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 void global_state_store_running(void)
 126 {
 127     const char *state = RunState_lookup[RUN_STATE_RUNNING];
 128     strncpy((char *)global_state.runstate,
 129            state, sizeof(global_state.runstate));
 130 }
 131
 132 static bool global_state_received(void)
 133 {
 134     return global_state.received;
 135 }
 136
 137 static RunState global_state_get_runstate(void)
 138 {
 139     return global_state.state;
 140 }
 141
 142 void global_state_set_optional(void)
 143 {
 144     global_state.optional = true;
 145 }
 146
 147 static bool global_state_needed(void *opaque)
 148 {
 149     GlobalState *s = opaque;
 150     char *runstate = (char *)s->runstate;
 151
 152     /* If it is not optional, it is mandatory */
 153
 154     if (s->optional == false) {
 155         return true;
 156     }
 157
 158     /* If state is running or paused, it is not needed */
 159
 160     if (strcmp(runstate, "running") == 0 ||
 161         strcmp(runstate, "paused") == 0) {
 162         return false;
 163     }
 164
 165     /* for any other state it is needed */
 166     return true;
 167 }
 168
 169 static int global_state_post_load(void *opaque, int version_id)
 170 {
 171     GlobalState *s = opaque;
 172     Error *local_err = NULL;
 173     int r;
 174     char *runstate = (char *)s->runstate;
 175
 176     s->received = true;
 177     trace_migrate_global_state_post_load(runstate);
 178
 179     r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE_MAX,
 180                                 -1, &local_err);
 181
 182     if (r == -1) {
 183         if (local_err) {
 184             error_report_err(local_err);
 185         }
 186         return -EINVAL;
 187     }
 188     s->state = r;
 189
 190     return 0;
 191 }
 192
 193 static void global_state_pre_save(void *opaque)
 194 {
 195     GlobalState *s = opaque;
 196
 197     trace_migrate_global_state_pre_save((char *)s->runstate);
 198     s->size = strlen((char *)s->runstate) + 1;
 199 }
 200
 201 static const VMStateDescription vmstate_globalstate = {
 202     .name = "globalstate",
 203     .version_id = 1,
 204     .minimum_version_id = 1,
 205     .post_load = global_state_post_load,
 206     .pre_save = global_state_pre_save,
 207     .needed = global_state_needed,
 208     .fields = (VMStateField[]) {
 209         VMSTATE_UINT32(size, GlobalState),
 210         VMSTATE_BUFFER(runstate, GlobalState),
 211         VMSTATE_END_OF_LIST()
 212     },
 213 };
 214
 215 void register_global_state(void)
 216 {
 217     /* We would use it independently that we receive it */
 218     strcpy((char *)&global_state.runstate, "");
 219     global_state.received = false;
 220     vmstate_register(NULL, 0, &vmstate_globalstate, &global_state);
 221 }
 222
 223 static void migrate_generate_event(int new_state)
 224 {
 225     if (migrate_use_events()) {
 226         qapi_event_send_migration(new_state, &error_abort);
 227     }
 228 }
 229
 230 /*
 231  * Called on -incoming with a defer: uri.
 232  * The migration can be started later after any parameters have been
 233  * changed.
