migration/migration.c

   1 /*
   2  * QEMU live migration
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu-common.h"
  17 #include "qemu/error-report.h"
  18 #include "qemu/main-loop.h"
  19 #include "migration/migration.h"
  20 #include "migration/qemu-file.h"
  21 #include "sysemu/sysemu.h"
  22 #include "block/block.h"
  23 #include "qapi/qmp/qerror.h"
  24 #include "qemu/sockets.h"
  25 #include "qemu/rcu.h"
  26 #include "migration/block.h"
  27 #include "qemu/thread.h"
  28 #include "qmp-commands.h"
  29 #include "trace.h"
  30 #include "qapi/util.h"
  31 #include "qapi-event.h"
  32
  33 #define MAX_THROTTLE  (32 << 20)      /* Migration speed throttling */
  34
  35 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
  36  * data. */
  37 #define BUFFER_DELAY     100
  38 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
  39
  40 /* Default compression thread count */
  41 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
  42 /* Default decompression thread count, usually decompression is at
  43  * least 4 times as fast as compression.*/
  44 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
  45 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
  46 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
  47 /* Define default autoconverge cpu throttle migration parameters */
  48 #define DEFAULT_MIGRATE_X_CPU_THROTTLE_INITIAL 20
  49 #define DEFAULT_MIGRATE_X_CPU_THROTTLE_INCREMENT 10
  50
  51 /* Migration XBZRLE default cache size */
  52 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
  53
  54 static NotifierList migration_state_notifiers =
  55     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  56
  57 static bool deferred_incoming;
  58
  59 /* When we add fault tolerance, we could have several
  60    migrations at once.  For now we don't need to add
  61    dynamic creation of migration */
  62
  63 /* For outgoing */
  64 MigrationState *migrate_get_current(void)
  65 {
  66     static MigrationState current_migration = {
  67         .state = MIGRATION_STATUS_NONE,
  68         .bandwidth_limit = MAX_THROTTLE,
  69         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
  70         .mbps = -1,
  71         .parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] =
  72                 DEFAULT_MIGRATE_COMPRESS_LEVEL,
  73         .parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
  74                 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
  75         .parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
  76                 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
  77         .parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL] =
  78                 DEFAULT_MIGRATE_X_CPU_THROTTLE_INITIAL,
  79         .parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT] =
  80                 DEFAULT_MIGRATE_X_CPU_THROTTLE_INCREMENT,
  81     };
  82
  83     return &current_migration;
  84 }
  85
  86 /* For incoming */
  87 static MigrationIncomingState *mis_current;
  88
  89 MigrationIncomingState *migration_incoming_get_current(void)
  90 {
  91     return mis_current;
  92 }
  93
  94 MigrationIncomingState *migration_incoming_state_new(QEMUFile* f)
  95 {
  96     mis_current = g_new0(MigrationIncomingState, 1);
  97     mis_current->file = f;
  98     QLIST_INIT(&mis_current->loadvm_handlers);
  99
 100     return mis_current;
 101 }
 102
 103 void migration_incoming_state_destroy(void)
 104 {
 105     loadvm_free_handlers(mis_current);
 106     g_free(mis_current);
 107     mis_current = NULL;
 108 }
 109
 110
 111 typedef struct {
 112     bool optional;
 113     uint32_t size;
 114     uint8_t runstate[100];
 115     RunState state;
 116     bool received;
 117 } GlobalState;
 118
 119 static GlobalState global_state;
 120
 121 int global_state_store(void)
 122 {
 123     if (!runstate_store((char *)global_state.runstate,
 124                         sizeof(global_state.runstate))) {
 125         error_report("runstate name too big: %s", global_state.runstate);
 126         trace_migrate_state_too_big();
 127         return -EINVAL;
 128     }
 129     return 0;
 130 }
 131
 132 void global_state_store_running(void)
 133 {
 134     const char *state = RunState_lookup[RUN_STATE_RUNNING];
 135     strncpy((char *)global_state.runstate,
 136            state, sizeof(global_state.runstate));
 137 }
 138
 139 static bool global_state_received(void)
 140 {
 141     return global_state.received;
