_get_runstate(void)
 131 {
 132     return global_state.state;
 133 }
 134
 135 void global_state_set_optional(void)
 136 {
 137     global_state.optional = true;
 138 }
 139
 140 static bool global_state_needed(void *opaque)
 141 {
 142     GlobalState *s = opaque;
 143     char *runstate = (char *)s->runstate;
 144
 145     /* If it is not optional, it is mandatory */
 146
 147     if (s->optional == false) {
 148         return true;
 149     }
 150
 151     /* If state is running or paused, it is not needed */
 152
 153     if (strcmp(runstate, "running") == 0 ||
 154         strcmp(runstate, "paused") == 0) {
 155         return false;
 156     }
 157
 158     /* for any other state it is needed */
 159     return true;
 160 }
 161
 162 static int global_state_post_load(void *opaque, int version_id)
 163 {
 164     GlobalState *s = opaque;
 165     Error *local_err = NULL;
 166     int r;
 167     char *runstate = (char *)s->runstate;
 168
 169     s->received = true;
 170     trace_migrate_global_state_post_load(runstate);
 171
 172     r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE_MAX,
 173                                 -1, &local_err);
 174
 175     if (r == -1) {
 176         if (local_err) {
 177             error_report_err(local_err);
 178         }
 179         return -EINVAL;
 180     }
 181     s->state = r;
 182
 183     return 0;
 184 }
 185
 186 static void global_state_pre_save(void *opaque)
 187 {
 188     GlobalState *s = opaque;
 189
 190     trace_migrate_global_state_pre_save((char *)s->runstate);
 191     s->size = strlen((char *)s->runstate) + 1;
 192 }
 193
 194 static const VMStateDescription vmstate_globalstate = {
 195     .name = "globalstate",
 196     .version_id = 1,
 197     .minimum_version_id = 1,
 198     .post_load = global_state_post_load,
 199     .pre_save = global_state_pre_save,
 200     .needed = global_state_needed,
 201     .fields = (VMStateField[]) {
 202         VMSTATE_UINT32(size, GlobalState),
 203         VMSTATE_BUFFER(runstate, GlobalState),
 204         VMSTATE_END_OF_LIST()
 205     },
 206 };
 207
 208 void register_global_state(void)
 209 {
 210     /* We would use it independently that we receive it */
 211     strcpy((char *)&global_state.runstate, "");
 212     global_state.received = false;
 213     vmstate_register(NULL, 0, &vmstate_globalstate, &global_state);
 214 }
 215
 216 static void migrate_generate_event(int new_state)
 217 {
 218     if (migrate_use_events()) {
 219         qapi_event_send_migration(new_state, &error_abort);
 220     }
 221 }
 222
 223 /*
 224  * Called on -incoming with a defer: uri.
 225  * The migration can be started later after any parameters have been
 226  * changed.
 227  */
 228 static void deferred_incoming_migration(Error **errp)
 229 {
 230     if (deferred_incoming) {
 231         error_setg(errp, "Incoming migration already deferred");
 232     }
 233     deferred_incoming = true;
 234 }
 235
 236 void qemu_start_incoming_migration(const char *uri, Error **errp)
 237 {
 238     const char *p;
 239
 240     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
 241     if (!strcmp(uri, "defer")) {
 242         deferred_incoming_migration(errp);
 243     } else if (strstart(uri, "tcp:", &p)) {
 244         tcp_start_incoming_migration(p, errp);
 245 #ifdef CONFIG_RDMA
 246     } else if (strstart(uri, "rdma:", &p)) {
 247         rdma_start_incoming_migration(p, errp);
 248 #endif
 249 #if !defined(WIN32)
 250     } else if (strstart(uri, "exec:", &p)) {
 251         exec_start_incoming_migration(p, errp);
 252     } else if (strstart(uri, "unix:", &p)) {
 253         unix_start_incoming_migration(p, errp);
 254     } else if (strstart(uri, "fd:", &p)) {
 255         fd_start_incoming_migration(p, errp);
 256 #endif
 257     } else {
 258         error_setg(errp, "unknown migration protocol: %s", uri);
 259     }
 260 }
 261
 262 static void process_incoming_migration_co(void *opaque)
 263 {
 264     QEMUFile *f = opaque;
 265     Error *local_err = NULL;
 266     int ret;
 267
 268     migration_incoming_state_new(f);
