migration/colo.c

   1 /*
   2  * COarse-grain LOck-stepping Virtual Machines for Non-stop Service (COLO)
   3  * (a.k.a. Fault Tolerance or Continuous Replication)
   4  *
   5  * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
   6  * Copyright (c) 2016 FUJITSU LIMITED
   7  * Copyright (c) 2016 Intel Corporation
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or
  10  * later.  See the COPYING file in the top-level directory.
  11  */
  12
  13 #include "qemu/osdep.h"
  14 #include "sysemu/sysemu.h"
  15 #include "qapi/error.h"
  16 #include "qapi/qapi-commands-migration.h"
  17 #include "migration.h"
  18 #include "qemu-file.h"
  19 #include "savevm.h"
  20 #include "migration/colo.h"
  21 #include "block.h"
  22 #include "io/channel-buffer.h"
  23 #include "trace.h"
  24 #include "qemu/error-report.h"
  25 #include "qemu/main-loop.h"
  26 #include "qemu/rcu.h"
  27 #include "migration/failover.h"
  28 #include "migration/ram.h"
  29 #ifdef CONFIG_REPLICATION
  30 #include "block/replication.h"
  31 #endif
  32 #include "net/colo-compare.h"
  33 #include "net/colo.h"
  34 #include "block/block.h"
  35 #include "qapi/qapi-events-migration.h"
  36 #include "sysemu/cpus.h"
  37 #include "sysemu/runstate.h"
  38 #include "net/filter.h"
  39
  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         local_err = NULL;
  95     }
  96
  97     /* Notify all filters of all NIC to do checkpoint */
  98     colo_notify_filters_event(COLO_EVENT_FAILOVER, &local_err);
  99     if (local_err) {
 100         error_report_err(local_err);
 101     }
 102
 103     if (!autostart) {
 104         error_report("\"-S\" qemu option will be ignored in secondary side");
 105         /* recover runstate to normal migration finish state */
 106         autostart = true;
 107     }
 108     /*
 109      * Make sure COLO incoming thread not block in recv or send,
 110      * If mis->from_src_file and mis->to_src_file use the same fd,
 111      * The second shutdown() will return -1, we ignore this value,
 112      * It is harmless.
 113      */
 114     if (mis->from_src_file) {
 115         qemu_file_shutdown(mis->from_src_file);
 116     }
 117     if (mis->to_src_file) {
 118         qemu_file_shutdown(mis->to_src_file);
 119     }
 120
 121     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 122                                    FAILOVER_STATUS_COMPLETED);
 123     if (old_state != FAILOVER_STATUS_ACTIVE) {
 124         error_report("Incorrect state (%s) while doing failover for "
 125                      "secondary VM", FailoverStatus_str(old_state));
 126         return;
 127     }
 128     /* Notify COLO incoming thread that failover work is finished */
 129     qemu_sem_post(&mis->colo_incoming_sem);
 130
 131     /* For Secondary VM, jump to incoming co */
 132     if (mis->migration_incoming_co) {
 133         qemu_coroutine_enter(mis->migration_incoming_co);
 134     }
 135 #else
 136     abort();
 137 #endif
 138 }
 139
 140 static void primary_vm_do_failover(void)
 141 {
 142 #ifdef CONFIG_REPLICATION
 143     MigrationState *s = migrate_get_current();
 144     int old_state;
 145     Error *local_err = NULL;
 146
 147     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
 148                       MIGRATION_STATUS_COMPLETED);
 149     /*
 150      * kick COLO thread which might wait at
 151      * qemu_sem_wait(&s->colo_checkpoint_sem).
 152      */
 153     colo_checkpoint_notify(s);
 154
 155     /*
 156      * Wake up COLO thread which may blocked in recv() or send(),
 157      * The s->rp_state.from_dst_file and s->to_dst_file may use the
 158      * same fd, but we still shutdown the fd for twice, it is harmless.
