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     qemu_announce_self();
 298
 299     /* Make sure all file formats flush their mutable metadata */
 300     bdrv_invalidate_cache_all(&local_err);
 301     if (local_err) {
 302         migrate_generate_event(MIGRATION_STATUS_FAILED);
 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      * This must happen after any state changes since as soon as an external
 325      * observer sees this event they might start to prod at the VM assuming
 326      * it's ready to use.
 327      */
 328     migrate_generate_event(MIGRATION_STATUS_COMPLETED);
 329 }
 330
 331 void process_incoming_migration(QEMUFile *f)
 332 {
 333     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
 334     int fd = qemu_get_fd(f);
 335
 336     assert(fd != -1);
 337     migrate_decompress_threads_create();
 338     qemu_set_nonblock(fd);
 339     qemu_coroutine_enter(co, f);
 340 }
 341
 342 /* amount of nanoseconds we are willing to wait for migration to be down.
 343  * the choice of nanoseconds is because it is the maximum resolution that
 344  * get_clock() can achieve. It is an internal measure. All user-visible
 345  * units must be in seconds */
 346 static uint64_t max_downtime = 300000000;
 347
 348 uint64_t migrate_max_downtime(void)
 349 {
 350     return max_downtime;
 351 }
 352
 353 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 354 {
 355     MigrationCapabilityStatusList *head = NULL;
 356     MigrationCapabilityStatusList *caps;
 357     MigrationState *s = migrate_get_current();
 358     int i;
 359
 360     caps = NULL; /* silence compiler warning */
 361     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 362         if (head == NULL) {
 363             head = g_malloc0(sizeof(*caps));
 364             caps = head;
 365         } else {
 366             caps->next = g_malloc0(sizeof(*caps));
 367             caps = caps->next;
 368         }
 369         caps->value =
 370             g_malloc(sizeof(*caps->value));
 371         caps->value->capability = i;
 372         caps->value->state = s->enabled_capabilities[i];
 373     }
 374
 375     return head;
 376 }
 377
 378 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 379 {
 380     MigrationParameters *params;
 381     MigrationState *s = migrate_get_current();
 382
 383     params = g_malloc0(sizeof(*params));
 384     params->compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 385     params->compress_threads =
 386             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 387     params->decompress_threads =
 388             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 389     params->x_cpu_throttle_initial =
 390             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL];
 391     params->x_cpu_throttle_increment =
 392             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT];
 393
 394     return params;
 395 }
 396
 397 static void get_xbzrle_cache_stats(MigrationInfo *info)
 398 {
 399     if (migrate_use_xbzrle()) {
 400         info->has_xbzrle_cache = true;
 401         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 402         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 403         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 404         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 405         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 406         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
 407         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 408     }
 409 }
 410
 411 MigrationInfo *qmp_query_migrate(Error **errp)
 412 {
 413     MigrationInfo *info = g_malloc0(sizeof(*info));
 414     MigrationState *s = migrate_get_current();
 415
 416     switch (s->state) {
 417     case MIGRATION_STATUS_NONE:
 418         /* no migration has happened ever */
 419         break;
 420     case MIGRATION_STATUS_SETUP:
 421         info->has_status = true;
 422         info->has_total_time = false;
 423         break;
 424     case MIGRATION_STATUS_ACTIVE:
 425     case MIGRATION_STATUS_CANCELLING:
 426         info->has_status = true;
 427         info->has_total_time = true;
 428         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 429             - s->total_time;
 430         info->has_expected_downtime = true;
 431         info->expected_downtime = s->expected_downtime;
 432         info->has_setup_time = true;