 269     migrate_generate_event(MIGRATION_STATUS_ACTIVE);
 270     ret = qemu_loadvm_state(f);
 271
 272     qemu_fclose(f);
 273     free_xbzrle_decoded_buf();
 274     migration_incoming_state_destroy();
 275
 276     if (ret < 0) {
 277         migrate_generate_event(MIGRATION_STATUS_FAILED);
 278         error_report("load of migration failed: %s", strerror(-ret));
 279         migrate_decompress_threads_join();
 280         exit(EXIT_FAILURE);
 281     }
 282     migrate_generate_event(MIGRATION_STATUS_COMPLETED);
 283     qemu_announce_self();
 284
 285     /* Make sure all file formats flush their mutable metadata */
 286     bdrv_invalidate_cache_all(&local_err);
 287     if (local_err) {
 288         error_report_err(local_err);
 289         migrate_decompress_threads_join();
 290         exit(EXIT_FAILURE);
 291     }
 292
 293     /* If global state section was not received or we are in running
 294        state, we need to obey autostart. Any other state is set with
 295        runstate_set. */
 296
 297     if (!global_state_received() ||
 298         global_state_get_runstate() == RUN_STATE_RUNNING) {
 299         if (autostart) {
 300             vm_start();
 301         } else {
 302             runstate_set(RUN_STATE_PAUSED);
 303         }
 304     } else {
 305         runstate_set(global_state_get_runstate());
 306     }
 307     migrate_decompress_threads_join();
 308 }
 309
 310 void process_incoming_migration(QEMUFile *f)
 311 {
 312     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
 313     int fd = qemu_get_fd(f);
 314
 315     assert(fd != -1);
 316     migrate_decompress_threads_create();
 317     qemu_set_nonblock(fd);
 318     qemu_coroutine_enter(co, f);
 319 }
 320
 321 /* amount of nanoseconds we are willing to wait for migration to be down.
 322  * the choice of nanoseconds is because it is the maximum resolution that
 323  * get_clock() can achieve. It is an internal measure. All user-visible
 324  * units must be in seconds */
 325 static uint64_t max_downtime = 300000000;
 326
 327 uint64_t migrate_max_downtime(void)
 328 {
 329     return max_downtime;
 330 }
 331
 332 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 333 {
 334     MigrationCapabilityStatusList *head = NULL;
 335     MigrationCapabilityStatusList *caps;
 336     MigrationState *s = migrate_get_current();
 337     int i;
 338
 339     caps = NULL; /* silence compiler warning */
 340     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 341         if (head == NULL) {
 342             head = g_malloc0(sizeof(*caps));
 343             caps = head;
 344         } else {
 345             caps->next = g_malloc0(sizeof(*caps));
 346             caps = caps->next;
 347         }
 348         caps->value =
 349             g_malloc(sizeof(*caps->value));
 350         caps->value->capability = i;
 351         caps->value->state = s->enabled_capabilities[i];
 352     }
 353
 354     return head;
 355 }
 356
 357 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 358 {
 359     MigrationParameters *params;
 360     MigrationState *s = migrate_get_current();
 361
 362     params = g_malloc0(sizeof(*params));
 363     params->compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 364     params->compress_threads =
 365             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 366     params->decompress_threads =
 367             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 368
 369     return params;
 370 }
 371
 372 static void get_xbzrle_cache_stats(MigrationInfo *info)
 373 {
 374     if (migrate_use_xbzrle()) {
 375         info->has_xbzrle_cache = true;
 376         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 377         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 378         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 379         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 380         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 381         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
 382         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 383     }
