migration/colo.c

   1 /*
   2  * COarse-grain LOck-stepping Virtual Machines for Non-stop Service (COLO)
   3  * (a.k.a. Fault Tolerance or Continuous Replication)
   4  *
   5  * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
   6  * Copyright (c) 2016 FUJITSU LIMITED
   7  * Copyright (c) 2016 Intel Corporation
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or
  10  * later.  See the COPYING file in the top-level directory.
  11  */
  12
  13 #include "qemu/osdep.h"
  14 #include "sysemu/sysemu.h"
  15 #include "qapi/error.h"
  16 #include "qapi/qapi-commands-migration.h"
  17 #include "qemu-file-channel.h"
  18 #include "migration.h"
  19 #include "qemu-file.h"
  20 #include "savevm.h"
  21 #include "migration/colo.h"
  22 #include "block.h"
  23 #include "io/channel-buffer.h"
  24 #include "trace.h"
  25 #include "qemu/error-report.h"
  26 #include "migration/failover.h"
  27 #ifdef CONFIG_REPLICATION
  28 #include "replication.h"
  29 #endif
  30 #include "net/colo-compare.h"
  31 #include "net/colo.h"
  32 #include "block/block.h"
  33 #include "qapi/qapi-events-migration.h"
  34 #include "qapi/qmp/qerror.h"
  35 #include "sysemu/cpus.h"
  36 #include "net/filter.h"
  37
  38 static bool vmstate_loading;
  39 static Notifier packets_compare_notifier;
  40
  41 /* User need to know colo mode after COLO failover */
  42 static COLOMode last_colo_mode;
  43
  44 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
  45
  46 bool migration_in_colo_state(void)
  47 {
  48     MigrationState *s = migrate_get_current();
  49
  50     return (s->state == MIGRATION_STATUS_COLO);
  51 }
  52
  53 bool migration_incoming_in_colo_state(void)
  54 {
  55     MigrationIncomingState *mis = migration_incoming_get_current();
  56
  57     return mis && (mis->state == MIGRATION_STATUS_COLO);
  58 }
  59
  60 static bool colo_runstate_is_stopped(void)
  61 {
  62     return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
  63 }
  64
  65 static void secondary_vm_do_failover(void)
  66 {
  67 /* COLO needs enable block-replication */
  68 #ifdef CONFIG_REPLICATION
  69     int old_state;
  70     MigrationIncomingState *mis = migration_incoming_get_current();
  71     Error *local_err = NULL;
  72
  73     /* Can not do failover during the process of VM's loading VMstate, Or
  74      * it will break the secondary VM.
  75      */
  76     if (vmstate_loading) {
  77         old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
  78                         FAILOVER_STATUS_RELAUNCH);
  79         if (old_state != FAILOVER_STATUS_ACTIVE) {
  80             error_report("Unknown error while do failover for secondary VM,"
  81                          "old_state: %s", FailoverStatus_str(old_state));
  82         }
  83         return;
  84     }
  85
  86     migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
  87                       MIGRATION_STATUS_COMPLETED);
  88
  89     replication_stop_all(true, &local_err);
  90     if (local_err) {
  91         error_report_err(local_err);
  92     }
  93
  94     /* Notify all filters of all NIC to do checkpoint */
  95     colo_notify_filters_event(COLO_EVENT_FAILOVER, &local_err);
  96     if (local_err) {
  97         error_report_err(local_err);
  98     }
  99
 100     if (!autostart) {
 101         error_report("\"-S\" qemu option will be ignored in secondary side");
 102         /* recover runstate to normal migration finish state */
 103         autostart = true;
 104     }
 105     /*
 106      * Make sure COLO incoming thread not block in recv or send,
 107      * If mis->from_src_file and mis->to_src_file use the same fd,
 108      * The second shutdown() will return -1, we ignore this value,
 109      * It is harmless.