 234  */
 235 static void deferred_incoming_migration(Error **errp)
 236 {
 237     if (deferred_incoming) {
 238         error_setg(errp, "Incoming migration already deferred");
 239     }
 240     deferred_incoming = true;
 241 }
 242
 243 void qemu_start_incoming_migration(const char *uri, Error **errp)
 244 {
 245     const char *p;
 246
 247     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
 248     if (!strcmp(uri, "defer")) {
 249         deferred_incoming_migration(errp);
 250     } else if (strstart(uri, "tcp:", &p)) {
 251         tcp_start_incoming_migration(p, errp);
 252 #ifdef CONFIG_RDMA
 253     } else if (strstart(uri, "rdma:", &p)) {
 254         rdma_start_incoming_migration(p, errp);
 255 #endif
 256 #if !defined(WIN32)
 257     } else if (strstart(uri, "exec:", &p)) {
 258         exec_start_incoming_migration(p, errp);
 259     } else if (strstart(uri, "unix:", &p)) {
 260         unix_start_incoming_migration(p, errp);
 261     } else if (strstart(uri, "fd:", &p)) {
 262         fd_start_incoming_migration(p, errp);
 263 #endif
 264     } else {
 265         error_setg(errp, "unknown migration protocol: %s", uri);
 266     }
 267 }
 268
 269 static void process_incoming_migration_co(void *opaque)
 270 {
 271     QEMUFile *f = opaque;
 272     Error *local_err = NULL;
 273     int ret;
 274
 275     migration_incoming_state_new(f);
 276     migrate_generate_event(MIGRATION_STATUS_ACTIVE);
 277     ret = qemu_loadvm_state(f);
 278
 279     qemu_fclose(f);
 280     free_xbzrle_decoded_buf();
 281     migration_incoming_state_destroy();
 282
 283     if (ret < 0) {
 284         migrate_generate_event(MIGRATION_STATUS_FAILED);
 285         error_report("load of migration failed: %s", strerror(-ret));
 286         migrate_decompress_threads_join();
 287         exit(EXIT_FAILURE);
 288     }
 289     migrate_generate_event(MIGRATION_STATUS_COMPLETED);
 290     qemu_announce_self();
 291
 292     /* Make sure all file formats flush their mutable metadata */
 293     bdrv_invalidate_cache_all(&local_err);
 294     if (local_err) {
 295         error_report_err(local_err);
 296         migrate_decompress_threads_join();
 297         exit(EXIT_FAILURE);
 298     }
 299
 300     /* If global state section was not received or we are in running
 301        state, we need to obey autostart. Any other state is set with
 302        runstate_set. */
 303
 304     if (!global_state_received() ||
 305         global_state_get_runstate() == RUN_STATE_RUNNING) {
 306         if (autostart) {
 307             vm_start();
 308         } else {
 309             runstate_set(RUN_STATE_PAUSED);
 310         }
 311     } else {
 312         runstate_set(global_state_get_runstate());
 313     }
 314     migrate_decompress_threads_join();
 315 }
 316
 317 void process_incoming_migration(QEMUFile *f)
 318 {
 319     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
 320     int fd = qemu_get_fd(f);
 321
 322     assert(fd != -1);
 323     migrate_decompress_threads_create();
 324     qemu_set_nonblock(fd);
 325     qemu_coroutine_enter(co, f);
 326 }
 327
 328 /* amount of nanoseconds we are willing to wait for migration to be down.
 329  * the choice of nanoseconds is because it is the maximum resolution that
 330  * get_clock() can achieve. It is an internal measure. All user-visible
 331  * units must be in seconds */
 332 static uint64_t max_downtime = 300000000;
 333
 334 uint64_t migrate_max_downtime(void)
 335 {
 336     return max_downtime;
 337 }
 338
 339 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 340 {
 341     MigrationCapabilityStatusList *head = NULL;
 342     MigrationCapabilityStatusList *caps;
 343     MigrationState *s = migrate_get_current();
 344     int i;
 345
 346     caps = NULL; /* silence compiler warning */
 347     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 348         if (head == NULL) {
 349             head = g_malloc0(sizeof(*caps));
 350             caps = head;
 351         } else {
 352             caps->next = g_malloc0(sizeof(*caps));
 353             caps = caps->next;
 354         }
 355         caps->value =
 356             g_malloc(sizeof(*caps->value));
 357         caps->value->capability = i;
 358         caps->value->state = s->enabled_capabilities[i];
 359     }
 360
 361     return head;
 362 }
 363
 364 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 365 {
 366     MigrationParameters *params;
 367     MigrationState *s = migrate_get_current();
 368
 369     params = g_malloc0(sizeof(*params));
