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 "qemu/main-loop.h"
  27 #include "qemu/rcu.h"
  28 #include "migration/failover.h"
  29 #include "migration/ram.h"
  30 #ifdef CONFIG_REPLICATION
  31 #include "replication.h"
  32 #endif
  33 #include "net/colo-compare.h"
  34 #include "net/colo.h"
  35 #include "block/block.h"
  36 #include "qapi/qapi-events-migration.h"
  37 #include "qapi/qmp/qerror.h"
  38 #include "sysemu/cpus.h"
  39 #include "sysemu/runstate.h"
  40 #include "net/filter.h"
  41
  42 static bool vmstate_loading;
  43 static Notifier packets_compare_notifier;
  44
  45 /* User need to know colo mode after COLO failover */
  46 static COLOMode last_colo_mode;
  47
  48 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
  49
  50 bool migration_in_colo_state(void)
  51 {
  52     MigrationState *s = migrate_get_current();
  53
  54     return (s->state == MIGRATION_STATUS_COLO);
  55 }
  56
  57 bool migration_incoming_in_colo_state(void)
  58 {
  59     MigrationIncomingState *mis = migration_incoming_get_current();
  60
  61     return mis && (mis->state == MIGRATION_STATUS_COLO);
  62 }
  63
  64 static bool colo_runstate_is_stopped(void)
  65 {
  66     return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
  67 }
  68
  69 static void secondary_vm_do_failover(void)
  70 {
  71 /* COLO needs enable block-replication */
  72 #ifdef CONFIG_REPLICATION
  73     int old_state;
  74     MigrationIncomingState *mis = migration_incoming_get_current();
  75     Error *local_err = NULL;
  76
  77     /* Can not do failover during the process of VM's loading VMstate, Or
  78      * it will break the secondary VM.
  79      */
  80     if (vmstate_loading) {
  81         old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
  82                         FAILOVER_STATUS_RELAUNCH);
  83         if (old_state != FAILOVER_STATUS_ACTIVE) {
  84             error_report("Unknown error while do failover for secondary VM,"
  85                          "old_state: %s", FailoverStatus_str(old_state));
  86         }
  87         return;
  88     }
  89
  90     migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
  91                       MIGRATION_STATUS_COMPLETED);
  92
  93     replication_stop_all(true, &local_err);
  94     if (local_err) {
  95         error_report_err(local_err);
  96         local_err = NULL;
  97     }
  98
  99     /* Notify all filters of all NIC to do checkpoint */
 100     colo_notify_filters_event(COLO_EVENT_FAILOVER, &local_err);
 101     if (local_err) {
 102         error_report_err(local_err);
 103     }
 104
 105     if (!autostart) {
 106         error_report("\"-S\" qemu option will be ignored in secondary side");
 107         /* recover runstate to normal migration finish state */
 108         autostart = true;
 109     }
 110     /*
 111      * Make sure COLO incoming thread not block in recv or send,
 112      * If mis->from_src_file and mis->to_src_file use the same fd,
 113      * The second shutdown() will return -1, we ignore this value,
 114      * It is harmless.
 115      */
 116     if (mis->from_src_file) {
 117         qemu_file_shutdown(mis->from_src_file);
 118     }
 119     if (mis->to_src_file) {
 120         qemu_file_shutdown(mis->to_src_file);
 121     }
 122
 123     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 124                                    FAILOVER_STATUS_COMPLETED);
 125     if (old_state != FAILOVER_STATUS_ACTIVE) {
 126         error_report("Incorrect state (%s) while doing failover for "
 127                      "secondary VM", FailoverStatus_str(old_state));
 128         return;
 129     }
 130     /* Notify COLO incoming thread that failover work is finished */
 131     qemu_sem_post(&mis->colo_incoming_sem);
 132
 133     /* For Secondary VM, jump to incoming co */
 134     if (mis->migration_incoming_co) {
 135         qemu_coroutine_enter(mis->migration_incoming_co);
 136     }
 137 #else
 138     abort();
 139 #endif
 140 }
 141
 142 static void primary_vm_do_failover(void)
 143 {
 144 #ifdef CONFIG_REPLICATION
 145     MigrationState *s = migrate_get_current();
 146     int old_state;
 147     Error *local_err = NULL;
 148
 149     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
 150                       MIGRATION_STATUS_COMPLETED);
 151     /*
 152      * kick COLO thread which might wait at
 153      * qemu_sem_wait(&s->colo_checkpoint_sem).