 384 }
 385
 386 MigrationInfo *qmp_query_migrate(Error **errp)
 387 {
 388     MigrationInfo *info = g_malloc0(sizeof(*info));
 389     MigrationState *s = migrate_get_current();
 390
 391     switch (s->state) {
 392     case MIGRATION_STATUS_NONE:
 393         /* no migration has happened ever */
 394         break;
 395     case MIGRATION_STATUS_SETUP:
 396         info->has_status = true;
 397         info->has_total_time = false;
 398         break;
 399     case MIGRATION_STATUS_ACTIVE:
 400     case MIGRATION_STATUS_CANCELLING:
 401         info->has_status = true;
 402         info->has_total_time = true;
 403         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 404             - s->total_time;
 405         info->has_expected_downtime = true;
 406         info->expected_downtime = s->expected_downtime;
 407         info->has_setup_time = true;
 408         info->setup_time = s->setup_time;
 409
 410         info->has_ram = true;
 411         info->ram = g_malloc0(sizeof(*info->ram));
 412         info->ram->transferred = ram_bytes_transferred();
 413         info->ram->remaining = ram_bytes_remaining();
 414         info->ram->total = ram_bytes_total();
 415         info->ram->duplicate = dup_mig_pages_transferred();
 416         info->ram->skipped = skipped_mig_pages_transferred();
 417         info->ram->normal = norm_mig_pages_transferred();
 418         info->ram->normal_bytes = norm_mig_bytes_transferred();
 419         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 420         info->ram->mbps = s->mbps;
 421         info->ram->dirty_sync_count = s->dirty_sync_count;
 422
 423         if (blk_mig_active()) {
 424             info->has_disk = true;
 425             info->disk = g_malloc0(sizeof(*info->disk));
 426             info->disk->transferred = blk_mig_bytes_transferred();
 427             info->disk->remaining = blk_mig_bytes_remaining();
 428             info->disk->total = blk_mig_bytes_total();
 429         }
 430
 431         get_xbzrle_cache_stats(info);
 432         break;
 433     case MIGRATION_STATUS_COMPLETED:
 434         get_xbzrle_cache_stats(info);
 435
 436         info->has_status = true;
 437         info->has_total_time = true;
 438         info->total_time = s->total_time;
 439         info->has_downtime = true;
 440         info->downtime = s->downtime;
 441         info->has_setup_time = true;
 442         info->setup_time = s->setup_time;
 443
 444         info->has_ram = true;
 445         info->ram = g_malloc0(sizeof(*info->ram));
 446         info->ram->transferred = ram_bytes_transferred();
 447         info->ram->remaining = 0;
 448         info->ram->total = ram_bytes_total();
 449         info->ram->duplicate = dup_mig_pages_transferred();
 450         info->ram->skipped = skipped_mig_pages_transferred();
 451         info->ram->normal = norm_mig_pages_transferred();
 452         info->ram->normal_bytes = norm_mig_bytes_transferred();
 453         info->ram->mbps = s->mbps;
 454         info->ram->dirty_sync_count = s->dirty_sync_count;
 455         break;
 456     case MIGRATION_STATUS_FAILED:
 457         info->has_status = true;
 458         break;
 459     case MIGRATION_STATUS_CANCELLED:
 460         info->has_status = true;
 461         break;
 462     }
 463     info->status = s->state;
 464
 465     return info;
 466 }
 467
 468 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 469                                   Error **errp)
 470 {
 471     MigrationState *s = migrate_get_current();
 472     MigrationCapabilityStatusList *cap;
 473
 474     if (s->state == MIGRATION_STATUS_ACTIVE ||
 475         s->state == MIGRATION_STATUS_SETUP) {
 476         error_setg(errp, QERR_MIGRATION_ACTIVE);
 477         return;
 478     }
 479
 480     for (cap = params; cap; cap = cap->next) {
 481         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 482     }
 483 }
 484
 485 void qmp_migrate_set_parameters(bool has_compress_level,
 486                                 int64_t compress_level,
 487                                 bool has_compress_threads,
