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