 154      */
 155     colo_checkpoint_notify(migrate_get_current());
 156
 157     /*
 158      * Wake up COLO thread which may blocked in recv() or send(),
 159      * The s->rp_state.from_dst_file and s->to_dst_file may use the
 160      * same fd, but we still shutdown the fd for twice, it is harmless.
 161      */
 162     if (s->to_dst_file) {
 163         qemu_file_shutdown(s->to_dst_file);
 164     }
 165     if (s->rp_state.from_dst_file) {
 166         qemu_file_shutdown(s->rp_state.from_dst_file);
 167     }
 168
 169     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 170                                    FAILOVER_STATUS_COMPLETED);
 171     if (old_state != FAILOVER_STATUS_ACTIVE) {
 172         error_report("Incorrect state (%s) while doing failover for Primary VM",
 173                      FailoverStatus_str(old_state));
 174         return;
 175     }
 176
 177     replication_stop_all(true, &local_err);
 178     if (local_err) {
 179         error_report_err(local_err);
 180         local_err = NULL;
 181     }
 182
 183     /* Notify COLO thread that failover work is finished */
 184     qemu_sem_post(&s->colo_exit_sem);
 185 #else
 186     abort();
 187 #endif
 188 }
 189
 190 COLOMode get_colo_mode(void)
 191 {
 192     if (migration_in_colo_state()) {
 193         return COLO_MODE_PRIMARY;
 194     } else if (migration_incoming_in_colo_state()) {
 195         return COLO_MODE_SECONDARY;
 196     } else {
 197         return COLO_MODE_NONE;
 198     }
 199 }
 200
 201 void colo_do_failover(void)
 202 {
 203     /* Make sure VM stopped while failover happened. */
 204     if (!colo_runstate_is_stopped()) {
 205         vm_stop_force_state(RUN_STATE_COLO);
 206     }
 207
 208     switch (get_colo_mode()) {
 209     case COLO_MODE_PRIMARY:
 210         primary_vm_do_failover();
 211         break;
 212     case COLO_MODE_SECONDARY:
 213         secondary_vm_do_failover();
 214         break;
 215     default:
 216         error_report("colo_do_failover failed because the colo mode"
 217                      " could not be obtained");
