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     replication_do_checkpoint_all(errp);
 267     /* Notify all filters of all NIC to do checkpoint */
 268     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, errp);
 269 }
 270 #endif
 271
 272 COLOStatus *qmp_query_colo_status(Error **errp)
 273 {
 274     COLOStatus *s = g_new0(COLOStatus, 1);
 275
 276     s->mode = get_colo_mode();
 277     s->last_mode = last_colo_mode;
 278
 279     switch (failover_get_state()) {
 280     case FAILOVER_STATUS_NONE:
 281         s->reason = COLO_EXIT_REASON_NONE;
 282         break;
 283     case FAILOVER_STATUS_COMPLETED:
 284         s->reason = COLO_EXIT_REASON_REQUEST;
 285         break;
 286     default:
 287         if (migration_in_colo_state()) {
 288             s->reason = COLO_EXIT_REASON_PROCESSING;
 289         } else {
 290             s->reason = COLO_EXIT_REASON_ERROR;
 291         }
 292     }
 293
 294     return s;
 295 }
 296
 297 static void colo_send_message(QEMUFile *f, COLOMessage msg,
 298                               Error **errp)
 299 {
 300     int ret;
 301
 302     if (msg >= COLO_MESSAGE__MAX) {
 303         error_setg(errp, "%s: Invalid message", __func__);
 304         return;
 305     }
 306     qemu_put_be32(f, msg);
 307     qemu_fflush(f);
 308
 309     ret = qemu_file_get_error(f);
 310     if (ret < 0) {
 311         error_setg_errno(errp, -ret, "Can't send COLO message");
 312     }
 313     trace_colo_send_message(COLOMessage_str(msg));
 314 }
 315
 316 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
 317                                     uint64_t value, Error **errp)
 318 {
 319     Error *local_err = NULL;
 320     int ret;
 321
 322     colo_send_message(f, msg, &local_err);
 323     if (local_err) {
 324         error_propagate(errp, local_err);
 325         return;
 326     }
 327     qemu_put_be64(f, value);
 328     qemu_fflush(f);
 329
 330     ret = qemu_file_get_error(f);
 331     if (ret < 0) {
 332         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
 333                          COLOMessage_str(msg));
 334     }
 335 }
 336
 337 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
 338 {
 339     COLOMessage msg;
 340     int ret;
 341
 342     msg = qemu_get_be32(f);
 343     ret = qemu_file_get_error(f);
 344     if (ret < 0) {
 345         error_setg_errno(errp, -ret, "Can't receive COLO message");
 346         return msg;
 347     }
 348     if (msg >= COLO_MESSAGE__MAX) {
 349         error_setg(errp, "%s: Invalid message", __func__);
 350         return msg;
 351     }
 352     trace_colo_receive_message(COLOMessage_str(msg));
 353     return msg;
 354 }
 355
 356 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
 357                                        Error **errp)
 358 {
 359     COLOMessage msg;
 360     Error *local_err = NULL;
 361
 362     msg = colo_receive_message(f, &local_err);
 363     if (local_err) {
 364         error_propagate(errp, local_err);
 365         return;
 366     }
 367     if (msg != expect_msg) {
 368         error_setg(errp, "Unexpected COLO message %d, expected %d",
 369                           msg, expect_msg);
 370     }
 371 }
 372
 373 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
 374                                            Error **errp)
 375 {
 376     Error *local_err = NULL;
 377     uint64_t value;
 378     int ret;
 379
 380     colo_receive_check_message(f, expect_msg, &local_err);
 381     if (local_err) {
 382         error_propagate(errp, local_err);
 383         return 0;
 384     }
 385
 386     value = qemu_get_be64(f);
 387     ret = qemu_file_get_error(f);
 388     if (ret < 0) {
 389         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
 390                          COLOMessage_str(expect_msg));
 391     }
 392     return value;
 393 }
 394
 395 static int colo_do_checkpoint_transaction(MigrationState *s,
 396                                           QIOChannelBuffer *bioc,
 397                                           QEMUFile *fb)
 398 {
 399     Error *local_err = NULL;
 400     int ret = -1;
 401
 402     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
 403                       &local_err);
 404     if (local_err) {
 405         goto out;
 406     }
 407
 408     colo_receive_check_message(s->rp_state.from_dst_file,
 409                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
 410     if (local_err) {
 411         goto out;
 412     }
 413     /* Reset channel-buffer directly */
 414     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 415     bioc->usage = 0;
 416
 417     qemu_mutex_lock_iothread();
 418     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 419         qemu_mutex_unlock_iothread();
 420         goto out;
 421     }
 422     vm_stop_force_state(RUN_STATE_COLO);
 423     qemu_mutex_unlock_iothread();
 424     trace_colo_vm_state_change("run", "stop");
 425     /*
 426      * Failover request bh could be called after vm_stop_force_state(),
 427      * So we need check failover_request_is_active() again.
