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 "migration.h"
  18 #include "qemu-file.h"
  19 #include "savevm.h"
  20 #include "migration/colo.h"
  21 #include "block.h"
  22 #include "io/channel-buffer.h"
  23 #include "trace.h"
  24 #include "qemu/error-report.h"
  25 #include "qemu/main-loop.h"
  26 #include "qemu/rcu.h"
  27 #include "migration/failover.h"
  28 #include "migration/ram.h"
  29 #ifdef CONFIG_REPLICATION
  30 #include "block/replication.h"
  31 #endif
  32 #include "net/colo-compare.h"
  33 #include "net/colo.h"
  34 #include "block/block.h"
  35 #include "qapi/qapi-events-migration.h"
  36 #include "sysemu/cpus.h"
  37 #include "sysemu/runstate.h"
  38 #include "net/filter.h"
  39 #include "options.h"
  40
  41 static bool vmstate_loading;
  42 static Notifier packets_compare_notifier;
  43
  44 /* User need to know colo mode after COLO failover */
  45 static COLOMode last_colo_mode;
  46
  47 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
  48
  49 bool migration_in_colo_state(void)
  50 {
  51     MigrationState *s = migrate_get_current();
  52
  53     return (s->state == MIGRATION_STATUS_COLO);
  54 }
  55
  56 bool migration_incoming_in_colo_state(void)
  57 {
  58     MigrationIncomingState *mis = migration_incoming_get_current();
  59
  60     return mis && (mis->state == MIGRATION_STATUS_COLO);
  61 }
  62
  63 static bool colo_runstate_is_stopped(void)
  64 {
  65     return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
  66 }
  67
  68 static void colo_checkpoint_notify(void *opaque)
  69 {
  70     MigrationState *s = opaque;
  71     int64_t next_notify_time;
  72
  73     qemu_event_set(&s->colo_checkpoint_event);
  74     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
  75     next_notify_time = s->colo_checkpoint_time + migrate_checkpoint_delay();
  76     timer_mod(s->colo_delay_timer, next_notify_time);
  77 }
  78
  79 void colo_checkpoint_delay_set(void)
  80 {
  81     if (migration_in_colo_state()) {
  82         colo_checkpoint_notify(migrate_get_current());
  83     }
  84 }
  85
  86 static void secondary_vm_do_failover(void)
  87 {
  88 /* COLO needs enable block-replication */
  89 #ifdef CONFIG_REPLICATION
  90     int old_state;
  91     MigrationIncomingState *mis = migration_incoming_get_current();
  92     Error *local_err = NULL;
  93
  94     /* Can not do failover during the process of VM's loading VMstate, Or
  95      * it will break the secondary VM.
  96      */
  97     if (vmstate_loading) {
  98         old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
  99                         FAILOVER_STATUS_RELAUNCH);
 100         if (old_state != FAILOVER_STATUS_ACTIVE) {
 101             error_report("Unknown error while do failover for secondary VM,"
 102                          "old_state: %s", FailoverStatus_str(old_state));
 103         }
 104         return;
 105     }
 106
 107     migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
 108                       MIGRATION_STATUS_COMPLETED);
 109
 110     replication_stop_all(true, &local_err);
 111     if (local_err) {
 112         error_report_err(local_err);
 113         local_err = NULL;
 114     }
 115
 116     /* Notify all filters of all NIC to do checkpoint */
 117     colo_notify_filters_event(COLO_EVENT_FAILOVER, &local_err);
 118     if (local_err) {
 119         error_report_err(local_err);
 120     }
 121
 122     if (!autostart) {
 123         error_report("\"-S\" qemu option will be ignored in secondary side");
 124         /* recover runstate to normal migration finish state */
 125         autostart = true;
 126     }
 127     /*
 128      * Make sure COLO incoming thread not block in recv or send,
 129      * If mis->from_src_file and mis->to_src_file use the same fd,
 130      * The second shutdown() will return -1, we ignore this value,
 131      * It is harmless.