 488                                 int64_t compress_threads,
 489                                 bool has_decompress_threads,
 490                                 int64_t decompress_threads, Error **errp)
 491 {
 492     MigrationState *s = migrate_get_current();
 493
 494     if (has_compress_level && (compress_level < 0 || compress_level > 9)) {
 495         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
 496                    "is invalid, it should be in the range of 0 to 9");
 497         return;
 498     }
 499     if (has_compress_threads &&
 500             (compress_threads < 1 || compress_threads > 255)) {
 501         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 502                    "compress_threads",
 503                    "is invalid, it should be in the range of 1 to 255");
 504         return;
 505     }
 506     if (has_decompress_threads &&
 507             (decompress_threads < 1 || decompress_threads > 255)) {
 508         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 509                    "decompress_threads",
 510                    "is invalid, it should be in the range of 1 to 255");
 511         return;
 512     }
 513
 514     if (has_compress_level) {
 515         s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 516     }
 517     if (has_compress_threads) {
 518         s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] = compress_threads;
 519     }
 520     if (has_decompress_threads) {
 521         s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 522                                                     decompress_threads;
 523     }
 524 }
 525
 526 /* shared migration helpers */
 527
 528 static void migrate_set_state(MigrationState *s, int old_state, int new_state)
 529 {
 530     if (atomic_cmpxchg(&s->state, old_state, new_state) == old_state) {
 531         trace_migrate_set_state(new_state);
 532         migrate_generate_event(new_state);
 533     }
 534 }
 535
 536 static void migrate_fd_cleanup(void *opaque)
 537 {
 538     MigrationState *s = opaque;
 539
 540     qemu_bh_delete(s->cleanup_bh);
 541     s->cleanup_bh = NULL;
 542
 543     if (s->file) {
 544         trace_migrate_fd_cleanup();
 545         qemu_mutex_unlock_iothread();
 546         qemu_thread_join(&s->thread);
 547         qemu_mutex_lock_iothread();
 548
 549         migrate_compress_threads_join();
 550         qemu_fclose(s->file);
 551         s->file = NULL;
 552     }
 553
 554     assert(s->state != MIGRATION_STATUS_ACTIVE);
 555
 556     if (s->state != MIGRATION_STATUS_COMPLETED) {
 557         qemu_savevm_state_cancel();
 558         if (s->state == MIGRATION_STATUS_CANCELLING) {
 559             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
 560                               MIGRATION_STATUS_CANCELLED);
 561         }
 562     }
 563
 564     notifier_list_notify(&migration_state_notifiers, s);
 565 }
 566
 567 void migrate_fd_error(MigrationState *s)
 568 {
 569     trace_migrate_fd_error();
 570     assert(s->file == NULL);
 571     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 572     notifier_list_notify(&migration_state_notifiers, s);
 573 }
 574
 575 static void migrate_fd_cancel(MigrationState *s)
 576 {
 577     int old_state ;
 578     QEMUFile *f = migrate_get_current()->file;
 579     trace_migrate_fd_cancel();
 580
 581     do {
 582         old_state = s->state;
 583         if (old_state != MIGRATION_STATUS_SETUP &&
 584             old_state != MIGRATION_STATUS_ACTIVE) {
 585             break;
 586         }
 587         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
 588     } while (s->state != MIGRATION_STATUS_CANCELLING);
 589
 590     /*
 591      * If we're unlucky the migration code might be stuck somewhere in a
 592      * send/write while the network has failed and is waiting to timeout;
 593      * if we've got shutdown(2) available then we can force it to quit.
 594      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
 595      * called in a bh, so there is no race against this cancel.