 142 }
 143
 144 static RunState global_state_get_runstate(void)
 145 {
 146     return global_state.state;
 147 }
 148
 149 void global_state_set_optional(void)
 150 {
 151     global_state.optional = true;
 152 }
 153
 154 static bool global_state_needed(void *opaque)
 155 {
 156     GlobalState *s = opaque;
 157     char *runstate = (char *)s->runstate;
 158
 159     /* If it is not optional, it is mandatory */
 160
 161     if (s->optional == false) {
 162         return true;
 163     }
 164
 165     /* If state is running or paused, it is not needed */
 166
 167     if (strcmp(runstate, "running") == 0 ||
 168         strcmp(runstate, "paused") == 0) {
 169         return false;
 170     }
 171
 172     /* for any other state it is needed */
 173     return true;
 174 }
 175
 176 static int global_state_post_load(void *opaque, int version_id)
 177 {
 178     GlobalState *s = opaque;
 179     Error *local_err = NULL;
 180     int r;
 181     char *runstate = (char *)s->runstate;
 182
 183     s->received = true;
 184     trace_migrate_global_state_post_load(runstate);
 185
 186     r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE_MAX,
 187                                 -1, &local_err);
 188
 189     if (r == -1) {
 190         if (local_err) {
 191             error_report_err(local_err);
 192         }
 193         return -EINVAL;
 194     }
 195     s->state = r;
 196
 197     return 0;
 198 }
 199
 200 static void global_state_pre_save(void *opaque)
 201 {
 202     GlobalState *s = opaque;
 203
 204     trace_migrate_global_state_pre_save((char *)s->runstate);
 205     s->size = strlen((char *)s->runstate) + 1;
 206 }
 207
 208 static const VMStateDescription vmstate_globalstate = {
 209     .name = "globalstate",
 210     .version_id = 1,
 211     .minimum_version_id = 1,
 212     .post_load = global_state_post_load,
 213     .pre_save = global_state_pre_save,
 214     .needed = global_state_needed,
 215     .fields = (VMStateField[]) {
 216         VMSTATE_UINT32(size, GlobalState),
 217         VMSTATE_BUFFER(runstate, GlobalState),
 218         VMSTATE_END_OF_LIST()
 219     },
 220 };
 221
 222 void register_global_state(void)
 223 {
 224     /* We would use it independently that we receive it */
 225     strcpy((char *)&global_state.runstate, "");
 226     global_state.received = false;
 227     vmstate_register(NULL, 0, &vmstate_globalstate, &global_state);
 228 }
 229
 230 static void migrate_generate_event(int new_state)
 231 {
 232     if (migrate_use_events()) {
 233         qapi_event_send_migration(new_state, &error_abort);
 234     }
 235 }
 236
 237 /*
 238  * Called on -incoming with a defer: uri.
 239  * The migration can be started later after any parameters have been
 240  * changed.
 241  */
 242 static void deferred_incoming_migration(Error **errp)
 243 {
 244     if (deferred_incoming) {
 245         error_setg(errp, "Incoming migration already deferred");
 246     }
 247     deferred_incoming = true;
 248 }
 249
 250 void qemu_start_incoming_migration(const char *uri, Error **errp)
 251 {
 252     const char *p;
 253
 254     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
 255     if (!strcmp(uri, "defer")) {
 256         deferred_incoming_migration(errp);
 257     } else if (strstart(uri, "tcp:", &p)) {
 258         tcp_start_incoming_migration(p, errp);
 259 #ifdef CONFIG_RDMA
 260     } else if (strstart(uri, "rdma:", &p)) {
 261         rdma_start_incoming_migration(p, errp);
 262 #endif
 263 #if !defined(WIN32)
 264     } else if (strstart(uri, "exec:", &p)) {
 265         exec_start_incoming_migration(p, errp);
 266     } else if (strstart(uri, "unix:", &p)) {
 267         unix_start_incoming_migration(p, errp);
 268     } else if (strstart(uri, "fd:", &p)) {
 269         fd_start_incoming_migration(p, errp);
 270 #endif
 271     } else {
 272         error_setg(errp, "unknown migration protocol: %s", uri);
 273     }
 274 }
 275
 276 static void process_incoming_migration_co(void *opaque)
 277 {
 278     QEMUFile *f = opaque;
 279     Error *local_err = NULL;
 280     int ret;
 281
 282     migration_incoming_state_new(f);
 283     migrate_generate_event(MIGRATION_STATUS_ACTIVE);
 284     ret = qemu_loadvm_state(f);
 285
 286     qemu_fclose(f);
 287     free_xbzrle_decoded_buf();
 288     migration_incoming_state_destroy();
 289
 290     if (ret < 0) {
 291         migrate_generate_event(MIGRATION_STATUS_FAILED);