 110      */
 111     if (mis->from_src_file) {
 112         qemu_file_shutdown(mis->from_src_file);
 113     }
 114     if (mis->to_src_file) {
 115         qemu_file_shutdown(mis->to_src_file);
 116     }
 117
 118     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 119                                    FAILOVER_STATUS_COMPLETED);
 120     if (old_state != FAILOVER_STATUS_ACTIVE) {
 121         error_report("Incorrect state (%s) while doing failover for "
 122                      "secondary VM", FailoverStatus_str(old_state));
 123         return;
 124     }
 125     /* Notify COLO incoming thread that failover work is finished */
 126     qemu_sem_post(&mis->colo_incoming_sem);
 127
 128     /* For Secondary VM, jump to incoming co */
 129     if (mis->migration_incoming_co) {
 130         qemu_coroutine_enter(mis->migration_incoming_co);
 131     }
 132 #else
 133     abort();
 134 #endif
 135 }
 136
 137 static void primary_vm_do_failover(void)
 138 {
 139 #ifdef CONFIG_REPLICATION
 140     MigrationState *s = migrate_get_current();
 141     int old_state;
 142     Error *local_err = NULL;
 143
 144     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
 145                       MIGRATION_STATUS_COMPLETED);
 146     /*
 147      * kick COLO thread which might wait at
 148      * qemu_sem_wait(&s->colo_checkpoint_sem).
 149      */
 150     colo_checkpoint_notify(migrate_get_current());
 151
 152     /*
 153      * Wake up COLO thread which may blocked in recv() or send(),
 154      * The s->rp_state.from_dst_file and s->to_dst_file may use the
 155      * same fd, but we still shutdown the fd for twice, it is harmless.
 156      */
 157     if (s->to_dst_file) {
 158         qemu_file_shutdown(s->to_dst_file);
 159     }
 160     if (s->rp_state.from_dst_file) {
 161         qemu_file_shutdown(s->rp_state.from_dst_file);
 162     }
 163
 164     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 165                                    FAILOVER_STATUS_COMPLETED);
 166     if (old_state != FAILOVER_STATUS_ACTIVE) {
 167         error_report("Incorrect state (%s) while doing failover for Primary VM",
 168                      FailoverStatus_str(old_state));
 169         return;
 170     }
 171
 172     replication_stop_all(true, &local_err);
 173     if (local_err) {
 174         error_report_err(local_err);
 175         local_err = NULL;
 176     }
 177
 178     /* Notify COLO thread that failover work is finished */
 179     qemu_sem_post(&s->colo_exit_sem);
 180 #else
 181     abort();
 182 #endif
 183 }
 184
 185 COLOMode get_colo_mode(void)
 186 {
 187     if (migration_in_colo_state()) {
 188         return COLO_MODE_PRIMARY;
 189     } else if (migration_incoming_in_colo_state()) {
 190         return COLO_MODE_SECONDARY;
 191     } else {
 192         return COLO_MODE_NONE;
 193     }
 194 }
 195
 196 void colo_do_failover(void)
 197 {
 198     /* Make sure VM stopped while failover happened. */
 199     if (!colo_runstate_is_stopped()) {
 200         vm_stop_force_state(RUN_STATE_COLO);
 201     }
 202
 203     switch (get_colo_mode()) {
 204     case COLO_MODE_PRIMARY:
 205         primary_vm_do_failover();
 206         break;
 207     case COLO_MODE_SECONDARY:
 208         secondary_vm_do_failover();
 209         break;
 210     default:
 211         error_report("colo_do_failover failed because the colo mode"
 212                      " could not be obtained");
