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 #include "qom/cpu.h"
  33
  34 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
  35
  36 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
  37  * data. */
  38 #define BUFFER_DELAY     100
  39 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
  40
  41 /* Default compression thread count */
  42 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
  43 /* Default decompression thread count, usually decompression is at
  44  * least 4 times as fast as compression.*/
  45 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
  46 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
  47 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
  48 /* Define default autoconverge cpu throttle migration parameters */
  49 #define DEFAULT_MIGRATE_X_CPU_THROTTLE_INITIAL 20
  50 #define DEFAULT_MIGRATE_X_CPU_THROTTLE_INCREMENT 10
  51
  52 /* Migration XBZRLE default cache size */
  53 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
  54
  55 static NotifierList migration_state_notifiers =
  56     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  57
  58 static bool deferred_incoming;
  59
  60 /* When we add fault tolerance, we could have several
  61    migrations at once.  For now we don't need to add
  62    dynamic creation of migration */
  63
  64 /* For outgoing */
  65 MigrationState *migrate_get_current(void)
  66 {
  67     static MigrationState current_migration = {
  68         .state = MIGRATION_STATUS_NONE,
  69         .bandwidth_limit = MAX_THROTTLE,
  70         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
  71         .mbps = -1,
  72         .parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] =
  73                 DEFAULT_MIGRATE_COMPRESS_LEVEL,
  74         .parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
  75                 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
  76         .parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
  77                 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
  78         .parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL] =
  79                 DEFAULT_MIGRATE_X_CPU_THROTTLE_INITIAL,
  80         .parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT] =
  81                 DEFAULT_MIGRATE_X_CPU_THROTTLE_INCREMENT,
  82     };
  83
  84     return &current_migration;
  85 }
  86
  87 /* For incoming */
  88 static MigrationIncomingState *mis_current;
  89
  90 MigrationIncomingState *migration_incoming_get_current(void)
  91 {
  92     return mis_current;
  93 }
  94
  95 MigrationIncomingState *migration_incoming_state_new(QEMUFile* f)
  96 {
  97     mis_current = g_new0(MigrationIncomingState, 1);
  98     mis_current->file = f;
  99     QLIST_INIT(&mis_current->loadvm_handlers);
 100
 101     return mis_current;
 102 }
 103
 104 void migration_incoming_state_destroy(void)
 105 {
 106     loadvm_free_handlers(mis_current);
 107     g_free(mis_current);
 108     mis_current = NULL;
 109 }
 110
 111
 112 typedef struct {
 113     bool optional;
 114     uint32_t size;
 115     uint8_t runstate[100];
 116     RunState state;
 117     bool received;
 118 } GlobalState;
 119
 120 static GlobalState global_state;
 121
 122 int global_state_store(void)
 123 {
 124     if (!runstate_store((char *)global_state.runstate,
 125                         sizeof(global_state.runstate))) {
 126         error_report("runstate name too big: %s", global_state.runstate);
 127         trace_migrate_state_too_big();
 128         return -EINVAL;
 129     }
 130     return 0;
 131 }
 132
 133 void global_state_store_running(void)
 134 {
 135     const char *state = RunState_lookup[RUN_STATE_RUNNING];
 136     strncpy((char *)global_state.runstate,
 137            state, sizeof(global_state.runstate));
 138 }
 139
 140 static bool global_state_received(void)
 141 {
 142     return global_state.received;
 143 }
 144
 145 static RunState global_state_get_runstate(void)
 146 {
 147     return global_state.state;
 148 }
 149
 150 void global_state_set_optional(void)
 151 {
 152     global_state.optional = true;
 153 }
 154
 155 static bool global_state_needed(void *opaque)
 156 {
 157     GlobalState *s = opaque;
 158     char *runstate = (char *)s->runstate;
 159
 160     /* If it is not optional, it is mandatory */
 161
 162     if (s->optional == false) {
 163         return true;
 164     }
 165
 166     /* If state is running or paused, it is not needed */
 167
 168     if (strcmp(runstate, "running") == 0 ||
 169         strcmp(runstate, "paused") == 0) {
 170         return false;
 171     }
 172
 173     /* for any other state it is needed */
 174     return true;
 175 }
 176
 177 static int global_state_post_load(void *opaque, int version_id)
 178 {
 179     GlobalState *s = opaque;
 180     Error *local_err = NULL;
 181     int r;
 182     char *runstate = (char *)s->runstate;
 183