 292         error_report("load of migration failed: %s", strerror(-ret));
 293         migrate_decompress_threads_join();
 294         exit(EXIT_FAILURE);
 295     }
 296     migrate_generate_event(MIGRATION_STATUS_COMPLETED);
 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         error_report_err(local_err);
 303         migrate_decompress_threads_join();
 304         exit(EXIT_FAILURE);
 305     }
 306
 307     /* If global state section was not received or we are in running
 308        state, we need to obey autostart. Any other state is set with
 309        runstate_set. */
 310
 311     if (!global_state_received() ||
 312         global_state_get_runstate() == RUN_STATE_RUNNING) {
 313         if (autostart) {
 314             vm_start();
 315         } else {
 316             runstate_set(RUN_STATE_PAUSED);
 317         }
 318     } else {
 319         runstate_set(global_state_get_runstate());
 320     }
 321     migrate_decompress_threads_join();
 322 }
 323
 324 void process_incoming_migration(QEMUFile *f)
 325 {
 326     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
 327     int fd = qemu_get_fd(f);
 328
 329     assert(fd != -1);
 330     migrate_decompress_threads_create();
 331     qemu_set_nonblock(fd);
 332     qemu_coroutine_enter(co, f);
 333 }
 334
 335 /* amount of nanoseconds we are willing to wait for migration to be down.
 336  * the choice of nanoseconds is because it is the maximum resolution that
 337  * get_clock() can achieve. It is an internal measure. All user-visible
 338  * units must be in seconds */
 339 static uint64_t max_downtime = 300000000;
 340
 341 uint64_t migrate_max_downtime(void)
 342 {
 343     return max_downtime;
 344 }
 345
 346 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 347 {
 348     MigrationCapabilityStatusList *head = NULL;
 349     MigrationCapabilityStatusList *caps;
 350     MigrationState *s = migrate_get_current();
 351     int i;
 352
 353     caps = NULL; /* silence compiler warning */
 354     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 355         if (head == NULL) {
 356             head = g_malloc0(sizeof(*caps));
 357             caps = head;
 358         } else {
 359             caps->next = g_malloc0(sizeof(*caps));
 360             caps = caps->next;
 361         }
 362         caps->value =
 363             g_malloc(sizeof(*caps->value));
 364         caps->value->capability = i;
 365         caps->value->state = s->enabled_capabilities[i];
 366     }
 367
 368     return head;
 369 }
 370
 371 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 372 {
 373     MigrationParameters *params;
 374     MigrationState *s = migrate_get_current();
 375
 376     params = g_malloc0(sizeof(*params));
 377     params->compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 378     params->compress_threads =
 379             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 380     params->decompress_threads =
 381             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 382     params->x_cpu_throttle_initial =
 383             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL];
 384     params->x_cpu_throttle_increment =
 385             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT];
 386
 387     return params;
 388 }
 389
 390 static void get_xbzrle_cache_stats(MigrationInfo *info)
 391 {
 392     if (migrate_use_xbzrle()) {
 393         info->has_xbzrle_cache = true;
 394         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 395         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 396         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 397         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 398         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 399         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
 400         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 401     }
 402 }
 403
 404 MigrationInfo *qmp_query_migrate(Error **errp)
 405 {
 406     MigrationInfo *info = g_malloc0(sizeof(*info));
 407     MigrationState *s = migrate_get_current();
 408
 409     switch (s->state) {
 410     case MIGRATION_STATUS_NONE:
 411         /* no migration has happened ever */
 412         break;
 413     case MIGRATION_STATUS_SETUP:
 414         info->has_status = true;
 415         info->has_total_time = false;
 416         break;
 417     case MIGRATION_STATUS_ACTIVE:
 418     case MIGRATION_STATUS_CANCELLING:
 419         info->has_status = true;