 428      */
 429     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 430         goto out;
 431     }
 432
 433     colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
 434     if (local_err) {
 435         goto out;
 436     }
 437
 438     /* Disable block migration */
 439     migrate_set_block_enabled(false, &local_err);
 440     qemu_mutex_lock_iothread();
 441
 442 #ifdef CONFIG_REPLICATION
 443     replication_do_checkpoint_all(&local_err);
 444     if (local_err) {
 445         qemu_mutex_unlock_iothread();
 446         goto out;
 447     }
 448 #else
 449         abort();
 450 #endif
 451
 452     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
 453     if (local_err) {
 454         qemu_mutex_unlock_iothread();
 455         goto out;
 456     }
 457     /* Note: device state is saved into buffer */
 458     ret = qemu_save_device_state(fb);
 459
 460     qemu_mutex_unlock_iothread();
 461     if (ret < 0) {
 462         goto out;
 463     }
 464     /*
 465      * Only save VM's live state, which not including device state.
 466      * TODO: We may need a timeout mechanism to prevent COLO process
 467      * to be blocked here.
 468      */
 469     qemu_savevm_live_state(s->to_dst_file);
 470
 471     qemu_fflush(fb);
 472
 473     /*
 474      * We need the size of the VMstate data in Secondary side,
 475      * With which we can decide how much data should be read.
 476      */
 477     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
 478                             bioc->usage, &local_err);
 479     if (local_err) {
 480         goto out;
 481     }
 482
 483     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
 484     qemu_fflush(s->to_dst_file);
 485     ret = qemu_file_get_error(s->to_dst_file);
 486     if (ret < 0) {
 487         goto out;
 488     }
 489
 490     colo_receive_check_message(s->rp_state.from_dst_file,
 491                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
 492     if (local_err) {
 493         goto out;
 494     }
 495
 496     colo_receive_check_message(s->rp_state.from_dst_file,
 497                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
 498     if (local_err) {
 499         goto out;
 500     }
 501
 502     ret = 0;
 503
 504     qemu_mutex_lock_iothread();
 505     vm_start();
 506     qemu_mutex_unlock_iothread();
 507     trace_colo_vm_state_change("stop", "run");
 508
 509 out:
 510     if (local_err) {
 511         error_report_err(local_err);
 512     }
 513     return ret;
 514 }
 515
 516 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
 517 {
 518     colo_checkpoint_notify(data);
 519 }
 520
 521 static void colo_process_checkpoint(MigrationState *s)
 522 {
 523     QIOChannelBuffer *bioc;
 524     QEMUFile *fb = NULL;
 525     int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 526     Error *local_err = NULL;
 527     int ret;
 528
 529     last_colo_mode = get_colo_mode();
 530     if (last_colo_mode != COLO_MODE_PRIMARY) {
 531         error_report("COLO mode must be COLO_MODE_PRIMARY");
 532         return;
 533     }
 534
 535     failover_init_state();
 536
 537     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
 538     if (!s->rp_state.from_dst_file) {
 539         error_report("Open QEMUFile from_dst_file failed");
 540         goto out;
 541     }
 542
 543     packets_compare_notifier.notify = colo_compare_notify_checkpoint;
 544     colo_compare_register_notifier(&packets_compare_notifier);
 545
 546     /*
 547      * Wait for Secondary finish loading VM states and enter COLO
 548      * restore.