 596      */
 597     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
 598         qemu_file_shutdown(f);
 599     }
 600 }
 601
 602 void add_migration_state_change_notifier(Notifier *notify)
 603 {
 604     notifier_list_add(&migration_state_notifiers, notify);
 605 }
 606
 607 void remove_migration_state_change_notifier(Notifier *notify)
 608 {
 609     notifier_remove(notify);
 610 }
 611
 612 bool migration_in_setup(MigrationState *s)
 613 {
 614     return s->state == MIGRATION_STATUS_SETUP;
 615 }
 616
 617 bool migration_has_finished(MigrationState *s)
 618 {
 619     return s->state == MIGRATION_STATUS_COMPLETED;
 620 }
 621
 622 bool migration_has_failed(MigrationState *s)
 623 {
 624     return (s->state == MIGRATION_STATUS_CANCELLED ||
 625             s->state == MIGRATION_STATUS_FAILED);
 626 }
 627
 628 static MigrationState *migrate_init(const MigrationParams *params)
 629 {
 630     MigrationState *s = migrate_get_current();
 631     int64_t bandwidth_limit = s->bandwidth_limit;
 632     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 633     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 634     int compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 635     int compress_thread_count =
 636             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 637     int decompress_thread_count =
 638             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 639
 640     memcpy(enabled_capabilities, s->enabled_capabilities,
 641            sizeof(enabled_capabilities));
 642
 643     memset(s, 0, sizeof(*s));
 644     s->params = *params;
 645     memcpy(s->enabled_capabilities, enabled_capabilities,
 646            sizeof(enabled_capabilities));
 647     s->xbzrle_cache_size = xbzrle_cache_size;
 648
 649     s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 650     s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
 651                compress_thread_count;
 652     s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 653                decompress_thread_count;
 654     s->bandwidth_limit = bandwidth_limit;
 655     migrate_set_state(s, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
 656
 657     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 658     return s;
 659 }
 660
 661 static GSList *migration_blockers;
 662
 663 void migrate_add_blocker(Error *reason)
 664 {
 665     migration_blockers = g_slist_prepend(migration_blockers, reason);
 666 }
 667
 668 void migrate_del_blocker(Error *reason)
 669 {
 670     migration_blockers = g_slist_remove(migration_blockers, reason);
 671 }
 672
 673 void qmp_migrate_incoming(const char *uri, Error **errp)
 674 {
 675     Error *local_err = NULL;
 676     static bool once = true;
 677
 678     if (!deferred_incoming) {
 679         error_setg(errp, "For use with '-incoming defer'");
 680         return;
 681     }
 682     if (!once) {
 683         error_setg(errp, "The incoming migration has already been started");
 684     }
 685
 686     qemu_start_incoming_migration(uri, &local_err);
 687
 688     if (local_err) {
 689         error_propagate(errp, local_err);
 690         return;
 691     }
 692
 693     once = false;
 694 }
 695
 696 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 697                  bool has_inc, bool inc, bool has_detach, bool detach,
 698                  Error **errp)
 699 {
 700     Error *local_err = NULL;
 701     MigrationState *s = migrate_get_current();
 702     MigrationParams params;
 703     const char *p;
 704
 705     params.blk = has_blk && blk;
 706     params.shared = has_inc && inc;
 707
 708     if (s->state == MIGRATION_STATUS_ACTIVE ||
 709         s->state == MIGRATION_STATUS_SETUP ||
 710         s->state == MIGRATION_STATUS_CANCELLING) {
 711         error_setg(errp, QERR_MIGRATION_ACTIVE);
 712         return;
 713     }
 714     if (runstate_check(RUN_STATE_INMIGRATE)) {
 715         error_setg(errp, "Guest is waiting for an incoming migration");
 716         return;
 717     }
 718
 719     if (qemu_savevm_state_blocked(errp)) {
 720         return;
 721     }
 722
 723     if (migration_blockers) {
 724         *errp = error_copy(migration_blockers->data);
 725         return;
 726     }
 727
 728     /* We are starting a new migration, so we want to start in a clean
 729        state.  This change is only needed if previous migration
 730        failed/was cancelled.  We don't use migrate_set_state() because
 731        we are setting the initial state, not changing it. */
 732     s->state = MIGRATION_STATUS_NONE;
 733
 734     s = migrate_init(&params);
 735
 736     if (strstart(uri, "tcp:", &p)) {
 737         tcp_start_outgoing_migration(s, p, &local_err);
 738 #ifdef CONFIG_RDMA
 739     } else if (strstart(uri, "rdma:", &p)) {
 740         rdma_start_outgoing_migration(s, p, &local_err);
 741 #endif
 742 #if !defined(WIN32)