 433         info->setup_time = s->setup_time;
 434
 435         info->has_ram = true;
 436         info->ram = g_malloc0(sizeof(*info->ram));
 437         info->ram->transferred = ram_bytes_transferred();
 438         info->ram->remaining = ram_bytes_remaining();
 439         info->ram->total = ram_bytes_total();
 440         info->ram->duplicate = dup_mig_pages_transferred();
 441         info->ram->skipped = skipped_mig_pages_transferred();
 442         info->ram->normal = norm_mig_pages_transferred();
 443         info->ram->normal_bytes = norm_mig_bytes_transferred();
 444         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 445         info->ram->mbps = s->mbps;
 446         info->ram->dirty_sync_count = s->dirty_sync_count;
 447
 448         if (blk_mig_active()) {
 449             info->has_disk = true;
 450             info->disk = g_malloc0(sizeof(*info->disk));
 451             info->disk->transferred = blk_mig_bytes_transferred();
 452             info->disk->remaining = blk_mig_bytes_remaining();
 453             info->disk->total = blk_mig_bytes_total();
 454         }
 455
 456         if (cpu_throttle_active()) {
 457             info->has_x_cpu_throttle_percentage = true;
 458             info->x_cpu_throttle_percentage = cpu_throttle_get_percentage();
 459         }
 460
 461         get_xbzrle_cache_stats(info);
 462         break;
 463     case MIGRATION_STATUS_COMPLETED:
 464         get_xbzrle_cache_stats(info);
 465
 466         info->has_status = true;
 467         info->has_total_time = true;
 468         info->total_time = s->total_time;
 469         info->has_downtime = true;
 470         info->downtime = s->downtime;
 471         info->has_setup_time = true;
 472         info->setup_time = s->setup_time;
 473
 474         info->has_ram = true;
 475         info->ram = g_malloc0(sizeof(*info->ram));
 476         info->ram->transferred = ram_bytes_transferred();
 477         info->ram->remaining = 0;
 478         info->ram->total = ram_bytes_total();
 479         info->ram->duplicate = dup_mig_pages_transferred();
 480         info->ram->skipped = skipped_mig_pages_transferred();
 481         info->ram->normal = norm_mig_pages_transferred();
 482         info->ram->normal_bytes = norm_mig_bytes_transferred();
 483         info->ram->mbps = s->mbps;
 484         info->ram->dirty_sync_count = s->dirty_sync_count;
 485         break;
 486     case MIGRATION_STATUS_FAILED:
 487         info->has_status = true;
 488         break;
 489     case MIGRATION_STATUS_CANCELLED:
 490         info->has_status = true;
 491         break;
 492     }
 493     info->status = s->state;
 494
 495     return info;
 496 }
 497
 498 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 499                                   Error **errp)
 500 {
 501     MigrationState *s = migrate_get_current();
 502     MigrationCapabilityStatusList *cap;
 503
 504     if (s->state == MIGRATION_STATUS_ACTIVE ||
 505         s->state == MIGRATION_STATUS_SETUP) {
 506         error_setg(errp, QERR_MIGRATION_ACTIVE);
 507         return;
 508     }
 509
 510     for (cap = params; cap; cap = cap->next) {
 511         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 512     }
 513 }
 514
 515 void qmp_migrate_set_parameters(bool has_compress_level,
 516                                 int64_t compress_level,
 517                                 bool has_compress_threads,
 518                                 int64_t compress_threads,
 519                                 bool has_decompress_threads,
 520                                 int64_t decompress_threads,
 521                                 bool has_x_cpu_throttle_initial,
 522                                 int64_t x_cpu_throttle_initial,
 523                                 bool has_x_cpu_throttle_increment,
 524                                 int64_t x_cpu_throttle_increment, Error **errp)
 525 {
 526     MigrationState *s = migrate_get_current();
 527
 528     if (has_compress_level && (compress_level < 0 || compress_level > 9)) {
 529         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
 530                    "is invalid, it should be in the range of 0 to 9");
 531         return;
 532     }
 533     if (has_compress_threads &&
 534             (compress_threads < 1 || compress_threads > 255)) {
 535         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 536                    "compress_threads",