 132      */
 133     if (mis->from_src_file) {
 134         qemu_file_shutdown(mis->from_src_file);
 135     }
 136     if (mis->to_src_file) {
 137         qemu_file_shutdown(mis->to_src_file);
 138     }
 139
 140     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 141                                    FAILOVER_STATUS_COMPLETED);
 142     if (old_state != FAILOVER_STATUS_ACTIVE) {
 143         error_report("Incorrect state (%s) while doing failover for "
 144                      "secondary VM", FailoverStatus_str(old_state));
 145         return;
 146     }
 147     /* Notify COLO incoming thread that failover work is finished */
 148     qemu_sem_post(&mis->colo_incoming_sem);
 149
 150     /* For Secondary VM, jump to incoming co */
 151     if (mis->migration_incoming_co) {
 152         qemu_coroutine_enter(mis->migration_incoming_co);
 153     }
 154 #else
 155     abort();
 156 #endif
 157 }
 158
 159 static void primary_vm_do_failover(void)
 160 {
 161 #ifdef CONFIG_REPLICATION
 162     MigrationState *s = migrate_get_current();
 163     int old_state;
 164     Error *local_err = NULL;
 165
 166     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
 167                       MIGRATION_STATUS_COMPLETED);
 168     /*
 169      * kick COLO thread which might wait at
 170      * qemu_sem_wait(&s->colo_checkpoint_sem).
 171      */
 172     colo_checkpoint_notify(s);
 173
 174     /*
 175      * Wake up COLO thread which may blocked in recv() or send(),
 176      * The s->rp_state.from_dst_file and s->to_dst_file may use the
 177      * same fd, but we still shutdown the fd for twice, it is harmless.
 178      */
 179     if (s->to_dst_file) {
 180         qemu_file_shutdown(s->to_dst_file);
 181     }
 182     if (s->rp_state.from_dst_file) {
 183         qemu_file_shutdown(s->rp_state.from_dst_file);
 184     }
 185
 186     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 187                                    FAILOVER_STATUS_COMPLETED);
 188     if (old_state != FAILOVER_STATUS_ACTIVE) {
 189         error_report("Incorrect state (%s) while doing failover for Primary VM",
 190                      FailoverStatus_str(old_state));
 191         return;
 192     }
 193
 194     replication_stop_all(true, &local_err);
 195     if (local_err) {
 196         error_report_err(local_err);
 197         local_err = NULL;
 198     }
 199
 200     /* Notify COLO thread that failover work is finished */
 201     qemu_sem_post(&s->colo_exit_sem);
 202 #else
 203     abort();
 204 #endif
 205 }
 206
 207 COLOMode get_colo_mode(void)
 208 {
 209     if (migration_in_colo_state()) {
 210         return COLO_MODE_PRIMARY;
 211     } else if (migration_incoming_in_colo_state()) {
 212         return COLO_MODE_SECONDARY;
 213     } else {
 214         return COLO_MODE_NONE;
 215     }
 216 }
 217
 218 void colo_do_failover(void)
 219 {
 220     /* Make sure VM stopped while failover happened. */
 221     if (!colo_runstate_is_stopped()) {
 222         vm_stop_force_state(RUN_STATE_COLO);
 223     }
 224
 225     switch (last_colo_mode = get_colo_mode()) {
 226     case COLO_MODE_PRIMARY:
 227         primary_vm_do_failover();
 228         break;
 229     case COLO_MODE_SECONDARY:
 230         secondary_vm_do_failover();
 231         break;
 232     default:
 233         error_report("colo_do_failover failed because the colo mode"
 234                      " could not be obtained");