 213     }
 214 }
 215
 216 #ifdef CONFIG_REPLICATION
 217 void qmp_xen_set_replication(bool enable, bool primary,
 218                              bool has_failover, bool failover,
 219                              Error **errp)
 220 {
 221     ReplicationMode mode = primary ?
 222                            REPLICATION_MODE_PRIMARY :
 223                            REPLICATION_MODE_SECONDARY;
 224
 225     if (has_failover && enable) {
 226         error_setg(errp, "Parameter 'failover' is only for"
 227                    " stopping replication");
 228         return;
 229     }
 230
 231     if (enable) {
 232         replication_start_all(mode, errp);
 233     } else {
 234         if (!has_failover) {
 235             failover = NULL;
 236         }
 237         replication_stop_all(failover, failover ? NULL : errp);
 238     }
 239 }
 240
 241 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
 242 {
 243     Error *err = NULL;
 244     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
 245
 246     replication_get_error_all(&err);
 247     if (err) {
 248         s->error = true;
 249         s->has_desc = true;
 250         s->desc = g_strdup(error_get_pretty(err));
 251     } else {
 252         s->error = false;
 253     }
 254
 255     error_free(err);
 256     return s;
 257 }
 258
 259 void qmp_xen_colo_do_checkpoint(Error **errp)
 260 {
 261     replication_do_checkpoint_all(errp);
 262     /* Notify all filters of all NIC to do checkpoint */
 263     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, errp);
 264 }
 265 #endif
 266
 267 COLOStatus *qmp_query_colo_status(Error **errp)
 268 {
 269     COLOStatus *s = g_new0(COLOStatus, 1);
 270
 271     s->mode = get_colo_mode();
 272     s->last_mode = last_colo_mode;
 273
 274     switch (failover_get_state()) {
 275     case FAILOVER_STATUS_NONE:
 276         s->reason = COLO_EXIT_REASON_NONE;
 277         break;
 278     case FAILOVER_STATUS_COMPLETED:
 279         s->reason = COLO_EXIT_REASON_REQUEST;
 280         break;
 281     default:
 282         if (migration_in_colo_state()) {
 283             s->reason = COLO_EXIT_REASON_PROCESSING;
 284         } else {
 285             s->reason = COLO_EXIT_REASON_ERROR;
 286         }
 287     }
 288
 289     return s;
 290 }
 291
 292 static void colo_send_message(QEMUFile *f, COLOMessage msg,
 293                               Error **errp)
 294 {
 295     int ret;
 296
 297     if (msg >= COLO_MESSAGE__MAX) {
 298         error_setg(errp, "%s: Invalid message", __func__);
 299         return;
 300     }
 301     qemu_put_be32(f, msg);
 302     qemu_fflush(f);
 303
 304     ret = qemu_file_get_error(f);
 305     if (ret < 0) {
 306         error_setg_errno(errp, -ret, "Can't send COLO message");
 307     }
 308     trace_colo_send_message(COLOMessage_str(msg));
 309 }
 310
 311 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
 312                                     uint64_t value, Error **errp)
 313 {
 314     Error *local_err = NULL;
 315     int ret;
 316
 317     colo_send_message(f, msg, &local_err);
 318     if (local_err) {
 319         error_propagate(errp, local_err);
 320         return;
 321     }
 322     qemu_put_be64(f, value);
 323     qemu_fflush(f);
 324
 325     ret = qemu_file_get_error(f);
 326     if (ret < 0) {
 327         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
 328                          COLOMessage_str(msg));
 329     }
 330 }
 331
 332 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
 333 {
 334     COLOMessage msg;
 335     int ret;
 336
 337     msg = qemu_get_be32(f);
 338     ret = qemu_file_get_error(f);
 339     if (ret < 0) {
 340         error_setg_errno(errp, -ret, "Can't receive COLO message");
 341         return msg;
 342     }
 343     if (msg >= COLO_MESSAGE__MAX) {
 344         error_setg(errp, "%s: Invalid message", __func__);
 345         return msg;
 346     }
 347     trace_colo_receive_message(COLOMessage_str(msg));
 348     return msg;
 349 }
 350
 351 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
 352                                        Error **errp)
 353 {
 354     COLOMessage msg;
 355     Error *local_err = NULL;
 356
 357     msg = colo_receive_message(f, &local_err);
 358     if (local_err) {
 359         error_propagate(errp, local_err);
 360         return;
 361     }
 362     if (msg != expect_msg) {
 363         error_setg(errp, "Unexpected COLO message %d, expected %d",
 364                           msg, expect_msg);
 365     }
 366 }
 367
 368 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
 369                                            Error **errp)
 370 {
 371     Error *local_err = NULL;
 372     uint64_t value;
 373     int ret;
 374
 375     colo_receive_check_message(f, expect_msg, &local_err);
 376     if (local_err) {
 377         error_propagate(errp, local_err);
 378         return 0;
 379     }
 380
 381     value = qemu_get_be64(f);
 382     ret = qemu_file_get_error(f);
 383     if (ret < 0) {
