migration/colo.c

   1 /*
   2  * COarse-grain LOck-stepping Virtual Machines for Non-stop Service (COLO)
   3  * (a.k.a. Fault Tolerance or Continuous Replication)
   4  *
   5  * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
   6  * Copyright (c) 2016 FUJITSU LIMITED
   7  * Copyright (c) 2016 Intel Corporation
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or
  10  * later.  See the COPYING file in the top-level directory.
  11  */
  12
  13 #include "qemu/osdep.h"
  14 #include "sysemu/sysemu.h"
  15 #include "qapi/error.h"
  16 #include "qapi/qapi-commands-migration.h"
  17 #include "qemu-file-channel.h"
  18 #include "migration.h"
  19 #include "qemu-file.h"
  20 #include "savevm.h"
  21 #include "migration/colo.h"
  22 #include "block.h"
  23 #include "io/channel-buffer.h"
  24 #include "trace.h"
  25 #include "qemu/error-report.h"
  26 #include "migration/failover.h"
  27 #include "replication.h"
  28 #include "net/colo-compare.h"
  29 #include "net/colo.h"
  30 #include "block/block.h"
  31 #include "qapi/qapi-events-migration.h"
  32 #include "qapi/qmp/qerror.h"
  33 #include "sysemu/cpus.h"
  34 #include "net/filter.h"
  35
  36 static bool vmstate_loading;
  37 static Notifier packets_compare_notifier;
  38
  39 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
  40
  41 bool migration_in_colo_state(void)
  42 {
  43     MigrationState *s = migrate_get_current();
  44
  45     return (s->state == MIGRATION_STATUS_COLO);
  46 }
  47
  48 bool migration_incoming_in_colo_state(void)
  49 {
  50     MigrationIncomingState *mis = migration_incoming_get_current();
  51
  52     return mis && (mis->state == MIGRATION_STATUS_COLO);
  53 }
  54
  55 static bool colo_runstate_is_stopped(void)
  56 {
  57     return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
  58 }
  59
  60 static void secondary_vm_do_failover(void)
  61 {
  62     int old_state;
  63     MigrationIncomingState *mis = migration_incoming_get_current();
  64     Error *local_err = NULL;
  65
  66     /* Can not do failover during the process of VM's loading VMstate, Or
  67      * it will break the secondary VM.
  68      */
  69     if (vmstate_loading) {
  70         old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
  71                         FAILOVER_STATUS_RELAUNCH);
  72         if (old_state != FAILOVER_STATUS_ACTIVE) {
  73             error_report("Unknown error while do failover for secondary VM,"
  74                          "old_state: %s", FailoverStatus_str(old_state));
  75         }
  76         return;
  77     }
  78
  79     migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
  80                       MIGRATION_STATUS_COMPLETED);
  81
  82     replication_stop_all(true, &local_err);
  83     if (local_err) {
  84         error_report_err(local_err);
  85     }
  86
  87     /* Notify all filters of all NIC to do checkpoint */
  88     colo_notify_filters_event(COLO_EVENT_FAILOVER, &local_err);
  89     if (local_err) {
  90         error_report_err(local_err);
  91     }
  92
  93     if (!autostart) {
  94         error_report("\"-S\" qemu option will be ignored in secondary side");
  95         /* recover runstate to normal migration finish state */
  96         autostart = true;
  97     }
  98     /*
  99      * Make sure COLO incoming thread not block in recv or send,
 100      * If mis->from_src_file and mis->to_src_file use the same fd,
 101      * The second shutdown() will return -1, we ignore this value,
 102      * It is harmless.
 103      */
 104     if (mis->from_src_file) {
 105         qemu_file_shutdown(mis->from_src_file);
 106     }
 107     if (mis->to_src_file) {
 108         qemu_file_shutdown(mis->to_src_file);
 109     }
 110
 111     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 112                                    FAILOVER_STATUS_COMPLETED);
 113     if (old_state != FAILOVER_STATUS_ACTIVE) {
 114         error_report("Incorrect state (%s) while doing failover for "
 115                      "secondary VM", FailoverStatus_str(old_state));
 116         return;
 117     }
 118     /* Notify COLO incoming thread that failover work is finished */
 119     qemu_sem_post(&mis->colo_incoming_sem);
 120     /* For Secondary VM, jump to incoming co */
 121     if (mis->migration_incoming_co) {
 122         qemu_coroutine_enter(mis->migration_incoming_co);
 123     }
 124 }
 125
 126 static void primary_vm_do_failover(void)
 127 {
 128     MigrationState *s = migrate_get_current();
 129     int old_state;
 130     Error *local_err = NULL;
 131
 132     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
 133                       MIGRATION_STATUS_COMPLETED);
 134     /*
 135      * kick COLO thread which might wait at
 136      * qemu_sem_wait(&s->colo_checkpoint_sem).
