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     RunState state;
 108     bool received;
 109 } GlobalState;
 110
 111 static GlobalState global_state;
 112
 113 static int global_state_store(void)
 114 {
 115     if (!runstate_store((char *)global_state.runstate,
 116                         sizeof(global_state.runstate))) {
 117         error_report("runstate name too big: %s", global_state.runstate);
 118         trace_migrate_state_too_big();
 119         return -EINVAL;
 120     }
 121     return 0;
 122 }
 123
 124 static bool global_state_received(void)
 125 {
 126     return global_state.received;
 127 }
 128
 129 static RunState global_state_get_runstate(void)
 130 {
 131     return global_state.state;
 132 }
 133
 134 void global_state_set_optional(void)
 135 {
 136     global_state.optional = true;
 137 }
 138
 139 static bool global_state_needed(void *opaque)
 140 {
 141     GlobalState *s = opaque;
 142     char *runstate = (char *)s->runstate;
 143
 144     /* If it is not optional, it is mandatory */
 145
 146     if (s->optional == false) {
 147         return true;
 148     }
 149
 150     /* If state is running or paused, it is not needed */
 151
 152     if (strcmp(runstate, "running") == 0 ||
 153         strcmp(runstate, "paused") == 0) {
 154         return false;
 155     }
 156
 157     /* for any other state it is needed */
 158     return true;
 159 }
 160
 161 static int global_state_post_load(void *opaque, int version_id)
 162 {
 163     GlobalState *s = opaque;
 164     Error *local_err = NULL;
 165     int r;
 166     char *runstate = (char *)s->runstate;
 167
 168     s->received = true;
 169     trace_migrate_global_state_post_load(runstate);
 170
 171     r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE_MAX,
 172                                 -1, &local_err);
 173
 174     if (r == -1) {
 175         if (local_err) {
 176             error_report_err(local_err);
 177         }
 178         return -EINVAL;
 179     }
 180     s->state = r;
 181
 182     return 0;
 183 }
 184
 185 static void global_state_pre_save(void *opaque)
 186 {
 187     GlobalState *s = opaque;
 188
 189     trace_migrate_global_state_pre_save((char *)s->runstate);
 190     s->size = strlen((char *)s->runstate) + 1;
 191 }
 192
 193 static const VMStateDescription vmstate_globalstate = {
 194     .name = "globalstate",
 195     .version_id = 1,
 196     .minimum_version_id = 1,
 197     .post_load = global_state_post_load,
 198     .pre_save = global_state_pre_save,
 199     .needed = global_state_needed,
 200     .fields = (VMStateField[]) {
 201         VMSTATE_UINT32(size, GlobalState),
 202         VMSTATE_BUFFER(runstate, GlobalState),
 203         VMSTATE_END_OF_LIST()
 204     },
 205 };
 206
 207 void register_global_state(void)
 208 {
 209     /* We would use it independently that we receive it */
 210     strcpy((char *)&global_state.runstate, "");
 211     global_state.received = false;
 212     vmstate_register(NULL, 0, &vmstate_globalstate, &global_state);
 213 }
 214
 215 static void migrate_generate_event(int new_state)
 216 {
 217     if (migrate_use_events()) {
 218         qapi_event_send_migration(new_state, &error_abort);
 219         trace_migrate_set_state(new_state);
 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         migrate_generate_event(new_state);
 532     }
 533 }
 534
 535 static void migrate_fd_cleanup(void *opaque)
 536 {
 537     MigrationState *s = opaque;
 538
 539     qemu_bh_delete(s->cleanup_bh);
 540     s->cleanup_bh = NULL;
 541
 542     if (s->file) {
 543         trace_migrate_fd_cleanup();
 544         qemu_mutex_unlock_iothread();
 545         qemu_thread_join(&s->thread);
 546         qemu_mutex_lock_iothread();
 547
 548         migrate_compress_threads_join();
 549         qemu_fclose(s->file);
 550         s->file = NULL;
 551     }
 552
 553     assert(s->state != MIGRATION_STATUS_ACTIVE);
 554
 555     if (s->state != MIGRATION_STATUS_COMPLETED) {
 556         qemu_savevm_state_cancel();
 557         if (s->state == MIGRATION_STATUS_CANCELLING) {
 558             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
 559                               MIGRATION_STATUS_CANCELLED);
 560         }
 561     }
 562
 563     notifier_list_notify(&migration_state_notifiers, s);
 564 }
 565
 566 void migrate_fd_error(MigrationState *s)
 567 {
 568     trace_migrate_fd_error();
 569     assert(s->file == NULL);
 570     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 571     notifier_list_notify(&migration_state_notifiers, s);
 572 }
 573
 574 static void migrate_fd_cancel(MigrationState *s)
 575 {
 576     int old_state ;
 577     QEMUFile *f = migrate_get_current()->file;
 578     trace_migrate_fd_cancel();
 579
 580     do {
 581         old_state = s->state;
 582         if (old_state != MIGRATION_STATUS_SETUP &&
 583             old_state != MIGRATION_STATUS_ACTIVE) {
 584             break;
 585         }
 586         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
 587     } while (s->state != MIGRATION_STATUS_CANCELLING);
 588
 589     /*
 590      * If we're unlucky the migration code might be stuck somewhere in a
 591      * send/write while the network has failed and is waiting to timeout;
 592      * if we've got shutdown(2) available then we can force it to quit.