 159      */
 160     if (s->to_dst_file) {
 161         qemu_file_shutdown(s->to_dst_file);
 162     }
 163     if (s->rp_state.from_dst_file) {
 164         qemu_file_shutdown(s->rp_state.from_dst_file);
 165     }
 166
 167     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 168                                    FAILOVER_STATUS_COMPLETED);
 169     if (old_state != FAILOVER_STATUS_ACTIVE) {
 170         error_report("Incorrect state (%s) while doing failover for Primary VM",
 171                      FailoverStatus_str(old_state));
 172         return;
 173     }
 174
 175     replication_stop_all(true, &local_err);
 176     if (local_err) {
 177         error_report_err(local_err);
 178         local_err = NULL;
 179     }
 180
 181     /* Notify COLO thread that failover work is finished */
 182     qemu_sem_post(&s->colo_exit_sem);
 183 #else
 184     abort();
 185 #endif
 186 }
 187
 188 COLOMode get_colo_mode(void)
 189 {
 190     if (migration_in_colo_state()) {
 191         return COLO_MODE_PRIMARY;
 192     } else if (migration_incoming_in_colo_state()) {
 193         return COLO_MODE_SECONDARY;
 194     } else {
 195         return COLO_MODE_NONE;
 196     }
 197 }
 198
 199 void colo_do_failover(void)
 200 {
 201     /* Make sure VM stopped while failover happened. */
 202     if (!colo_runstate_is_stopped()) {
 203         vm_stop_force_state(RUN_STATE_COLO);
 204     }
 205
 206     switch (last_colo_mode = get_colo_mode()) {
 207     case COLO_MODE_PRIMARY:
 208         primary_vm_do_failover();
 209         break;
 210     case COLO_MODE_SECONDARY:
 211         secondary_vm_do_failover();
 212         break;
 213     default:
 214         error_report("colo_do_failover failed because the colo mode"
 215                      " could not be obtained");
 216     }
 217 }
 218
 219 #ifdef CONFIG_REPLICATION
 220 void qmp_xen_set_replication(bool enable, bool primary,
 221                              bool has_failover, bool failover,
 222                              Error **errp)
 223 {
 224     ReplicationMode mode = primary ?
 225                            REPLICATION_MODE_PRIMARY :
 226                            REPLICATION_MODE_SECONDARY;
 227
 228     if (has_failover && enable) {
 229         error_setg(errp, "Parameter 'failover' is only for"
 230                    " stopping replication");
 231         return;
 232     }
 233
 234     if (enable) {
 235         replication_start_all(mode, errp);
 236     } else {
 237         if (!has_failover) {
 238             failover = NULL;
 239         }
 240         replication_stop_all(failover, failover ? NULL : errp);
 241     }
 242 }
 243
 244 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
 245 {
 246     Error *err = NULL;
 247     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
 248
 249     replication_get_error_all(&err);
 250     if (err) {
 251         s->error = 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     Error *err = NULL;
 264
 265     replication_do_checkpoint_all(&err);
 266     if (err) {
 267         error_propagate(errp, err);
 268         return;
 269     }
 270     /* Notify all filters of all NIC to do checkpoint */
 271     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, errp);
 272 }
 273 #endif
 274
 275 COLOStatus *qmp_query_colo_status(Error **errp)
 276 {
 277     COLOStatus *s = g_new0(COLOStatus, 1);
 278
 279     s->mode = get_colo_mode();
 280     s->last_mode = last_colo_mode;
 281
 282     switch (failover_get_state()) {
 283     case FAILOVER_STATUS_NONE:
 284         s->reason = COLO_EXIT_REASON_NONE;
 285         break;
 286     case FAILOVER_STATUS_COMPLETED:
 287         s->reason = COLO_EXIT_REASON_REQUEST;
 288         break;
 289     default:
 290         if (migration_in_colo_state()) {
 291             s->reason = COLO_EXIT_REASON_PROCESSING;
 292         } else {
 293             s->reason = COLO_EXIT_REASON_ERROR;
 294         }
 295     }
 296
 297     return s;
 298 }
 299
 300 static void colo_send_message(QEMUFile *f, COLOMessage msg,
 301                               Error **errp)
 302 {
 303     int ret;
 304
 305     if (msg >= COLO_MESSAGE__MAX) {
 306         error_setg(errp, "%s: Invalid message", __func__);
 307         return;
 308     }
 309     qemu_put_be32(f, msg);
 310     qemu_fflush(f);
 311
 312     ret = qemu_file_get_error(f);
 313     if (ret < 0) {
 314         error_setg_errno(errp, -ret, "Can't send COLO message");
 315     }
 316     trace_colo_send_message(COLOMessage_str(msg));
 317 }
 318
 319 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
 320                                     uint64_t value, Error **errp)
 321 {
 322     Error *local_err = NULL;
 323     int ret;
 324
 325     colo_send_message(f, msg, &local_err);
 326     if (local_err) {
 327         error_propagate(errp, local_err);
 328         return;
 329     }
 330     qemu_put_be64(f, value);
 331     qemu_fflush(f);
 332
 333     ret = qemu_file_get_error(f);
 334     if (ret < 0) {