 184     s->received = true;
 185     trace_migrate_global_state_post_load(runstate);
 186
 187     r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE_MAX,
 188                                 -1, &local_err);
 189
 190     if (r == -1) {
 191         if (local_err) {
 192             error_report_err(local_err);
 193         }
 194         return -EINVAL;
 195     }
 196     s->state = r;
 197
 198     return 0;
 199 }
 200
 201 static void global_state_pre_save(void *opaque)
 202 {
 203     GlobalState *s = opaque;
 204
 205     trace_migrate_global_state_pre_save((char *)s->runstate);
 206     s->size = strlen((char *)s->runstate) + 1;
 207 }
 208
 209 static const VMStateDescription vmstate_globalstate = {
 210     .name = "globalstate",
 211     .version_id = 1,
 212     .minimum_version_id = 1,
 213     .post_load = global_state_post_load,
 214     .pre_save = global_state_pre_save,
 215     .needed = global_state_needed,
 216     .fields = (VMStateField[]) {
 217         VMSTATE_UINT32(size, GlobalState),
 218         VMSTATE_BUFFER(runstate, GlobalState),
 219         VMSTATE_END_OF_LIST()
 220     },
 221 };
 222
 223 void register_global_state(void)
 224 {
 225     /* We would use it independently that we receive it */
 226     strcpy((char *)&global_state.runstate, "");
 227     global_state.received = false;
 228     vmstate_register(NULL, 0, &vmstate_globalstate, &global_state);
 229 }
 230
 231 static void migrate_generate_event(int new_state)
 232 {
 233     if (migrate_use_events()) {
 234         qapi_event_send_migration(new_state, &error_abort);
 235     }
 236 }
 237
 238 /*
 239  * Called on -incoming with a defer: uri.
 240  * The migration can be started later after any parameters have been
 241  * changed.
 242  */
 243 static void deferred_incoming_migration(Error **errp)
 244 {
 245     if (deferred_incoming) {
 246         error_setg(errp, "Incoming migration already deferred");
 247     }
 248     deferred_incoming = true;
 249 }
 250
 251 void qemu_start_incoming_migration(const char *uri, Error **errp)
 252 {
 253     const char *p;
 254
 255     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
 256     if (!strcmp(uri, "defer")) {
 257         deferred_incoming_migration(errp);
 258     } else if (strstart(uri, "tcp:", &p)) {
 259         tcp_start_incoming_migration(p, errp);
 260 #ifdef CONFIG_RDMA
 261     } else if (strstart(uri, "rdma:", &p)) {
 262         rdma_start_incoming_migration(p, errp);
 263 #endif
 264 #if !defined(WIN32)
 265     } else if (strstart(uri, "exec:", &p)) {
 266         exec_start_incoming_migration(p, errp);
 267     } else if (strstart(uri, "unix:", &p)) {
 268         unix_start_incoming_migration(p, errp);
 269     } else if (strstart(uri, "fd:", &p)) {
 270         fd_start_incoming_migration(p, errp);
 271 #endif
 272     } else {
 273         error_setg(errp, "unknown migration protocol: %s", uri);
 274     }
 275 }
 276
 277 static void process_incoming_migration_co(void *opaque)
 278 {
 279     QEMUFile *f = opaque;
 280     Error *local_err = NULL;
 281     int ret;
 282
 283     migration_incoming_state_new(f);
 284     migrate_generate_event(MIGRATION_STATUS_ACTIVE);
 285     ret = qemu_loadvm_state(f);
 286
 287     qemu_fclose(f);
 288     free_xbzrle_decoded_buf();
 289     migration_incoming_state_destroy();
 290
 291     if (ret < 0) {
 292         migrate_generate_event(MIGRATION_STATUS_FAILED);
 293         error_report("load of migration failed: %s", strerror(-ret));
 294         migrate_decompress_threads_join();
 295         exit(EXIT_FAILURE);
 296     }
 297     migrate_generate_event(MIGRATION_STATUS_COMPLETED);
 298     qemu_announce_self();
 299
 300     /* Make sure all file formats flush their mutable metadata */
 301     bdrv_invalidate_cache_all(&local_err);
 302     if (local_err) {
 303         error_report_err(local_err);
 304         migrate_decompress_threads_join();
 305         exit(EXIT_FAILURE);
 306     }
 307
 308     /* If global state section was not received or we are in running
 309        state, we need to obey autostart. Any other state is set with
 310        runstate_set. */
 311
 312     if (!global_state_received() ||
 313         global_state_get_runstate() == RUN_STATE_RUNNING) {
 314         if (autostart) {
 315             vm_start();
 316         } else {
 317             runstate_set(RUN_STATE_PAUSED);
 318         }
 319     } else {
 320         runstate_set(global_state_get_runstate());
 321     }
 322     migrate_decompress_threads_join();
 323 }
 324
 325 void process_incoming_migration(QEMUFile *f)
 326 {
 327     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
 328     int fd = qemu_get_fd(f);
 329
 330     assert(fd != -1);
 331     migrate_decompress_threads_create();
 332     qemu_set_nonblock(fd);
 333     qemu_coroutine_enter(co, f);
 334 }
 335
 336 /* amount of nanoseconds we are willing to wait for migration to be down.