 593      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
 594      * called in a bh, so there is no race against this cancel.
 595      */
 596     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
 597         qemu_file_shutdown(f);
 598     }
 599 }
 600
 601 void add_migration_state_change_notifier(Notifier *notify)
 602 {
 603     notifier_list_add(&migration_state_notifiers, notify);
 604 }
 605
 606 void remove_migration_state_change_notifier(Notifier *notify)
 607 {
 608     notifier_remove(notify);
 609 }
 610
 611 bool migration_in_setup(MigrationState *s)
 612 {
 613     return s->state == MIGRATION_STATUS_SETUP;
 614 }
 615
 616 bool migration_has_finished(MigrationState *s)
 617 {
 618     return s->state == MIGRATION_STATUS_COMPLETED;
 619 }
 620
 621 bool migration_has_failed(MigrationState *s)
 622 {
 623     return (s->state == MIGRATION_STATUS_CANCELLED ||
 624             s->state == MIGRATION_STATUS_FAILED);
 625 }
 626
 627 static MigrationState *migrate_init(const MigrationParams *params)
 628 {
 629     MigrationState *s = migrate_get_current();
 630     int64_t bandwidth_limit = s->bandwidth_limit;
 631     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 632     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 633     int compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 634     int compress_thread_count =
 635             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 636     int decompress_thread_count =
 637             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 638
 639     memcpy(enabled_capabilities, s->enabled_capabilities,
 640            sizeof(enabled_capabilities));
 641
 642     memset(s, 0, sizeof(*s));
 643     s->params = *params;
 644     memcpy(s->enabled_capabilities, enabled_capabilities,
 645            sizeof(enabled_capabilities));
 646     s->xbzrle_cache_size = xbzrle_cache_size;
 647
 648     s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 649     s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
 650                compress_thread_count;
 651     s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 652                decompress_thread_count;
 653     s->bandwidth_limit = bandwidth_limit;
 654     migrate_set_state(s, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
 655
 656     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 657     return s;
 658 }
 659
 660 static GSList *migration_blockers;
 661
 662 void migrate_add_blocker(Error *reason)
 663 {
 664     migration_blockers = g_slist_prepend(migration_blockers, reason);
 665 }
 666
 667 void migrate_del_blocker(Error *reason)
 668 {
 669     migration_blockers = g_slist_remove(migration_blockers, reason);
 670 }
 671
 672 void qmp_migrate_incoming(const char *uri, Error **errp)
 673 {
 674     Error *local_err = NULL;
 675     static bool once = true;
 676
 677     if (!deferred_incoming) {
 678         error_setg(errp, "For use with '-incoming defer'");
 679         return;
 680     }
 681     if (!once) {
 682         error_setg(errp, "The incoming migration has already been started");
 683     }
 684
 685     qemu_start_incoming_migration(uri, &local_err);
 686
 687     if (local_err) {
 688         error_propagate(errp, local_err);
 689         return;
 690     }
 691
 692     once = false;
 693 }
 694
 695 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 696                  bool has_inc, bool inc, bool has_detach, bool detach,
 697                  Error **errp)
 698 {
 699     Error *local_err = NULL;
 700     MigrationState *s = migrate_get_current();
 701     MigrationParams params;
 702     const char *p;
 703
 704     params.blk = has_blk && blk;
 705     params.shared = has_inc && inc;
 706
 707     if (s->state == MIGRATION_STATUS_ACTIVE ||
 708         s->state == MIGRATION_STATUS_SETUP ||
 709         s->state == MIGRATION_STATUS_CANCELLING) {
 710         error_setg(errp, QERR_MIGRATION_ACTIVE);
 711         return;
 712     }
 713     if (runstate_check(RUN_STATE_INMIGRATE)) {
 714         error_setg(errp, "Guest is waiting for an incoming migration");
 715         return;
 716     }
 717
 718     if (qemu_savevm_state_blocked(errp)) {
 719         return;
 720     }
 721
 722     if (migration_blockers) {
 723         *errp = error_copy(migration_blockers->data);
 724         return;
 725     }
 726
 727     /* We are starting a new migration, so we want to start in a clean
 728        state.  This change is only needed if previous migration
 729        failed/was cancelled.  We don't use migrate_set_state() because
 730        we are setting the initial state, not changing it. */
 731     s->state = MIGRATION_STATUS_NONE;
 732
 733     s = migrate_init(&params);
 734
 735     if (strstart(uri, "tcp:", &p)) {
 736         tcp_start_outgoing_migration(s, p, &local_err);
 737 #ifdef CONFIG_RDMA
 738     } else if (strstart(uri, "rdma:", &p)) {
