xref: /openbmc/qemu/migration/colo.c (revision 745a4f5e)
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 #include "replication.h"
28 #include "net/colo-compare.h"
29 #include "net/colo.h"
30 #include "block/block.h"
31 #include "qapi/qapi-events-migration.h"
32 #include "qapi/qmp/qerror.h"
33 #include "sysemu/cpus.h"
34 #include "net/filter.h"
35 
36 static bool vmstate_loading;
37 static Notifier packets_compare_notifier;
38 
39 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
40 
41 bool migration_in_colo_state(void)
42 {
43     MigrationState *s = migrate_get_current();
44 
45     return (s->state == MIGRATION_STATUS_COLO);
46 }
47 
48 bool migration_incoming_in_colo_state(void)
49 {
50     MigrationIncomingState *mis = migration_incoming_get_current();
51 
52     return mis && (mis->state == MIGRATION_STATUS_COLO);
53 }
54 
55 static bool colo_runstate_is_stopped(void)
56 {
57     return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
58 }
59 
60 static void secondary_vm_do_failover(void)
61 {
62 /* COLO needs enable block-replication */
63 #ifdef CONFIG_REPLICATION
64     int old_state;
65     MigrationIncomingState *mis = migration_incoming_get_current();
66     Error *local_err = NULL;
67 
68     /* Can not do failover during the process of VM's loading VMstate, Or
69      * it will break the secondary VM.
70      */
71     if (vmstate_loading) {
72         old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
73                         FAILOVER_STATUS_RELAUNCH);
74         if (old_state != FAILOVER_STATUS_ACTIVE) {
75             error_report("Unknown error while do failover for secondary VM,"
76                          "old_state: %s", FailoverStatus_str(old_state));
77         }
78         return;
79     }
80 
81     migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
82                       MIGRATION_STATUS_COMPLETED);
83 
84     replication_stop_all(true, &local_err);
85     if (local_err) {
86         error_report_err(local_err);
87     }
88 
89     /* Notify all filters of all NIC to do checkpoint */
90     colo_notify_filters_event(COLO_EVENT_FAILOVER, &local_err);
91     if (local_err) {
92         error_report_err(local_err);
93     }
94 
95     if (!autostart) {
96         error_report("\"-S\" qemu option will be ignored in secondary side");
97         /* recover runstate to normal migration finish state */
98         autostart = true;
99     }
100     /*
101      * Make sure COLO incoming thread not block in recv or send,
102      * If mis->from_src_file and mis->to_src_file use the same fd,
103      * The second shutdown() will return -1, we ignore this value,
104      * It is harmless.
105      */
106     if (mis->from_src_file) {
107         qemu_file_shutdown(mis->from_src_file);
108     }
109     if (mis->to_src_file) {
110         qemu_file_shutdown(mis->to_src_file);
111     }
112 
113     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
114                                    FAILOVER_STATUS_COMPLETED);
115     if (old_state != FAILOVER_STATUS_ACTIVE) {
116         error_report("Incorrect state (%s) while doing failover for "
117                      "secondary VM", FailoverStatus_str(old_state));
118         return;
119     }
120     /* Notify COLO incoming thread that failover work is finished */
121     qemu_sem_post(&mis->colo_incoming_sem);
122     /* For Secondary VM, jump to incoming co */
123     if (mis->migration_incoming_co) {
124         qemu_coroutine_enter(mis->migration_incoming_co);
125     }
126 #else
127     abort();
128 #endif
129 }
130 
131 static void primary_vm_do_failover(void)
132 {
133 #ifdef CONFIG_REPLICATION
134     MigrationState *s = migrate_get_current();
135     int old_state;
136     Error *local_err = NULL;
137 
138     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
139                       MIGRATION_STATUS_COMPLETED);
140     /*
141      * kick COLO thread which might wait at
142      * qemu_sem_wait(&s->colo_checkpoint_sem).