 137      */
 138     colo_checkpoint_notify(migrate_get_current());
 139
 140     /*
 141      * Wake up COLO thread which may blocked in recv() or send(),
 142      * The s->rp_state.from_dst_file and s->to_dst_file may use the
 143      * same fd, but we still shutdown the fd for twice, it is harmless.
 144      */
 145     if (s->to_dst_file) {
 146         qemu_file_shutdown(s->to_dst_file);
 147     }
 148     if (s->rp_state.from_dst_file) {
 149         qemu_file_shutdown(s->rp_state.from_dst_file);
 150     }
 151
 152     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 153                                    FAILOVER_STATUS_COMPLETED);
 154     if (old_state != FAILOVER_STATUS_ACTIVE) {
 155         error_report("Incorrect state (%s) while doing failover for Primary VM",
 156                      FailoverStatus_str(old_state));
 157         return;
 158     }
 159
 160     replication_stop_all(true, &local_err);
 161     if (local_err) {
 162         error_report_err(local_err);
 163         local_err = NULL;
 164     }
 165
 166     /* Notify COLO thread that failover work is finished */
 167     qemu_sem_post(&s->colo_exit_sem);
 168 }
 169
 170 COLOMode get_colo_mode(void)
 171 {
 172     if (migration_in_colo_state()) {
 173         return COLO_MODE_PRIMARY;
 174     } else if (migration_incoming_in_colo_state()) {
 175         return COLO_MODE_SECONDARY;
 176     } else {
 177         return COLO_MODE_NONE;
 178     }
 179 }
 180
 181 void colo_do_failover(MigrationState *s)
 182 {
 183     /* Make sure VM stopped while failover happened. */
 184     if (!colo_runstate_is_stopped()) {
 185         vm_stop_force_state(RUN_STATE_COLO);
 186     }
 187
 188     if (get_colo_mode() == COLO_MODE_PRIMARY) {
 189         primary_vm_do_failover();
 190     } else {
 191         secondary_vm_do_failover();
 192     }
 193 }
 194
 195 void qmp_xen_set_replication(bool enable, bool primary,
 196                              bool has_failover, bool failover,
 197                              Error **errp)
 198 {
 199 #ifdef CONFIG_REPLICATION
 200     ReplicationMode mode = primary ?
 201                            REPLICATION_MODE_PRIMARY :
 202                            REPLICATION_MODE_SECONDARY;
 203
 204     if (has_failover && enable) {
 205         error_setg(errp, "Parameter 'failover' is only for"
 206                    " stopping replication");
 207         return;
 208     }
 209
 210     if (enable) {
 211         replication_start_all(mode, errp);
 212     } else {
 213         if (!has_failover) {
 214             failover = NULL;
 215         }
 216         replication_stop_all(failover, failover ? NULL : errp);
 217     }
 218 #else
 219     abort();
 220 #endif
 221 }
 222
 223 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
 224 {
 225 #ifdef CONFIG_REPLICATION
 226     Error *err = NULL;
 227     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
 228
 229     replication_get_error_all(&err);
 230     if (err) {
 231         s->error = true;
 232         s->has_desc = true;
 233         s->desc = g_strdup(error_get_pretty(err));
 234     } else {
 235         s->error = false;
 236     }
 237
 238     error_free(err);
 239     return s;
 240 #else
 241     abort();
 242 #endif
 243 }
 244
 245 void qmp_xen_colo_do_checkpoint(Error **errp)
 246 {
 247 #ifdef CONFIG_REPLICATION
 248     replication_do_checkpoint_all(errp);
 249 #else
 250     abort();
 251 #endif
 252 }
 253
 254 COLOStatus *qmp_query_colo_status(Error **errp)
 255 {
 256     COLOStatus *s = g_new0(COLOStatus, 1);
 257
 258     s->mode = get_colo_mode();
 259
 260     switch (failover_get_state()) {
 261     case FAILOVER_STATUS_NONE:
 262         s->reason = COLO_EXIT_REASON_NONE;
 263         break;
 264     case FAILOVER_STATUS_REQUIRE:
 265         s->reason = COLO_EXIT_REASON_REQUEST;
 266         break;
 267     default:
 268         s->reason = COLO_EXIT_REASON_ERROR;
 269     }
 270
 271     return s;
 272 }
 273
 274 static void colo_send_message(QEMUFile *f, COLOMessage msg,
 275                               Error **errp)
 276 {
 277     int ret;
 278
 279     if (msg >= COLO_MESSAGE__MAX) {
 280         error_setg(errp, "%s: Invalid message", __func__);
 281         return;
 282     }
 283     qemu_put_be32(f, msg);
 284     qemu_fflush(f);
 285
 286     ret = qemu_file_get_error(f);
 287     if (ret < 0) {
 288         error_setg_errno(errp, -ret, "Can't send COLO message");
 289     }
 290     trace_colo_send_message(COLOMessage_str(msg));
 291 }
 292