 739         rdma_start_outgoing_migration(s, p, &local_err);
 740 #endif
 741 #if !defined(WIN32)
 742     } else if (strstart(uri, "exec:", &p)) {
 743         exec_start_outgoing_migration(s, p, &local_err);
 744     } else if (strstart(uri, "unix:", &p)) {
 745         unix_start_outgoing_migration(s, p, &local_err);
 746     } else if (strstart(uri, "fd:", &p)) {
 747         fd_start_outgoing_migration(s, p, &local_err);
 748 #endif
 749     } else {
 750         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
 751                    "a valid migration protocol");
 752         migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 753         return;
 754     }
 755
 756     if (local_err) {
 757         migrate_fd_error(s);
 758         error_propagate(errp, local_err);
 759         return;
 760     }
 761 }
 762
 763 void qmp_migrate_cancel(Error **errp)
 764 {
 765     migrate_fd_cancel(migrate_get_current());
 766 }
 767
 768 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 769 {
 770     MigrationState *s = migrate_get_current();
 771     int64_t new_size;
 772
 773     /* Check for truncation */
 774     if (value != (size_t)value) {
 775         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 776                    "exceeding address space");
 777         return;
 778     }
 779
 780     /* Cache should not be larger than guest ram size */
 781     if (value > ram_bytes_total()) {
 782         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 783                    "exceeds guest ram size ");
 784         return;
 785     }
 786
 787     new_size = xbzrle_cache_resize(value);
 788     if (new_size < 0) {
 789         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 790                    "is smaller than page size");
 791         return;
 792     }
 793
 794     s->xbzrle_cache_size = new_size;
 795 }
 796
 797 int64_t qmp_query_migrate_cache_size(Error **errp)
 798 {
 799     return migrate_xbzrle_cache_size();
 800 }
 801
 802 void qmp_migrate_set_speed(int64_t value, Error **errp)
 803 {
 804     MigrationState *s;
 805
 806     if (value < 0) {
 807         value = 0;
 808     }
 809     if (value > SIZE_MAX) {
 810         value = SIZE_MAX;
 811     }
 812
 813     s = migrate_get_current();
 814     s->bandwidth_limit = value;
 815     if (s->file) {
 816         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
 817     }
 818 }
 819
 820 void qmp_migrate_set_downtime(double value, Error **errp)
 821 {
 822     value *= 1e9;
 823     value = MAX(0, MIN(UINT64_MAX, value));
 824     max_downtime = (uint64_t)value;
 825 }
 826
 827 bool migrate_auto_converge(void)
 828 {
 829     MigrationState *s;
 830
 831     s = migrate_get_current();
 832
 833     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
 834 }
 835
 836 bool migrate_zero_blocks(void)
 837 {
 838     MigrationState *s;
 839
 840     s = migrate_get_current();
 841
 842     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
 843 }
 844
 845 bool migrate_use_compression(void)
 846 {
 847     MigrationState *s;
 848
 849     s = migrate_get_current();
 850
 851     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
 852 }
 853
 854 int migrate_compress_level(void)
 855 {
 856     MigrationState *s;
 857
 858     s = migrate_get_current();
 859
 860     return s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 861 }
 862
 863 int migrate_compress_threads(void)
 864 {
 865     MigrationState *s;
 866
 867     s = migrate_get_current();
 868
 869     return s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 870 }
 871
 872 int migrate_decompress_threads(void)
 873 {
 874     MigrationState *s;
 875
 876     s = migrate_get_current();
 877
 878     return s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 879 }
 880
 881 bool migrate_use_events(void)
 882 {
 883     MigrationState *s;
 884
 885     s = migrate_get_current();
 886
 887     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
 888 }
 889
 890 int migrate_use_xbzrle(void)
 891 {
 892     MigrationState *s;
 893
 894     s = migrate_get_current();
 895
 896     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 897 }
 898
 899 int64_t migrate_xbzrle_cache_size(void)
 900 {
 901     MigrationState *s;
 902
 903     s = migrate_get_current();
 904
 905     return s->xbzrle_cache_size;
 906 }
 907
 908 /* migration thread support */
 909
 910 static void *migration_thread(void *opaque)
 911 {
 912     MigrationState *s = opaque;
 913     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 914     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 915     int64_t initial_bytes = 0;
 916     int64_t max_size = 0;
 917     int64_t start_time = initial_time;
 918     bool old_vm_running = false;
 919
 920     qemu_savevm_state_header(s->file);
 921     qemu_savevm_state_begin(s->file, &s->params);
 922