 218     }
 219 }
 220
 221 #ifdef CONFIG_REPLICATION
 222 void qmp_xen_set_replication(bool enable, bool primary,
 223                              bool has_failover, bool failover,
 224                              Error **errp)
 225 {
 226     ReplicationMode mode = primary ?
 227                            REPLICATION_MODE_PRIMARY :
 228                            REPLICATION_MODE_SECONDARY;
 229
 230     if (has_failover && enable) {
 231         error_setg(errp, "Parameter 'failover' is only for"
 232                    " stopping replication");
 233         return;
 234     }
 235
 236     if (enable) {
 237         replication_start_all(mode, errp);
 238     } else {
 239         if (!has_failover) {
 240             failover = NULL;
 241         }
 242         replication_stop_all(failover, failover ? NULL : errp);
 243     }
 244 }
 245
 246 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
 247 {
 248     Error *err = NULL;
 249     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
 250
 251     replication_get_error_all(&err);
 252     if (err) {
 253         s->error = true;
 254         s->has_desc = true;
 255         s->desc = g_strdup(error_get_pretty(err));
 256     } else {
 257         s->error = false;
 258     }
 259
 260     error_free(err);
 261     return s;
 262 }
 263
 264 void qmp_xen_colo_do_checkpoint(Error **errp)
 265 {
 266     Error *err = NULL;
 267
 268     replication_do_checkpoint_all(&err);
 269     if (err) {
 270         error_propagate(errp, err);
 271         return;
 272     }
 273     /* Notify all filters of all NIC to do checkpoint */
 274     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, errp);
 275 }
 276 #endif
 277
 278 COLOStatus *qmp_query_colo_status(Error **errp)
 279 {
 280     COLOStatus *s = g_new0(COLOStatus, 1);
 281
 282     s->mode = get_colo_mode();
 283     s->last_mode = last_colo_mode;
 284
 285     switch (failover_get_state()) {
 286     case FAILOVER_STATUS_NONE:
 287         s->reason = COLO_EXIT_REASON_NONE;
 288         break;
 289     case FAILOVER_STATUS_COMPLETED:
 290         s->reason = COLO_EXIT_REASON_REQUEST;
 291         break;
 292     default:
 293         if (migration_in_colo_state()) {
 294             s->reason = COLO_EXIT_REASON_PROCESSING;
 295         } else {
 296             s->reason = COLO_EXIT_REASON_ERROR;
 297         }
 298     }
 299
 300     return s;
 301 }
 302
 303 static void colo_send_message(QEMUFile *f, COLOMessage msg,
 304                               Error **errp)
 305 {
 306     int ret;
 307
 308     if (msg >= COLO_MESSAGE__MAX) {
 309         error_setg(errp, "%s: Invalid message", __func__);
 310         return;
 311     }
 312     qemu_put_be32(f, msg);
 313     qemu_fflush(f);
 314
 315     ret = qemu_file_get_error(f);
 316     if (ret < 0) {
 317         error_setg_errno(errp, -ret, "Can't send COLO message");
 318     }
 319     trace_colo_send_message(COLOMessage_str(msg));
 320 }
 321
 322 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
 323                                     uint64_t value, Error **errp)
 324 {
 325     Error *local_err = NULL;
 326     int ret;
 327
 328     colo_send_message(f, msg, &local_err);
 329     if (local_err) {
 330         error_propagate(errp, local_err);
 331         return;
 332     }
 333     qemu_put_be64(f, value);
 334     qemu_fflush(f);
 335
 336     ret = qemu_file_get_error(f);
 337     if (ret < 0) {
 338         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
 339                          COLOMessage_str(msg));
 340     }
 341 }
 342
 343 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
 344 {
 345     COLOMessage msg;
 346     int ret;
 347
 348     msg = qemu_get_be32(f);
 349     ret = qemu_file_get_error(f);
 350     if (ret < 0) {
 351         error_setg_errno(errp, -ret, "Can't receive COLO message");
 352         return msg;
 353     }
 354     if (msg >= COLO_MESSAGE__MAX) {
 355         error_setg(errp, "%s: Invalid message", __func__);
 356         return msg;
 357     }
 358     trace_colo_receive_message(COLOMessage_str(msg));
 359     return msg;
 360 }
 361
 362 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
 363                                        Error **errp)
 364 {
 365     COLOMessage msg;
 366     Error *local_err = NULL;
 367
 368     msg = colo_receive_message(f, &local_err);
 369     if (local_err) {
 370         error_propagate(errp, local_err);
 371         return;
 372     }
 373     if (msg != expect_msg) {
 374         error_setg(errp, "Unexpected COLO message %d, expected %d",
 375                           msg, expect_msg);
 376     }
 377 }
 378
 379 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
 380                                            Error **errp)
 381 {
 382     Error *local_err = NULL;
 383     uint64_t value;
 384     int ret;
 385
 386     colo_receive_check_message(f, expect_msg, &local_err);
 387     if (local_err) {
 388         error_propagate(errp, local_err);
 389         return 0;
 390     }
 391
 392     value = qemu_get_be64(f);
 393     ret = qemu_file_get_error(f);
 394     if (ret < 0) {
 395         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
 396                          COLOMessage_str(expect_msg));
 397     }
 398     return value;
 399 }
 400
 401 static int colo_do_checkpoint_transaction(MigrationState *s,
 402                                           QIOChannelBuffer *bioc,
 403                                           QEMUFile *fb)
 404 {
 405     Error *local_err = NULL;
 406     int ret = -1;
 407
 408     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
 409                       &local_err);
 410     if (local_err) {
 411         goto out;
 412     }
 413
 414     colo_receive_check_message(s->rp_state.from_dst_file,
 415                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
 416     if (local_err) {
 417         goto out;
 418     }
 419     /* Reset channel-buffer directly */
 420     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 421     bioc->usage = 0;
 422
 423     qemu_mutex_lock_iothread();
 424     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 425         qemu_mutex_unlock_iothread();
 426         goto out;
 427     }
 428     vm_stop_force_state(RUN_STATE_COLO);
 429     qemu_mutex_unlock_iothread();
 430     trace_colo_vm_state_change("run", "stop");
 431     /*
 432      * Failover request bh could be called after vm_stop_force_state(),
 433      * So we need check failover_request_is_active() again.