 549      */
 550     colo_receive_check_message(s->rp_state.from_dst_file,
 551                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
 552     if (local_err) {
 553         goto out;
 554     }
 555     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 556     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
 557     object_unref(OBJECT(bioc));
 558
 559     qemu_mutex_lock_iothread();
 560 #ifdef CONFIG_REPLICATION
 561     replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
 562     if (local_err) {
 563         qemu_mutex_unlock_iothread();
 564         goto out;
 565     }
 566 #else
 567         abort();
 568 #endif
 569
 570     vm_start();
 571     qemu_mutex_unlock_iothread();
 572     trace_colo_vm_state_change("stop", "run");
 573
 574     timer_mod(s->colo_delay_timer,
 575             current_time + s->parameters.x_checkpoint_delay);
 576
 577     while (s->state == MIGRATION_STATUS_COLO) {
 578         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 579             error_report("failover request");
 580             goto out;
 581         }
 582
 583         qemu_sem_wait(&s->colo_checkpoint_sem);
 584
 585         if (s->state != MIGRATION_STATUS_COLO) {
 586             goto out;
 587         }
 588         ret = colo_do_checkpoint_transaction(s, bioc, fb);
 589         if (ret < 0) {
 590             goto out;
 591         }
 592     }
 593
 594 out:
 595     /* Throw the unreported error message after exited from loop */
 596     if (local_err) {
 597         error_report_err(local_err);
 598     }
 599
 600     if (fb) {
 601         qemu_fclose(fb);
 602     }
 603
 604     /*
 605      * There are only two reasons we can get here, some error happened
 606      * or the user triggered failover.
 607      */
 608     switch (failover_get_state()) {
 609     case FAILOVER_STATUS_COMPLETED:
 610         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 611                                   COLO_EXIT_REASON_REQUEST);
 612         break;
 613     default:
 614         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 615                                   COLO_EXIT_REASON_ERROR);
 616     }
 617
 618     /* Hope this not to be too long to wait here */
 619     qemu_sem_wait(&s->colo_exit_sem);
 620     qemu_sem_destroy(&s->colo_exit_sem);
 621
 622     /*
 623      * It is safe to unregister notifier after failover finished.
 624      * Besides, colo_delay_timer and colo_checkpoint_sem can't be
 625      * released befor unregister notifier, or there will be use-after-free
 626      * error.
 627      */
 628     colo_compare_unregister_notifier(&packets_compare_notifier);
 629     timer_del(s->colo_delay_timer);
 630     timer_free(s->colo_delay_timer);
 631     qemu_sem_destroy(&s->colo_checkpoint_sem);
 632
 633     /*
 634      * Must be called after failover BH is completed,
 635      * Or the failover BH may shutdown the wrong fd that
 636      * re-used by other threads after we release here.