 337  * the choice of nanoseconds is because it is the maximum resolution that
 338  * get_clock() can achieve. It is an internal measure. All user-visible
 339  * units must be in seconds */
 340 static uint64_t max_downtime = 300000000;
 341
 342 uint64_t migrate_max_downtime(void)
 343 {
 344     return max_downtime;
 345 }
 346
 347 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 348 {
 349     MigrationCapabilityStatusList *head = NULL;
 350     MigrationCapabilityStatusList *caps;
 351     MigrationState *s = migrate_get_current();
 352     int i;
 353
 354     caps = NULL; /* silence compiler warning */
 355     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 356         if (head == NULL) {
 357             head = g_malloc0(sizeof(*caps));
 358             caps = head;
 359         } else {
 360             caps->next = g_malloc0(sizeof(*caps));
 361             caps = caps->next;
 362         }
 363         caps->value =
 364             g_malloc(sizeof(*caps->value));
 365         caps->value->capability = i;
 366         caps->value->state = s->enabled_capabilities[i];
 367     }
 368
 369     return head;
 370 }
 371
 372 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 373 {
 374     MigrationParameters *params;
 375     MigrationState *s = migrate_get_current();
 376
 377     params = g_malloc0(sizeof(*params));
 378     params->compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 379     params->compress_threads =
 380             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 381     params->decompress_threads =
 382             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 383     params->x_cpu_throttle_initial =
 384             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL];
 385     params->x_cpu_throttle_increment =
 386             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT];
 387
 388     return params;
 389 }
 390
 391 static void get_xbzrle_cache_stats(MigrationInfo *info)
 392 {
 393     if (migrate_use_xbzrle()) {
 394         info->has_xbzrle_cache = true;
 395         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 396         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 397         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 398         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 399         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 400         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
 401         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 402     }
 403 }
 404
 405 MigrationInfo *qmp_query_migrate(Error **errp)
 406 {
 407     MigrationInfo *info = g_malloc0(sizeof(*info));
 408     MigrationState *s = migrate_get_current();
 409
 410     switch (s->state) {
 411     case MIGRATION_STATUS_NONE:
 412         /* no migration has happened ever */
 413         break;
 414     case MIGRATION_STATUS_SETUP:
 415         info->has_status = true;
 416         info->has_total_time = false;
 417         break;
 418     case MIGRATION_STATUS_ACTIVE:
 419     case MIGRATION_STATUS_CANCELLING:
 420         info->has_status = true;
 421         info->has_total_time = true;
 422         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 423             - s->total_time;
 424         info->has_expected_downtime = true;
 425         info->expected_downtime = s->expected_downtime;
 426         info->has_setup_time = true;
 427         info->setup_time = s->setup_time;
 428
 429         info->has_ram = true;
 430         info->ram = g_malloc0(sizeof(*info->ram));
 431         info->ram->transferred = ram_bytes_transferred();
 432         info->ram->remaining = ram_bytes_remaining();
 433         info->ram->total = ram_bytes_total();
 434         info->ram->duplicate = dup_mig_pages_transferred();
 435         info->ram->skipped = skipped_mig_pages_transferred();
 436         info->ram->normal = norm_mig_pages_transferred();
 437         info->ram->normal_bytes = norm_mig_bytes_transferred();
 438         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 439         info->ram->mbps = s->mbps;