 370     params->compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 371     params->compress_threads =
 372             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 373     params->decompress_threads =
 374             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 375
 376     return params;
 377 }
 378
 379 static void get_xbzrle_cache_stats(MigrationInfo *info)
 380 {
 381     if (migrate_use_xbzrle()) {
 382         info->has_xbzrle_cache = true;
 383         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 384         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 385         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 386         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 387         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 388         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
 389         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 390     }
 391 }
 392
 393 MigrationInfo *qmp_query_migrate(Error **errp)
 394 {
 395     MigrationInfo *info = g_malloc0(sizeof(*info));
 396     MigrationState *s = migrate_get_current();
 397
 398     switch (s->state) {
 399     case MIGRATION_STATUS_NONE:
 400         /* no migration has happened ever */
 401         break;
 402     case MIGRATION_STATUS_SETUP:
 403         info->has_status = true;
 404         info->has_total_time = false;
 405         break;
 406     case MIGRATION_STATUS_ACTIVE:
 407     case MIGRATION_STATUS_CANCELLING:
 408         info->has_status = true;
 409         info->has_total_time = true;
 410         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 411             - s->total_time;
 412         info->has_expected_downtime = true;
 413         info->expected_downtime = s->expected_downtime;
 414         info->has_setup_time = true;
 415         info->setup_time = s->setup_time;
 416
 417         info->has_ram = true;
 418         info->ram = g_malloc0(sizeof(*info->ram));
 419         info->ram->transferred = ram_bytes_transferred();
 420         info->ram->remaining = ram_bytes_remaining();
 421         info->ram->total = ram_bytes_total();
 422         info->ram->duplicate = dup_mig_pages_transferred();
 423         info->ram->skipped = skipped_mig_pages_transferred();
 424         info->ram->normal = norm_mig_pages_transferred();
 425         info->ram->normal_bytes = norm_mig_bytes_transferred();
 426         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 427         info->ram->mbps = s->mbps;
 428         info->ram->dirty_sync_count = s->dirty_sync_count;
 429
 430         if (blk_mig_active()) {
 431             info->has_disk = true;
 432             info->disk = g_malloc0(sizeof(*info->disk));
 433             info->disk->transferred = blk_mig_bytes_transferred();
 434             info->disk->remaining = blk_mig_bytes_remaining();
 435             info->disk->total = blk_mig_bytes_total();
 436         }
 437
 438         get_xbzrle_cache_stats(info);
 439         break;
 440     case MIGRATION_STATUS_COMPLETED:
 441         get_xbzrle_cache_stats(info);
 442
 443         info->has_status = true;
 444         info->has_total_time = true;
 445         info->total_time = s->total_time;
 446         info->has_downtime = true;
 447         info->downtime = s->downtime;
 448         info->has_setup_time = true;
 449         info->setup_time = s->setup_time;
 450
 451         info->has_ram = true;
 452         info->ram = g_malloc0(sizeof(*info->ram));
 453         info->ram->transferred = ram_bytes_transferred();
 454         info->ram->remaining = 0;
 455         info->ram->total = ram_bytes_total();
 456         info->ram->duplicate = dup_mig_pages_transferred();
 457         info->ram->skipped = skipped_mig_pages_transferred();
 458         info->ram->normal = norm_mig_pages_transferred();
 459         info->ram->normal_bytes = norm_mig_bytes_transferred();
 460         info->ram->mbps = s->mbps;
 461         info->ram->dirty_sync_count = s->dirty_sync_count;
 462         break;
 463     case MIGRATION_STATUS_FAILED:
 464         info->has_status = true;
 465         break;
 466     case MIGRATION_STATUS_CANCELLED:
 467         info->has_status = true;
 468         break;
 469     }
 470     info->status = s->state;
 471
 472     return info;
 473 }
 474
 475 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 476                                   Error **errp)
 477 {
 478     MigrationState *s = migrate_get_current();
 479     MigrationCapabilityStatusList *cap;
 480
 481     if (s->state == MIGRATION_STATUS_ACTIVE ||