 637      */
 638     if (s->rp_state.from_dst_file) {
 639         qemu_fclose(s->rp_state.from_dst_file);
 640     }
 641 }
 642
 643 void colo_checkpoint_notify(void *opaque)
 644 {
 645     MigrationState *s = opaque;
 646     int64_t next_notify_time;
 647
 648     qemu_sem_post(&s->colo_checkpoint_sem);
 649     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 650     next_notify_time = s->colo_checkpoint_time +
 651                     s->parameters.x_checkpoint_delay;
 652     timer_mod(s->colo_delay_timer, next_notify_time);
 653 }
 654
 655 void migrate_start_colo_process(MigrationState *s)
 656 {
 657     qemu_mutex_unlock_iothread();
 658     qemu_sem_init(&s->colo_checkpoint_sem, 0);
 659     s->colo_delay_timer =  timer_new_ms(QEMU_CLOCK_HOST,
 660                                 colo_checkpoint_notify, s);
 661
 662     qemu_sem_init(&s->colo_exit_sem, 0);
 663     migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
 664                       MIGRATION_STATUS_COLO);
 665     colo_process_checkpoint(s);
 666     qemu_mutex_lock_iothread();
 667 }
 668
 669 static void colo_incoming_process_checkpoint(MigrationIncomingState *mis,
 670                       QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
 671 {
 672     uint64_t total_size;
 673     uint64_t value;
 674     Error *local_err = NULL;
 675     int ret;
 676
 677     qemu_mutex_lock_iothread();
 678     vm_stop_force_state(RUN_STATE_COLO);
 679     trace_colo_vm_state_change("run", "stop");
 680     qemu_mutex_unlock_iothread();
 681
 682     /* FIXME: This is unnecessary for periodic checkpoint mode */
 683     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
 684                  &local_err);
 685     if (local_err) {
 686         error_propagate(errp, local_err);
 687         return;
 688     }
 689
 690     colo_receive_check_message(mis->from_src_file,
 691                        COLO_MESSAGE_VMSTATE_SEND, &local_err);
 692     if (local_err) {
 693         error_propagate(errp, local_err);
 694         return;
 695     }
 696
 697     qemu_mutex_lock_iothread();
 698     cpu_synchronize_all_pre_loadvm();
 699     ret = qemu_loadvm_state_main(mis->from_src_file, mis);
 700     qemu_mutex_unlock_iothread();
 701
 702     if (ret < 0) {
 703         error_setg(errp, "Load VM's live state (ram) error");
 704         return;
 705     }
 706
 707     value = colo_receive_message_value(mis->from_src_file,
 708                              COLO_MESSAGE_VMSTATE_SIZE, &local_err);
 709     if (local_err) {
 710         error_propagate(errp, local_err);
 711         return;
 712     }
 713
 714     /*
 715      * Read VM device state data into channel buffer,
 716      * It's better to re-use the memory allocated.
 717      * Here we need to handle the channel buffer directly.
 718      */
 719     if (value > bioc->capacity) {
 720         bioc->capacity = value;
 721         bioc->data = g_realloc(bioc->data, bioc->capacity);
 722     }
 723     total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
 724     if (total_size != value) {
 725         error_setg(errp, "Got %" PRIu64 " VMState data, less than expected"
 726                     " %" PRIu64, total_size, value);
 727         return;
 728     }
 729     bioc->usage = total_size;
 730     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 731
 732     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
 733                  &local_err);
 734     if (local_err) {
 735         error_propagate(errp, local_err);
 736         return;
 737     }
 738
 739     qemu_mutex_lock_iothread();
 740     vmstate_loading = true;
 741     ret = qemu_load_device_state(fb);
 742     if (ret < 0) {
 743         error_setg(errp, "COLO: load device state failed");
 744         qemu_mutex_unlock_iothread();
 745         return;
 746     }
 747
 748 #ifdef CONFIG_REPLICATION
 749     replication_get_error_all(&local_err);
 750     if (local_err) {
 751         error_propagate(errp, local_err);
 752         qemu_mutex_unlock_iothread();
 753         return;
 754     }
 755
 756     /* discard colo disk buffer */
 757     replication_do_checkpoint_all(&local_err);
 758     if (local_err) {
 759         error_propagate(errp, local_err);
 760         qemu_mutex_unlock_iothread();
 761         return;
 762     }
 763 #else
 764     abort();
 765 #endif
 766     /* Notify all filters of all NIC to do checkpoint */
 767     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
 768
 769     if (local_err) {
 770         error_propagate(errp, local_err);
 771         qemu_mutex_unlock_iothread();
 772         return;
 773     }
 774
 775     vmstate_loading = false;
 776     vm_start();