 537                    "is invalid, it should be in the range of 1 to 255");
 538         return;
 539     }
 540     if (has_decompress_threads &&
 541             (decompress_threads < 1 || decompress_threads > 255)) {
 542         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 543                    "decompress_threads",
 544                    "is invalid, it should be in the range of 1 to 255");
 545         return;
 546     }
 547     if (has_x_cpu_throttle_initial &&
 548             (x_cpu_throttle_initial < 1 || x_cpu_throttle_initial > 99)) {
 549         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 550                    "x_cpu_throttle_initial",
 551                    "an integer in the range of 1 to 99");
 552     }
 553     if (has_x_cpu_throttle_increment &&
 554             (x_cpu_throttle_increment < 1 || x_cpu_throttle_increment > 99)) {
 555         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 556                    "x_cpu_throttle_increment",
 557                    "an integer in the range of 1 to 99");
 558     }
 559
 560     if (has_compress_level) {
 561         s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 562     }
 563     if (has_compress_threads) {
 564         s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] = compress_threads;
 565     }
 566     if (has_decompress_threads) {
 567         s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 568                                                     decompress_threads;
 569     }
 570     if (has_x_cpu_throttle_initial) {
 571         s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL] =
 572                                                     x_cpu_throttle_initial;
 573     }
 574
 575     if (has_x_cpu_throttle_increment) {
 576         s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT] =
 577                                                     x_cpu_throttle_increment;
 578     }
 579 }
 580
 581 /* shared migration helpers */
 582
 583 static void migrate_set_state(MigrationState *s, int old_state, int new_state)
 584 {
 585     if (atomic_cmpxchg(&s->state, old_state, new_state) == old_state) {
 586         trace_migrate_set_state(new_state);
 587         migrate_generate_event(new_state);
 588     }
 589 }
 590
 591 static void migrate_fd_cleanup(void *opaque)
 592 {
 593     MigrationState *s = opaque;
 594
 595     qemu_bh_delete(s->cleanup_bh);
 596     s->cleanup_bh = NULL;
 597
 598     if (s->file) {
 599         trace_migrate_fd_cleanup();
 600         qemu_mutex_unlock_iothread();
 601         qemu_thread_join(&s->thread);
 602         qemu_mutex_lock_iothread();
 603
 604         migrate_compress_threads_join();
 605         qemu_fclose(s->file);
 606         s->file = NULL;
 607     }
 608
 609     assert(s->state != MIGRATION_STATUS_ACTIVE);
 610
 611     if (s->state != MIGRATION_STATUS_COMPLETED) {
 612         qemu_savevm_state_cancel();
 613         if (s->state == MIGRATION_STATUS_CANCELLING) {
 614             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
 615                               MIGRATION_STATUS_CANCELLED);
 616         }
 617     }
 618
 619     notifier_list_notify(&migration_state_notifiers, s);
 620 }
 621
 622 void migrate_fd_error(MigrationState *s)
 623 {
 624     trace_migrate_fd_error();
 625     assert(s->file == NULL);
 626     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 627     notifier_list_notify(&migration_state_notifiers, s);
 628 }
 629
 630 static void migrate_fd_cancel(MigrationState *s)
 631 {
 632     int old_state ;
 633     QEMUFile *f = migrate_get_current()->file;
 634     trace_migrate_fd_cancel();
 635
 636     do {
 637         old_state = s->state;
 638         if (old_state != MIGRATION_STATUS_SETUP &&
 639             old_state != MIGRATION_STATUS_ACTIVE) {
 640             break;
 641         }
 642         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
 643     } while (s->state != MIGRATION_STATUS_CANCELLING);
 644
 645     /*
 646      * If we're unlucky the migration code might be stuck somewhere in a
 647      * send/write while the network has failed and is waiting to timeout;
 648      * if we've got shutdown(2) available then we can force it to quit.
 649      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
 650      * called in a bh, so there is no race against this cancel.