 235     }
 236 }
 237
 238 #ifdef CONFIG_REPLICATION
 239 void qmp_xen_set_replication(bool enable, bool primary,
 240                              bool has_failover, bool failover,
 241                              Error **errp)
 242 {
 243     ReplicationMode mode = primary ?
 244                            REPLICATION_MODE_PRIMARY :
 245                            REPLICATION_MODE_SECONDARY;
 246
 247     if (has_failover && enable) {
 248         error_setg(errp, "Parameter 'failover' is only for"
 249                    " stopping replication");
 250         return;
 251     }
 252
 253     if (enable) {
 254         replication_start_all(mode, errp);
 255     } else {
 256         if (!has_failover) {
 257             failover = NULL;
 258         }
 259         replication_stop_all(failover, failover ? NULL : errp);
 260     }
 261 }
 262
 263 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
 264 {
 265     Error *err = NULL;
 266     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
 267
 268     replication_get_error_all(&err);
 269     if (err) {
 270         s->error = true;
 271         s->desc = g_strdup(error_get_pretty(err));
 272     } else {
 273         s->error = false;
 274     }
 275
 276     error_free(err);
 277     return s;
 278 }
 279
 280 void qmp_xen_colo_do_checkpoint(Error **errp)
 281 {
 282     Error *err = NULL;
 283
 284     replication_do_checkpoint_all(&err);
 285     if (err) {
 286         error_propagate(errp, err);
 287         return;
 288     }
 289     /* Notify all filters of all NIC to do checkpoint */
 290     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, errp);
 291 }
 292 #endif
 293
 294 COLOStatus *qmp_query_colo_status(Error **errp)
 295 {
 296     COLOStatus *s = g_new0(COLOStatus, 1);
 297
 298     s->mode = get_colo_mode();
 299     s->last_mode = last_colo_mode;
 300
 301     switch (failover_get_state()) {
 302     case FAILOVER_STATUS_NONE:
 303         s->reason = COLO_EXIT_REASON_NONE;
 304         break;
 305     case FAILOVER_STATUS_COMPLETED:
 306         s->reason = COLO_EXIT_REASON_REQUEST;
 307         break;
 308     default:
 309         if (migration_in_colo_state()) {
 310             s->reason = COLO_EXIT_REASON_PROCESSING;
 311         } else {
 312             s->reason = COLO_EXIT_REASON_ERROR;
 313         }
 314     }
 315
 316     return s;
 317 }
 318
 319 static void colo_send_message(QEMUFile *f, COLOMessage msg,
 320                               Error **errp)
 321 {
 322     int ret;
 323
 324     if (msg >= COLO_MESSAGE__MAX) {
 325         error_setg(errp, "%s: Invalid message", __func__);
 326         return;
 327     }
 328     qemu_put_be32(f, msg);
 329     qemu_fflush(f);
 330
 331     ret = qemu_file_get_error(f);
 332     if (ret < 0) {
 333         error_setg_errno(errp, -ret, "Can't send COLO message");
 334     }
 335     trace_colo_send_message(COLOMessage_str(msg));
 336 }
 337
 338 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
 339                                     uint64_t value, Error **errp)
 340 {
 341     Error *local_err = NULL;
 342     int ret;
 343
 344     colo_send_message(f, msg, &local_err);
 345     if (local_err) {
 346         error_propagate(errp, local_err);
 347         return;
 348     }
 349     qemu_put_be64(f, value);
 350     qemu_fflush(f);
 351
 352     ret = qemu_file_get_error(f);
 353     if (ret < 0) {
 354         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
 355                          COLOMessage_str(msg));
 356     }
 357 }
 358
 359 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
 360 {
 361     COLOMessage msg;
 362     int ret;
 363
 364     msg = qemu_get_be32(f);
 365     ret = qemu_file_get_error(f);
 366     if (ret < 0) {
 367         error_setg_errno(errp, -ret, "Can't receive COLO message");
 368         return msg;
 369     }
 370     if (msg >= COLO_MESSAGE__MAX) {
 371         error_setg(errp, "%s: Invalid message", __func__);
 372         return msg;
 373     }
 374     trace_colo_receive_message(COLOMessage_str(msg));
 375     return msg;
 376 }
 377
 378 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
 379                                        Error **errp)
 380 {
 381     COLOMessage msg;
 382     Error *local_err = NULL;
 383
 384     msg = colo_receive_message(f, &local_err);
 385     if (local_err) {
 386         error_propagate(errp, local_err);
 387         return;
 388     }
 389     if (msg != expect_msg) {
 390         error_setg(errp, "Unexpected COLO message %d, expected %d",
 391                           msg, expect_msg);
 392     }
 393 }
 394
 395 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
 396                                            Error **errp)
 397 {
 398     Error *local_err = NULL;
 399     uint64_t value;
 400     int ret;
 401
 402     colo_receive_check_message(f, expect_msg, &local_err);
 403     if (local_err) {
 404         error_propagate(errp, local_err);
 405         return 0;
 406     }
 407