 440         info->ram->dirty_sync_count = s->dirty_sync_count;
 441
 442         if (blk_mig_active()) {
 443             info->has_disk = true;
 444             info->disk = g_malloc0(sizeof(*info->disk));
 445             info->disk->transferred = blk_mig_bytes_transferred();
 446             info->disk->remaining = blk_mig_bytes_remaining();
 447             info->disk->total = blk_mig_bytes_total();
 448         }
 449
 450         if (cpu_throttle_active()) {
 451             info->has_x_cpu_throttle_percentage = true;
 452             info->x_cpu_throttle_percentage = cpu_throttle_get_percentage();
 453         }
 454
 455         get_xbzrle_cache_stats(info);
 456         break;
 457     case MIGRATION_STATUS_COMPLETED:
 458         get_xbzrle_cache_stats(info);
 459
 460         info->has_status = true;
 461         info->has_total_time = true;
 462         info->total_time = s->total_time;
 463         info->has_downtime = true;
 464         info->downtime = s->downtime;
 465         info->has_setup_time = true;
 466         info->setup_time = s->setup_time;
 467
 468         info->has_ram = true;
 469         info->ram = g_malloc0(sizeof(*info->ram));
 470         info->ram->transferred = ram_bytes_transferred();
 471         info->ram->remaining = 0;
 472         info->ram->total = ram_bytes_total();
 473         info->ram->duplicate = dup_mig_pages_transferred();
 474         info->ram->skipped = skipped_mig_pages_transferred();
 475         info->ram->normal = norm_mig_pages_transferred();
 476         info->ram->normal_bytes = norm_mig_bytes_transferred();
 477         info->ram->mbps = s->mbps;
 478         info->ram->dirty_sync_count = s->dirty_sync_count;
 479         break;
 480     case MIGRATION_STATUS_FAILED:
 481         info->has_status = true;
 482         break;
 483     case MIGRATION_STATUS_CANCELLED:
 484         info->has_status = true;
 485         break;
 486     }
 487     info->status = s->state;
 488
 489     return info;
 490 }
 491
 492 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 493                                   Error **errp)
 494 {
 495     MigrationState *s = migrate_get_current();
 496     MigrationCapabilityStatusList *cap;
 497
 498     if (s->state == MIGRATION_STATUS_ACTIVE ||
 499         s->state == MIGRATION_STATUS_SETUP) {
 500         error_setg(errp, QERR_MIGRATION_ACTIVE);
 501         return;
 502     }
 503
 504     for (cap = params; cap; cap = cap->next) {
 505         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 506     }
 507 }
 508
 509 void qmp_migrate_set_parameters(bool has_compress_level,
 510                                 int64_t compress_level,
 511                                 bool has_compress_threads,
 512                                 int64_t compress_threads,
 513                                 bool has_decompress_threads,
 514                                 int64_t decompress_threads,
 515                                 bool has_x_cpu_throttle_initial,
 516                                 int64_t x_cpu_throttle_initial,
 517                                 bool has_x_cpu_throttle_increment,
 518                                 int64_t x_cpu_throttle_increment, Error **errp)
 519 {
 520     MigrationState *s = migrate_get_current();
 521
 522     if (has_compress_level && (compress_level < 0 || compress_level > 9)) {
 523         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
 524                    "is invalid, it should be in the range of 0 to 9");
 525         return;
 526     }
 527     if (has_compress_threads &&
 528             (compress_threads < 1 || compress_threads > 255)) {
 529         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 530                    "compress_threads",
 531                    "is invalid, it should be in the range of 1 to 255");
 532         return;
 533     }
 534     if (has_decompress_threads &&
 535             (decompress_threads < 1 || decompress_threads > 255)) {
 536         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 537                    "decompress_threads",