 293 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
 294                                     uint64_t value, Error **errp)
 295 {
 296     Error *local_err = NULL;
 297     int ret;
 298
 299     colo_send_message(f, msg, &local_err);
 300     if (local_err) {
 301         error_propagate(errp, local_err);
 302         return;
 303     }
 304     qemu_put_be64(f, value);
 305     qemu_fflush(f);
 306
 307     ret = qemu_file_get_error(f);
 308     if (ret < 0) {
 309         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
 310                          COLOMessage_str(msg));
 311     }
 312 }
 313
 314 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
 315 {
 316     COLOMessage msg;
 317     int ret;
 318
 319     msg = qemu_get_be32(f);
 320     ret = qemu_file_get_error(f);
 321     if (ret < 0) {
 322         error_setg_errno(errp, -ret, "Can't receive COLO message");
 323         return msg;
 324     }
 325     if (msg >= COLO_MESSAGE__MAX) {
 326         error_setg(errp, "%s: Invalid message", __func__);
 327         return msg;
 328     }
 329     trace_colo_receive_message(COLOMessage_str(msg));
 330     return msg;
 331 }
 332
 333 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
 334                                        Error **errp)
 335 {
 336     COLOMessage msg;
 337     Error *local_err = NULL;
 338
 339     msg = colo_receive_message(f, &local_err);
 340     if (local_err) {
 341         error_propagate(errp, local_err);
 342         return;
 343     }
 344     if (msg != expect_msg) {
 345         error_setg(errp, "Unexpected COLO message %d, expected %d",
 346                           msg, expect_msg);
 347     }
 348 }
 349
 350 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
 351                                            Error **errp)
 352 {
 353     Error *local_err = NULL;
 354     uint64_t value;
 355     int ret;
 356
 357     colo_receive_check_message(f, expect_msg, &local_err);
 358     if (local_err) {
 359         error_propagate(errp, local_err);
 360         return 0;
 361     }
 362
 363     value = qemu_get_be64(f);
 364     ret = qemu_file_get_error(f);
 365     if (ret < 0) {
 366         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
 367                          COLOMessage_str(expect_msg));
 368     }
 369     return value;
 370 }
 371
 372 static int colo_do_checkpoint_transaction(MigrationState *s,
 373                                           QIOChannelBuffer *bioc,
 374                                           QEMUFile *fb)
 375 {
 376     Error *local_err = NULL;
 377     int ret = -1;
 378
 379     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
 380                       &local_err);
 381     if (local_err) {
 382         goto out;
 383     }
 384
 385     colo_receive_check_message(s->rp_state.from_dst_file,
 386                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
 387     if (local_err) {
 388         goto out;
 389     }
 390     /* Reset channel-buffer directly */
 391     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 392     bioc->usage = 0;
 393
 394     qemu_mutex_lock_iothread();
 395     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 396         qemu_mutex_unlock_iothread();
 397         goto out;
 398     }
 399     vm_stop_force_state(RUN_STATE_COLO);
 400     qemu_mutex_unlock_iothread();
 401     trace_colo_vm_state_change("run", "stop");
 402     /*
 403      * Failover request bh could be called after vm_stop_force_state(),
 404      * So we need check failover_request_is_active() again.
 405      */
 406     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 407         goto out;
 408     }
 409
 410     colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
 411     if (local_err) {
 412         goto out;
 413     }
 414
 415     /* Disable block migration */
 416     migrate_set_block_enabled(false, &local_err);
 417     qemu_mutex_lock_iothread();
 418     replication_do_checkpoint_all(&local_err);
 419     if (local_err) {
 420         qemu_mutex_unlock_iothread();
 421         goto out;
 422     }
 423
 424     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
 425     if (local_err) {
 426         qemu_mutex_unlock_iothread();
 427         goto out;
 428     }
 429     /* Note: device state is saved into buffer */
 430     ret = qemu_save_device_state(fb);
 431
 432     qemu_mutex_unlock_iothread();
 433     if (ret < 0) {
 434         goto out;
 435     }
 436     /*
 437      * Only save VM's live state, which not including device state.