143      */
144     colo_checkpoint_notify(migrate_get_current());
145 
146     /*
147      * Wake up COLO thread which may blocked in recv() or send(),
148      * The s->rp_state.from_dst_file and s->to_dst_file may use the
149      * same fd, but we still shutdown the fd for twice, it is harmless.
150      */
151     if (s->to_dst_file) {
152         qemu_file_shutdown(s->to_dst_file);
153     }
154     if (s->rp_state.from_dst_file) {
155         qemu_file_shutdown(s->rp_state.from_dst_file);
156     }
157 
158     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
159                                    FAILOVER_STATUS_COMPLETED);
160     if (old_state != FAILOVER_STATUS_ACTIVE) {
161         error_report("Incorrect state (%s) while doing failover for Primary VM",
162                      FailoverStatus_str(old_state));
163         return;
164     }
165 
166     replication_stop_all(true, &local_err);
167     if (local_err) {
168         error_report_err(local_err);
169         local_err = NULL;
170     }
171 
172     /* Notify COLO thread that failover work is finished */
173     qemu_sem_post(&s->colo_exit_sem);
174 #else
175     abort();
176 #endif
177 }
178 
179 COLOMode get_colo_mode(void)
180 {
181     if (migration_in_colo_state()) {
182         return COLO_MODE_PRIMARY;
183     } else if (migration_incoming_in_colo_state()) {
184         return COLO_MODE_SECONDARY;
185     } else {
186         return COLO_MODE_NONE;
187     }
188 }
189 
190 void colo_do_failover(MigrationState *s)
191 {
192     /* Make sure VM stopped while failover happened. */
193     if (!colo_runstate_is_stopped()) {
194         vm_stop_force_state(RUN_STATE_COLO);
195     }
196 
197     if (get_colo_mode() == COLO_MODE_PRIMARY) {
198         primary_vm_do_failover();
199     } else {
200         secondary_vm_do_failover();
201     }
202 }
203 
204 void qmp_xen_set_replication(bool enable, bool primary,
205                              bool has_failover, bool failover,
206                              Error **errp)
207 {
208 #ifdef CONFIG_REPLICATION
209     ReplicationMode mode = primary ?
210                            REPLICATION_MODE_PRIMARY :
211                            REPLICATION_MODE_SECONDARY;
212 
213     if (has_failover && enable) {
214         error_setg(errp, "Parameter 'failover' is only for"
215                    " stopping replication");
216         return;
217     }
218 
219     if (enable) {
220         replication_start_all(mode, errp);
221     } else {
222         if (!has_failover) {
223             failover = NULL;
224         }
225         replication_stop_all(failover, failover ? NULL : errp);
226     }
227 #else
228     abort();
229 #endif
230 }
231 
232 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
233 {
234 #ifdef CONFIG_REPLICATION
235     Error *err = NULL;
236     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
237 
238     replication_get_error_all(&err);
239     if (err) {
240         s->error = true;
241         s->has_desc = true;
242         s->desc = g_strdup(error_get_pretty(err));
243     } else {
244         s->error = false;
245     }
246 
247     error_free(err);
248     return s;
249 #else
250     abort();
251 #endif
252 }
253 
254 void qmp_xen_colo_do_checkpoint(Error **errp)
255 {
256 #ifdef CONFIG_REPLICATION
257     replication_do_checkpoint_all(errp);
258 #else
259     abort();
260 #endif
261 }
262 
263 COLOStatus *qmp_query_colo_status(Error **errp)
264 {
265     COLOStatus *s = g_new0(COLOStatus, 1);
266 
267     s->mode = get_colo_mode();
268 
269     switch (failover_get_state()) {
270     case FAILOVER_STATUS_NONE:
271         s->reason = COLO_EXIT_REASON_NONE;
272         break;
273     case FAILOVER_STATUS_REQUIRE:
274         s->reason = COLO_EXIT_REASON_REQUEST;
275         break;
276     default:
277         s->reason = COLO_EXIT_REASON_ERROR;
278     }
279 
280     return s;
281 }
282 
283 static void colo_send_message(QEMUFile *f, COLOMessage msg,