 743     } else if (strstart(uri, "exec:", &p)) {
 744         exec_start_outgoing_migration(s, p, &local_err);
 745     } else if (strstart(uri, "unix:", &p)) {
 746         unix_start_outgoing_migration(s, p, &local_err);
 747     } else if (strstart(uri, "fd:", &p)) {
 748         fd_start_outgoing_migration(s, p, &local_err);
 749 #endif
 750     } else {
 751         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
 752                    "a valid migration protocol");
 753         migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 754         return;
 755     }
 756
 757     if (local_err) {
 758         migrate_fd_error(s);
 759         error_propagate(errp, local_err);
 760         return;
 761     }
 762 }
 763
 764 void qmp_migrate_cancel(Error **errp)
 765 {
 766     migrate_fd_cancel(migrate_get_current());
 767 }
 768
 769 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 770 {
 771     MigrationState *s = migrate_get_current();
 772     int64_t new_size;
 773
 774     /* Check for truncation */
 775     if (value != (size_t)value) {
 776         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 777                    "exceeding address space");
 778         return;
 779     }
 780
 781     /* Cache should not be larger than guest ram size */
 782     if (value > ram_bytes_total()) {
 783         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 784                    "exceeds guest ram size ");
 785         return;
 786     }
 787
 788     new_size = xbzrle_cache_resize(value);
 789     if (new_size < 0) {
 790         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 791                    "is smaller than page size");
 792         return;
 793     }
 794
 795     s->xbzrle_cache_size = new_size;
 796 }
 797
 798 int64_t qmp_query_migrate_cache_size(Error **errp)
 799 {
 800     return migrate_xbzrle_cache_size();
 801 }
 802
 803 void qmp_migrate_set_speed(int64_t value, Error **errp)
 804 {
 805     MigrationState *s;
 806
 807     if (value < 0) {
 808         value = 0;
 809     }
 810     if (value > SIZE_MAX) {
 811         value = SIZE_MAX;
 812     }
 813
 814     s = migrate_get_current();
 815     s->bandwidth_limit = value;
 816     if (s->file) {
 817         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
 818     }
 819 }
 820
 821 void qmp_migrate_set_downtime(double value, Error **errp)
 822 {
 823     value *= 1e9;
 824     value = MAX(0, MIN(UINT64_MAX, value));
 825     max_downtime = (uint64_t)value;
 826 }
 827
 828 bool migrate_auto_converge(void)
 829 {
 830     MigrationState *s;
 831
 832     s = migrate_get_current();
 833
 834     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
 835 }
 836
 837 bool migrate_zero_blocks(void)
 838 {
 839     MigrationState *s;
 840
 841     s = migrate_get_current();
 842
 843     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
 844 }
 845
 846 bool migrate_use_compression(void)
 847 {
 848     MigrationState *s;
 849
 850     s = migrate_get_current();
 851
 852     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
 853 }
 854
 855 int migrate_compress_level(void)
 856 {
 857     MigrationState *s;
 858
 859     s = migrate_get_current();
 860
 861     return s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 862 }
 863
 864 int migrate_compress_threads(void)
 865 {
 866     MigrationState *s;
 867
 868     s = migrate_get_current();
 869
 870     return s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 871 }
 872
 873 int migrate_decompress_threads(void)
 874 {
 875     MigrationState *s;
 876
 877     s = migrate_get_current();
 878
 879     return s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 880 }
 881
 882 bool migrate_use_events(void)
 883 {
 884     MigrationState *s;
 885
 886     s = migrate_get_current();
 887
 888     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
 889 }
 890
 891 int migrate_use_xbzrle(void)
 892 {
 893     MigrationState *s;
 894
 895     s = migrate_get_current();
 896
 897     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 898 }
 899
 900 int64_t migrate_xbzrle_cache_size(void)
 901 {
 902     MigrationState *s;
 903
 904     s = migrate_get_current();
 905
 906     return s->xbzrle_cache_size;
 907 }
 908
 909 /* migration thread support */
 910
 911 static void *migration_thread(void *opaque)
 912 {
 913     MigrationState *s = opaque;
 914     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 915     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 916     int64_t initial_bytes = 0;
 917     int64_t max_size = 0;
 918     int64_t start_time = initial_time;
 919     bool old_vm_running = false;
 920
 921     rcu_register_thread();
 922
 923     qemu_savevm_state_header(s->file);
 924     qemu_savevm_state_begin(s->file, &s->params);
 925
 926     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