 420         info->has_total_time = true;
 421         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 422             - s->total_time;
 423         info->has_expected_downtime = true;
 424         info->expected_downtime = s->expected_downtime;
 425         info->has_setup_time = true;
 426         info->setup_time = s->setup_time;
 427
 428         info->has_ram = true;
 429         info->ram = g_malloc0(sizeof(*info->ram));
 430         info->ram->transferred = ram_bytes_transferred();
 431         info->ram->remaining = ram_bytes_remaining();
 432         info->ram->total = ram_bytes_total();
 433         info->ram->duplicate = dup_mig_pages_transferred();
 434         info->ram->skipped = skipped_mig_pages_transferred();
 435         info->ram->normal = norm_mig_pages_transferred();
 436         info->ram->normal_bytes = norm_mig_bytes_transferred();
 437         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 438         info->ram->mbps = s->mbps;
 439         info->ram->dirty_sync_count = s->dirty_sync_count;
 440
 441         if (blk_mig_active()) {
 442             info->has_disk = true;
 443             info->disk = g_malloc0(sizeof(*info->disk));
 444             info->disk->transferred = blk_mig_bytes_transferred();
 445             info->disk->remaining = blk_mig_bytes_remaining();
 446             info->disk->total = blk_mig_bytes_total();
 447         }
 448
 449         get_xbzrle_cache_stats(info);
 450         break;
 451     case MIGRATION_STATUS_COMPLETED:
 452         get_xbzrle_cache_stats(info);
 453
 454         info->has_status = true;
 455         info->has_total_time = true;
 456         info->total_time = s->total_time;
 457         info->has_downtime = true;
 458         info->downtime = s->downtime;
 459         info->has_setup_time = true;
 460         info->setup_time = s->setup_time;
 461
 462         info->has_ram = true;
 463         info->ram = g_malloc0(sizeof(*info->ram));
 464         info->ram->transferred = ram_bytes_transferred();
 465         info->ram->remaining = 0;
 466         info->ram->total = ram_bytes_total();
 467         info->ram->duplicate = dup_mig_pages_transferred();
 468         info->ram->skipped = skipped_mig_pages_transferred();
 469         info->ram->normal = norm_mig_pages_transferred();
 470         info->ram->normal_bytes = norm_mig_bytes_transferred();
 471         info->ram->mbps = s->mbps;
 472         info->ram->dirty_sync_count = s->dirty_sync_count;
 473         break;
 474     case MIGRATION_STATUS_FAILED:
 475         info->has_status = true;
 476         break;
 477     case MIGRATION_STATUS_CANCELLED:
 478         info->has_status = true;
 479         break;
 480     }
 481     info->status = s->state;
 482
 483     return info;
 484 }
 485
 486 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 487                                   Error **errp)
 488 {
 489     MigrationState *s = migrate_get_current();
 490     MigrationCapabilityStatusList *cap;
 491
 492     if (s->state == MIGRATION_STATUS_ACTIVE ||
 493         s->state == MIGRATION_STATUS_SETUP) {
 494         error_setg(errp, QERR_MIGRATION_ACTIVE);
 495         return;
 496     }
 497
 498     for (cap = params; cap; cap = cap->next) {
 499         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 500     }
 501 }
 502
 503 void qmp_migrate_set_parameters(bool has_compress_level,
 504                                 int64_t compress_level,
 505                                 bool has_compress_threads,
 506                                 int64_t compress_threads,
 507                                 bool has_decompress_threads,
 508                                 int64_t decompress_threads,
 509                                 bool has_x_cpu_throttle_initial,
 510                                 int64_t x_cpu_throttle_initial,
 511                                 bool has_x_cpu_throttle_increment,
 512                                 int64_t x_cpu_throttle_increment, Error **errp)
 513 {
 514     MigrationState *s = migrate_get_current();
 515
 516     if (has_compress_level && (compress_level < 0 || compress_level > 9)) {
 517         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
 518                    "is invalid, it should be in the range of 0 to 9");
 519         return;
 520     }
 521     if (has_compress_threads &&
 522             (compress_threads < 1 || compress_threads > 255)) {