 651      */
 652     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
 653         qemu_file_shutdown(f);
 654     }
 655 }
 656
 657 void add_migration_state_change_notifier(Notifier *notify)
 658 {
 659     notifier_list_add(&migration_state_notifiers, notify);
 660 }
 661
 662 void remove_migration_state_change_notifier(Notifier *notify)
 663 {
 664     notifier_remove(notify);
 665 }
 666
 667 bool migration_in_setup(MigrationState *s)
 668 {
 669     return s->state == MIGRATION_STATUS_SETUP;
 670 }
 671
 672 bool migration_has_finished(MigrationState *s)
 673 {
 674     return s->state == MIGRATION_STATUS_COMPLETED;
 675 }
 676
 677 bool migration_has_failed(MigrationState *s)
 678 {
 679     return (s->state == MIGRATION_STATUS_CANCELLED ||
 680             s->state == MIGRATION_STATUS_FAILED);
 681 }
 682
 683 static MigrationState *migrate_init(const MigrationParams *params)
 684 {
 685     MigrationState *s = migrate_get_current();
 686     int64_t bandwidth_limit = s->bandwidth_limit;
 687     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 688     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 689     int compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 690     int compress_thread_count =
 691             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 692     int decompress_thread_count =
 693             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 694     int x_cpu_throttle_initial =
 695             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL];
 696     int x_cpu_throttle_increment =
 697             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT];
 698
 699     memcpy(enabled_capabilities, s->enabled_capabilities,
 700            sizeof(enabled_capabilities));
 701
 702     memset(s, 0, sizeof(*s));
 703     s->params = *params;
 704     memcpy(s->enabled_capabilities, enabled_capabilities,
 705            sizeof(enabled_capabilities));
 706     s->xbzrle_cache_size = xbzrle_cache_size;
 707
 708     s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 709     s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
 710                compress_thread_count;
 711     s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 712                decompress_thread_count;
 713     s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL] =
 714                 x_cpu_throttle_initial;
 715     s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT] =
 716                 x_cpu_throttle_increment;
 717     s->bandwidth_limit = bandwidth_limit;
 718     migrate_set_state(s, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
 719
 720     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 721     return s;
 722 }
 723
 724 static GSList *migration_blockers;
 725
 726 void migrate_add_blocker(Error *reason)
 727 {
 728     migration_blockers = g_slist_prepend(migration_blockers, reason);
 729 }
 730
 731 void migrate_del_blocker(Error *reason)
 732 {
 733     migration_blockers = g_slist_remove(migration_blockers, reason);
 734 }
 735
 736 void qmp_migrate_incoming(const char *uri, Error **errp)
 737 {
 738     Error *local_err = NULL;
 739     static bool once = true;
 740
 741     if (!deferred_incoming) {
 742         error_setg(errp, "For use with '-incoming defer'");
 743         return;
 744     }
 745     if (!once) {
 746         error_setg(errp, "The incoming migration has already been started");
 747     }
 748
 749     qemu_start_incoming_migration(uri, &local_err);
 750
 751     if (local_err) {
 752         error_propagate(errp, local_err);
 753         return;
 754     }
 755
 756     once = false;
 757 }
 758
 759 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 760                  bool has_inc, bool inc, bool has_detach, bool detach,
 761                  Error **errp)
 762 {
 763     Error *local_err = NULL;
 764     MigrationState *s = migrate_get_current();
 765     MigrationParams params;
 766     const char *p;
 767
 768     params.blk = has_blk && blk;
 769     params.shared = has_inc && inc;
 770
 771     if (s->state == MIGRATION_STATUS_ACTIVE ||
 772         s->state == MIGRATION_STATUS_SETUP ||
 773         s->state == MIGRATION_STATUS_CANCELLING) {
 774         error_setg(errp, QERR_MIGRATION_ACTIVE);
 775         return;
 776     }
 777     if (runstate_check(RUN_STATE_INMIGRATE)) {
 778         error_setg(errp, "Guest is waiting for an incoming migration");
 779         return;
 780     }
 781
 782     if (qemu_savevm_state_blocked(errp)) {
 783         return;
 784     }
 785
 786     if (migration_blockers) {
 787         *errp = error_copy(migration_blockers->data);
 788         return;
 789     }
 790
 791     /* We are starting a new migration, so we want to start in a clean
 792        state.  This change is only needed if previous migration
 793        failed/was cancelled.  We don't use migrate_set_state() because