 434      */
 435     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 436         goto out;
 437     }
 438
 439     /* Disable block migration */
 440     migrate_set_block_enabled(false, &local_err);
 441     if (local_err) {
 442         goto out;
 443     }
 444     qemu_mutex_lock_iothread();
 445
 446 #ifdef CONFIG_REPLICATION
 447     replication_do_checkpoint_all(&local_err);
 448     if (local_err) {
 449         qemu_mutex_unlock_iothread();
 450         goto out;
 451     }
 452 #else
 453         abort();
 454 #endif
 455
 456     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
 457     if (local_err) {
 458         qemu_mutex_unlock_iothread();
 459         goto out;
 460     }
 461     /* Note: device state is saved into buffer */
 462     ret = qemu_save_device_state(fb);
 463
 464     qemu_mutex_unlock_iothread();
 465     if (ret < 0) {
 466         goto out;
 467     }
 468     /*
 469      * Only save VM's live state, which not including device state.
 470      * TODO: We may need a timeout mechanism to prevent COLO process
 471      * to be blocked here.
 472      */
 473     qemu_savevm_live_state(s->to_dst_file);
 474
 475     qemu_fflush(fb);
 476
 477     /*
 478      * We need the size of the VMstate data in Secondary side,
 479      * With which we can decide how much data should be read.
 480      */
 481     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
 482                             bioc->usage, &local_err);
 483     if (local_err) {
 484         goto out;
 485     }
 486
 487     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
 488     qemu_fflush(s->to_dst_file);
 489     ret = qemu_file_get_error(s->to_dst_file);
 490     if (ret < 0) {
 491         goto out;
 492     }
 493
 494     colo_receive_check_message(s->rp_state.from_dst_file,
 495                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
 496     if (local_err) {
 497         goto out;
 498     }
 499
 500     qemu_event_reset(&s->colo_checkpoint_event);
 501     colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
 502     if (local_err) {
 503         goto out;
 504     }
 505
 506     colo_receive_check_message(s->rp_state.from_dst_file,
 507                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
 508     if (local_err) {
 509         goto out;
 510     }
 511
 512     ret = 0;
 513
 514     qemu_mutex_lock_iothread();
 515     vm_start();
 516     qemu_mutex_unlock_iothread();
 517     trace_colo_vm_state_change("stop", "run");
 518
 519 out:
 520     if (local_err) {
 521         error_report_err(local_err);
 522     }
 523     return ret;
 524 }
 525
 526 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
 527 {
 528     colo_checkpoint_notify(data);
 529 }
 530
 531 static void colo_process_checkpoint(MigrationState *s)
 532 {
 533     QIOChannelBuffer *bioc;
 534     QEMUFile *fb = NULL;
 535     int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 536     Error *local_err = NULL;
 537     int ret;
 538
 539     last_colo_mode = get_colo_mode();
 540     if (last_colo_mode != COLO_MODE_PRIMARY) {
 541         error_report("COLO mode must be COLO_MODE_PRIMARY");
 542         return;
 543     }
 544
 545     failover_init_state();
 546
 547     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
 548     if (!s->rp_state.from_dst_file) {
 549         error_report("Open QEMUFile from_dst_file failed");
 550         goto out;
 551     }
 552
 553     packets_compare_notifier.notify = colo_compare_notify_checkpoint;
 554     colo_compare_register_notifier(&packets_compare_notifier);
 555
 556     /*
 557      * Wait for Secondary finish loading VM states and enter COLO
 558      * restore.