 523         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 524                    "compress_threads",
 525                    "is invalid, it should be in the range of 1 to 255");
 526         return;
 527     }
 528     if (has_decompress_threads &&
 529             (decompress_threads < 1 || decompress_threads > 255)) {
 530         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 531                    "decompress_threads",
 532                    "is invalid, it should be in the range of 1 to 255");
 533         return;
 534     }
 535     if (has_x_cpu_throttle_initial &&
 536             (x_cpu_throttle_initial < 1 || x_cpu_throttle_initial > 99)) {
 537         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 538                    "x_cpu_throttle_initial",
 539                    "an integer in the range of 1 to 99");
 540     }
 541     if (has_x_cpu_throttle_increment &&
 542             (x_cpu_throttle_increment < 1 || x_cpu_throttle_increment > 99)) {
 543         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 544                    "x_cpu_throttle_increment",
 545                    "an integer in the range of 1 to 99");
 546     }
 547
 548     if (has_compress_level) {
 549         s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 550     }
 551     if (has_compress_threads) {
 552         s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] = compress_threads;
 553     }
 554     if (has_decompress_threads) {
 555         s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 556                                                     decompress_threads;
 557     }
 558     if (has_x_cpu_throttle_initial) {
 559         s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL] =
 560                                                     x_cpu_throttle_initial;
 561     }
 562
 563     if (has_x_cpu_throttle_increment) {
 564         s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT] =
 565                                                     x_cpu_throttle_increment;
 566     }
 567 }
 568
 569 /* shared migration helpers */
 570
 571 static void migrate_set_state(MigrationState *s, int old_state, int new_state)
 572 {
 573     if (atomic_cmpxchg(&s->state, old_state, new_state) == old_state) {
 574         trace_migrate_set_state(new_state);
 575         migrate_generate_event(new_state);
 576     }
 577 }
 578
 579 static void migrate_fd_cleanup(void *opaque)
 580 {
 581     MigrationState *s = opaque;
 582
 583     qemu_bh_delete(s->cleanup_bh);
 584     s->cleanup_bh = NULL;
 585
 586     if (s->file) {
 587         trace_migrate_fd_cleanup();
 588         qemu_mutex_unlock_iothread();
 589         qemu_thread_join(&s->thread);
 590         qemu_mutex_lock_iothread();
 591
 592         migrate_compress_threads_join();
 593         qemu_fclose(s->file);
 594         s->file = NULL;
 595     }
 596
 597     assert(s->state != MIGRATION_STATUS_ACTIVE);
 598
 599     if (s->state != MIGRATION_STATUS_COMPLETED) {
 600         qemu_savevm_state_cancel();
 601         if (s->state == MIGRATION_STATUS_CANCELLING) {
 602             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
 603                               MIGRATION_STATUS_CANCELLED);
 604         }
 605     }
 606
 607     notifier_list_notify(&migration_state_notifiers, s);
 608 }
 609
 610 void migrate_fd_error(MigrationState *s)
 611 {
 612     trace_migrate_fd_error();
 613     assert(s->file == NULL);
 614     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 615     notifier_list_notify(&migration_state_notifiers, s);
 616 }
 617
 618 static void migrate_fd_cancel(MigrationState *s)
 619 {
 620     int old_state ;
 621     QEMUFile *f = migrate_get_current()->file;
 622     trace_migrate_fd_cancel();
 623
 624     do {
 625         old_state = s->state;
 626         if (old_state != MIGRATION_STATUS_SETUP &&
 627             old_state != MIGRATION_STATUS_ACTIVE) {
 628             break;
 629         }
 630         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
 631     } while (s->state != MIGRATION_STATUS_CANCELLING);
 632
 633     /*
 634      * If we're unlucky the migration code might be stuck somewhere in a
 635      * send/write while the network has failed and is waiting to timeout;
 636      * if we've got shutdown(2) available then we can force it to quit.
 637      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
 638      * called in a bh, so there is no race against this cancel.