 538                    "is invalid, it should be in the range of 1 to 255");
 539         return;
 540     }
 541     if (has_x_cpu_throttle_initial &&
 542             (x_cpu_throttle_initial < 1 || x_cpu_throttle_initial > 99)) {
 543         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 544                    "x_cpu_throttle_initial",
 545                    "an integer in the range of 1 to 99");
 546     }
 547     if (has_x_cpu_throttle_increment &&
 548             (x_cpu_throttle_increment < 1 || x_cpu_throttle_increment > 99)) {
 549         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 550                    "x_cpu_throttle_increment",
 551                    "an integer in the range of 1 to 99");
 552     }
 553
 554     if (has_compress_level) {
 555         s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 556     }
 557     if (has_compress_threads) {
 558         s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] = compress_threads;
 559     }
 560     if (has_decompress_threads) {
 561         s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 562                                                     decompress_threads;
 563     }
 564     if (has_x_cpu_throttle_initial) {
 565         s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL] =
 566                                                     x_cpu_throttle_initial;
 567     }
 568
 569     if (has_x_cpu_throttle_increment) {
 570         s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT] =
 571                                                     x_cpu_throttle_increment;
 572     }
 573 }
 574
 575 /* shared migration helpers */
 576
 577 static void migrate_set_state(MigrationState *s, int old_state, int new_state)
 578 {
 579     if (atomic_cmpxchg(&s->state, old_state, new_state) == old_state) {
 580         trace_migrate_set_state(new_state);
 581         migrate_generate_event(new_state);
 582     }
 583 }
 584
 585 static void migrate_fd_cleanup(void *opaque)
 586 {
 587     MigrationState *s = opaque;
 588
 589     qemu_bh_delete(s->cleanup_bh);
 590     s->cleanup_bh = NULL;
 591
 592     if (s->file) {
 593         trace_migrate_fd_cleanup();
 594         qemu_mutex_unlock_iothread();
 595         qemu_thread_join(&s->thread);
 596         qemu_mutex_lock_iothread();
 597
 598         migrate_compress_threads_join();
 599         qemu_fclose(s->file);
 600         s->file = NULL;
 601     }
 602
 603     assert(s->state != MIGRATION_STATUS_ACTIVE);
 604
 605     if (s->state != MIGRATION_STATUS_COMPLETED) {
 606         qemu_savevm_state_cancel();
 607         if (s->state == MIGRATION_STATUS_CANCELLING) {
 608             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
 609                               MIGRATION_STATUS_CANCELLED);
 610         }
 611     }
 612
 613     notifier_list_notify(&migration_state_notifiers, s);
 614 }
 615
 616 void migrate_fd_error(MigrationState *s)
 617 {
 618     trace_migrate_fd_error();
 619     assert(s->file == NULL);
 620     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 621     notifier_list_notify(&migration_state_notifiers, s);
 622 }
 623
 624 static void migrate_fd_cancel(MigrationState *s)
 625 {
 626     int old_state ;
 627     QEMUFile *f = migrate_get_current()->file;
 628     trace_migrate_fd_cancel();
 629
 630     do {
 631         old_state = s->state;
 632         if (old_state != MIGRATION_STATUS_SETUP &&
 633             old_state != MIGRATION_STATUS_ACTIVE) {
 634             break;
 635         }
 636         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
 637     } while (s->state != MIGRATION_STATUS_CANCELLING);
 638
 639     /*
 640      * If we're unlucky the migration code might be stuck somewhere in a
 641      * send/write while the network has failed and is waiting to timeout;
 642      * if we've got shutdown(2) available then we can force it to quit.
 643      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
 644      * called in a bh, so there is no race against this cancel.