 482         s->state == MIGRATION_STATUS_SETUP) {
 483         error_setg(errp, QERR_MIGRATION_ACTIVE);
 484         return;
 485     }
 486
 487     for (cap = params; cap; cap = cap->next) {
 488         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 489     }
 490 }
 491
 492 void qmp_migrate_set_parameters(bool has_compress_level,
 493                                 int64_t compress_level,
 494                                 bool has_compress_threads,
 495                                 int64_t compress_threads,
 496                                 bool has_decompress_threads,
 497                                 int64_t decompress_threads, Error **errp)
 498 {
 499     MigrationState *s = migrate_get_current();
 500
 501     if (has_compress_level && (compress_level < 0 || compress_level > 9)) {
 502         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
 503                    "is invalid, it should be in the range of 0 to 9");
 504         return;
 505     }
 506     if (has_compress_threads &&
 507             (compress_threads < 1 || compress_threads > 255)) {
 508         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 509                    "compress_threads",
 510                    "is invalid, it should be in the range of 1 to 255");
 511         return;
 512     }
 513     if (has_decompress_threads &&
 514             (decompress_threads < 1 || decompress_threads > 255)) {
 515         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 516                    "decompress_threads",
 517                    "is invalid, it should be in the range of 1 to 255");
 518         return;
 519     }
 520
 521     if (has_compress_level) {
 522         s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 523     }
 524     if (has_compress_threads) {
 525         s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] = compress_threads;
 526     }
 527     if (has_decompress_threads) {
 528         s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 529                                                     decompress_threads;
 530     }
 531 }
 532
 533 /* shared migration helpers */
 534
 535 static void migrate_set_state(MigrationState *s, int old_state, int new_state)
 536 {
 537     if (atomic_cmpxchg(&s->state, old_state, new_state) == old_state) {
 538         trace_migrate_set_state(new_state);
 539         migrate_generate_event(new_state);
 540     }
 541 }
 542
 543 static void migrate_fd_cleanup(void *opaque)
 544 {
 545     MigrationState *s = opaque;
 546
 547     qemu_bh_delete(s->cleanup_bh);
 548     s->cleanup_bh = NULL;
 549
 550     if (s->file) {
 551         trace_migrate_fd_cleanup();
 552         qemu_mutex_unlock_iothread();
 553         qemu_thread_join(&s->thread);
 554         qemu_mutex_lock_iothread();
 555
 556         migrate_compress_threads_join();
 557         qemu_fclose(s->file);
 558         s->file = NULL;
 559     }
 560
 561     assert(s->state != MIGRATION_STATUS_ACTIVE);
 562
 563     if (s->state != MIGRATION_STATUS_COMPLETED) {
 564         qemu_savevm_state_cancel();
 565         if (s->state == MIGRATION_STATUS_CANCELLING) {
 566             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
 567                               MIGRATION_STATUS_CANCELLED);
 568         }
 569     }
 570
 571     notifier_list_notify(&migration_state_notifiers, s);
 572 }
 573
 574 void migrate_fd_error(MigrationState *s)
 575 {
 576     trace_migrate_fd_error();
 577     assert(s->file == NULL);
 578     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 579     notifier_list_notify(&migration_state_notifiers, s);
 580 }
 581
 582 static void migrate_fd_cancel(MigrationState *s)
 583 {
 584     int old_state ;
 585     QEMUFile *f = migrate_get_current()->file;
 586     trace_migrate_fd_cancel();
 587
 588     do {
 589         old_state = s->state;
 590         if (old_state != MIGRATION_STATUS_SETUP &&
 591             old_state != MIGRATION_STATUS_ACTIVE) {
 592             break;
 593         }
 594         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
 595     } while (s->state != MIGRATION_STATUS_CANCELLING);
 596
 597     /*
 598      * If we're unlucky the migration code might be stuck somewhere in a
 599      * send/write while the network has failed and is waiting to timeout;
 600      * if we've got shutdown(2) available then we can force it to quit.
 601      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
 602      * called in a bh, so there is no race against this cancel.