 794        we are setting the initial state, not changing it. */
 795     s->state = MIGRATION_STATUS_NONE;
 796
 797     s = migrate_init(&params);
 798
 799     if (strstart(uri, "tcp:", &p)) {
 800         tcp_start_outgoing_migration(s, p, &local_err);
 801 #ifdef CONFIG_RDMA
 802     } else if (strstart(uri, "rdma:", &p)) {
 803         rdma_start_outgoing_migration(s, p, &local_err);
 804 #endif
 805 #if !defined(WIN32)
 806     } else if (strstart(uri, "exec:", &p)) {
 807         exec_start_outgoing_migration(s, p, &local_err);
 808     } else if (strstart(uri, "unix:", &p)) {
 809         unix_start_outgoing_migration(s, p, &local_err);
 810     } else if (strstart(uri, "fd:", &p)) {
 811         fd_start_outgoing_migration(s, p, &local_err);
 812 #endif
 813     } else {
 814         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
 815                    "a valid migration protocol");
 816         migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 817         return;
 818     }
 819
 820     if (local_err) {
 821         migrate_fd_error(s);
 822         error_propagate(errp, local_err);
 823         return;
 824     }
 825 }
 826
 827 void qmp_migrate_cancel(Error **errp)
 828 {
 829     migrate_fd_cancel(migrate_get_current());
 830 }
 831
 832 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 833 {
 834     MigrationState *s = migrate_get_current();
 835     int64_t new_size;
 836
 837     /* Check for truncation */
 838     if (value != (size_t)value) {
 839         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 840                    "exceeding address space");
 841         return;
 842     }
 843
 844     /* Cache should not be larger than guest ram size */
 845     if (value > ram_bytes_total()) {
 846         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 847                    "exceeds guest ram size ");
 848         return;
 849     }
 850
 851     new_size = xbzrle_cache_resize(value);
 852     if (new_size < 0) {
 853         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 854                    "is smaller than page size");
 855         return;
 856     }
 857
 858     s->xbzrle_cache_size = new_size;
 859 }
 860
 861 int64_t qmp_query_migrate_cache_size(Error **errp)
 862 {
 863     return migrate_xbzrle_cache_size();
 864 }
 865
 866 void qmp_migrate_set_speed(int64_t value, Error **errp)
 867 {
 868     MigrationState *s;
 869
 870     if (value < 0) {
 871         value = 0;
 872     }
 873     if (value > SIZE_MAX) {
 874         value = SIZE_MAX;
 875     }
 876
 877     s = migrate_get_current();
 878     s->bandwidth_limit = value;
 879     if (s->file) {
 880         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
 881     }
 882 }
 883
 884 void qmp_migrate_set_downtime(double value, Error **errp)
 885 {
 886     value *= 1e9;
 887     value = MAX(0, MIN(UINT64_MAX, value));
 888     max_downtime = (uint64_t)value;
 889 }
 890
 891 bool migrate_auto_converge(void)
 892 {
 893     MigrationState *s;
 894
 895     s = migrate_get_current();
 896
 897     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
 898 }
 899
 900 bool migrate_zero_blocks(void)
 901 {
 902     MigrationState *s;
 903
 904     s = migrate_get_current();
 905
 906     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
 907 }
 908
 909 bool migrate_use_compression(void)
 910 {
 911     MigrationState *s;
 912
 913     s = migrate_get_current();
 914
 915     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
 916 }
 917
 918 int migrate_compress_level(void)
 919 {
 920     MigrationState *s;
 921
 922     s = migrate_get_current();
 923
 924     return s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 925 }
 926
 927 int migrate_compress_threads(void)
 928 {
 929     MigrationState *s;
 930
 931     s = migrate_get_current();
 932
 933     return s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 934 }
 935
 936 int migrate_decompress_threads(void)
 937 {
 938     MigrationState *s;
 939
 940     s = migrate_get_current();
 941
 942     return s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 943 }
 944
 945 bool migrate_use_events(void)
 946 {
 947     MigrationState *s;
 948
 949     s = migrate_get_current();
 950
 951     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
 952 }
 953
 954 int migrate_use_xbzrle(void)
 955 {
 956     MigrationState *s;
 957
 958     s = migrate_get_current();
 959
 960     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 961 }
 962
 963 int64_t migrate_xbzrle_cache_size(void)
 964 {
 965     MigrationState *s;
 966
 967     s = migrate_get_current();
 968
 969     return s->xbzrle_cache_size;
 970 }
 971
 972 /**
 973  * migration_completion: Used by migration_thread when there's not much left.
 974  *   The caller 'breaks' the loop when this returns.