 923     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
 924     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
 925
 926     while (s->state == MIGRATION_STATUS_ACTIVE) {
 927         int64_t current_time;
 928         uint64_t pending_size;
 929
 930         if (!qemu_file_rate_limit(s->file)) {
 931             pending_size = qemu_savevm_state_pending(s->file, max_size);
 932             trace_migrate_pending(pending_size, max_size);
 933             if (pending_size && pending_size >= max_size) {
 934                 qemu_savevm_state_iterate(s->file);
 935             } else {
 936                 int ret;
 937
 938                 qemu_mutex_lock_iothread();
 939                 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 940                 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 941                 old_vm_running = runstate_is_running();
 942
 943                 ret = global_state_store();
 944                 if (!ret) {
 945                     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 946                     if (ret >= 0) {
 947                         qemu_file_set_rate_limit(s->file, INT64_MAX);
 948                         qemu_savevm_state_complete(s->file);
 949                     }
 950                 }
 951                 qemu_mutex_unlock_iothread();
 952
 953                 if (ret < 0) {
 954                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 955                                       MIGRATION_STATUS_FAILED);
 956                     break;
 957                 }
 958
 959                 if (!qemu_file_get_error(s->file)) {
 960                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 961                                       MIGRATION_STATUS_COMPLETED);
 962                     break;
 963                 }
 964             }
 965         }
 966
 967         if (qemu_file_get_error(s->file)) {
 968             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
 969                               MIGRATION_STATUS_FAILED);
 970             break;
 971         }
 972         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 973         if (current_time >= initial_time + BUFFER_DELAY) {
 974             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
 975             uint64_t time_spent = current_time - initial_time;
 976             double bandwidth = transferred_bytes / time_spent;
 977             max_size = bandwidth * migrate_max_downtime() / 1000000;
 978
 979             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
 980                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
 981
 982             trace_migrate_transferred(transferred_bytes, time_spent,
 983                                       bandwidth, max_size);
 984             /* if we haven't sent anything, we don't want to recalculate
 985                10000 is a small enough number for our purposes */
 986             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
 987                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
 988             }
 989
 990             qemu_file_reset_rate_limit(s->file);
 991             initial_time = current_time;
 992             initial_bytes = qemu_ftell(s->file);
 993         }
 994         if (qemu_file_rate_limit(s->file)) {
 995             /* usleep expects microseconds */
 996             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
 997         }
 998     }
 999
1000     qemu_mutex_lock_iothread();
1001     if (s->state == MIGRATION_STATUS_COMPLETED) {
1002         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1003         uint64_t transferred_bytes = qemu_ftell(s->file);
1004         s->total_time = end_time - s->total_time;
1005         s->downtime = end_time - start_time;
1006         if (s->total_time) {
1007             s->mbps = (((double) transferred_bytes * 8.0) /
1008                        ((double) s->total_time)) / 1000;
1009         }
1010         runstate_set(RUN_STATE_POSTMIGRATE);
1011     } else {
1012         if (old_vm_running) {
1013             vm_start();
1014         }
1015     }
1016     qemu_bh_schedule(s->cleanup_bh);
1017     qemu_mutex_unlock_iothread();
1018
1019     return NULL;
1020 }
1021
1022 void migrate_fd_connect(MigrationState *s)
1023 {
1024     /* This is a best 1st approximation. ns to ms */
1025     s->expected_downtime = max_downtime/1000000;
1026     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
1027
1028     qemu_file_set_rate_limit(s->file,
1029                              s->bandwidth_limit / XFER_LIMIT_RATIO);
1030
1031     /* Notify before starting migration thread */
1032     notifier_list_notify(&migration_state_notifiers, s);
1033
1034     migrate_compress_threads_create();
1035     qemu_thread_create(&s->thread, "migration", migration_thread, s,
1036                        QEMU_THREAD_JOINABLE);
1037 }