 559      */
 560     colo_receive_check_message(s->rp_state.from_dst_file,
 561                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
 562     if (local_err) {
 563         goto out;
 564     }
 565     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 566     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
 567     object_unref(OBJECT(bioc));
 568
 569     qemu_mutex_lock_iothread();
 570 #ifdef CONFIG_REPLICATION
 571     replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
 572     if (local_err) {
 573         qemu_mutex_unlock_iothread();
 574         goto out;
 575     }
 576 #else
 577         abort();
 578 #endif
 579
 580     vm_start();
 581     qemu_mutex_unlock_iothread();
 582     trace_colo_vm_state_change("stop", "run");
 583
 584     timer_mod(s->colo_delay_timer,
 585             current_time + s->parameters.x_checkpoint_delay);
 586
 587     while (s->state == MIGRATION_STATUS_COLO) {
 588         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 589             error_report("failover request");
 590             goto out;
 591         }
 592
 593         qemu_event_wait(&s->colo_checkpoint_event);
 594
 595         if (s->state != MIGRATION_STATUS_COLO) {
 596             goto out;
 597         }
 598         ret = colo_do_checkpoint_transaction(s, bioc, fb);
 599         if (ret < 0) {
 600             goto out;
 601         }
 602     }
 603
 604 out:
 605     /* Throw the unreported error message after exited from loop */
 606     if (local_err) {
 607         error_report_err(local_err);
 608     }
 609
 610     if (fb) {
 611         qemu_fclose(fb);
 612     }
 613
 614     /*
 615      * There are only two reasons we can get here, some error happened
 616      * or the user triggered failover.
 617      */
 618     switch (failover_get_state()) {
 619     case FAILOVER_STATUS_COMPLETED:
 620         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 621                                   COLO_EXIT_REASON_REQUEST);
 622         break;
 623     default:
 624         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 625                                   COLO_EXIT_REASON_ERROR);
 626     }
 627
 628     /* Hope this not to be too long to wait here */
 629     qemu_sem_wait(&s->colo_exit_sem);
 630     qemu_sem_destroy(&s->colo_exit_sem);
 631
 632     /*
 633      * It is safe to unregister notifier after failover finished.
 634      * Besides, colo_delay_timer and colo_checkpoint_sem can't be
 635      * released before unregister notifier, or there will be use-after-free
 636      * error.
 637      */
 638     colo_compare_unregister_notifier(&packets_compare_notifier);
 639     timer_del(s->colo_delay_timer);
 640     timer_free(s->colo_delay_timer);
 641     qemu_event_destroy(&s->colo_checkpoint_event);
 642
 643     /*
 644      * Must be called after failover BH is completed,
 645      * Or the failover BH may shutdown the wrong fd that
 646      * re-used by other threads after we release here.