 335         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
 336                          COLOMessage_str(msg));
 337     }
 338 }
 339
 340 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
 341 {
 342     COLOMessage msg;
 343     int ret;
 344
 345     msg = qemu_get_be32(f);
 346     ret = qemu_file_get_error(f);
 347     if (ret < 0) {
 348         error_setg_errno(errp, -ret, "Can't receive COLO message");
 349         return msg;
 350     }
 351     if (msg >= COLO_MESSAGE__MAX) {
 352         error_setg(errp, "%s: Invalid message", __func__);
 353         return msg;
 354     }
 355     trace_colo_receive_message(COLOMessage_str(msg));
 356     return msg;
 357 }
 358
 359 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
 360                                        Error **errp)
 361 {
 362     COLOMessage msg;
 363     Error *local_err = NULL;
 364
 365     msg = colo_receive_message(f, &local_err);
 366     if (local_err) {
 367         error_propagate(errp, local_err);
 368         return;
 369     }
 370     if (msg != expect_msg) {
 371         error_setg(errp, "Unexpected COLO message %d, expected %d",
 372                           msg, expect_msg);
 373     }
 374 }
 375
 376 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
 377                                            Error **errp)
 378 {
 379     Error *local_err = NULL;
 380     uint64_t value;
 381     int ret;
 382
 383     colo_receive_check_message(f, expect_msg, &local_err);
 384     if (local_err) {
 385         error_propagate(errp, local_err);
 386         return 0;
 387     }
 388
 389     value = qemu_get_be64(f);
 390     ret = qemu_file_get_error(f);
 391     if (ret < 0) {
 392         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
 393                          COLOMessage_str(expect_msg));
 394     }
 395     return value;
 396 }
 397
 398 static int colo_do_checkpoint_transaction(MigrationState *s,
 399                                           QIOChannelBuffer *bioc,
 400                                           QEMUFile *fb)
 401 {
 402     Error *local_err = NULL;
 403     int ret = -1;
 404
 405     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
 406                       &local_err);
 407     if (local_err) {
 408         goto out;
 409     }
 410
 411     colo_receive_check_message(s->rp_state.from_dst_file,
 412                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
 413     if (local_err) {
 414         goto out;
 415     }
 416     /* Reset channel-buffer directly */
 417     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 418     bioc->usage = 0;
 419
 420     qemu_mutex_lock_iothread();
 421     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 422         qemu_mutex_unlock_iothread();
 423         goto out;
 424     }
 425     vm_stop_force_state(RUN_STATE_COLO);
 426     qemu_mutex_unlock_iothread();
 427     trace_colo_vm_state_change("run", "stop");
 428     /*
 429      * Failover request bh could be called after vm_stop_force_state(),
 430      * So we need check failover_request_is_active() again.
 431      */
 432     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 433         goto out;
 434     }
 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     if (migrate_auto_converge()) {
 461         mig_throttle_counter_reset();
 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     qemu_event_reset(&s->colo_checkpoint_event);
 496     colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
 497     if (local_err) {
 498         goto out;
 499     }
 500
 501     colo_receive_check_message(s->rp_state.from_dst_file,
 502                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
 503     if (local_err) {
 504         goto out;
 505     }
 506
 507     ret = 0;
 508
 509     qemu_mutex_lock_iothread();
 510     vm_start();
 511     qemu_mutex_unlock_iothread();
 512     trace_colo_vm_state_change("stop", "run");
 513
 514 out:
 515     if (local_err) {
 516         error_report_err(local_err);
 517     }
 518     return ret;
 519 }
 520
 521 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
 522 {
 523     colo_checkpoint_notify(data);
 524 }
 525
 526 static void colo_process_checkpoint(MigrationState *s)
 527 {
 528     QIOChannelBuffer *bioc;
 529     QEMUFile *fb = NULL;
 530     Error *local_err = NULL;
 531     int ret;
 532
 533     if (get_colo_mode() != COLO_MODE_PRIMARY) {
 534         error_report("COLO mode must be COLO_MODE_PRIMARY");
 535         return;
 536     }
 537
 538     failover_init_state();
 539
 540     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
 541     if (!s->rp_state.from_dst_file) {
 542         error_report("Open QEMUFile from_dst_file failed");
 543         goto out;
 544     }
 545
 546     packets_compare_notifier.notify = colo_compare_notify_checkpoint;
 547     colo_compare_register_notifier(&packets_compare_notifier);
 548
 549     /*
 550      * Wait for Secondary finish loading VM states and enter COLO
 551      * restore.