 639      */
 640     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
 641         qemu_file_shutdown(f);
 642     }
 643 }
 644
 645 void add_migration_state_change_notifier(Notifier *notify)
 646 {
 647     notifier_list_add(&migration_state_notifiers, notify);
 648 }
 649
 650 void remove_migration_state_change_notifier(Notifier *notify)
 651 {
 652     notifier_remove(notify);
 653 }
 654
 655 bool migration_in_setup(MigrationState *s)
 656 {
 657     return s->state == MIGRATION_STATUS_SETUP;
 658 }
 659
 660 bool migration_has_finished(MigrationState *s)
 661 {
 662     return s->state == MIGRATION_STATUS_COMPLETED;
 663 }
 664
 665 bool migration_has_failed(MigrationState *s)
 666 {
 667     return (s->state == MIGRATION_STATUS_CANCELLED ||
 668             s->state == MIGRATION_STATUS_FAILED);
 669 }
 670
 671 static MigrationState *migrate_init(const MigrationParams *params)
 672 {
 673     MigrationState *s = migrate_get_current();
 674     int64_t bandwidth_limit = s->bandwidth_limit;
 675     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 676     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 677     int compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 678     int compress_thread_count =
 679             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 680     int decompress_thread_count =
 681             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 682     int x_cpu_throttle_initial =
 683             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL];
 684     int x_cpu_throttle_increment =
 685             s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT];
 686
 687     memcpy(enabled_capabilities, s->enabled_capabilities,
 688            sizeof(enabled_capabilities));
 689
 690     memset(s, 0, sizeof(*s));
 691     s->params = *params;
 692     memcpy(s->enabled_capabilities, enabled_capabilities,
 693            sizeof(enabled_capabilities));
 694     s->xbzrle_cache_size = xbzrle_cache_size;
 695
 696     s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
 697     s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
 698                compress_thread_count;
 699     s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
 700                decompress_thread_count;
 701     s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL] =
 702                 x_cpu_throttle_initial;
 703     s->parameters[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT] =
 704                 x_cpu_throttle_increment;
 705     s->bandwidth_limit = bandwidth_limit;
 706     migrate_set_state(s, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
 707
 708     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 709     return s;
 710 }
 711
 712 static GSList *migration_blockers;
 713
 714 void migrate_add_blocker(Error *reason)
 715 {
 716     migration_blockers = g_slist_prepend(migration_blockers, reason);
 717 }
 718
 719 void migrate_del_blocker(Error *reason)
 720 {
 721     migration_blockers = g_slist_remove(migration_blockers, reason);
 722 }
 723
 724 void qmp_migrate_incoming(const char *uri, Error **errp)
 725 {
 726     Error *local_err = NULL;
 727     static bool once = true;
 728
 729     if (!deferred_incoming) {
 730         error_setg(errp, "For use with '-incoming defer'");
 731         return;
 732     }
 733     if (!once) {
 734         error_setg(errp, "The incoming migration has already been started");
 735     }
 736
 737     qemu_start_incoming_migration(uri, &local_err);
 738
 739     if (local_err) {
 740         error_propagate(errp, local_err);
 741         return;
 742     }
 743
 744     once = false;
 745 }
 746
 747 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 748                  bool has_inc, bool inc, bool has_detach, bool detach,
 749                  Error **errp)
 750 {
 751     Error *local_err = NULL;
 752     MigrationState *s = migrate_get_current();
 753     MigrationParams params;
 754     const char *p;
 755
 756     params.blk = has_blk && blk;
 757     params.shared = has_inc && inc;
 758
 759     if (s->state == MIGRATION_STATUS_ACTIVE ||
 760         s->state == MIGRATION_STATUS_SETUP ||
 761         s->state == MIGRATION_STATUS_CANCELLING) {
 762         error_setg(errp, QERR_MIGRATION_ACTIVE);
 763         return;
 764     }
 765     if (runstate_check(RUN_STATE_INMIGRATE)) {
 766         error_setg(errp, "Guest is waiting for an incoming migration");
 767         return;
 768     }
 769
 770     if (qemu_savevm_state_blocked(errp)) {
 771         return;
 772     }
 773
 774     if (migration_blockers) {
 775         *errp = error_copy(migration_blockers->data);
 776         return;
 777     }
 778
 779     /* We are starting a new migration, so we want to start in a clean
 780        state.  This change is only needed if previous migration
 781        failed/was cancelled.  We don't use migrate_set_state() because