 647      */
 648     if (s->rp_state.from_dst_file) {
 649         qemu_fclose(s->rp_state.from_dst_file);
 650     }
 651 }
 652
 653 void colo_checkpoint_notify(void *opaque)
 654 {
 655     MigrationState *s = opaque;
 656     int64_t next_notify_time;
 657
 658     qemu_event_set(&s->colo_checkpoint_event);
 659     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 660     next_notify_time = s->colo_checkpoint_time +
 661                     s->parameters.x_checkpoint_delay;
 662     timer_mod(s->colo_delay_timer, next_notify_time);
 663 }
 664
 665 void migrate_start_colo_process(MigrationState *s)
 666 {
 667     qemu_mutex_unlock_iothread();
 668     qemu_event_init(&s->colo_checkpoint_event, false);
 669     s->colo_delay_timer =  timer_new_ms(QEMU_CLOCK_HOST,
 670                                 colo_checkpoint_notify, s);
 671
 672     qemu_sem_init(&s->colo_exit_sem, 0);
 673     migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
 674                       MIGRATION_STATUS_COLO);
 675     colo_process_checkpoint(s);
 676     qemu_mutex_lock_iothread();
 677 }
 678
 679 static void colo_incoming_process_checkpoint(MigrationIncomingState *mis,
 680                       QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
 681 {
 682     uint64_t total_size;
 683     uint64_t value;
 684     Error *local_err = NULL;
 685     int ret;
 686
 687     qemu_mutex_lock_iothread();
 688     vm_stop_force_state(RUN_STATE_COLO);
 689     trace_colo_vm_state_change("run", "stop");
 690     qemu_mutex_unlock_iothread();
 691
 692     /* FIXME: This is unnecessary for periodic checkpoint mode */
 693     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
 694                  &local_err);
 695     if (local_err) {
 696         error_propagate(errp, local_err);
 697         return;
 698     }
 699
 700     colo_receive_check_message(mis->from_src_file,
 701                        COLO_MESSAGE_VMSTATE_SEND, &local_err);
 702     if (local_err) {
 703         error_propagate(errp, local_err);
 704         return;
 705     }
 706
 707     qemu_mutex_lock_iothread();
 708     cpu_synchronize_all_states();
 709     ret = qemu_loadvm_state_main(mis->from_src_file, mis);
 710     qemu_mutex_unlock_iothread();
 711
 712     if (ret < 0) {
 713         error_setg(errp, "Load VM's live state (ram) error");
 714         return;
 715     }
 716
 717     value = colo_receive_message_value(mis->from_src_file,
 718                              COLO_MESSAGE_VMSTATE_SIZE, &local_err);
 719     if (local_err) {
 720         error_propagate(errp, local_err);
 721         return;
 722     }
 723
 724     /*
 725      * Read VM device state data into channel buffer,
 726      * It's better to re-use the memory allocated.
 727      * Here we need to handle the channel buffer directly.
 728      */
 729     if (value > bioc->capacity) {
 730         bioc->capacity = value;
 731         bioc->data = g_realloc(bioc->data, bioc->capacity);
 732     }
 733     total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
 734     if (total_size != value) {
 735         error_setg(errp, "Got %" PRIu64 " VMState data, less than expected"
 736                     " %" PRIu64, total_size, value);
 737         return;
 738     }
 739     bioc->usage = total_size;
 740     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 741
 742     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
 743                  &local_err);
 744     if (local_err) {
 745         error_propagate(errp, local_err);
 746         return;
 747     }
 748
 749     qemu_mutex_lock_iothread();
 750     vmstate_loading = true;
 751     colo_flush_ram_cache();
 752     ret = qemu_load_device_state(fb);
 753     if (ret < 0) {
 754         error_setg(errp, "COLO: load device state failed");
 755         vmstate_loading = false;
 756         qemu_mutex_unlock_iothread();
 757         return;
 758     }
 759
 760 #ifdef CONFIG_REPLICATION
 761     replication_get_error_all(&local_err);
 762     if (local_err) {
 763         error_propagate(errp, local_err);
 764         vmstate_loading = false;
 765         qemu_mutex_unlock_iothread();
 766         return;
 767     }
 768
 769     /* discard colo disk buffer */
 770     replication_do_checkpoint_all(&local_err);
 771     if (local_err) {
 772         error_propagate(errp, local_err);
 773         vmstate_loading = false;
 774         qemu_mutex_unlock_iothread();
 775         return;
 776     }
 777 #else
 778     abort();
 779 #endif
 780     /* Notify all filters of all NIC to do checkpoint */
 781     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
 782
 783     if (local_err) {
 784         error_propagate(errp, local_err);
 785         vmstate_loading = false;
 786         qemu_mutex_unlock_iothread();
 787         return;
 788     }
 789