 552      */
 553     colo_receive_check_message(s->rp_state.from_dst_file,
 554                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
 555     if (local_err) {
 556         goto out;
 557     }
 558     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 559     fb = qemu_file_new_output(QIO_CHANNEL(bioc));
 560     object_unref(OBJECT(bioc));
 561
 562     qemu_mutex_lock_iothread();
 563 #ifdef CONFIG_REPLICATION
 564     replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
 565     if (local_err) {
 566         qemu_mutex_unlock_iothread();
 567         goto out;
 568     }
 569 #else
 570         abort();
 571 #endif
 572
 573     vm_start();
 574     qemu_mutex_unlock_iothread();
 575     trace_colo_vm_state_change("stop", "run");
 576
 577     timer_mod(s->colo_delay_timer, qemu_clock_get_ms(QEMU_CLOCK_HOST) +
 578               s->parameters.x_checkpoint_delay);
 579
 580     while (s->state == MIGRATION_STATUS_COLO) {
 581         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 582             error_report("failover request");
 583             goto out;
 584         }
 585
 586         qemu_event_wait(&s->colo_checkpoint_event);
 587
 588         if (s->state != MIGRATION_STATUS_COLO) {
 589             goto out;
 590         }
 591         ret = colo_do_checkpoint_transaction(s, bioc, fb);
 592         if (ret < 0) {
 593             goto out;
 594         }
 595     }
 596
 597 out:
 598     /* Throw the unreported error message after exited from loop */
 599     if (local_err) {
 600         error_report_err(local_err);
 601     }
 602
 603     if (fb) {
 604         qemu_fclose(fb);
 605     }
 606
 607     /*
 608      * There are only two reasons we can get here, some error happened
 609      * or the user triggered failover.
 610      */
 611     switch (failover_get_state()) {
 612     case FAILOVER_STATUS_COMPLETED:
 613         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 614                                   COLO_EXIT_REASON_REQUEST);
 615         break;
 616     default:
 617         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 618                                   COLO_EXIT_REASON_ERROR);
 619     }
 620
 621     /* Hope this not to be too long to wait here */
 622     qemu_sem_wait(&s->colo_exit_sem);
 623     qemu_sem_destroy(&s->colo_exit_sem);
 624
 625     /*
 626      * It is safe to unregister notifier after failover finished.
 627      * Besides, colo_delay_timer and colo_checkpoint_sem can't be
 628      * released before unregister notifier, or there will be use-after-free
 629      * error.
 630      */
 631     colo_compare_unregister_notifier(&packets_compare_notifier);
 632     timer_free(s->colo_delay_timer);
 633     qemu_event_destroy(&s->colo_checkpoint_event);
 634
 635     /*
 636      * Must be called after failover BH is completed,
 637      * Or the failover BH may shutdown the wrong fd that
 638      * re-used by other threads after we release here.
