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