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 #include "block/replication.h"
  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 "sysemu/cpus.h"
  35 #include "sysemu/runstate.h"
  36 #include "net/filter.h"
  37 #include "options.h"
  38
  39 static bool vmstate_loading;
  40 static Notifier packets_compare_notifier;
  41
  42 /* User need to know colo mode after COLO failover */
  43 static COLOMode last_colo_mode;
  44
  45 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
  46
  47 bool migration_in_colo_state(void)
  48 {
  49     MigrationState *s = migrate_get_current();
  50
  51     return (s->state == MIGRATION_STATUS_COLO);
  52 }
  53
  54 bool migration_incoming_in_colo_state(void)
  55 {
  56     MigrationIncomingState *mis = migration_incoming_get_current();
  57
  58     return mis && (mis->state == MIGRATION_STATUS_COLO);
  59 }
  60
  61 static bool colo_runstate_is_stopped(void)
  62 {
  63     return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
  64 }
  65
  66 static void colo_checkpoint_notify(void)
  67 {
  68     MigrationState *s = migrate_get_current();
  69     int64_t next_notify_time;
  70
  71     qemu_event_set(&s->colo_checkpoint_event);
  72     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
  73     next_notify_time = s->colo_checkpoint_time + migrate_checkpoint_delay();
  74     timer_mod(s->colo_delay_timer, next_notify_time);
  75 }
  76
  77 static void colo_checkpoint_notify_timer(void *opaque)
  78 {
  79     colo_checkpoint_notify();
  80 }
  81
  82 void colo_checkpoint_delay_set(void)
  83 {
  84     if (migration_in_colo_state()) {
  85         colo_checkpoint_notify();
  86     }
  87 }
  88
  89 static void secondary_vm_do_failover(void)
  90 {
  91 /* COLO needs enable block-replication */
  92     int old_state;
  93     MigrationIncomingState *mis = migration_incoming_get_current();
  94     Error *local_err = NULL;
  95
  96     /* Can not do failover during the process of VM's loading VMstate, Or
  97      * it will break the secondary VM.
  98      */
  99     if (vmstate_loading) {
 100         old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 101                         FAILOVER_STATUS_RELAUNCH);
 102         if (old_state != FAILOVER_STATUS_ACTIVE) {
 103             error_report("Unknown error while do failover for secondary VM,"
 104                          "old_state: %s", FailoverStatus_str(old_state));
 105         }
 106         return;
 107     }
 108
 109     migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
 110                       MIGRATION_STATUS_COMPLETED);
 111
 112     replication_stop_all(true, &local_err);
 113     if (local_err) {
 114         error_report_err(local_err);
 115         local_err = NULL;
 116     }
 117
 118     /* Notify all filters of all NIC to do checkpoint */
 119     colo_notify_filters_event(COLO_EVENT_FAILOVER, &local_err);
 120     if (local_err) {
 121         error_report_err(local_err);
 122     }
 123
 124     if (!autostart) {
 125         error_report("\"-S\" qemu option will be ignored in secondary side");
 126         /* recover runstate to normal migration finish state */
 127         autostart = true;
 128     }
 129     /*
 130      * Make sure COLO incoming thread not block in recv or send,
 131      * If mis->from_src_file and mis->to_src_file use the same fd,
 132      * The second shutdown() will return -1, we ignore this value,
 133      * It is harmless.
 134      */
 135     if (mis->from_src_file) {
 136         qemu_file_shutdown(mis->from_src_file);
 137     }
 138     if (mis->to_src_file) {
 139         qemu_file_shutdown(mis->to_src_file);
 140     }
 141
 142     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 143                                    FAILOVER_STATUS_COMPLETED);
 144     if (old_state != FAILOVER_STATUS_ACTIVE) {
 145         error_report("Incorrect state (%s) while doing failover for "
 146                      "secondary VM", FailoverStatus_str(old_state));
 147         return;
 148     }
 149     /* Notify COLO incoming thread that failover work is finished */
 150     qemu_sem_post(&mis->colo_incoming_sem);
 151
 152     /* For Secondary VM, jump to incoming co */
 153     if (mis->colo_incoming_co) {
 154         qemu_coroutine_enter(mis->colo_incoming_co);
 155     }
 156 }
 157
 158 static void primary_vm_do_failover(void)
 159 {
 160     MigrationState *s = migrate_get_current();
 161     int old_state;
 162     Error *local_err = NULL;
 163
 164     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
 165                       MIGRATION_STATUS_COMPLETED);
 166     /*
 167      * kick COLO thread which might wait at
 168      * qemu_sem_wait(&s->colo_checkpoint_sem).