 645      */
 646     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
 647         qemu_file_shutdown(f);
 648     }
 649 }
 650
 651 void add_migration_state_change_notifier(Notifier *notify)
 652 {
 653     notifier_list_add(&migration_state_notifiers, notify);
 654 }
 655
 656 void remove_migration_state_change_notifier(Notifier *notify)
 657 {
 658     notifier_remove(notify);
 659 }
 660
 661 bool migration_in_setup(MigrationState *s)
 662 {
 663     return s->state == MIGRATION_STATUS_SETUP;
 664 }
 665
 666 bool migration_has_finished(MigrationState *s)
 667 {
 668     return s->state == MIGRATION_STATUS_COMPLETED;
 669 }
 670
 671 bool migration_has_failed(MigrationState *s)
 672 {
 673     return (s->state == MIGRATION_STATUS_CANCELLED ||
 674             s->state == MIGRATION_STATUS_FAILED);
 675 }
 676
 677 static MigrationState *migrate_init(const MigrationParams *params)
 678 {
 679     MigrationState *s = migrate_get_current();
 680     int64_t bandwidth_limit = s->bandwidth_limit;
 681     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 682     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 683     int compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 684     int compress_thread_count =
 685             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 686     int decompress_thread_count =
 687             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 688     int x_cpu_throttle_initial =
 689             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL];
 690     int x_cpu_throttle_increment =
 691             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT];
 692
 693     memcpy(enabled_capabilities, s->enabled_capabilities,
 694            sizeof(enabled_capabilities));
 695
 696     memset(s, 0, sizeof(*s));
 697     s->params = *params;
 698     memcpy(s->enabled_capabilities, enabled_capabilities,
 699            sizeof(enabled_capabilities));
 700     s->xbzrle_cache_size = xbzrle_cache_size;
 701
 702     s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 703     s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
 704                compress_thread_count;
 705     s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 706                decompress_thread_count;
 707     s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL] =
 708                 x_cpu_throttle_initial;
 709     s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT] =
 710                 x_cpu_throttle_increment;
 711     s->bandwidth_limit = bandwidth_limit;
 712     migrate_set_state(s, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
 713
 714     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 715     return s;
 716 }
 717
 718 static GSList *migration_blockers;
 719
 720 void migrate_add_blocker(Error *reason)
 721 {
 722     migration_blockers = g_slist_prepend(migration_blockers, reason);
 723 }
 724
 725 void migrate_del_blocker(Error *reason)
 726 {
 727     migration_blockers = g_slist_remove(migration_blockers, reason);
 728 }
 729
 730 void qmp_migrate_incoming(const char *uri, Error **errp)
 731 {
 732     Error *local_err = NULL;
 733     static bool once = true;
 734
 735     if (!deferred_incoming) {
 736         error_setg(errp, "For use with '-incoming defer'");
 737         return;
 738     }
 739     if (!once) {
 740         error_setg(errp, "The incoming migration has already been started");
 741     }
 742
 743     qemu_start_incoming_migration(uri, &local_err);
 744
 745     if (local_err) {
 746         error_propagate(errp, local_err);
 747         return;
 748     }
 749
 750     once = false;
 751 }
 752
 753 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 754                  bool has_inc, bool inc, bool has_detach, bool detach,
 755                  Error **errp)
 756 {
 757     Error *local_err = NULL;
 758     MigrationState *s = migrate_get_current();
 759     MigrationParams params;
 760     const char *p;
 761
 762     params.blk = has_blk && blk;
 763     params.shared = has_inc && inc;
 764
 765     if (s->state == MIGRATION_STATUS_ACTIVE ||
 766         s->state == MIGRATION_STATUS_SETUP ||
 767         s->state == MIGRATION_STATUS_CANCELLING) {
 768         error_setg(errp, QERR_MIGRATION_ACTIVE);
 769         return;
 770     }
 771     if (runstate_check(RUN_STATE_INMIGRATE)) {
 772         error_setg(errp, "Guest is waiting for an incoming migration");
 773         return;
 774     }
 775
 776     if (qemu_savevm_state_blocked(errp)) {
 777         return;
 778     }
 779
 780     if (migration_blockers) {
 781         *errp = error_copy(migration_blockers->data);
 782         return;
 783     }
 784
 785     /* We are starting a new migration, so we want to start in a clean
 786        state.  This change is only needed if previous migration
 787        failed/was cancelled.  We don't use migrate_set_state() because