 603      */
 604     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
 605         qemu_file_shutdown(f);
 606     }
 607 }
 608
 609 void add_migration_state_change_notifier(Notifier *notify)
 610 {
 611     notifier_list_add(&migration_state_notifiers, notify);
 612 }
 613
 614 void remove_migration_state_change_notifier(Notifier *notify)
 615 {
 616     notifier_remove(notify);
 617 }
 618
 619 bool migration_in_setup(MigrationState *s)
 620 {
 621     return s->state == MIGRATION_STATUS_SETUP;
 622 }
 623
 624 bool migration_has_finished(MigrationState *s)
 625 {
 626     return s->state == MIGRATION_STATUS_COMPLETED;
 627 }
 628
 629 bool migration_has_failed(MigrationState *s)
 630 {
 631     return (s->state == MIGRATION_STATUS_CANCELLED ||
 632             s->state == MIGRATION_STATUS_FAILED);
 633 }
 634
 635 static MigrationState *migrate_init(const MigrationParams *params)
 636 {
 637     MigrationState *s = migrate_get_current();
 638     int64_t bandwidth_limit = s->bandwidth_limit;
 639     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 640     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 641     int compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 642     int compress_thread_count =
 643             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 644     int decompress_thread_count =
 645             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 646
 647     memcpy(enabled_capabilities, s->enabled_capabilities,
 648            sizeof(enabled_capabilities));
 649
 650     memset(s, 0, sizeof(*s));
 651     s->params = *params;
 652     memcpy(s->enabled_capabilities, enabled_capabilities,
 653            sizeof(enabled_capabilities));
 654     s->xbzrle_cache_size = xbzrle_cache_size;
 655
 656     s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 657     s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
 658                compress_thread_count;
 659     s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 660                decompress_thread_count;
 661     s->bandwidth_limit = bandwidth_limit;
 662     migrate_set_state(s, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
 663
 664     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 665     return s;
 666 }
 667
 668 static GSList *migration_blockers;
 669
 670 void migrate_add_blocker(Error *reason)
 671 {
 672     migration_blockers = g_slist_prepend(migration_blockers, reason);
 673 }
 674
 675 void migrate_del_blocker(Error *reason)
 676 {
 677     migration_blockers = g_slist_remove(migration_blockers, reason);
 678 }
 679
 680 void qmp_migrate_incoming(const char *uri, Error **errp)
 681 {
 682     Error *local_err = NULL;
 683     static bool once = true;
 684
 685     if (!deferred_incoming) {
 686         error_setg(errp, "For use with '-incoming defer'");
 687         return;
 688     }
 689     if (!once) {
 690         error_setg(errp, "The incoming migration has already been started");
 691     }
 692
 693     qemu_start_incoming_migration(uri, &local_err);
 694
 695     if (local_err) {
 696         error_propagate(errp, local_err);
 697         return;
 698     }
 699
 700     once = false;
 701 }
 702
 703 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 704                  bool has_inc, bool inc, bool has_detach, bool detach,
 705                  Error **errp)
 706 {
 707     Error *local_err = NULL;
 708     MigrationState *s = migrate_get_current();
 709     MigrationParams params;
 710     const char *p;
 711
 712     params.blk = has_blk && blk;
 713     params.shared = has_inc && inc;
 714
 715     if (s->state == MIGRATION_STATUS_ACTIVE ||
 716         s->state == MIGRATION_STATUS_SETUP ||
 717         s->state == MIGRATION_STATUS_CANCELLING) {
 718         error_setg(errp, QERR_MIGRATION_ACTIVE);
 719         return;
 720     }
 721     if (runstate_check(RUN_STATE_INMIGRATE)) {
 722         error_setg(errp, "Guest is waiting for an incoming migration");
 723         return;
 724     }
 725
 726     if (qemu_savevm_state_blocked(errp)) {
 727         return;
 728     }
 729
 730     if (migration_blockers) {
 731         *errp = error_copy(migration_blockers->data);
 732         return;
 733     }
 734
 735     /* We are starting a new migration, so we want to start in a clean
 736        state.  This change is only needed if previous migration
 737        failed/was cancelled.  We don't use migrate_set_state() because
 738        we are setting the initial state, not changing it. */
 739     s->state = MIGRATION_STATUS_NONE;
 740
 741     s = migrate_init(&params);
 742
 743     if (strstart(uri, "tcp:", &p)) {
 744         tcp_start_outgoing_migration(s, p, &local_err);
 745 #ifdef CONFIG_RDMA
 746     } else if (strstart(uri, "rdma:", &p)) {
 747         rdma_start_outgoing_migration(s, p, &local_err);
 748 #endif
 749 #if !defined(WIN32)