 975  *
 976  * @s: Current migration state
 977  * @*old_vm_running: Pointer to old_vm_running flag
 978  * @*start_time: Pointer to time to update
 979  */
 980 static void migration_completion(MigrationState *s, bool *old_vm_running,
 981                                  int64_t *start_time)
 982 {
 983     int ret;
 984
 985     qemu_mutex_lock_iothread();
 986     *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 987     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 988     *old_vm_running = runstate_is_running();
 989
 990     ret = global_state_store();
 991     if (!ret) {
 992         ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 993         if (ret >= 0) {
 994             qemu_file_set_rate_limit(s->file, INT64_MAX);
 995             qemu_savevm_state_complete(s->file);
 996         }
 997     }
 998     qemu_mutex_unlock_iothread();
 999
1000     if (ret < 0) {
1001         goto fail;
1002     }
1003
1004     if (qemu_file_get_error(s->file)) {
1005         trace_migration_completion_file_err();
1006         goto fail;
1007     }
1008
1009     migrate_set_state(s, MIGRATION_STATUS_ACTIVE, MIGRATION_STATUS_COMPLETED);
1010     return;
1011
1012 fail:
1013     migrate_set_state(s, MIGRATION_STATUS_ACTIVE, MIGRATION_STATUS_FAILED);
1014 }
1015
1016 /* migration thread support */
1017
1018 static void *migration_thread(void *opaque)
1019 {
1020     MigrationState *s = opaque;
1021     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1022     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1023     int64_t initial_bytes = 0;
1024     int64_t max_size = 0;
1025     int64_t start_time = initial_time;
1026     bool old_vm_running = false;
1027
1028     rcu_register_thread();
1029
1030     qemu_savevm_state_header(s->file);
1031     qemu_savevm_state_begin(s->file, &s->params);
1032
1033     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1034     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
1035
1036     while (s->state == MIGRATION_STATUS_ACTIVE) {
1037         int64_t current_time;
1038         uint64_t pending_size;
1039
1040         if (!qemu_file_rate_limit(s->file)) {
1041             pending_size = qemu_savevm_state_pending(s->file, max_size);
1042             trace_migrate_pending(pending_size, max_size);
1043             if (pending_size && pending_size >= max_size) {
1044                 qemu_savevm_state_iterate(s->file);
1045             } else {
1046                 trace_migration_thread_low_pending(pending_size);
1047                 migration_completion(s, &old_vm_running, &start_time);
1048                 break;
1049             }
1050         }
1051
1052         if (qemu_file_get_error(s->file)) {
1053             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
1054                               MIGRATION_STATUS_FAILED);
1055             break;
1056         }
1057         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1058         if (current_time >= initial_time + BUFFER_DELAY) {
1059             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
1060             uint64_t time_spent = current_time - initial_time;
1061             double bandwidth = transferred_bytes / time_spent;
1062             max_size = bandwidth * migrate_max_downtime() / 1000000;
1063
1064             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
1065                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
1066
1067             trace_migrate_transferred(transferred_bytes, time_spent,
1068                                       bandwidth, max_size);
1069             /* if we haven't sent anything, we don't want to recalculate
1070                10000 is a small enough number for our purposes */
1071             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
1072                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
1073             }
1074
1075             qemu_file_reset_rate_limit(s->file);
1076             initial_time = current_time;
1077             initial_bytes = qemu_ftell(s->file);
1078         }
1079         if (qemu_file_rate_limit(s->file)) {
1080             /* usleep expects microseconds */
1081             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
1082         }
1083     }
1084
1085     /* If we enabled cpu throttling for auto-converge, turn it off. */
1086     cpu_throttle_stop();
1087
1088     qemu_mutex_lock_iothread();
1089     if (s->state == MIGRATION_STATUS_COMPLETED) {
1090         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1091         uint64_t transferred_bytes = qemu_ftell(s->file);
1092         s->total_time = end_time - s->total_time;
1093         s->downtime = end_time - start_time;
1094         if (s->total_time) {
1095             s->mbps = (((double) transferred_bytes * 8.0) /
1096                        ((double) s->total_time)) / 1000;
1097         }
1098         runstate_set(RUN_STATE_POSTMIGRATE);
1099     } else {
1100         if (old_vm_running) {
1101             vm_start();
1102         }
1103     }
1104     qemu_bh_schedule(s->cleanup_bh);
1105     qemu_mutex_unlock_iothread();
1106
1107     rcu_unregister_thread();
1108     return NULL;
1109 }
1110
1111 void migrate_fd_connect(MigrationState *s)
1112 {
1113     /* This is a best 1st approximation. ns to ms */
1114     s->expected_downtime = max_downtime/1000000;
1115     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
1116
1117     qemu_file_set_rate_limit(s->file,
1118                              s->bandwidth_limit / XFER_LIMIT_RATIO);
1119
1120     /* Notify before starting migration thread */
1121     notifier_list_notify(&migration_state_notifiers, s);
1122
1123     migrate_compress_threads_create();
1124     qemu_thread_create(&s->thread, "migration", migration_thread, s,
1125                        QEMU_THREAD_JOINABLE);
1126 }