 384         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
 385                          COLOMessage_str(expect_msg));
 386     }
 387     return value;
 388 }
 389
 390 static int colo_do_checkpoint_transaction(MigrationState *s,
 391                                           QIOChannelBuffer *bioc,
 392                                           QEMUFile *fb)
 393 {
 394     Error *local_err = NULL;
 395     int ret = -1;
 396
 397     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
 398                       &local_err);
 399     if (local_err) {
 400         goto out;
 401     }
 402
 403     colo_receive_check_message(s->rp_state.from_dst_file,
 404                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
 405     if (local_err) {
 406         goto out;
 407     }
 408     /* Reset channel-buffer directly */
 409     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 410     bioc->usage = 0;
 411
 412     qemu_mutex_lock_iothread();
 413     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 414         qemu_mutex_unlock_iothread();
 415         goto out;
 416     }
 417     vm_stop_force_state(RUN_STATE_COLO);
 418     qemu_mutex_unlock_iothread();
 419     trace_colo_vm_state_change("run", "stop");
 420     /*
 421      * Failover request bh could be called after vm_stop_force_state(),
 422      * So we need check failover_request_is_active() again.
 423      */
 424     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 425         goto out;
 426     }
 427
 428     colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
 429     if (local_err) {
 430         goto out;
 431     }
 432
 433     /* Disable block migration */
 434     migrate_set_block_enabled(false, &local_err);
 435     qemu_mutex_lock_iothread();
 436
 437 #ifdef CONFIG_REPLICATION
 438     replication_do_checkpoint_all(&local_err);
 439     if (local_err) {
 440         qemu_mutex_unlock_iothread();
 441         goto out;
 442     }
 443 #else
 444         abort();
 445 #endif
 446
 447     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
 448     if (local_err) {
 449         qemu_mutex_unlock_iothread();
 450         goto out;
 451     }
 452     /* Note: device state is saved into buffer */
 453     ret = qemu_save_device_state(fb);
 454
 455     qemu_mutex_unlock_iothread();
 456     if (ret < 0) {
 457         goto out;
 458     }
 459     /*
 460      * Only save VM's live state, which not including device state.
 461      * TODO: We may need a timeout mechanism to prevent COLO process
 462      * to be blocked here.
 463      */
 464     qemu_savevm_live_state(s->to_dst_file);
 465
 466     qemu_fflush(fb);
 467
 468     /*
 469      * We need the size of the VMstate data in Secondary side,
 470      * With which we can decide how much data should be read.
 471      */
 472     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
 473                             bioc->usage, &local_err);
 474     if (local_err) {
 475         goto out;
 476     }
 477
 478     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
 479     qemu_fflush(s->to_dst_file);
 480     ret = qemu_file_get_error(s->to_dst_file);
 481     if (ret < 0) {
 482         goto out;
 483     }
 484
 485     colo_receive_check_message(s->rp_state.from_dst_file,
 486                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
 487     if (local_err) {
 488         goto out;
 489     }
 490
 491     colo_receive_check_message(s->rp_state.from_dst_file,
 492                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
 493     if (local_err) {
 494         goto out;
 495     }
 496
 497     ret = 0;
 498
 499     qemu_mutex_lock_iothread();
 500     vm_start();
 501     qemu_mutex_unlock_iothread();
 502     trace_colo_vm_state_change("stop", "run");
 503
 504 out:
 505     if (local_err) {
 506         error_report_err(local_err);
 507     }
 508     return ret;
 509 }
 510
 511 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
 512 {
 513     colo_checkpoint_notify(data);
 514 }
 515
 516 static void colo_process_checkpoint(MigrationState *s)
 517 {
 518     QIOChannelBuffer *bioc;
 519     QEMUFile *fb = NULL;
 520     int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 521     Error *local_err = NULL;
 522     int ret;
 523
 524     last_colo_mode = get_colo_mode();
 525     if (last_colo_mode != COLO_MODE_PRIMARY) {
 526         error_report("COLO mode must be COLO_MODE_PRIMARY");
 527         return;
 528     }
 529
 530     failover_init_state();
 531
 532     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
 533     if (!s->rp_state.from_dst_file) {
 534         error_report("Open QEMUFile from_dst_file failed");
 535         goto out;
 536     }
 537
 538     packets_compare_notifier.notify = colo_compare_notify_checkpoint;
 539     colo_compare_register_notifier(&packets_compare_notifier);
 540
 541     /*
 542      * Wait for Secondary finish loading VM states and enter COLO
 543      * restore.