284                               Error **errp)
285 {
286     int ret;
287 
288     if (msg >= COLO_MESSAGE__MAX) {
289         error_setg(errp, "%s: Invalid message", __func__);
290         return;
291     }
292     qemu_put_be32(f, msg);
293     qemu_fflush(f);
294 
295     ret = qemu_file_get_error(f);
296     if (ret < 0) {
297         error_setg_errno(errp, -ret, "Can't send COLO message");
298     }
299     trace_colo_send_message(COLOMessage_str(msg));
300 }
301 
302 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
303                                     uint64_t value, Error **errp)
304 {
305     Error *local_err = NULL;
306     int ret;
307 
308     colo_send_message(f, msg, &local_err);
309     if (local_err) {
310         error_propagate(errp, local_err);
311         return;
312     }
313     qemu_put_be64(f, value);
314     qemu_fflush(f);
315 
316     ret = qemu_file_get_error(f);
317     if (ret < 0) {
318         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
319                          COLOMessage_str(msg));
320     }
321 }
322 
323 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
324 {
325     COLOMessage msg;
326     int ret;
327 
328     msg = qemu_get_be32(f);
329     ret = qemu_file_get_error(f);
330     if (ret < 0) {
331         error_setg_errno(errp, -ret, "Can't receive COLO message");
332         return msg;
333     }
334     if (msg >= COLO_MESSAGE__MAX) {
335         error_setg(errp, "%s: Invalid message", __func__);
336         return msg;
337     }
338     trace_colo_receive_message(COLOMessage_str(msg));
339     return msg;
340 }
341 
342 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
343                                        Error **errp)
344 {
345     COLOMessage msg;
346     Error *local_err = NULL;
347 
348     msg = colo_receive_message(f, &local_err);
349     if (local_err) {
350         error_propagate(errp, local_err);
351         return;
352     }
353     if (msg != expect_msg) {
354         error_setg(errp, "Unexpected COLO message %d, expected %d",
355                           msg, expect_msg);
356     }
357 }
358 
359 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
360                                            Error **errp)
361 {
362     Error *local_err = NULL;
363     uint64_t value;
364     int ret;
365 
366     colo_receive_check_message(f, expect_msg, &local_err);
367     if (local_err) {
368         error_propagate(errp, local_err);
369         return 0;
370     }
371 
372     value = qemu_get_be64(f);
373     ret = qemu_file_get_error(f);
374     if (ret < 0) {
375         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
376                          COLOMessage_str(expect_msg));
377     }
378     return value;
379 }
380 
381 static int colo_do_checkpoint_transaction(MigrationState *s,
382                                           QIOChannelBuffer *bioc,
383                                           QEMUFile *fb)
384 {
385     Error *local_err = NULL;
386     int ret = -1;
387 
388     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
389                       &local_err);
390     if (local_err) {
391         goto out;
392     }
393 
394     colo_receive_check_message(s->rp_state.from_dst_file,
395                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
396     if (local_err) {
397         goto out;
398     }
399     /* Reset channel-buffer directly */
400     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
401     bioc->usage = 0;
402 
403     qemu_mutex_lock_iothread();
404     if (failover_get_state() != FAILOVER_STATUS_NONE) {
405         qemu_mutex_unlock_iothread();
406         goto out;
407     }
408     vm_stop_force_state(RUN_STATE_COLO);
409     qemu_mutex_unlock_iothread();
410     trace_colo_vm_state_change("run", "stop");
411     /*
412      * Failover request bh could be called after vm_stop_force_state(),
413      * So we need check failover_request_is_active() again.