 438      * TODO: We may need a timeout mechanism to prevent COLO process
 439      * to be blocked here.
 440      */
 441     qemu_savevm_live_state(s->to_dst_file);
 442
 443     qemu_fflush(fb);
 444
 445     /*
 446      * We need the size of the VMstate data in Secondary side,
 447      * With which we can decide how much data should be read.
 448      */
 449     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
 450                             bioc->usage, &local_err);
 451     if (local_err) {
 452         goto out;
 453     }
 454
 455     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
 456     qemu_fflush(s->to_dst_file);
 457     ret = qemu_file_get_error(s->to_dst_file);
 458     if (ret < 0) {
 459         goto out;
 460     }
 461
 462     colo_receive_check_message(s->rp_state.from_dst_file,
 463                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
 464     if (local_err) {
 465         goto out;
 466     }
 467
 468     colo_receive_check_message(s->rp_state.from_dst_file,
 469                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
 470     if (local_err) {
 471         goto out;
 472     }
 473
 474     ret = 0;
 475
 476     qemu_mutex_lock_iothread();
 477     vm_start();
 478     qemu_mutex_unlock_iothread();
 479     trace_colo_vm_state_change("stop", "run");
 480
 481 out:
 482     if (local_err) {
 483         error_report_err(local_err);
 484     }
 485     return ret;
 486 }
 487
 488 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
 489 {
 490     colo_checkpoint_notify(data);
 491 }
 492
 493 static void colo_process_checkpoint(MigrationState *s)
 494 {
 495     QIOChannelBuffer *bioc;
 496     QEMUFile *fb = NULL;
 497     int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 498     Error *local_err = NULL;
 499     int ret;
 500
 501     failover_init_state();
 502
 503     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
 504     if (!s->rp_state.from_dst_file) {
 505         error_report("Open QEMUFile from_dst_file failed");
 506         goto out;
 507     }
 508
 509     packets_compare_notifier.notify = colo_compare_notify_checkpoint;
 510     colo_compare_register_notifier(&packets_compare_notifier);
 511
 512     /*
 513      * Wait for Secondary finish loading VM states and enter COLO
 514      * restore.
 515      */
 516     colo_receive_check_message(s->rp_state.from_dst_file,
 517                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
 518     if (local_err) {
 519         goto out;
 520     }
 521     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 522     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
 523     object_unref(OBJECT(bioc));
 524
 525     qemu_mutex_lock_iothread();
 526     replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
 527     if (local_err) {
 528         qemu_mutex_unlock_iothread();
 529         goto out;
 530     }
 531
 532     vm_start();
 533     qemu_mutex_unlock_iothread();
 534     trace_colo_vm_state_change("stop", "run");
 535
 536     timer_mod(s->colo_delay_timer,
 537             current_time + s->parameters.x_checkpoint_delay);
 538
 539     while (s->state == MIGRATION_STATUS_COLO) {
 540         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 541             error_report("failover request");
 542             goto out;
 543         }
 544
 545         qemu_sem_wait(&s->colo_checkpoint_sem);
 546
 547         if (s->state != MIGRATION_STATUS_COLO) {
 548             goto out;
 549         }
 550         ret = colo_do_checkpoint_transaction(s, bioc, fb);
 551         if (ret < 0) {
 552             goto out;
 553         }
 554     }
 555
 556 out:
 557     /* Throw the unreported error message after exited from loop */
 558     if (local_err) {
 559         error_report_err(local_err);
 560     }
 561
 562     if (fb) {
 563         qemu_fclose(fb);
 564     }
 565
 566     /*
 567      * There are only two reasons we can get here, some error happened
 568      * or the user triggered failover.
 569      */
 570     switch (failover_get_state()) {
 571     case FAILOVER_STATUS_NONE:
 572         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 573                                   COLO_EXIT_REASON_ERROR);
 574         break;
 575     case FAILOVER_STATUS_REQUIRE:
 576         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 577                                   COLO_EXIT_REASON_REQUEST);
 578         break;
 579     default:
 580         abort();
 581     }
 582
 583     /* Hope this not to be too long to wait here */
 584     qemu_sem_wait(&s->colo_exit_sem);
 585     qemu_sem_destroy(&s->colo_exit_sem);
 586
 587     /*
 588      * It is safe to unregister notifier after failover finished.