 544      */
 545     colo_receive_check_message(s->rp_state.from_dst_file,
 546                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
 547     if (local_err) {
 548         goto out;
 549     }
 550     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 551     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
 552     object_unref(OBJECT(bioc));
 553
 554     qemu_mutex_lock_iothread();
 555 #ifdef CONFIG_REPLICATION
 556     replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
 557     if (local_err) {
 558         qemu_mutex_unlock_iothread();
 559         goto out;
 560     }
 561 #else
 562         abort();
 563 #endif
 564
 565     vm_start();
 566     qemu_mutex_unlock_iothread();
 567     trace_colo_vm_state_change("stop", "run");
 568
 569     timer_mod(s->colo_delay_timer,
 570             current_time + s->parameters.x_checkpoint_delay);
 571
 572     while (s->state == MIGRATION_STATUS_COLO) {
 573         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 574             error_report("failover request");
 575             goto out;
 576         }
 577
 578         qemu_sem_wait(&s->colo_checkpoint_sem);
 579
 580         if (s->state != MIGRATION_STATUS_COLO) {
 581             goto out;
 582         }
 583         ret = colo_do_checkpoint_transaction(s, bioc, fb);
 584         if (ret < 0) {
 585             goto out;
 586         }
 587     }
 588
 589 out:
 590     /* Throw the unreported error message after exited from loop */
 591     if (local_err) {
 592         error_report_err(local_err);
 593     }
 594
 595     if (fb) {
 596         qemu_fclose(fb);
 597     }
 598
 599     /*
 600      * There are only two reasons we can get here, some error happened
 601      * or the user triggered failover.
 602      */
 603     switch (failover_get_state()) {
 604     case FAILOVER_STATUS_COMPLETED:
 605         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 606                                   COLO_EXIT_REASON_REQUEST);
 607         break;
 608     default:
 609         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 610                                   COLO_EXIT_REASON_ERROR);
 611     }
 612
 613     /* Hope this not to be too long to wait here */
 614     qemu_sem_wait(&s->colo_exit_sem);
 615     qemu_sem_destroy(&s->colo_exit_sem);
 616
 617     /*
 618      * It is safe to unregister notifier after failover finished.
 619      * Besides, colo_delay_timer and colo_checkpoint_sem can't be
 620      * released befor unregister notifier, or there will be use-after-free
 621      * error.
 622      */
 623     colo_compare_unregister_notifier(&packets_compare_notifier);
 624     timer_del(s->colo_delay_timer);
 625     timer_free(s->colo_delay_timer);
 626     qemu_sem_destroy(&s->colo_checkpoint_sem);
 627
 628     /*
 629      * Must be called after failover BH is completed,
 630      * Or the failover BH may shutdown the wrong fd that
 631      * re-used by other threads after we release here.