414      */
415     if (failover_get_state() != FAILOVER_STATUS_NONE) {
416         goto out;
417     }
418 
419     colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
420     if (local_err) {
421         goto out;
422     }
423 
424     /* Disable block migration */
425     migrate_set_block_enabled(false, &local_err);
426     qemu_mutex_lock_iothread();
427 
428 #ifdef CONFIG_REPLICATION
429     replication_do_checkpoint_all(&local_err);
430     if (local_err) {
431         qemu_mutex_unlock_iothread();
432         goto out;
433     }
434 #else
435         abort();
436 #endif
437 
438     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
439     if (local_err) {
440         qemu_mutex_unlock_iothread();
441         goto out;
442     }
443     /* Note: device state is saved into buffer */
444     ret = qemu_save_device_state(fb);
445 
446     qemu_mutex_unlock_iothread();
447     if (ret < 0) {
448         goto out;
449     }
450     /*
451      * Only save VM's live state, which not including device state.
452      * TODO: We may need a timeout mechanism to prevent COLO process
453      * to be blocked here.
454      */
455     qemu_savevm_live_state(s->to_dst_file);
456 
457     qemu_fflush(fb);
458 
459     /*
460      * We need the size of the VMstate data in Secondary side,
461      * With which we can decide how much data should be read.
462      */
463     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
464                             bioc->usage, &local_err);
465     if (local_err) {
466         goto out;
467     }
468 
469     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
470     qemu_fflush(s->to_dst_file);
471     ret = qemu_file_get_error(s->to_dst_file);
472     if (ret < 0) {
473         goto out;
474     }
475 
476     colo_receive_check_message(s->rp_state.from_dst_file,
477                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
478     if (local_err) {
479         goto out;
480     }
481 
482     colo_receive_check_message(s->rp_state.from_dst_file,
483                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
484     if (local_err) {
485         goto out;
486     }
487 
488     ret = 0;
489 
490     qemu_mutex_lock_iothread();
491     vm_start();
492     qemu_mutex_unlock_iothread();
493     trace_colo_vm_state_change("stop", "run");
494 
495 out:
496     if (local_err) {
497         error_report_err(local_err);
498     }
499     return ret;
500 }
501 
502 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
503 {
504     colo_checkpoint_notify(data);
505 }
506 
507 static void colo_process_checkpoint(MigrationState *s)
508 {
509     QIOChannelBuffer *bioc;
510     QEMUFile *fb = NULL;
511     int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
512     Error *local_err = NULL;
513     int ret;
514 
515     failover_init_state();
516 
517     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
518     if (!s->rp_state.from_dst_file) {
519         error_report("Open QEMUFile from_dst_file failed");
520         goto out;
521     }
522 
523     packets_compare_notifier.notify = colo_compare_notify_checkpoint;
524     colo_compare_register_notifier(&packets_compare_notifier);
525 
526     /*
527      * Wait for Secondary finish loading VM states and enter COLO
528      * restore.
529      */
530     colo_receive_check_message(s->rp_state.from_dst_file,
531                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
532     if (local_err) {
533         goto out;
534     }
535     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
536     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
537     object_unref(OBJECT(bioc));
538 
539     qemu_mutex_lock_iothread();
540 #ifdef CONFIG_REPLICATION
541     replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
542     if (local_err) {
543         qemu_mutex_unlock_iothread();
544         goto out;
545     }
546 #else
547         abort();
548 #endif
549 
550     vm_start();
551     qemu_mutex_unlock_iothread();
552     trace_colo_vm_state_change("stop", "run");
553 
554     timer_mod(s->colo_delay_timer,
555             current_time + s->parameters.x_checkpoint_delay);
556 
557     while (s->state == MIGRATION_STATUS_COLO) {
558         if (failover_get_state() != FAILOVER_STATUS_NONE) {
559             error_report("failover request");
560             goto out;
561         }
562 
563         qemu_sem_wait(&s->colo_checkpoint_sem);
564 
565         if (s->state != MIGRATION_STATUS_COLO) {
566             goto out;
567         }
568         ret = colo_do_checkpoint_transaction(s, bioc, fb);
569         if (ret < 0) {
570             goto out;
571         }
572     }
573 
574 out:
575     /* Throw the unreported error message after exited from loop */
576     if (local_err) {
577         error_report_err(local_err);
578     }
579 
580     if (fb) {
581         qemu_fclose(fb);
582     }
583 
584     /*
585      * There are only two reasons we can get here, some error happened
586      * or the user triggered failover.
