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