 589      * Besides, colo_delay_timer and colo_checkpoint_sem can't be
 590      * released befor unregister notifier, or there will be use-after-free
 591      * error.
 592      */
 593     colo_compare_unregister_notifier(&packets_compare_notifier);
 594     timer_del(s->colo_delay_timer);
 595     timer_free(s->colo_delay_timer);
 596     qemu_sem_destroy(&s->colo_checkpoint_sem);
 597
 598     /*
 599      * Must be called after failover BH is completed,
 600      * Or the failover BH may shutdown the wrong fd that
 601      * re-used by other threads after we release here.
 602      */
 603     if (s->rp_state.from_dst_file) {
 604         qemu_fclose(s->rp_state.from_dst_file);
 605     }
 606 }
 607
 608 void colo_checkpoint_notify(void *opaque)
 609 {
 610     MigrationState *s = opaque;
 611     int64_t next_notify_time;
 612
 613     qemu_sem_post(&s->colo_checkpoint_sem);
 614     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 615     next_notify_time = s->colo_checkpoint_time +
 616                     s->parameters.x_checkpoint_delay;
 617     timer_mod(s->colo_delay_timer, next_notify_time);
 618 }
 619
 620 void migrate_start_colo_process(MigrationState *s)
 621 {
 622     qemu_mutex_unlock_iothread();
 623     qemu_sem_init(&s->colo_checkpoint_sem, 0);
 624     s->colo_delay_timer =  timer_new_ms(QEMU_CLOCK_HOST,
 625                                 colo_checkpoint_notify, s);
 626
 627     qemu_sem_init(&s->colo_exit_sem, 0);
 628     migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
 629                       MIGRATION_STATUS_COLO);
 630     colo_process_checkpoint(s);
 631     qemu_mutex_lock_iothread();
 632 }
 633
 634 static void colo_wait_handle_message(QEMUFile *f, int *checkpoint_request,
 635                                      Error **errp)
 636 {
 637     COLOMessage msg;
 638     Error *local_err = NULL;
 639
 640     msg = colo_receive_message(f, &local_err);
 641     if (local_err) {
 642         error_propagate(errp, local_err);
 643         return;
 644     }
 645
 646     switch (msg) {
 647     case COLO_MESSAGE_CHECKPOINT_REQUEST:
 648         *checkpoint_request = 1;
 649         break;
 650     default:
 651         *checkpoint_request = 0;
 652         error_setg(errp, "Got unknown COLO message: %d", msg);
 653         break;
 654     }
 655 }
 656
 657 void *colo_process_incoming_thread(void *opaque)
 658 {
 659     MigrationIncomingState *mis = opaque;
 660     QEMUFile *fb = NULL;
 661     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
 662     uint64_t total_size;
 663     uint64_t value;
 664     Error *local_err = NULL;
 665     int ret;
 666
 667     rcu_register_thread();
 668     qemu_sem_init(&mis->colo_incoming_sem, 0);
 669
 670     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 671                       MIGRATION_STATUS_COLO);
 672
 673     failover_init_state();
 674
 675     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
 676     if (!mis->to_src_file) {
 677         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
 678         goto out;
 679     }
 680     /*
 681      * Note: the communication between Primary side and Secondary side
 682      * should be sequential, we set the fd to unblocked in migration incoming
 683      * coroutine, and here we are in the COLO incoming thread, so it is ok to
 684      * set the fd back to blocked.