 169      */
 170     colo_checkpoint_notify();
 171
 172     /*
 173      * Wake up COLO thread which may blocked in recv() or send(),
 174      * The s->rp_state.from_dst_file and s->to_dst_file may use the
 175      * same fd, but we still shutdown the fd for twice, it is harmless.
 176      */
 177     if (s->to_dst_file) {
 178         qemu_file_shutdown(s->to_dst_file);
 179     }
 180     if (s->rp_state.from_dst_file) {
 181         qemu_file_shutdown(s->rp_state.from_dst_file);
 182     }
 183
 184     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
 185                                    FAILOVER_STATUS_COMPLETED);
 186     if (old_state != FAILOVER_STATUS_ACTIVE) {
 187         error_report("Incorrect state (%s) while doing failover for Primary VM",
 188                      FailoverStatus_str(old_state));
 189         return;
 190     }
 191
 192     replication_stop_all(true, &local_err);
 193     if (local_err) {
 194         error_report_err(local_err);
 195         local_err = NULL;
 196     }
 197
 198     /* Notify COLO thread that failover work is finished */
 199     qemu_sem_post(&s->colo_exit_sem);
 200 }
 201
 202 COLOMode get_colo_mode(void)
 203 {
 204     if (migration_in_colo_state()) {
 205         return COLO_MODE_PRIMARY;
 206     } else if (migration_incoming_in_colo_state()) {
 207         return COLO_MODE_SECONDARY;
 208     } else {
 209         return COLO_MODE_NONE;
 210     }
 211 }
 212
 213 void colo_do_failover(void)
 214 {
 215     /* Make sure VM stopped while failover happened. */
 216     if (!colo_runstate_is_stopped()) {
 217         vm_stop_force_state(RUN_STATE_COLO);
 218     }
 219
 220     switch (last_colo_mode = get_colo_mode()) {
 221     case COLO_MODE_PRIMARY:
 222         primary_vm_do_failover();
 223         break;
 224     case COLO_MODE_SECONDARY:
 225         secondary_vm_do_failover();
 226         break;
 227     default:
 228         error_report("colo_do_failover failed because the colo mode"
 229                      " could not be obtained");