 408     value = qemu_get_be64(f);
 409     ret = qemu_file_get_error(f);
 410     if (ret < 0) {
 411         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
 412                          COLOMessage_str(expect_msg));
 413     }
 414     return value;
 415 }
 416
 417 static int colo_do_checkpoint_transaction(MigrationState *s,
 418                                           QIOChannelBuffer *bioc,
 419                                           QEMUFile *fb)
 420 {
 421     Error *local_err = NULL;
 422     int ret = -1;
 423
 424     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
 425                       &local_err);
 426     if (local_err) {
 427         goto out;
 428     }
 429
 430     colo_receive_check_message(s->rp_state.from_dst_file,
 431                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
 432     if (local_err) {
 433         goto out;
 434     }
 435     /* Reset channel-buffer directly */
 436     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 437     bioc->usage = 0;
 438
 439     qemu_mutex_lock_iothread();
 440     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 441         qemu_mutex_unlock_iothread();
 442         goto out;
 443     }
 444     vm_stop_force_state(RUN_STATE_COLO);
 445     qemu_mutex_unlock_iothread();
 446     trace_colo_vm_state_change("run", "stop");
 447     /*
 448      * Failover request bh could be called after vm_stop_force_state(),
 449      * So we need check failover_request_is_active() again.
 450      */
 451     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 452         goto out;
 453     }
 454     qemu_mutex_lock_iothread();
 455
 456 #ifdef CONFIG_REPLICATION
 457     replication_do_checkpoint_all(&local_err);
 458     if (local_err) {
 459         qemu_mutex_unlock_iothread();
 460         goto out;
 461     }
 462 #else
 463         abort();
 464 #endif
 465
 466     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
 467     if (local_err) {
 468         qemu_mutex_unlock_iothread();
 469         goto out;
 470     }
 471     /* Note: device state is saved into buffer */
 472     ret = qemu_save_device_state(fb);
 473
 474     qemu_mutex_unlock_iothread();
 475     if (ret < 0) {
 476         goto out;
 477     }
 478
 479     if (migrate_auto_converge()) {
 480         mig_throttle_counter_reset();
 481     }
 482     /*
 483      * Only save VM's live state, which not including device state.
 484      * TODO: We may need a timeout mechanism to prevent COLO process
 485      * to be blocked here.
 486      */
 487     qemu_savevm_live_state(s->to_dst_file);
 488
 489     qemu_fflush(fb);
 490
 491     /*
 492      * We need the size of the VMstate data in Secondary side,
 493      * With which we can decide how much data should be read.
 494      */
 495     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
 496                             bioc->usage, &local_err);
 497     if (local_err) {
 498         goto out;
 499     }
 500
 501     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
 502     qemu_fflush(s->to_dst_file);
 503     ret = qemu_file_get_error(s->to_dst_file);
 504     if (ret < 0) {
 505         goto out;
 506     }
 507
 508     colo_receive_check_message(s->rp_state.from_dst_file,
 509                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
 510     if (local_err) {
 511         goto out;
 512     }
 513
 514     qemu_event_reset(&s->colo_checkpoint_event);
 515     colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
 516     if (local_err) {
 517         goto out;
 518     }
 519
 520     colo_receive_check_message(s->rp_state.from_dst_file,
 521                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
 522     if (local_err) {
 523         goto out;
 524     }
 525
 526     ret = 0;
 527
 528     qemu_mutex_lock_iothread();
 529     vm_start();
 530     qemu_mutex_unlock_iothread();
 531     trace_colo_vm_state_change("stop", "run");
 532
 533 out:
 534     if (local_err) {
 535         error_report_err(local_err);
 536     }
 537     return ret;
 538 }
 539
 540 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
 541 {
 542     colo_checkpoint_notify(data);
 543 }
 544
 545 static void colo_process_checkpoint(MigrationState *s)
 546 {
 547     QIOChannelBuffer *bioc;
 548     QEMUFile *fb = NULL;
 549     Error *local_err = NULL;
 550     int ret;
 551
 552     if (get_colo_mode() != COLO_MODE_PRIMARY) {
 553         error_report("COLO mode must be COLO_MODE_PRIMARY");
 554         return;
 555     }
 556
 557     failover_init_state();
 558
 559     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
 560     if (!s->rp_state.from_dst_file) {
 561         error_report("Open QEMUFile from_dst_file failed");
 562         goto out;
 563     }
 564
 565     packets_compare_notifier.notify = colo_compare_notify_checkpoint;
 566     colo_compare_register_notifier(&packets_compare_notifier);
 567
 568     /*
 569      * Wait for Secondary finish loading VM states and enter COLO
 570      * restore.