 639      */
 640     if (s->rp_state.from_dst_file) {
 641         qemu_fclose(s->rp_state.from_dst_file);
 642         s->rp_state.from_dst_file = NULL;
 643     }
 644 }
 645
 646 void colo_checkpoint_notify(void *opaque)
 647 {
 648     MigrationState *s = opaque;
 649     int64_t next_notify_time;
 650
 651     qemu_event_set(&s->colo_checkpoint_event);
 652     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 653     next_notify_time = s->colo_checkpoint_time +
 654                     s->parameters.x_checkpoint_delay;
 655     timer_mod(s->colo_delay_timer, next_notify_time);
 656 }
 657
 658 void migrate_start_colo_process(MigrationState *s)
 659 {
 660     qemu_mutex_unlock_iothread();
 661     qemu_event_init(&s->colo_checkpoint_event, false);
 662     s->colo_delay_timer =  timer_new_ms(QEMU_CLOCK_HOST,
 663                                 colo_checkpoint_notify, s);
 664
 665     qemu_sem_init(&s->colo_exit_sem, 0);
 666     colo_process_checkpoint(s);
 667     qemu_mutex_lock_iothread();
 668 }
 669
 670 static void colo_incoming_process_checkpoint(MigrationIncomingState *mis,
 671                       QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
 672 {
 673     uint64_t total_size;
 674     uint64_t value;
 675     Error *local_err = NULL;
 676     int ret;
 677
 678     qemu_mutex_lock_iothread();
 679     vm_stop_force_state(RUN_STATE_COLO);
 680     qemu_mutex_unlock_iothread();
 681     trace_colo_vm_state_change("run", "stop");
 682
 683     /* FIXME: This is unnecessary for periodic checkpoint mode */
 684     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
 685                  &local_err);
 686     if (local_err) {
 687         error_propagate(errp, local_err);
 688         return;
 689     }
 690
 691     colo_receive_check_message(mis->from_src_file,
 692                        COLO_MESSAGE_VMSTATE_SEND, &local_err);
 693     if (local_err) {
 694         error_propagate(errp, local_err);
 695         return;
 696     }
 697
 698     qemu_mutex_lock_iothread();
 699     cpu_synchronize_all_states();
 700     ret = qemu_loadvm_state_main(mis->from_src_file, mis);
 701     qemu_mutex_unlock_iothread();
 702
 703     if (ret < 0) {
 704         error_setg(errp, "Load VM's live state (ram) error");
 705         return;
 706     }
 707
 708     value = colo_receive_message_value(mis->from_src_file,
 709                              COLO_MESSAGE_VMSTATE_SIZE, &local_err);
 710     if (local_err) {
 711         error_propagate(errp, local_err);
 712         return;
 713     }
 714
 715     /*
 716      * Read VM device state data into channel buffer,
 717      * It's better to re-use the memory allocated.
 718      * Here we need to handle the channel buffer directly.
 719      */
 720     if (value > bioc->capacity) {
 721         bioc->capacity = value;
 722         bioc->data = g_realloc(bioc->data, bioc->capacity);
 723     }
 724     total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
 725     if (total_size != value) {
 726         error_setg(errp, "Got %" PRIu64 " VMState data, less than expected"
 727                     " %" PRIu64, total_size, value);
 728         return;
 729     }
 730     bioc->usage = total_size;
 731     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 732
 733     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
 734                  &local_err);
 735     if (local_err) {
 736         error_propagate(errp, local_err);
 737         return;
 738     }
 739
 740     qemu_mutex_lock_iothread();
 741     vmstate_loading = true;
 742     colo_flush_ram_cache();
 743     ret = qemu_load_device_state(fb);
 744     if (ret < 0) {
 745         error_setg(errp, "COLO: load device state failed");
 746         vmstate_loading = false;
 747         qemu_mutex_unlock_iothread();
 748         return;
 749     }
 750
 751 #ifdef CONFIG_REPLICATION
 752     replication_get_error_all(&local_err);
 753     if (local_err) {
 754         error_propagate(errp, local_err);
 755         vmstate_loading = false;
 756         qemu_mutex_unlock_iothread();
 757         return;
 758     }
 759
 760     /* discard colo disk buffer */
 761     replication_do_checkpoint_all(&local_err);
 762     if (local_err) {
 763         error_propagate(errp, local_err);
 764         vmstate_loading = false;
 765         qemu_mutex_unlock_iothread();
 766         return;
 767     }
 768 #else
 769     abort();
 770 #endif
 771     /* Notify all filters of all NIC to do checkpoint */
 772     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
 773
 774     if (local_err) {
 775         error_propagate(errp, local_err);
 776         vmstate_loading = false;
 777         qemu_mutex_unlock_iothread();
 778         return;
 779     }
 780
 781     vmstate_loading = false;
 782     vm_start();
 783     qemu_mutex_unlock_iothread();
 784     trace_colo_vm_state_change("stop", "run");
 785
 786     if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 787         return;