 230     }
 231 }
 232
 233 void qmp_xen_set_replication(bool enable, bool primary,
 234                              bool has_failover, bool failover,
 235                              Error **errp)
 236 {
 237     ReplicationMode mode = primary ?
 238                            REPLICATION_MODE_PRIMARY :
 239                            REPLICATION_MODE_SECONDARY;
 240
 241     if (has_failover && enable) {
 242         error_setg(errp, "Parameter 'failover' is only for"
 243                    " stopping replication");
 244         return;
 245     }
 246
 247     if (enable) {
 248         replication_start_all(mode, errp);
 249     } else {
 250         if (!has_failover) {
 251             failover = NULL;
 252         }
 253         replication_stop_all(failover, failover ? NULL : errp);
 254     }
 255 }
 256
 257 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
 258 {
 259     Error *err = NULL;
 260     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
 261
 262     replication_get_error_all(&err);
 263     if (err) {
 264         s->error = true;
 265         s->desc = g_strdup(error_get_pretty(err));
 266     } else {
 267         s->error = false;
 268     }
 269
 270     error_free(err);
 271     return s;
 272 }
 273
 274 void qmp_xen_colo_do_checkpoint(Error **errp)
 275 {
 276     Error *err = NULL;
 277
 278     replication_do_checkpoint_all(&err);
 279     if (err) {
 280         error_propagate(errp, err);
 281         return;
 282     }
 283     /* Notify all filters of all NIC to do checkpoint */
 284     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, errp);
 285 }
 286
 287 COLOStatus *qmp_query_colo_status(Error **errp)
 288 {
 289     COLOStatus *s = g_new0(COLOStatus, 1);
 290
 291     s->mode = get_colo_mode();
 292     s->last_mode = last_colo_mode;
 293
 294     switch (failover_get_state()) {
 295     case FAILOVER_STATUS_NONE:
 296         s->reason = COLO_EXIT_REASON_NONE;
 297         break;
 298     case FAILOVER_STATUS_COMPLETED:
 299         s->reason = COLO_EXIT_REASON_REQUEST;
 300         break;
 301     default:
 302         if (migration_in_colo_state()) {
 303             s->reason = COLO_EXIT_REASON_PROCESSING;
 304         } else {
 305             s->reason = COLO_EXIT_REASON_ERROR;
 306         }
 307     }
 308
 309     return s;
 310 }
 311
 312 static void colo_send_message(QEMUFile *f, COLOMessage msg,
 313                               Error **errp)
 314 {
 315     int ret;
 316
 317     if (msg >= COLO_MESSAGE__MAX) {
 318         error_setg(errp, "%s: Invalid message", __func__);
 319         return;
 320     }
 321     qemu_put_be32(f, msg);
 322     ret = qemu_fflush(f);
 323     if (ret < 0) {
 324         error_setg_errno(errp, -ret, "Can't send COLO message");
 325     }
 326     trace_colo_send_message(COLOMessage_str(msg));
 327 }
 328
 329 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
 330                                     uint64_t value, Error **errp)
 331 {
 332     Error *local_err = NULL;
 333     int ret;
 334
 335     colo_send_message(f, msg, &local_err);
 336     if (local_err) {
 337         error_propagate(errp, local_err);
 338         return;
 339     }
 340     qemu_put_be64(f, value);
 341     ret = qemu_fflush(f);
 342     if (ret < 0) {
 343         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
 344                          COLOMessage_str(msg));
 345     }
 346 }
 347
 348 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
 349 {
 350     COLOMessage msg;
 351     int ret;
 352
 353     msg = qemu_get_be32(f);
 354     ret = qemu_file_get_error(f);
 355     if (ret < 0) {
 356         error_setg_errno(errp, -ret, "Can't receive COLO message");
 357         return msg;
 358     }
 359     if (msg >= COLO_MESSAGE__MAX) {
 360         error_setg(errp, "%s: Invalid message", __func__);
 361         return msg;
 362     }
 363     trace_colo_receive_message(COLOMessage_str(msg));
 364     return msg;
 365 }
 366
 367 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
 368                                        Error **errp)
 369 {
 370     COLOMessage msg;
 371     Error *local_err = NULL;
 372
 373     msg = colo_receive_message(f, &local_err);
 374     if (local_err) {
 375         error_propagate(errp, local_err);
 376         return;
 377     }
 378     if (msg != expect_msg) {
 379         error_setg(errp, "Unexpected COLO message %d, expected %d",
 380                           msg, expect_msg);
 381     }
 382 }
 383
 384 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
 385                                            Error **errp)
 386 {
 387     Error *local_err = NULL;
 388     uint64_t value;
 389     int ret;
 390
 391     colo_receive_check_message(f, expect_msg, &local_err);
 392     if (local_err) {
 393         error_propagate(errp, local_err);
 394         return 0;
 395     }
 396
 397     value = qemu_get_be64(f);
 398     ret = qemu_file_get_error(f);
 399     if (ret < 0) {
 400         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
 401                          COLOMessage_str(expect_msg));
 402     }
 403     return value;
 404 }
 405
 406 static int colo_do_checkpoint_transaction(MigrationState *s,
 407                                           QIOChannelBuffer *bioc,
 408                                           QEMUFile *fb)
 409 {
 410     Error *local_err = NULL;
 411     int ret = -1;
 412
 413     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
 414                       &local_err);
 415     if (local_err) {
 416         goto out;
 417     }
 418
 419     colo_receive_check_message(s->rp_state.from_dst_file,
 420                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
 421     if (local_err) {
 422         goto out;
 423     }
 424     /* Reset channel-buffer directly */
 425     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 426     bioc->usage = 0;
 427
 428     bql_lock();
 429     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 430         bql_unlock();
 431         goto out;
 432     }
 433     vm_stop_force_state(RUN_STATE_COLO);
 434     bql_unlock();
 435     trace_colo_vm_state_change("run", "stop");
 436     /*
 437      * Failover request bh could be called after vm_stop_force_state(),
 438      * So we need check failover_request_is_active() again.