 927     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
 928
 929     while (s->state == MIGRATION_STATUS_ACTIVE) {
 930         int64_t current_time;
 931         uint64_t pending_size;
 932
 933         if (!qemu_file_rate_limit(s->file)) {
 934             pending_size = qemu_savevm_state_pending(s->file, max_size);
 935             trace_migrate_pending(pending_size, max_size);
 936             if (pending_size && pending_size >= max_size) {
 937                 qemu_savevm_state_iterate(s->file);
 938             } else {
 939                 int ret;
 940
 941                 qemu_mutex_lock_iothread();
 942                 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 943                 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 944                 old_vm_running = runstate_is_running();
 945
 946                 ret = global_state_store();
 947                 if (!ret) {
 948                     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 949                     if (ret >= 0) {
 950                         qemu_file_set_rate_limit(s->file, INT64_MAX);
 951                         qemu_savevm_state_complete(s->file);
 952                     }
 953                 }
 954                 qemu_mutex_unlock_iothread();
 955
 956                 if (ret < 0) {
 957                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 958                                       MIGRATION_STATUS_FAILED);
 959                     break;
 960                 }
 961
 962                 if (!qemu_file_get_error(s->file)) {
 963                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 964                                       MIGRATION_STATUS_COMPLETED);
 965                     break;
 966                 }
 967             }
 968         }
 969
 970         if (qemu_file_get_error(s->file)) {
 971             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 972                               MIGRATION_STATUS_FAILED);
 973             break;
 974         }
 975         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 976         if (current_time >= initial_time + BUFFER_DELAY) {
 977             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
 978             uint64_t time_spent = current_time - initial_time;
 979             double bandwidth = transferred_bytes / time_spent;
 980             max_size = bandwidth * migrate_max_downtime() / 1000000;
 981
 982             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
 983                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
 984
 985             trace_migrate_transferred(transferred_bytes, time_spent,
 986                                       bandwidth, max_size);
 987             /* if we haven't sent anything, we don't want to recalculate
 988                10000 is a small enough number for our purposes */
 989             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
 990                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
 991             }
 992
 993             qemu_file_reset_rate_limit(s->file);
 994             initial_time = current_time;
 995             initial_bytes = qemu_ftell(s->file);
 996         }
 997         if (qemu_file_rate_limit(s->file)) {
 998             /* usleep expects microseconds */
 999             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
1000         }
1001     }
1002
1003     qemu_mutex_lock_iothread();
1004     if (s->state == MIGRATION_STATUS_COMPLETED) {
1005         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1006         uint64_t transferred_bytes = qemu_ftell(s->file);
1007         s->total_time = end_time - s->total_time;
1008         s->downtime = end_time - start_time;
1009         if (s->total_time) {
1010             s->mbps = (((double) transferred_bytes * 8.0) /
1011                        ((double) s->total_time)) / 1000;
1012         }
1013         runstate_set(RUN_STATE_POSTMIGRATE);
1014     } else {
1015         if (old_vm_running) {
1016             vm_start();
1017         }
1018     }
1019     qemu_bh_schedule(s->cleanup_bh);
1020     qemu_mutex_unlock_iothread();
1021
1022     rcu_unregister_thread();
1023     return NULL;
1024 }
1025
1026 void migrate_fd_connect(MigrationState *s)
1027 {
1028     /* This is a best 1st approximation. ns to ms */
1029     s->expected_downtime = max_downtime/1000000;
1030     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
1031
1032     qemu_file_set_rate_limit(s->file,
1033                              s->bandwidth_limit / XFER_LIMIT_RATIO);
1034
1035     /* Notify before starting migration thread */
1036     notifier_list_notify(&migration_state_notifiers, s);
1037
1038     migrate_compress_threads_create();
1039     qemu_thread_create(&s->thread, "migration", migration_thread, s,
1040                        QEMU_THREAD_JOINABLE);
1041 }