 788     }
 789
 790     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
 791                  &local_err);
 792     error_propagate(errp, local_err);
 793 }
 794
 795 static void colo_wait_handle_message(MigrationIncomingState *mis,
 796                 QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
 797 {
 798     COLOMessage msg;
 799     Error *local_err = NULL;
 800
 801     msg = colo_receive_message(mis->from_src_file, &local_err);
 802     if (local_err) {
 803         error_propagate(errp, local_err);
 804         return;
 805     }
 806
 807     switch (msg) {
 808     case COLO_MESSAGE_CHECKPOINT_REQUEST:
 809         colo_incoming_process_checkpoint(mis, fb, bioc, errp);
 810         break;
 811     default:
 812         error_setg(errp, "Got unknown COLO message: %d", msg);
 813         break;
 814     }
 815 }
 816
 817 void colo_shutdown(void)
 818 {
 819     MigrationIncomingState *mis = NULL;
 820     MigrationState *s = NULL;
 821
 822     switch (get_colo_mode()) {
 823     case COLO_MODE_PRIMARY:
 824         s = migrate_get_current();
 825         qemu_event_set(&s->colo_checkpoint_event);
 826         qemu_sem_post(&s->colo_exit_sem);
 827         break;
 828     case COLO_MODE_SECONDARY:
 829         mis = migration_incoming_get_current();
 830         qemu_sem_post(&mis->colo_incoming_sem);
 831         break;
 832     default:
 833         break;
 834     }
 835 }
 836
 837 void *colo_process_incoming_thread(void *opaque)
 838 {
 839     MigrationIncomingState *mis = opaque;
 840     QEMUFile *fb = NULL;
 841     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
 842     Error *local_err = NULL;
 843
 844     rcu_register_thread();
 845     qemu_sem_init(&mis->colo_incoming_sem, 0);
 846
 847     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 848                       MIGRATION_STATUS_COLO);
 849
 850     if (get_colo_mode() != COLO_MODE_SECONDARY) {
 851         error_report("COLO mode must be COLO_MODE_SECONDARY");
 852         return NULL;
 853     }
 854
 855     failover_init_state();
 856
 857     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
 858     if (!mis->to_src_file) {
 859         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
 860         goto out;
 861     }
 862     /*
 863      * Note: the communication between Primary side and Secondary side
 864      * should be sequential, we set the fd to unblocked in migration incoming
 865      * coroutine, and here we are in the COLO incoming thread, so it is ok to
 866      * set the fd back to blocked.
 867      */
 868     qemu_file_set_blocking(mis->from_src_file, true);
 869
 870     colo_incoming_start_dirty_log();
 871
 872     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 873     fb = qemu_file_new_input(QIO_CHANNEL(bioc));
 874     object_unref(OBJECT(bioc));
 875
 876     qemu_mutex_lock_iothread();
 877 #ifdef CONFIG_REPLICATION
 878     replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
 879     if (local_err) {
 880         qemu_mutex_unlock_iothread();
 881         goto out;
 882     }
 883 #else
 884         abort();
 885 #endif
 886     vm_start();
 887     qemu_mutex_unlock_iothread();
 888     trace_colo_vm_state_change("stop", "run");
 889
 890     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
 891                       &local_err);
 892     if (local_err) {
 893         goto out;
 894     }
 895
 896     while (mis->state == MIGRATION_STATUS_COLO) {
 897         colo_wait_handle_message(mis, fb, bioc, &local_err);
 898         if (local_err) {
 899             error_report_err(local_err);
 900             break;
 901         }
 902
 903         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 904             failover_set_state(FAILOVER_STATUS_RELAUNCH,
 905                             FAILOVER_STATUS_NONE);
 906             failover_request_active(NULL);
 907             break;
 908         }
 909
 910         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 911             error_report("failover request");
 912             break;
 913         }
 914     }
 915
 916 out:
 917     /*
 918      * There are only two reasons we can get here, some error happened
 919      * or the user triggered failover.
 920      */
 921     switch (failover_get_state()) {
 922     case FAILOVER_STATUS_COMPLETED:
 923         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 924                                   COLO_EXIT_REASON_REQUEST);
 925         break;
 926     default:
 927         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 928                                   COLO_EXIT_REASON_ERROR);
 929     }
 930
 931     if (fb) {
 932         qemu_fclose(fb);
 933     }
 934
 935     /* Hope this not to be too long to loop here */
 936     qemu_sem_wait(&mis->colo_incoming_sem);
 937     qemu_sem_destroy(&mis->colo_incoming_sem);
 938
 939     rcu_unregister_thread();
 940     return NULL;
 941 }