 439      */
 440     if (failover_get_state() != FAILOVER_STATUS_NONE) {
 441         goto out;
 442     }
 443     bql_lock();
 444
 445     replication_do_checkpoint_all(&local_err);
 446     if (local_err) {
 447         bql_unlock();
 448         goto out;
 449     }
 450
 451     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
 452     if (local_err) {
 453         bql_unlock();
 454         goto out;
 455     }
 456     /* Note: device state is saved into buffer */
 457     ret = qemu_save_device_state(fb);
 458
 459     bql_unlock();
 460     if (ret < 0) {
 461         goto out;
 462     }
 463
 464     if (migrate_auto_converge()) {
 465         mig_throttle_counter_reset();
 466     }
 467     /*
 468      * Only save VM's live state, which not including device state.
 469      * TODO: We may need a timeout mechanism to prevent COLO process
 470      * to be blocked here.
 471      */
 472     qemu_savevm_live_state(s->to_dst_file);
 473
 474     qemu_fflush(fb);
 475
 476     /*
 477      * We need the size of the VMstate data in Secondary side,
 478      * With which we can decide how much data should be read.
 479      */
 480     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
 481                             bioc->usage, &local_err);
 482     if (local_err) {
 483         goto out;
 484     }
 485
 486     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
 487     ret = qemu_fflush(s->to_dst_file);
 488     if (ret < 0) {
 489         goto out;
 490     }
 491
 492     colo_receive_check_message(s->rp_state.from_dst_file,
 493                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
 494     if (local_err) {
 495         goto out;
 496     }
 497
 498     qemu_event_reset(&s->colo_checkpoint_event);
 499     colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
 500     if (local_err) {
 501         goto out;
 502     }
 503
 504     colo_receive_check_message(s->rp_state.from_dst_file,
 505                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
 506     if (local_err) {
 507         goto out;
 508     }
 509
 510     ret = 0;
 511
 512     bql_lock();
 513     vm_start();
 514     bql_unlock();
 515     trace_colo_vm_state_change("stop", "run");
 516
 517 out:
 518     if (local_err) {
 519         error_report_err(local_err);
 520     }
 521     return ret;
 522 }
 523
 524 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
 525 {
 526     colo_checkpoint_notify();
 527 }
 528
 529 static void colo_process_checkpoint(MigrationState *s)
 530 {
 531     QIOChannelBuffer *bioc;
 532     QEMUFile *fb = NULL;
 533     Error *local_err = NULL;
 534     int ret;
 535
 536     if (get_colo_mode() != COLO_MODE_PRIMARY) {
 537         error_report("COLO mode must be COLO_MODE_PRIMARY");
 538         return;
 539     }
 540
 541     failover_init_state();
 542
 543     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
 544     if (!s->rp_state.from_dst_file) {
 545         error_report("Open QEMUFile from_dst_file failed");
 546         goto out;
 547     }
 548
 549     packets_compare_notifier.notify = colo_compare_notify_checkpoint;
 550     colo_compare_register_notifier(&packets_compare_notifier);
 551
 552     /*
 553      * Wait for Secondary finish loading VM states and enter COLO
 554      * restore.