 632      */
 633     if (s->rp_state.from_dst_file) {
 634         qemu_fclose(s->rp_state.from_dst_file);
 635     }
 636 }
 637
 638 void colo_checkpoint_notify(void *opaque)
 639 {
 640     MigrationState *s = opaque;
 641     int64_t next_notify_time;
 642
 643     qemu_sem_post(&s->colo_checkpoint_sem);
 644     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 645     next_notify_time = s->colo_checkpoint_time +
 646                     s->parameters.x_checkpoint_delay;
 647     timer_mod(s->colo_delay_timer, next_notify_time);
 648 }
 649
 650 void migrate_start_colo_process(MigrationState *s)
 651 {
 652     qemu_mutex_unlock_iothread();
 653     qemu_sem_init(&s->colo_checkpoint_sem, 0);
 654     s->colo_delay_timer =  timer_new_ms(QEMU_CLOCK_HOST,
 655                                 colo_checkpoint_notify, s);
 656
 657     qemu_sem_init(&s->colo_exit_sem, 0);
 658     migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
 659                       MIGRATION_STATUS_COLO);
 660     colo_process_checkpoint(s);
 661     qemu_mutex_lock_iothread();
 662 }
 663
 664 static void colo_wait_handle_message(QEMUFile *f, int *checkpoint_request,
 665                                      Error **errp)
 666 {
 667     COLOMessage msg;
 668     Error *local_err = NULL;
 669
 670     msg = colo_receive_message(f, &local_err);
 671     if (local_err) {
 672         error_propagate(errp, local_err);
 673         return;
 674     }
 675
 676     switch (msg) {
 677     case COLO_MESSAGE_CHECKPOINT_REQUEST:
 678         *checkpoint_request = 1;
 679         break;
 680     default:
 681         *checkpoint_request = 0;
 682         error_setg(errp, "Got unknown COLO message: %d", msg);
 683         break;
 684     }
 685 }
 686
 687 void *colo_process_incoming_thread(void *opaque)
 688 {
 689     MigrationIncomingState *mis = opaque;
 690     QEMUFile *fb = NULL;
 691     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
 692     uint64_t total_size;
 693     uint64_t value;
 694     Error *local_err = NULL;
 695     int ret;
 696
 697     rcu_register_thread();
 698     qemu_sem_init(&mis->colo_incoming_sem, 0);
 699
 700     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 701                       MIGRATION_STATUS_COLO);
 702
 703     last_colo_mode = get_colo_mode();
 704     if (last_colo_mode != COLO_MODE_SECONDARY) {
 705         error_report("COLO mode must be COLO_MODE_SECONDARY");
 706         return NULL;
 707     }
 708
 709     failover_init_state();
 710
 711     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
 712     if (!mis->to_src_file) {
 713         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
 714         goto out;
 715     }
 716     /*
 717      * Note: the communication between Primary side and Secondary side
 718      * should be sequential, we set the fd to unblocked in migration incoming
 719      * coroutine, and here we are in the COLO incoming thread, so it is ok to
 720      * set the fd back to blocked.
 721      */
 722     qemu_file_set_blocking(mis->from_src_file, true);
 723
 724     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 725     fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
 726     object_unref(OBJECT(bioc));
 727
 728     qemu_mutex_lock_iothread();
 729 #ifdef CONFIG_REPLICATION
 730     replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
 731     if (local_err) {
 732         qemu_mutex_unlock_iothread();
 733         goto out;
 734     }
 735 #else
 736         abort();
 737 #endif
 738     vm_start();
 739     trace_colo_vm_state_change("stop", "run");
 740     qemu_mutex_unlock_iothread();
 741
 742     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
 743                       &local_err);
 744     if (local_err) {
 745         goto out;
 746     }
 747
 748     while (mis->state == MIGRATION_STATUS_COLO) {
 749         int request = 0;
 750
 751         colo_wait_handle_message(mis->from_src_file, &request, &local_err);
 752         if (local_err) {
 753             goto out;
 754         }
 755         assert(request);
 756         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 757             error_report("failover request");
 758             goto out;
 759         }
 760
 761         qemu_mutex_lock_iothread();
 762         vm_stop_force_state(RUN_STATE_COLO);
 763         trace_colo_vm_state_change("run", "stop");
 764         qemu_mutex_unlock_iothread();
 765
 766         /* FIXME: This is unnecessary for periodic checkpoint mode */
 767         colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
 768                      &local_err);
 769         if (local_err) {
 770             goto out;
 771         }
 772
 773         colo_receive_check_message(mis->from_src_file,
 774                            COLO_MESSAGE_VMSTATE_SEND, &local_err);
 775         if (local_err) {
 776             goto out;
 777         }
 778
 779         qemu_mutex_lock_iothread();
 780         cpu_synchronize_all_pre_loadvm();
 781         ret = qemu_loadvm_state_main(mis->from_src_file, mis);
 782         qemu_mutex_unlock_iothread();
 783
 784         if (ret < 0) {
 785             error_report("Load VM's live state (ram) error");
 786             goto out;
 787         }
 788
 789         value = colo_receive_message_value(mis->from_src_file,
 790                                  COLO_MESSAGE_VMSTATE_SIZE, &local_err);
 791         if (local_err) {
 792             goto out;
 793         }
 794
 795         /*
 796          * Read VM device state data into channel buffer,
 797          * It's better to re-use the memory allocated.