 571      */
 572     colo_receive_check_message(s->rp_state.from_dst_file,
 573                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
 574     if (local_err) {
 575         goto out;
 576     }
 577     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 578     fb = qemu_file_new_output(QIO_CHANNEL(bioc));
 579     object_unref(OBJECT(bioc));
 580
 581     qemu_mutex_lock_iothread();
 582 #ifdef CONFIG_REPLICATION
 583     replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
 584     if (local_err) {
 585         qemu_mutex_unlock_iothread();
 586         goto out;
 587     }
 588 #else
 589         abort();
 590 #endif
 591
 592     vm_start();
 593     qemu_mutex_unlock_iothread();
 594     trace_colo_vm_state_change("stop", "run");
 595
 596     timer_mod(s->colo_delay_timer, qemu_clock_get_ms(QEMU_CLOCK_HOST) +
 597               migrate_checkpoint_delay());
 598
 599     while (s->state == MIGRATION_STATUS_COLO) {
 600         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 601             error_report("failover request");
 602             goto out;
 603         }
 604
 605         qemu_event_wait(&s->colo_checkpoint_event);
 606
 607         if (s->state != MIGRATION_STATUS_COLO) {
 608             goto out;
 609         }
 610         ret = colo_do_checkpoint_transaction(s, bioc, fb);
 611         if (ret < 0) {
 612             goto out;
 613         }
 614     }
 615
 616 out:
 617     /* Throw the unreported error message after exited from loop */
 618     if (local_err) {
 619         error_report_err(local_err);
 620     }
 621
 622     if (fb) {
 623         qemu_fclose(fb);
 624     }
 625
 626     /*
 627      * There are only two reasons we can get here, some error happened
 628      * or the user triggered failover.
 629      */
 630     switch (failover_get_state()) {
 631     case FAILOVER_STATUS_COMPLETED:
 632         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 633                                   COLO_EXIT_REASON_REQUEST);
 634         break;
 635     default:
 636         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 637                                   COLO_EXIT_REASON_ERROR);
 638     }
 639
 640     /* Hope this not to be too long to wait here */
 641     qemu_sem_wait(&s->colo_exit_sem);
 642     qemu_sem_destroy(&s->colo_exit_sem);
 643
 644     /*
 645      * It is safe to unregister notifier after failover finished.
 646      * Besides, colo_delay_timer and colo_checkpoint_sem can't be
 647      * released before unregister notifier, or there will be use-after-free
 648      * error.
 649      */
 650     colo_compare_unregister_notifier(&packets_compare_notifier);
 651     timer_free(s->colo_delay_timer);
 652     qemu_event_destroy(&s->colo_checkpoint_event);
 653
 654     /*
 655      * Must be called after failover BH is completed,
 656      * Or the failover BH may shutdown the wrong fd that
 657      * re-used by other threads after we release here.