 555      */
 556     colo_receive_check_message(s->rp_state.from_dst_file,
 557                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
 558     if (local_err) {
 559         goto out;
 560     }
 561     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 562     fb = qemu_file_new_output(QIO_CHANNEL(bioc));
 563     object_unref(OBJECT(bioc));
 564
 565     bql_lock();
 566     replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
 567     if (local_err) {
 568         bql_unlock();
 569         goto out;
 570     }
 571
 572     vm_start();
 573     bql_unlock();
 574     trace_colo_vm_state_change("stop", "run");
 575
 576     timer_mod(s->colo_delay_timer, qemu_clock_get_ms(QEMU_CLOCK_HOST) +
 577               migrate_checkpoint_delay());
 578
 579     while (s->state == MIGRATION_STATUS_COLO) {
 580         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 581             error_report("failover request");
 582             goto out;
 583         }
 584
 585         qemu_event_wait(&s->colo_checkpoint_event);
 586
 587         if (s->state != MIGRATION_STATUS_COLO) {
 588             goto out;
 589         }
 590         ret = colo_do_checkpoint_transaction(s, bioc, fb);
 591         if (ret < 0) {
 592             goto out;
 593         }
 594     }
 595
 596 out:
 597     /* Throw the unreported error message after exited from loop */
 598     if (local_err) {
 599         error_report_err(local_err);
 600     }
 601
 602     if (fb) {
 603         qemu_fclose(fb);
 604     }
 605
 606     /*
 607      * There are only two reasons we can get here, some error happened
 608      * or the user triggered failover.
 609      */
 610     switch (failover_get_state()) {
 611     case FAILOVER_STATUS_COMPLETED:
 612         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 613                                   COLO_EXIT_REASON_REQUEST);
 614         break;
 615     default:
 616         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
 617                                   COLO_EXIT_REASON_ERROR);
 618     }
 619
 620     /* Hope this not to be too long to wait here */
 621     qemu_sem_wait(&s->colo_exit_sem);
 622     qemu_sem_destroy(&s->colo_exit_sem);
 623
 624     /*
 625      * It is safe to unregister notifier after failover finished.
 626      * Besides, colo_delay_timer and colo_checkpoint_sem can't be
 627      * released before unregister notifier, or there will be use-after-free
 628      * error.
 629      */
 630     colo_compare_unregister_notifier(&packets_compare_notifier);
 631     timer_free(s->colo_delay_timer);
 632     qemu_event_destroy(&s->colo_checkpoint_event);
 633
 634     /*
 635      * Must be called after failover BH is completed,
 636      * Or the failover BH may shutdown the wrong fd that
 637      * re-used by other threads after we release here.
 638      */
 639     if (s->rp_state.from_dst_file) {
 640         qemu_fclose(s->rp_state.from_dst_file);
 641         s->rp_state.from_dst_file = NULL;
 642     }
 643 }
 644
 645 void migrate_start_colo_process(MigrationState *s)
 646 {
 647     bql_unlock();
 648     qemu_event_init(&s->colo_checkpoint_event, false);
 649     s->colo_delay_timer =  timer_new_ms(QEMU_CLOCK_HOST,
 650                                 colo_checkpoint_notify_timer, NULL);
 651
 652     qemu_sem_init(&s->colo_exit_sem, 0);
 653     colo_process_checkpoint(s);
 654     bql_lock();
 655 }
 656
 657 static void colo_incoming_process_checkpoint(MigrationIncomingState *mis,
 658                       QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
 659 {
 660     uint64_t total_size;
 661     uint64_t value;
 662     Error *local_err = NULL;
 663     int ret;
 664
 665     bql_lock();
 666     vm_stop_force_state(RUN_STATE_COLO);
 667     bql_unlock();
 668     trace_colo_vm_state_change("run", "stop");
 669
 670     /* FIXME: This is unnecessary for periodic checkpoint mode */
 671     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
 672                  &local_err);
 673     if (local_err) {
 674         error_propagate(errp, local_err);
 675         return;
 676     }
 677
 678     colo_receive_check_message(mis->from_src_file,
 679                        COLO_MESSAGE_VMSTATE_SEND, &local_err);
 680     if (local_err) {
 681         error_propagate(errp, local_err);
 682         return;
 683     }
 684
 685     bql_lock();
 686     cpu_synchronize_all_states();
 687     ret = qemu_loadvm_state_main(mis->from_src_file, mis);
 688     bql_unlock();
 689
 690     if (ret < 0) {
 691         error_setg(errp, "Load VM's live state (ram) error");
 692         return;
 693     }
 694
 695     value = colo_receive_message_value(mis->from_src_file,
 696                              COLO_MESSAGE_VMSTATE_SIZE, &local_err);
 697     if (local_err) {
 698         error_propagate(errp, local_err);
 699         return;
 700     }
 701
 702     /*
 703      * Read VM device state data into channel buffer,
 704      * It's better to re-use the memory allocated.