 750     } else if (strstart(uri, "exec:", &p)) {
 751         exec_start_outgoing_migration(s, p, &local_err);
 752     } else if (strstart(uri, "unix:", &p)) {
 753         unix_start_outgoing_migration(s, p, &local_err);
 754     } else if (strstart(uri, "fd:", &p)) {
 755         fd_start_outgoing_migration(s, p, &local_err);
 756 #endif
 757     } else {
 758         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
 759                    "a valid migration protocol");
 760         migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 761         return;
 762     }
 763
 764     if (local_err) {
 765         migrate_fd_error(s);
 766         error_propagate(errp, local_err);
 767         return;
 768     }
 769 }
 770
 771 void qmp_migrate_cancel(Error **errp)
 772 {
 773     migrate_fd_cancel(migrate_get_current());
 774 }
 775
 776 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 777 {
 778     MigrationState *s = migrate_get_current();
 779     int64_t new_size;
 780
 781     /* Check for truncation */
 782     if (value != (size_t)value) {
 783         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 784                    "exceeding address space");
 785         return;
 786     }
 787
 788     /* Cache should not be larger than guest ram size */
 789     if (value > ram_bytes_total()) {
 790         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 791                    "exceeds guest ram size ");
 792         return;
 793     }
 794
 795     new_size = xbzrle_cache_resize(value);
 796     if (new_size < 0) {
 797         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 798                    "is smaller than page size");
 799         return;
 800     }
 801
 802     s->xbzrle_cache_size = new_size;
 803 }
 804
 805 int64_t qmp_query_migrate_cache_size(Error **errp)
 806 {
 807     return migrate_xbzrle_cache_size();
 808 }
 809
 810 void qmp_migrate_set_speed(int64_t value, Error **errp)
 811 {
 812     MigrationState *s;
 813
 814     if (value < 0) {
 815         value = 0;
 816     }
 817     if (value > SIZE_MAX) {
 818         value = SIZE_MAX;
 819     }
 820
 821     s = migrate_get_current();
 822     s->bandwidth_limit = value;
 823     if (s->file) {
 824         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
 825     }
 826 }
 827
 828 void qmp_migrate_set_downtime(double value, Error **errp)
 829 {
 830     value *= 1e9;
 831     value = MAX(0, MIN(UINT64_MAX, value));
 832     max_downtime = (uint64_t)value;
 833 }
 834
 835 bool migrate_auto_converge(void)
 836 {
 837     MigrationState *s;
 838
 839     s = migrate_get_current();
 840
 841     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
 842 }
 843
 844 bool migrate_zero_blocks(void)
 845 {
 846     MigrationState *s;
 847
 848     s = migrate_get_current();
 849
 850     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
 851 }
 852
 853 bool migrate_use_compression(void)
 854 {
 855     MigrationState *s;
 856
 857     s = migrate_get_current();
 858
 859     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
 860 }
 861
 862 int migrate_compress_level(void)
 863 {
 864     MigrationState *s;
 865
 866     s = migrate_get_current();
 867
 868     return s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 869 }
 870
 871 int migrate_compress_threads(void)
 872 {
 873     MigrationState *s;
 874
 875     s = migrate_get_current();
 876
 877     return s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 878 }
 879
 880 int migrate_decompress_threads(void)
 881 {
 882     MigrationState *s;
 883
 884     s = migrate_get_current();
 885
 886     return s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 887 }
 888
 889 bool migrate_use_events(void)
 890 {
 891     MigrationState *s;
 892
 893     s = migrate_get_current();
 894
 895     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
 896 }
 897
 898 int migrate_use_xbzrle(void)
 899 {
 900     MigrationState *s;
 901
 902     s = migrate_get_current();
 903
 904     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 905 }
 906
 907 int64_t migrate_xbzrle_cache_size(void)
 908 {
 909     MigrationState *s;
 910
 911     s = migrate_get_current();
 912
 913     return s->xbzrle_cache_size;
 914 }
 915
 916 /* migration thread support */
 917
 918 static void *migration_thread(void *opaque)
 919 {
 920     MigrationState *s = opaque;
 921     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 922     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 923     int64_t initial_bytes = 0;
 924     int64_t max_size = 0;
 925     int64_t start_time = initial_time;
 926     bool old_vm_running = false;
 927
 928     rcu_register_thread();
 929
 930     qemu_savevm_state_header(s->file);
 931     qemu_savevm_state_begin(s->file, &s->params);
 932
 933     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