 685      */
 686     qemu_file_set_blocking(mis->from_src_file, true);
 687
 688     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 689     fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
 690     object_unref(OBJECT(bioc));
 691
 692     qemu_mutex_lock_iothread();
 693     replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
 694     if (local_err) {
 695         qemu_mutex_unlock_iothread();
 696         goto out;
 697     }
 698     vm_start();
 699     trace_colo_vm_state_change("stop", "run");
 700     qemu_mutex_unlock_iothread();
 701
 702     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
 703                       &local_err);
 704     if (local_err) {
 705         goto out;
 706     }
 707
 708     while (mis->state == MIGRATION_STATUS_COLO) {
 709         int request = 0;
 710
 711         colo_wait_handle_message(mis->from_src_file, &request, &local_err);
 712         if (local_err) {
 713             goto out;
 714         }
 715         assert(request);
 716         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 717             error_report("failover request");
 718             goto out;
 719         }
 720
 721         qemu_mutex_lock_iothread();
 722         vm_stop_force_state(RUN_STATE_COLO);
 723         trace_colo_vm_state_change("run", "stop");
 724         qemu_mutex_unlock_iothread();
 725
 726         /* FIXME: This is unnecessary for periodic checkpoint mode */
 727         colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
 728                      &local_err);
 729         if (local_err) {
 730             goto out;
 731         }
 732
 733         colo_receive_check_message(mis->from_src_file,
 734                            COLO_MESSAGE_VMSTATE_SEND, &local_err);
 735         if (local_err) {
 736             goto out;
 737         }
 738
 739         qemu_mutex_lock_iothread();
 740         cpu_synchronize_all_pre_loadvm();
 741         ret = qemu_loadvm_state_main(mis->from_src_file, mis);
 742         qemu_mutex_unlock_iothread();
 743
 744         if (ret < 0) {
 745             error_report("Load VM's live state (ram) error");
 746             goto out;
 747         }
 748
 749         value = colo_receive_message_value(mis->from_src_file,
 750                                  COLO_MESSAGE_VMSTATE_SIZE, &local_err);
 751         if (local_err) {
 752             goto out;
 753         }
 754
 755         /*
 756          * Read VM device state data into channel buffer,
 757          * It's better to re-use the memory allocated.
 758          * Here we need to handle the channel buffer directly.
 759          */
 760         if (value > bioc->capacity) {
 761             bioc->capacity = value;
 762             bioc->data = g_realloc(bioc->data, bioc->capacity);
 763         }
 764         total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
 765         if (total_size != value) {
 766             error_report("Got %" PRIu64 " VMState data, less than expected"
 767                         " %" PRIu64, total_size, value);
 768             goto out;
 769         }
 770         bioc->usage = total_size;
 771         qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 772
 773         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
 774                      &local_err);
 775         if (local_err) {
 776             goto out;
 777         }
 778
 779         qemu_mutex_lock_iothread();
 780         vmstate_loading = true;
 781         ret = qemu_load_device_state(fb);
 782         if (ret < 0) {
 783             error_report("COLO: load device state failed");
 784             qemu_mutex_unlock_iothread();
 785             goto out;
 786         }
 787
 788         replication_get_error_all(&local_err);
 789         if (local_err) {
 790             qemu_mutex_unlock_iothread();
 791             goto out;
 792         }
 793         /* discard colo disk buffer */
 794         replication_do_checkpoint_all(&local_err);
 795         if (local_err) {
 796             qemu_mutex_unlock_iothread();
 797             goto out;
 798         }
 799
 800         /* Notify all filters of all NIC to do checkpoint */
 801         colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
 802
 803         if (local_err) {
 804             qemu_mutex_unlock_iothread();
 805             goto out;
 806         }
 807
 808         vmstate_loading = false;
 809         vm_start();
 810         trace_colo_vm_state_change("stop", "run");
 811         qemu_mutex_unlock_iothread();
 812
 813         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 814             failover_set_state(FAILOVER_STATUS_RELAUNCH,
 815                             FAILOVER_STATUS_NONE);
 816             failover_request_active(NULL);
 817             goto out;
 818         }
 819
 820         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
 821                      &local_err);
 822         if (local_err) {
 823             goto out;
 824         }
 825     }
 826
 827 out:
 828     vmstate_loading = false;
 829     /* Throw the unreported error message after exited from loop */
 830     if (local_err) {
 831         error_report_err(local_err);
 832     }
 833
 834     switch (failover_get_state()) {
 835     case FAILOVER_STATUS_NONE:
 836         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 837                                   COLO_EXIT_REASON_ERROR);
 838         break;
 839     case FAILOVER_STATUS_REQUIRE:
 840         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 841                                   COLO_EXIT_REASON_REQUEST);
 842         break;
 843     default:
 844         abort();
 845     }
 846
 847     if (fb) {
 848         qemu_fclose(fb);
 849     }
 850
 851     /* Hope this not to be too long to loop here */
 852     qemu_sem_wait(&mis->colo_incoming_sem);
 853     qemu_sem_destroy(&mis->colo_incoming_sem);
 854     /* Must be called after failover BH is completed */
 855     if (mis->to_src_file) {
 856         qemu_fclose(mis->to_src_file);
 857     }
 858     migration_incoming_disable_colo();
 859
 860     rcu_unregister_thread();
 861     return NULL;
 862 }