 798          * Here we need to handle the channel buffer directly.
 799          */
 800         if (value > bioc->capacity) {
 801             bioc->capacity = value;
 802             bioc->data = g_realloc(bioc->data, bioc->capacity);
 803         }
 804         total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
 805         if (total_size != value) {
 806             error_report("Got %" PRIu64 " VMState data, less than expected"
 807                         " %" PRIu64, total_size, value);
 808             goto out;
 809         }
 810         bioc->usage = total_size;
 811         qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 812
 813         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
 814                      &local_err);
 815         if (local_err) {
 816             goto out;
 817         }
 818
 819         qemu_mutex_lock_iothread();
 820         vmstate_loading = true;
 821         ret = qemu_load_device_state(fb);
 822         if (ret < 0) {
 823             error_report("COLO: load device state failed");
 824             qemu_mutex_unlock_iothread();
 825             goto out;
 826         }
 827
 828 #ifdef CONFIG_REPLICATION
 829         replication_get_error_all(&local_err);
 830         if (local_err) {
 831             qemu_mutex_unlock_iothread();
 832             goto out;
 833         }
 834
 835         /* discard colo disk buffer */
 836         replication_do_checkpoint_all(&local_err);
 837         if (local_err) {
 838             qemu_mutex_unlock_iothread();
 839             goto out;
 840         }
 841 #else
 842         abort();
 843 #endif
 844         /* Notify all filters of all NIC to do checkpoint */
 845         colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
 846
 847         if (local_err) {
 848             qemu_mutex_unlock_iothread();
 849             goto out;
 850         }
 851
 852         vmstate_loading = false;
 853         vm_start();
 854         trace_colo_vm_state_change("stop", "run");
 855         qemu_mutex_unlock_iothread();
 856
 857         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 858             failover_set_state(FAILOVER_STATUS_RELAUNCH,
 859                             FAILOVER_STATUS_NONE);
 860             failover_request_active(NULL);
 861             goto out;
 862         }
 863
 864         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
 865                      &local_err);
 866         if (local_err) {
 867             goto out;
 868         }
 869     }
 870
 871 out:
 872     vmstate_loading = false;
 873     /* Throw the unreported error message after exited from loop */
 874     if (local_err) {
 875         error_report_err(local_err);
 876     }
 877
 878     /*
 879      * There are only two reasons we can get here, some error happened
 880      * or the user triggered failover.
 881      */
 882     switch (failover_get_state()) {
 883     case FAILOVER_STATUS_COMPLETED:
 884         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 885                                   COLO_EXIT_REASON_REQUEST);
 886         break;
 887     default:
 888         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 889                                   COLO_EXIT_REASON_ERROR);
 890     }
 891
 892     if (fb) {
 893         qemu_fclose(fb);
 894     }
 895
 896     /* Hope this not to be too long to loop here */
 897     qemu_sem_wait(&mis->colo_incoming_sem);
 898     qemu_sem_destroy(&mis->colo_incoming_sem);
 899     /* Must be called after failover BH is completed */
 900     if (mis->to_src_file) {
 901         qemu_fclose(mis->to_src_file);
 902         mis->to_src_file = NULL;
 903     }
 904
 905     rcu_unregister_thread();
 906     return NULL;
 907 }