 777     trace_colo_vm_state_change("stop", "run");
 778     qemu_mutex_unlock_iothread();
 779
 780     if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 781         failover_set_state(FAILOVER_STATUS_RELAUNCH,
 782                         FAILOVER_STATUS_NONE);
 783         failover_request_active(NULL);
 784         return;
 785     }
 786
 787     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
 788                  &local_err);
 789     if (local_err) {
 790         error_propagate(errp, local_err);
 791     }
 792 }
 793
 794 static void colo_wait_handle_message(MigrationIncomingState *mis,
 795                 QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
 796 {
 797     COLOMessage msg;
 798     Error *local_err = NULL;
 799
 800     msg = colo_receive_message(mis->from_src_file, &local_err);
 801     if (local_err) {
 802         error_propagate(errp, local_err);
 803         return;
 804     }
 805
 806     switch (msg) {
 807     case COLO_MESSAGE_CHECKPOINT_REQUEST:
 808         colo_incoming_process_checkpoint(mis, fb, bioc, errp);
 809         break;
 810     default:
 811         error_setg(errp, "Got unknown COLO message: %d", msg);
 812         break;
 813     }
 814 }
 815
 816 void *colo_process_incoming_thread(void *opaque)
 817 {
 818     MigrationIncomingState *mis = opaque;
 819     QEMUFile *fb = NULL;
 820     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
 821     Error *local_err = NULL;
 822
 823     rcu_register_thread();
 824     qemu_sem_init(&mis->colo_incoming_sem, 0);
 825
 826     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 827                       MIGRATION_STATUS_COLO);
 828
 829     last_colo_mode = get_colo_mode();
 830     if (last_colo_mode != COLO_MODE_SECONDARY) {
 831         error_report("COLO mode must be COLO_MODE_SECONDARY");
 832         return NULL;
 833     }
 834
 835     failover_init_state();
 836
 837     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
 838     if (!mis->to_src_file) {
 839         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
 840         goto out;
 841     }
 842     /*
 843      * Note: the communication between Primary side and Secondary side
 844      * should be sequential, we set the fd to unblocked in migration incoming
 845      * coroutine, and here we are in the COLO incoming thread, so it is ok to
 846      * set the fd back to blocked.
 847      */
 848     qemu_file_set_blocking(mis->from_src_file, true);
 849
 850     colo_incoming_start_dirty_log();
 851
 852     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 853     fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
 854     object_unref(OBJECT(bioc));
 855
 856     qemu_mutex_lock_iothread();
 857 #ifdef CONFIG_REPLICATION
 858     replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
 859     if (local_err) {
 860         qemu_mutex_unlock_iothread();
 861         goto out;
 862     }
 863 #else
 864         abort();
 865 #endif
 866     vm_start();
 867     trace_colo_vm_state_change("stop", "run");
 868     qemu_mutex_unlock_iothread();
 869
 870     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
 871                       &local_err);
 872     if (local_err) {
 873         goto out;
 874     }
 875
 876     while (mis->state == MIGRATION_STATUS_COLO) {
 877         colo_wait_handle_message(mis, fb, bioc, &local_err);
 878         if (local_err) {
 879             error_report_err(local_err);
 880             break;
 881         }
 882         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 883             error_report("failover request");
 884             break;
 885         }
 886     }
 887
 888 out:
 889     vmstate_loading = false;
 890
 891     /*
 892      * There are only two reasons we can get here, some error happened
 893      * or the user triggered failover.
 894      */
 895     switch (failover_get_state()) {
 896     case FAILOVER_STATUS_COMPLETED:
 897         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 898                                   COLO_EXIT_REASON_REQUEST);
 899         break;
 900     default:
 901         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 902                                   COLO_EXIT_REASON_ERROR);
 903     }
 904
 905     if (fb) {
 906         qemu_fclose(fb);
 907     }
 908
 909     /* Hope this not to be too long to loop here */
 910     qemu_sem_wait(&mis->colo_incoming_sem);
 911     qemu_sem_destroy(&mis->colo_incoming_sem);
 912     /* Must be called after failover BH is completed */
 913     if (mis->to_src_file) {
 914         qemu_fclose(mis->to_src_file);
 915         mis->to_src_file = NULL;
 916     }
 917
 918     rcu_unregister_thread();
 919     return NULL;
 920 }