 788        we are setting the initial state, not changing it. */
 789     s->state = MIGRATION_STATUS_NONE;
 790
 791     s = migrate_init(&params);
 792
 793     if (strstart(uri, "tcp:", &p)) {
 794         tcp_start_outgoing_migration(s, p, &local_err);
 795 #ifdef CONFIG_RDMA
 796     } else if (strstart(uri, "rdma:", &p)) {
 797         rdma_start_outgoing_migration(s, p, &local_err);
 798 #endif
 799 #if !defined(WIN32)
 800     } else if (strstart(uri, "exec:", &p)) {
 801         exec_start_outgoing_migration(s, p, &local_err);
 802     } else if (strstart(uri, "unix:", &p)) {
 803         unix_start_outgoing_migration(s, p, &local_err);
 804     } else if (strstart(uri, "fd:", &p)) {
 805         fd_start_outgoing_migration(s, p, &local_err);
 806 #endif
 807     } else {
 808         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
 809                    "a valid migration protocol");
 810         migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 811         return;
 812     }
 813
 814     if (local_err) {
 815         migrate_fd_error(s);
 816         error_propagate(errp, local_err);
 817         return;
 818     }
 819 }
 820
 821 void qmp_migrate_cancel(Error **errp)
 822 {
 823     migrate_fd_cancel(migrate_get_current());
 824 }
 825
 826 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 827 {
 828     MigrationState *s = migrate_get_current();
 829     int64_t new_size;
 830
 831     /* Check for truncation */
 832     if (value != (size_t)value) {
 833         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 834                    "exceeding address space");
 835         return;
 836     }
 837
 838     /* Cache should not be larger than guest ram size */
 839     if (value > ram_bytes_total()) {
 840         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 841                    "exceeds guest ram size ");
 842         return;
 843     }
 844
 845     new_size = xbzrle_cache_resize(value);
 846     if (new_size < 0) {
 847         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 848                    "is smaller than page size");
 849         return;
 850     }
 851
 852     s->xbzrle_cache_size = new_size;
 853 }
 854
 855 int64_t qmp_query_migrate_cache_size(Error **errp)
 856 {
 857     return migrate_xbzrle_cache_size();
 858 }
 859
 860 void qmp_migrate_set_speed(int64_t value, Error **errp)
 861 {
 862     MigrationState *s;
 863
 864     if (value < 0) {
 865         value = 0;
 866     }
 867     if (value > SIZE_MAX) {
 868         value = SIZE_MAX;
 869     }
 870
 871     s = migrate_get_current();
 872     s->bandwidth_limit = value;
 873     if (s->file) {
 874         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
 875     }
 876 }
 877
 878 void qmp_migrate_set_downtime(double value, Error **errp)
 879 {
 880     value *= 1e9;
 881     value = MAX(0, MIN(UINT64_MAX, value));
 882     max_downtime = (uint64_t)value;
 883 }
 884
 885 bool migrate_auto_converge(void)
 886 {
 887     MigrationState *s;
 888
 889     s = migrate_get_current();
 890
 891     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
 892 }
 893
 894 bool migrate_zero_blocks(void)
 895 {
 896     MigrationState *s;
 897
 898     s = migrate_get_current();
 899
 900     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
 901 }
 902
 903 bool migrate_use_compression(void)
 904 {
 905     MigrationState *s;
 906
 907     s = migrate_get_current();
 908
 909     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
 910 }
 911
 912 int migrate_compress_level(void)
 913 {
 914     MigrationState *s;
 915
 916     s = migrate_get_current();
 917
 918     return s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 919 }
 920
 921 int migrate_compress_threads(void)
 922 {
 923     MigrationState *s;
 924
 925     s = migrate_get_current();
 926
 927     return s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 928 }
 929
 930 int migrate_decompress_threads(void)
 931 {
 932     MigrationState *s;
 933
 934     s = migrate_get_current();
 935
 936     return s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 937 }
 938
 939 bool migrate_use_events(void)
 940 {
 941     MigrationState *s;
 942
 943     s = migrate_get_current();
 944
 945     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
 946 }
 947
 948 int migrate_use_xbzrle(void)
 949 {
 950     MigrationState *s;
 951
 952     s = migrate_get_current();
 953
 954     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 955 }
 956
 957 int64_t migrate_xbzrle_cache_size(void)
 958 {
 959     MigrationState *s;
 960
 961     s = migrate_get_current();
 962
 963     return s->xbzrle_cache_size;
 964 }
 965
 966 /**
 967  * migration_completion: Used by migration_thread when there's not much left.
 968  *   The caller 'breaks' the loop when this returns.