 705      * Here we need to handle the channel buffer directly.
 706      */
 707     if (value > bioc->capacity) {
 708         bioc->capacity = value;
 709         bioc->data = g_realloc(bioc->data, bioc->capacity);
 710     }
 711     total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
 712     if (total_size != value) {
 713         error_setg(errp, "Got %" PRIu64 " VMState data, less than expected"
 714                     " %" PRIu64, total_size, value);
 715         return;
 716     }
 717     bioc->usage = total_size;
 718     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
 719
 720     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
 721                  &local_err);
 722     if (local_err) {
 723         error_propagate(errp, local_err);
 724         return;
 725     }
 726
 727     bql_lock();
 728     vmstate_loading = true;
 729     colo_flush_ram_cache();
 730     ret = qemu_load_device_state(fb);
 731     if (ret < 0) {
 732         error_setg(errp, "COLO: load device state failed");
 733         vmstate_loading = false;
 734         bql_unlock();
 735         return;
 736     }
 737
 738     replication_get_error_all(&local_err);
 739     if (local_err) {
 740         error_propagate(errp, local_err);
 741         vmstate_loading = false;
 742         bql_unlock();
 743         return;
 744     }
 745
 746     /* discard colo disk buffer */
 747     replication_do_checkpoint_all(&local_err);
 748     if (local_err) {
 749         error_propagate(errp, local_err);
 750         vmstate_loading = false;
 751         bql_unlock();
 752         return;
 753     }
 754     /* Notify all filters of all NIC to do checkpoint */
 755     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
 756
 757     if (local_err) {
 758         error_propagate(errp, local_err);
 759         vmstate_loading = false;
 760         bql_unlock();
 761         return;
 762     }
 763
 764     vmstate_loading = false;
 765     vm_start();
 766     bql_unlock();
 767     trace_colo_vm_state_change("stop", "run");
 768
 769     if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 770         return;
 771     }
 772
 773     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
 774                  &local_err);
 775     error_propagate(errp, local_err);
 776 }
 777
 778 static void colo_wait_handle_message(MigrationIncomingState *mis,
 779                 QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
 780 {
 781     COLOMessage msg;
 782     Error *local_err = NULL;
 783
 784     msg = colo_receive_message(mis->from_src_file, &local_err);
 785     if (local_err) {
 786         error_propagate(errp, local_err);
 787         return;
 788     }
 789
 790     switch (msg) {
 791     case COLO_MESSAGE_CHECKPOINT_REQUEST:
 792         colo_incoming_process_checkpoint(mis, fb, bioc, errp);
 793         break;
 794     default:
 795         error_setg(errp, "Got unknown COLO message: %d", msg);
 796         break;
 797     }
 798 }
 799
 800 void colo_shutdown(void)
 801 {
 802     MigrationIncomingState *mis = NULL;
 803     MigrationState *s = NULL;
 804
 805     switch (get_colo_mode()) {
 806     case COLO_MODE_PRIMARY:
 807         s = migrate_get_current();
 808         qemu_event_set(&s->colo_checkpoint_event);
 809         qemu_sem_post(&s->colo_exit_sem);
 810         break;
 811     case COLO_MODE_SECONDARY:
 812         mis = migration_incoming_get_current();
 813         qemu_sem_post(&mis->colo_incoming_sem);
 814         break;
 815     default:
 816         break;
 817     }
 818 }
 819
 820 static void *colo_process_incoming_thread(void *opaque)
 821 {
 822     MigrationIncomingState *mis = opaque;
 823     QEMUFile *fb = NULL;
 824     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
 825     Error *local_err = NULL;
 826
 827     rcu_register_thread();
 828     qemu_sem_init(&mis->colo_incoming_sem, 0);
 829
 830     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 831                       MIGRATION_STATUS_COLO);
 832
 833     if (get_colo_mode() != COLO_MODE_SECONDARY) {
 834         error_report("COLO mode must be COLO_MODE_SECONDARY");
 835         return NULL;
 836     }
 837
 838     failover_init_state();
 839
 840     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
 841     if (!mis->to_src_file) {
 842         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
 843         goto out;
 844     }
 845     /*
 846      * Note: the communication between Primary side and Secondary side
 847      * should be sequential, we set the fd to unblocked in migration incoming
 848      * coroutine, and here we are in the COLO incoming thread, so it is ok to
 849      * set the fd back to blocked.