 790     vmstate_loading = false;
 791     vm_start();
 792     trace_colo_vm_state_change("stop", "run");
 793     qemu_mutex_unlock_iothread();
 794
 795     if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 796         return;
 797     }
 798
 799     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
 800                  &local_err);
 801     error_propagate(errp, local_err);
 802 }
 803
 804 static void colo_wait_handle_message(MigrationIncomingState *mis,
 805                 QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
 806 {
 807     COLOMessage msg;
 808     Error *local_err = NULL;
 809
 810     msg = colo_receive_message(mis->from_src_file, &local_err);
 811     if (local_err) {
 812         error_propagate(errp, local_err);
 813         return;
 814     }
 815
 816     switch (msg) {
 817     case COLO_MESSAGE_CHECKPOINT_REQUEST:
 818         colo_incoming_process_checkpoint(mis, fb, bioc, errp);
 819         break;
 820     default:
 821         error_setg(errp, "Got unknown COLO message: %d", msg);
 822         break;
 823     }
 824 }
 825
 826 void *colo_process_incoming_thread(void *opaque)
 827 {
 828     MigrationIncomingState *mis = opaque;
 829     QEMUFile *fb = NULL;
 830     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
 831     Error *local_err = NULL;
 832
 833     rcu_register_thread();
 834     qemu_sem_init(&mis->colo_incoming_sem, 0);
 835
 836     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 837                       MIGRATION_STATUS_COLO);
 838
 839     last_colo_mode = get_colo_mode();
 840     if (last_colo_mode != COLO_MODE_SECONDARY) {
 841         error_report("COLO mode must be COLO_MODE_SECONDARY");
 842         return NULL;
 843     }
 844
 845     failover_init_state();
 846
 847     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
 848     if (!mis->to_src_file) {
 849         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
 850         goto out;
 851     }
 852     /*
 853      * Note: the communication between Primary side and Secondary side
 854      * should be sequential, we set the fd to unblocked in migration incoming
 855      * coroutine, and here we are in the COLO incoming thread, so it is ok to
 856      * set the fd back to blocked.
 857      */
 858     qemu_file_set_blocking(mis->from_src_file, true);
 859
 860     colo_incoming_start_dirty_log();
 861
 862     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 863     fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
 864     object_unref(OBJECT(bioc));
 865
 866     qemu_mutex_lock_iothread();
 867 #ifdef CONFIG_REPLICATION
 868     replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
 869     if (local_err) {
 870         qemu_mutex_unlock_iothread();
 871         goto out;
 872     }
 873 #else
 874         abort();
 875 #endif
 876     vm_start();
 877     trace_colo_vm_state_change("stop", "run");
 878     qemu_mutex_unlock_iothread();
 879
 880     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
 881                       &local_err);
 882     if (local_err) {
 883         goto out;
 884     }
 885
 886     while (mis->state == MIGRATION_STATUS_COLO) {
 887         colo_wait_handle_message(mis, fb, bioc, &local_err);
 888         if (local_err) {
 889             error_report_err(local_err);
 890             break;
 891         }
 892
 893         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 894             failover_set_state(FAILOVER_STATUS_RELAUNCH,
 895                             FAILOVER_STATUS_NONE);
 896             failover_request_active(NULL);
 897             break;
 898         }
 899
 900         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 901             error_report("failover request");
 902             break;
 903         }
 904     }
 905
 906 out:
 907     /*
 908      * There are only two reasons we can get here, some error happened
 909      * or the user triggered failover.
 910      */
 911     switch (failover_get_state()) {
 912     case FAILOVER_STATUS_COMPLETED:
 913         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 914                                   COLO_EXIT_REASON_REQUEST);
 915         break;
 916     default:
 917         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 918                                   COLO_EXIT_REASON_ERROR);
 919     }
 920
 921     if (fb) {
 922         qemu_fclose(fb);
 923     }
 924
 925     /* Hope this not to be too long to loop here */
 926     qemu_sem_wait(&mis->colo_incoming_sem);
 927     qemu_sem_destroy(&mis->colo_incoming_sem);
 928     /* Must be called after failover BH is completed */
 929     if (mis->to_src_file) {
 930         qemu_fclose(mis->to_src_file);
 931         mis->to_src_file = NULL;
 932     }
 933
 934     rcu_unregister_thread();
 935     return NULL;
 936 }