 782        we are setting the initial state, not changing it. */
 783     s->state = MIGRATION_STATUS_NONE;
 784
 785     s = migrate_init(&params);
 786
 787     if (strstart(uri, "tcp:", &p)) {
 788         tcp_start_outgoing_migration(s, p, &local_err);
 789 #ifdef CONFIG_RDMA
 790     } else if (strstart(uri, "rdma:", &p)) {
 791         rdma_start_outgoing_migration(s, p, &local_err);
 792 #endif
 793 #if !defined(WIN32)
 794     } else if (strstart(uri, "exec:", &p)) {
 795         exec_start_outgoing_migration(s, p, &local_err);
 796     } else if (strstart(uri, "unix:", &p)) {
 797         unix_start_outgoing_migration(s, p, &local_err);
 798     } else if (strstart(uri, "fd:", &p)) {
 799         fd_start_outgoing_migration(s, p, &local_err);
 800 #endif
 801     } else {
 802         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
 803                    "a valid migration protocol");
 804         migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_FAILED);
 805         return;
 806     }
 807
 808     if (local_err) {
 809         migrate_fd_error(s);
 810         error_propagate(errp, local_err);
 811         return;
 812     }
 813 }
 814
 815 void qmp_migrate_cancel(Error **errp)
 816 {
 817     migrate_fd_cancel(migrate_get_current());
 818 }
 819
 820 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 821 {
 822     MigrationState *s = migrate_get_current();
 823     int64_t new_size;
 824
 825     /* Check for truncation */
 826     if (value != (size_t)value) {
 827         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 828                    "exceeding address space");
 829         return;
 830     }
 831
 832     /* Cache should not be larger than guest ram size */
 833     if (value > ram_bytes_total()) {
 834         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 835                    "exceeds guest ram size ");
 836         return;
 837     }
 838
 839     new_size = xbzrle_cache_resize(value);
 840     if (new_size < 0) {
 841         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 842                    "is smaller than page size");
 843         return;
 844     }
 845
 846     s->xbzrle_cache_size = new_size;
 847 }
 848
 849 int64_t qmp_query_migrate_cache_size(Error **errp)
 850 {
 851     return migrate_xbzrle_cache_size();
 852 }
 853
 854 void qmp_migrate_set_speed(int64_t value, Error **errp)
 855 {
 856     MigrationState *s;
 857
 858     if (value < 0) {
 859         value = 0;
 860     }
 861     if (value > SIZE_MAX) {
 862         value = SIZE_MAX;
 863     }
 864
 865     s = migrate_get_current();
 866     s->bandwidth_limit = value;
 867     if (s->file) {
 868         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
 869     }
 870 }
 871
 872 void qmp_migrate_set_downtime(double value, Error **errp)
 873 {
 874     value *= 1e9;
 875     value = MAX(0, MIN(UINT64_MAX, value));
 876     max_downtime = (uint64_t)value;
 877 }
 878
 879 bool migrate_auto_converge(void)
 880 {
 881     MigrationState *s;
 882
 883     s = migrate_get_current();
 884
 885     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
 886 }
 887
 888 bool migrate_zero_blocks(void)
 889 {
 890     MigrationState *s;
 891
 892     s = migrate_get_current();
 893
 894     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
 895 }
 896
 897 bool migrate_use_compression(void)
 898 {
 899     MigrationState *s;
 900
 901     s = migrate_get_current();
 902
 903     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
 904 }
 905
 906 int migrate_compress_level(void)
 907 {
 908     MigrationState *s;
 909
 910     s = migrate_get_current();
 911
 912     return s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
 913 }
 914
 915 int migrate_compress_threads(void)
 916 {
 917     MigrationState *s;
 918
 919     s = migrate_get_current();
 920
 921     return s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
 922 }
 923
 924 int migrate_decompress_threads(void)
 925 {
 926     MigrationState *s;
 927
 928     s = migrate_get_current();
 929
 930     return s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
 931 }
 932
 933 bool migrate_use_events(void)
 934 {
 935     MigrationState *s;
 936
 937     s = migrate_get_current();
 938
 939     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
 940 }
 941
 942 int migrate_use_xbzrle(void)
 943 {
 944     MigrationState *s;
 945
 946     s = migrate_get_current();
 947
 948     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 949 }
 950
 951 int64_t migrate_xbzrle_cache_size(void)
 952 {
 953     MigrationState *s;
 954
 955     s = migrate_get_current();
 956
 957     return s->xbzrle_cache_size;
 958 }
 959
 960 /**
 961  * migration_completion: Used by migration_thread when there's not much left.
 962  *   The caller 'breaks' the loop when this returns.