587      */
588     switch (failover_get_state()) {
589     case FAILOVER_STATUS_NONE:
590         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
591                                   COLO_EXIT_REASON_ERROR);
592         break;
593     case FAILOVER_STATUS_REQUIRE:
594         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
595                                   COLO_EXIT_REASON_REQUEST);
596         break;
597     default:
598         abort();
599     }
600 
601     /* Hope this not to be too long to wait here */
602     qemu_sem_wait(&s->colo_exit_sem);
603     qemu_sem_destroy(&s->colo_exit_sem);
604 
605     /*
606      * It is safe to unregister notifier after failover finished.
607      * Besides, colo_delay_timer and colo_checkpoint_sem can't be
608      * released befor unregister notifier, or there will be use-after-free
609      * error.
610      */
611     colo_compare_unregister_notifier(&packets_compare_notifier);
612     timer_del(s->colo_delay_timer);
613     timer_free(s->colo_delay_timer);
614     qemu_sem_destroy(&s->colo_checkpoint_sem);
615 
616     /*
617      * Must be called after failover BH is completed,
618      * Or the failover BH may shutdown the wrong fd that
619      * re-used by other threads after we release here.
620      */
621     if (s->rp_state.from_dst_file) {
622         qemu_fclose(s->rp_state.from_dst_file);
623     }
624 }
625 
626 void colo_checkpoint_notify(void *opaque)
627 {
628     MigrationState *s = opaque;
629     int64_t next_notify_time;
630 
631     qemu_sem_post(&s->colo_checkpoint_sem);
632     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
633     next_notify_time = s->colo_checkpoint_time +
634                     s->parameters.x_checkpoint_delay;
635     timer_mod(s->colo_delay_timer, next_notify_time);
636 }
637 
638 void migrate_start_colo_process(MigrationState *s)
639 {
640     qemu_mutex_unlock_iothread();
641     qemu_sem_init(&s->colo_checkpoint_sem, 0);
642     s->colo_delay_timer =  timer_new_ms(QEMU_CLOCK_HOST,
643                                 colo_checkpoint_notify, s);
644 
645     qemu_sem_init(&s->colo_exit_sem, 0);
646     migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
647                       MIGRATION_STATUS_COLO);
648     colo_process_checkpoint(s);
649     qemu_mutex_lock_iothread();
650 }
651 
652 static void colo_wait_handle_message(QEMUFile *f, int *checkpoint_request,
653                                      Error **errp)
654 {
655     COLOMessage msg;
656     Error *local_err = NULL;
657 
658     msg = colo_receive_message(f, &local_err);
659     if (local_err) {
660         error_propagate(errp, local_err);
661         return;
662     }
663 
664     switch (msg) {
665     case COLO_MESSAGE_CHECKPOINT_REQUEST:
666         *checkpoint_request = 1;
667         break;
668     default:
669         *checkpoint_request = 0;
670         error_setg(errp, "Got unknown COLO message: %d", msg);
671         break;
672     }
673 }
674 
675 void *colo_process_incoming_thread(void *opaque)
676 {
677     MigrationIncomingState *mis = opaque;
678     QEMUFile *fb = NULL;
679     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
680     uint64_t total_size;
681     uint64_t value;
682     Error *local_err = NULL;
683     int ret;
684 
685     rcu_register_thread();
686     qemu_sem_init(&mis->colo_incoming_sem, 0);
687 
688     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
689                       MIGRATION_STATUS_COLO);
690 
691     failover_init_state();
692 
693     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
694     if (!mis->to_src_file) {
695         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
696         goto out;
697     }
698     /*
699      * Note: the communication between Primary side and Secondary side
700      * should be sequential, we set the fd to unblocked in migration incoming
701      * coroutine, and here we are in the COLO incoming thread, so it is ok to
702      * set the fd back to blocked.