 658      */
 659     if (s->rp_state.from_dst_file) {
 660         qemu_fclose(s->rp_state.from_dst_file);
 661         s->rp_state.from_dst_file = NULL;
 662     }
 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     colo_process_checkpoint(s);
 674     qemu_mutex_lock_iothread();
 675 }
 676
 677 static void colo_incoming_process_checkpoint(MigrationIncomingState *mis,
 678                       QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
 679 {
 680     uint64_t total_size;
 681     uint64_t value;
 682     Error *local_err = NULL;
 683     int ret;
 684
 685     qemu_mutex_lock_iothread();
 686     vm_stop_force_state(RUN_STATE_COLO);
 687     qemu_mutex_unlock_iothread();
 688     trace_colo_vm_state_change("run", "stop");
 689
 690     /* FIXME: This is unnecessary for periodic checkpoint mode */
 691     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
 692                  &local_err);
 693     if (local_err) {
 694         error_propagate(errp, local_err);
 695         return;
 696     }
 697
 698     colo_receive_check_message(mis->from_src_file,
 699                        COLO_MESSAGE_VMSTATE_SEND, &local_err);
 700     if (local_err) {
 701         error_propagate(errp, local_err);
 702         return;
 703     }
 704
 705     qemu_mutex_lock_iothread();
 706     cpu_synchronize_all_states();
 707     ret = qemu_loadvm_state_main(mis->from_src_file, mis);
 708     qemu_mutex_unlock_iothread();
 709
 710     if (ret < 0) {
 711         error_setg(errp, "Load VM's live state (ram) error");
 712         return;
 713     }
 714
 715     value = colo_receive_message_value(mis->from_src_file,
 716                              COLO_MESSAGE_VMSTATE_SIZE, &local_err);
 717     if (local_err) {
 718         error_propagate(errp, local_err);
 719         return;
 720     }
 721
 722     /*
 723      * Read VM device state data into channel buffer,
 724      * It's better to re-use the memory allocated.
 725      * Here we need to handle the channel buffer directly.
 726      */
 727     if (value > bioc->capacity) {
 728         bioc->capacity = value;
 729         bioc->data = g_realloc(bioc->data, bioc->capacity);
 730     }
 731     total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
 732     if (total_size != value) {
 733         error_setg(errp, "Got %" PRIu64 " VMState data, less than expected"
 734                     " %" PRIu64, total_size, value);
 735         return;
 736     }
 737     bioc->usage = total_size;
 738     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 739
 740     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
 741                  &local_err);
 742     if (local_err) {
 743         error_propagate(errp, local_err);
 744         return;
 745     }
 746
 747     qemu_mutex_lock_iothread();
 748     vmstate_loading = true;
 749     colo_flush_ram_cache();
 750     ret = qemu_load_device_state(fb);
 751     if (ret < 0) {
 752         error_setg(errp, "COLO: load device state failed");
 753         vmstate_loading = false;
 754         qemu_mutex_unlock_iothread();
 755         return;
 756     }
 757
 758 #ifdef CONFIG_REPLICATION
 759     replication_get_error_all(&local_err);
 760     if (local_err) {
 761         error_propagate(errp, local_err);
 762         vmstate_loading = false;
 763         qemu_mutex_unlock_iothread();
 764         return;
 765     }
 766
 767     /* discard colo disk buffer */
 768     replication_do_checkpoint_all(&local_err);
 769     if (local_err) {
 770         error_propagate(errp, local_err);
 771         vmstate_loading = false;
 772         qemu_mutex_unlock_iothread();
 773         return;
 774     }
 775 #else
 776     abort();
 777 #endif
 778     /* Notify all filters of all NIC to do checkpoint */
 779     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
 780
 781     if (local_err) {
 782         error_propagate(errp, local_err);
 783         vmstate_loading = false;
 784         qemu_mutex_unlock_iothread();
 785         return;
 786     }
 787
 788     vmstate_loading = false;
 789     vm_start();
 790     qemu_mutex_unlock_iothread();
 791     trace_colo_vm_state_change("stop", "run");
 792
 793     if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 794         return;
 795     }
 796
 797     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
 798                  &local_err);
 799     error_propagate(errp, local_err);
 800 }
 801
 802 static void colo_wait_handle_message(MigrationIncomingState *mis,
 803                 QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
 804 {
 805     COLOMessage msg;