 963  *
 964  * @s: Current migration state
 965  * @*old_vm_running: Pointer to old_vm_running flag
 966  * @*start_time: Pointer to time to update
 967  */
 968 static void migration_completion(MigrationState *s, bool *old_vm_running,
 969                                  int64_t *start_time)
 970 {
 971     int ret;
 972
 973     qemu_mutex_lock_iothread();
 974     *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
 975     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 976     *old_vm_running = runstate_is_running();
 977
 978     ret = global_state_store();
 979     if (!ret) {
 980         ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 981         if (ret >= 0) {
 982             qemu_file_set_rate_limit(s->file, INT64_MAX);
 983             qemu_savevm_state_complete(s->file);
 984         }
 985     }
 986     qemu_mutex_unlock_iothread();
 987
 988     if (ret < 0) {
 989         goto fail;
 990     }
 991
 992     if (qemu_file_get_error(s->file)) {
 993         trace_migration_completion_file_err();
 994         goto fail;
 995     }
 996
 997     migrate_set_state(s, MIGRATION_STATUS_ACTIVE, MIGRATION_STATUS_COMPLETED);
 998     return;
 999
1000 fail:
1001     migrate_set_state(s, MIGRATION_STATUS_ACTIVE, MIGRATION_STATUS_FAILED);
1002 }
1003
1004 /* migration thread support */
1005
1006 static void *migration_thread(void *opaque)
1007 {
1008     MigrationState *s = opaque;
1009     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1010     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1011     int64_t initial_bytes = 0;
1012     int64_t max_size = 0;
1013     int64_t start_time = initial_time;
1014     bool old_vm_running = false;
1015
1016     rcu_register_thread();
1017
1018     qemu_savevm_state_header(s->file);
1019     qemu_savevm_state_begin(s->file, &s->params);
1020
1021     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1022     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
1023
1024     while (s->state == MIGRATION_STATUS_ACTIVE) {
1025         int64_t current_time;
1026         uint64_t pending_size;
1027
1028         if (!qemu_file_rate_limit(s->file)) {
1029             pending_size = qemu_savevm_state_pending(s->file, max_size);
1030             trace_migrate_pending(pending_size, max_size);
1031             if (pending_size && pending_size >= max_size) {
1032                 qemu_savevm_state_iterate(s->file);
1033             } else {
1034                 trace_migration_thread_low_pending(pending_size);
1035                 migration_completion(s, &old_vm_running, &start_time);
1036                 break;
1037             }
1038         }
1039
1040         if (qemu_file_get_error(s->file)) {
1041             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
1042                               MIGRATION_STATUS_FAILED);
1043             break;
1044         }
1045         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1046         if (current_time >= initial_time + BUFFER_DELAY) {
1047             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
1048             uint64_t time_spent = current_time - initial_time;
1049             double bandwidth = transferred_bytes / time_spent;
1050             max_size = bandwidth * migrate_max_downtime() / 1000000;
1051
1052             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
1053                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
1054
1055             trace_migrate_transferred(transferred_bytes, time_spent,
1056                                       bandwidth, max_size);
1057             /* if we haven't sent anything, we don't want to recalculate
1058                10000 is a small enough number for our purposes */
1059             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
1060                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
1061             }
1062
1063             qemu_file_reset_rate_limit(s->file);
1064             initial_time = current_time;
1065             initial_bytes = qemu_ftell(s->file);
1066         }
1067         if (qemu_file_rate_limit(s->file)) {
1068             /* usleep expects microseconds */
1069             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
1070         }
1071     }
1072
1073     qemu_mutex_lock_iothread();
1074     if (s->state == MIGRATION_STATUS_COMPLETED) {
1075         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1076         uint64_t transferred_bytes = qemu_ftell(s->file);
1077         s->total_time = end_time - s->total_time;
1078         s->downtime = end_time - start_time;
1079         if (s->total_time) {
1080             s->mbps = (((double) transferred_bytes * 8.0) /
1081                        ((double) s->total_time)) / 1000;
1082         }
1083         runstate_set(RUN_STATE_POSTMIGRATE);
1084     } else {
1085         if (old_vm_running) {
1086             vm_start();
1087         }
1088     }
1089     qemu_bh_schedule(s->cleanup_bh);
1090     qemu_mutex_unlock_iothread();
1091
1092     rcu_unregister_thread();
1093     return NULL;
1094 }
1095
1096 void migrate_fd_connect(MigrationState *s)
1097 {
1098     /* This is a best 1st approximation. ns to ms */
1099     s->expected_downtime = max_downtime/1000000;
1100     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
1101
1102     qemu_file_set_rate_limit(s->file,
1103                              s->bandwidth_limit / XFER_LIMIT_RATIO);
1104
1105     /* Notify before starting migration thread */
1106     notifier_list_notify(&migration_state_notifiers, s);
1107
1108     migrate_compress_threads_create();
1109     qemu_thread_create(&s->thread, "migration", migration_thread, s,
1110                        QEMU_THREAD_JOINABLE);
1111 }