 934     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
 935
 936     while (s->state == MIGRATION_STATUS_ACTIVE) {
 937         int64_t current_time;
 938         uint64_t pending_size;
 939
 940         if (!qemu_file_rate_limit(s->file)) {
 941             pending_size = qemu_savevm_state_pending(s->file, max_size);
 942             trace_migrate_pending(pending_size, max_size);
 943             if (pending_size && pending_size >= max_size) {
 944                 qemu_savevm_state_iterate(s->file);
 945             } else {
 946                 int ret;
 947
 948                 qemu_mutex_lock_iothread();
 949                 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 950                 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 951                 old_vm_running = runstate_is_running();
 952
 953                 ret = global_state_store();
 954                 if (!ret) {
 955                     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 956                     if (ret >= 0) {
 957                         qemu_file_set_rate_limit(s->file, INT64_MAX);
 958                         qemu_savevm_state_complete(s->file);
 959                     }
 960                 }
 961                 qemu_mutex_unlock_iothread();
 962
 963                 if (ret < 0) {
 964                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 965                                       MIGRATION_STATUS_FAILED);
 966                     break;
 967                 }
 968
 969                 if (!qemu_file_get_error(s->file)) {
 970                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 971                                       MIGRATION_STATUS_COMPLETED);
 972                     break;
 973                 }
 974             }
 975         }
 976
 977         if (qemu_file_get_error(s->file)) {
 978             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 979                               MIGRATION_STATUS_FAILED);
 980             break;
 981         }
 982         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 983         if (current_time >= initial_time + BUFFER_DELAY) {
 984             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
 985             uint64_t time_spent = current_time - initial_time;
 986             double bandwidth = transferred_bytes / time_spent;
 987             max_size = bandwidth * migrate_max_downtime() / 1000000;
 988
 989             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
 990                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
 991
 992             trace_migrate_transferred(transferred_bytes, time_spent,
 993                                       bandwidth, max_size);
 994             /* if we haven't sent anything, we don't want to recalculate
 995                10000 is a small enough number for our purposes */
 996             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
 997                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
 998             }
 999
1000             qemu_file_reset_rate_limit(s->file);
1001             initial_time = current_time;
1002             initial_bytes = qemu_ftell(s->file);
1003         }
1004         if (qemu_file_rate_limit(s->file)) {
1005             /* usleep expects microseconds */
1006             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
1007         }
1008     }
1009
1010     qemu_mutex_lock_iothread();
1011     if (s->state == MIGRATION_STATUS_COMPLETED) {
1012         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1013         uint64_t transferred_bytes = qemu_ftell(s->file);
1014         s->total_time = end_time - s->total_time;
1015         s->downtime = end_time - start_time;
1016         if (s->total_time) {
1017             s->mbps = (((double) transferred_bytes * 8.0) /
1018                        ((double) s->total_time)) / 1000;
1019         }
1020         runstate_set(RUN_STATE_POSTMIGRATE);
1021     } else {
1022         if (old_vm_running) {
1023             vm_start();
1024         }
1025     }
1026     qemu_bh_schedule(s->cleanup_bh);
1027     qemu_mutex_unlock_iothread();
1028
1029     rcu_unregister_thread();
1030     return NULL;
1031 }
1032
1033 void migrate_fd_connect(MigrationState *s)
1034 {
1035     /* This is a best 1st approximation. ns to ms */
1036     s->expected_downtime = max_downtime/1000000;
1037     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
1038
1039     qemu_file_set_rate_limit(s->file,
1040                              s->bandwidth_limit / XFER_LIMIT_RATIO);
1041
1042     /* Notify before starting migration thread */
1043     notifier_list_notify(&migration_state_notifiers, s);
1044
1045     migrate_compress_threads_create();
1046     qemu_thread_create(&s->thread, "migration", migration_thread, s,
1047                        QEMU_THREAD_JOINABLE);
1048 }