 850      */
 851     qemu_file_set_blocking(mis->from_src_file, true);
 852
 853     colo_incoming_start_dirty_log();
 854
 855     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
 856     fb = qemu_file_new_input(QIO_CHANNEL(bioc));
 857     object_unref(OBJECT(bioc));
 858
 859     bql_lock();
 860     replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
 861     if (local_err) {
 862         bql_unlock();
 863         goto out;
 864     }
 865     vm_start();
 866     bql_unlock();
 867     trace_colo_vm_state_change("stop", "run");
 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
 882         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
 883             failover_set_state(FAILOVER_STATUS_RELAUNCH,
 884                             FAILOVER_STATUS_NONE);
 885             failover_request_active(NULL);
 886             break;
 887         }
 888
 889         if (failover_get_state() != FAILOVER_STATUS_NONE) {
 890             error_report("failover request");
 891             break;
 892         }
 893     }
 894
 895 out:
 896     /*
 897      * There are only two reasons we can get here, some error happened
 898      * or the user triggered failover.
 899      */
 900     switch (failover_get_state()) {
 901     case FAILOVER_STATUS_COMPLETED:
 902         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 903                                   COLO_EXIT_REASON_REQUEST);
 904         break;
 905     default:
 906         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
 907                                   COLO_EXIT_REASON_ERROR);
 908     }
 909
 910     if (fb) {
 911         qemu_fclose(fb);
 912     }
 913
 914     /* Hope this not to be too long to loop here */
 915     qemu_sem_wait(&mis->colo_incoming_sem);
 916     qemu_sem_destroy(&mis->colo_incoming_sem);
 917
 918     rcu_unregister_thread();
 919     return NULL;
 920 }
 921
 922 int coroutine_fn colo_incoming_co(void)
 923 {
 924     MigrationIncomingState *mis = migration_incoming_get_current();
 925     Error *local_err = NULL;
 926     QemuThread th;
 927
 928     assert(bql_locked());
 929
 930     if (!migration_incoming_colo_enabled()) {
 931         return 0;
 932     }
 933
 934     /* Make sure all file formats throw away their mutable metadata */
 935     bdrv_activate_all(&local_err);
 936     if (local_err) {
 937         error_report_err(local_err);
 938         return -EINVAL;
 939     }
 940
 941     qemu_thread_create(&th, "COLO incoming", colo_process_incoming_thread,
 942                        mis, QEMU_THREAD_JOINABLE);
 943
 944     mis->colo_incoming_co = qemu_coroutine_self();
 945     qemu_coroutine_yield();
 946     mis->colo_incoming_co = NULL;
 947
 948     bql_unlock();
 949     /* Wait checkpoint incoming thread exit before free resource */
 950     qemu_thread_join(&th);
 951     bql_lock();
 952
 953     /* We hold the global BQL, so it is safe here */
 954     colo_release_ram_cache();
 955
 956     return 0;
 957 }