 806     Error *local_err = NULL;
 807
 808     msg = colo_receive_message(mis->from_src_file, &local_err);
 809     if (local_err) {
 810         error_propagate(errp, local_err);
 811         return;
 812     }
 813
 814     switch (msg) {
 815     case COLO_MESSAGE_CHECKPOINT_REQUEST:
 816         colo_incoming_process_checkpoint(mis, fb, bioc, errp);
 817         break;
 818     default:
 819         error_setg(errp, "Got unknown COLO message: %d", msg);
 820         break;
 821     }
 822 }
 823
 824 void colo_shutdown(void)
 825 {
 826     MigrationIncomingState *mis = NULL;
 827     MigrationState *s = NULL;
 828
 829     switch (get_colo_mode()) {
 830     case COLO_MODE_PRIMARY:
 831         s = migrate_get_current();
 832         qemu_event_set(&s->colo_checkpoint_event);
 833         qemu_sem_post(&s->colo_exit_sem);
 834         break;
 835     case COLO_MODE_SECONDARY:
 836         mis = migration_incoming_get_current();
 837         qemu_sem_post(&mis->colo_incoming_sem);
 838         break;
 839     default:
 840         break;
 841     }
 842 }
 843
 844 void *colo_process_incoming_thread(void *opaque)
 845 {
 846     MigrationIncomingState *mis = opaque;
 847     QEMUFile *fb = NULL;
 848     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
 849     Error *local_err = NULL;
 850
 851     rcu_register_thread();
 852     qemu_sem_init(&mis->colo_incoming_sem, 0);
 853
 854     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 855                       MIGRATION_STATUS_COLO);
 856
 857     if (get_colo_mode() != COLO_MODE_SECONDARY) {
 858         error_report("COLO mode must be COLO_MODE_SECONDARY");
 859         return NULL;
 860     }
 861
 862     failover_init_state();
 863
 864     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
 865     if (!mis->to_src_file) {
 866         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
 867         goto out;
 868     }
 869     /*
 870      * Note: the communication between Primary side and Secondary side
 871      * should be sequential, we set the fd to unblocked in migration incoming
 872      * coroutine, and here we are in the COLO incoming thread, so it is ok to
 873      * set the fd back to blocked.
 874      */
 875     qemu_file_set_blocking(mis->from_src_file, true);
 876
 877     colo_incoming_start_dirty_log();
 878
 879     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 880     fb = qemu_file_new_input(QIO_CHANNEL(bioc));
 881     object_unref(OBJECT(bioc));
 882
 883     qemu_mutex_lock_iothread();
 884 #ifdef CONFIG_REPLICATION
 885     replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
 886     if (local_err) {
 887         qemu_mutex_unlock_iothread();
 888         goto out;
 889     }
 890 #else
 891         abort();
 892 #endif
 893     vm_start();
 894     qemu_mutex_unlock_iothread();
 895     trace_colo_vm_state_change("stop", "run");
 896
 897     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
 898                       &local_err);
 899     if (local_err) {
 900         goto out;
 901     }
 902
 903     while (mis->state == MIGRATION_STATUS_COLO) {
 904         colo_wait_handle_message(mis, fb, bioc, &local_err);
 905         if (local_err) {
 906             error_report_err(local_err);
 907             break;
 908         }
 909
 910         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 911             failover_set_state(FAILOVER_STATUS_RELAUNCH,
 912                             FAILOVER_STATUS_NONE);
 913             failover_request_active(NULL);
 914             break;
 915         }
 916
 917         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 918             error_report("failover request");
 919             break;
 920         }
 921     }
 922
 923 out:
 924     /*
 925      * There are only two reasons we can get here, some error happened
 926      * or the user triggered failover.
 927      */
 928     switch (failover_get_state()) {
 929     case FAILOVER_STATUS_COMPLETED:
 930         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 931                                   COLO_EXIT_REASON_REQUEST);
 932         break;
 933     default:
 934         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 935                                   COLO_EXIT_REASON_ERROR);
 936     }
 937
 938     if (fb) {
 939         qemu_fclose(fb);
 940     }
 941
 942     /* Hope this not to be too long to loop here */
 943     qemu_sem_wait(&mis->colo_incoming_sem);
 944     qemu_sem_destroy(&mis->colo_incoming_sem);
 945
 946     rcu_unregister_thread();
 947     return NULL;
 948 }