 969  *
 970  * @s: Current migration state
 971  * @*old_vm_running: Pointer to old_vm_running flag
 972  * @*start_time: Pointer to time to update
 973  */
 974 static void migration_completion(MigrationState *s, bool *old_vm_running,
 975                                  int64_t *start_time)
 976 {
 977     int ret;
 978
 979     qemu_mutex_lock_iothread();
 980     *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 981     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 982     *old_vm_running = runstate_is_running();
 983
 984     ret = global_state_store();
 985     if (!ret) {
 986         ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 987         if (ret >= 0) {
 988             qemu_file_set_rate_limit(s->file, INT64_MAX);
 989             qemu_savevm_state_complete(s->file);
 990         }
 991     }
 992     qemu_mutex_unlock_iothread();
 993
 994     if (ret < 0) {
 995         goto fail;
 996     }
 997
 998     if (qemu_file_get_error(s->file)) {
 999         trace_migration_completion_file_err();
1000         goto fail;
1001     }
1002
1003     migrate_set_state(s, MIGRATION_STATUS_ACTIVE, MIGRATION_STATUS_COMPLETED);
1004     return;
1005
1006 fail:
1007     migrate_set_state(s, MIGRATION_STATUS_ACTIVE, MIGRATION_STATUS_FAILED);
1008 }
1009
1010 /* migration thread support */
1011
1012 static void *migration_thread(void *opaque)
1013 {
1014     MigrationState *s = opaque;
1015     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1016     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1017     int64_t initial_bytes = 0;
1018     int64_t max_size = 0;
1019     int64_t start_time = initial_time;
1020     bool old_vm_running = false;
1021
1022     rcu_register_thread();
1023
1024     qemu_savevm_state_header(s->file);
1025     qemu_savevm_state_begin(s->file, &s->params);
1026
1027     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1028     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
1029
1030     while (s->state == MIGRATION_STATUS_ACTIVE) {
1031         int64_t current_time;
1032         uint64_t pending_size;
1033
1034         if (!qemu_file_rate_limit(s->file)) {
1035             pending_size = qemu_savevm_state_pending(s->file, max_size);
1036             trace_migrate_pending(pending_size, max_size);
1037             if (pending_size && pending_size >= max_size) {
1038                 qemu_savevm_state_iterate(s->file);
1039             } else {
1040                 trace_migration_thread_low_pending(pending_size);
1041                 migration_completion(s, &old_vm_running, &start_time);
1042                 break;
1043             }
1044         }
1045
1046         if (qemu_file_get_error(s->file)) {
1047             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
1048                               MIGRATION_STATUS_FAILED);
1049             break;
1050         }
1051         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1052         if (current_time >= initial_time + BUFFER_DELAY) {
1053             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
1054             uint64_t time_spent = current_time - initial_time;
1055             double bandwidth = transferred_bytes / time_spent;
1056             max_size = bandwidth * migrate_max_downtime() / 1000000;
1057
1058             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
1059                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
1060
1061             trace_migrate_transferred(transferred_bytes, time_spent,
1062                                       bandwidth, max_size);
1063             /* if we haven't sent anything, we don't want to recalculate
1064                10000 is a small enough number for our purposes */
1065             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
1066                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
1067             }
1068
1069             qemu_file_reset_rate_limit(s->file);
1070             initial_time = current_time;
1071             initial_bytes = qemu_ftell(s->file);
1072         }
1073         if (qemu_file_rate_limit(s->file)) {
1074             /* usleep expects microseconds */
1075             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
1076         }
1077     }
1078
1079     /* If we enabled cpu throttling for auto-converge, turn it off. */
1080     cpu_throttle_stop();
1081
1082     qemu_mutex_lock_iothread();
1083     if (s->state == MIGRATION_STATUS_COMPLETED) {
1084         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1085         uint64_t transferred_bytes = qemu_ftell(s->file);
1086         s->total_time = end_time - s->total_time;
1087         s->downtime = end_time - start_time;
1088         if (s->total_time) {
1089             s->mbps = (((double) transferred_bytes * 8.0) /
1090                        ((double) s->total_time)) / 1000;
1091         }
1092         runstate_set(RUN_STATE_POSTMIGRATE);
1093     } else {
1094         if (old_vm_running) {
1095             vm_start();
1096         }
1097     }
1098     qemu_bh_schedule(s->cleanup_bh);
1099     qemu_mutex_unlock_iothread();
1100
1101     rcu_unregister_thread();
1102     return NULL;
1103 }
1104
1105 void migrate_fd_connect(MigrationState *s)
1106 {
1107     /* This is a best 1st approximation. ns to ms */
1108     s->expected_downtime = max_downtime/1000000;
1109     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
1110
1111     qemu_file_set_rate_limit(s->file,
1112                              s->bandwidth_limit / XFER_LIMIT_RATIO);
1113
1114     /* Notify before starting migration thread */
1115     notifier_list_notify(&migration_state_notifiers, s);
1116
1117     migrate_compress_threads_create();
1118     qemu_thread_create(&s->thread, "migration", migration_thread, s,
1119                        QEMU_THREAD_JOINABLE);
1120 }