703      */
704     qemu_file_set_blocking(mis->from_src_file, true);
705 
706     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
707     fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
708     object_unref(OBJECT(bioc));
709 
710     qemu_mutex_lock_iothread();
711 #ifdef CONFIG_REPLICATION
712     replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
713     if (local_err) {
714         qemu_mutex_unlock_iothread();
715         goto out;
716     }
717 #else
718         abort();
719 #endif
720     vm_start();
721     trace_colo_vm_state_change("stop", "run");
722     qemu_mutex_unlock_iothread();
723 
724     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
725                       &local_err);
726     if (local_err) {
727         goto out;
728     }
729 
730     while (mis->state == MIGRATION_STATUS_COLO) {
731         int request = 0;
732 
733         colo_wait_handle_message(mis->from_src_file, &request, &local_err);
734         if (local_err) {
735             goto out;
736         }
737         assert(request);
738         if (failover_get_state() != FAILOVER_STATUS_NONE) {
739             error_report("failover request");
740             goto out;
741         }
742 
743         qemu_mutex_lock_iothread();
744         vm_stop_force_state(RUN_STATE_COLO);
745         trace_colo_vm_state_change("run", "stop");
746         qemu_mutex_unlock_iothread();
747 
748         /* FIXME: This is unnecessary for periodic checkpoint mode */
749         colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
750                      &local_err);
751         if (local_err) {
752             goto out;
753         }
754 
755         colo_receive_check_message(mis->from_src_file,
756                            COLO_MESSAGE_VMSTATE_SEND, &local_err);
757         if (local_err) {
758             goto out;
759         }
760 
761         qemu_mutex_lock_iothread();
762         cpu_synchronize_all_pre_loadvm();
763         ret = qemu_loadvm_state_main(mis->from_src_file, mis);
764         qemu_mutex_unlock_iothread();
765 
766         if (ret < 0) {
767             error_report("Load VM's live state (ram) error");
768             goto out;
769         }
770 
771         value = colo_receive_message_value(mis->from_src_file,
772                                  COLO_MESSAGE_VMSTATE_SIZE, &local_err);
773         if (local_err) {
774             goto out;
775         }
776 
777         /*
778          * Read VM device state data into channel buffer,
779          * It's better to re-use the memory allocated.
780          * Here we need to handle the channel buffer directly.
781          */
782         if (value > bioc->capacity) {
783             bioc->capacity = value;
784             bioc->data = g_realloc(bioc->data, bioc->capacity);
785         }
786         total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
787         if (total_size != value) {
788             error_report("Got %" PRIu64 " VMState data, less than expected"
789                         " %" PRIu64, total_size, value);
790             goto out;
791         }
792         bioc->usage = total_size;
793         qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
794 
795         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
796                      &local_err);
797         if (local_err) {
798             goto out;
799         }
800 
801         qemu_mutex_lock_iothread();
802         vmstate_loading = true;
803         ret = qemu_load_device_state(fb);
804         if (ret < 0) {
805             error_report("COLO: load device state failed");
806             qemu_mutex_unlock_iothread();
807             goto out;
808         }
809 
810 #ifdef CONFIG_REPLICATION
811         replication_get_error_all(&local_err);
812         if (local_err) {
813             qemu_mutex_unlock_iothread();
814             goto out;
815         }
816 
817         /* discard colo disk buffer */
818         replication_do_checkpoint_all(&local_err);
819         if (local_err) {
820             qemu_mutex_unlock_iothread();
821             goto out;
822         }
823 #else
824         abort();
825 #endif
826         /* Notify all filters of all NIC to do checkpoint */
827         colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
828 
829         if (local_err) {
830             qemu_mutex_unlock_iothread();
831             goto out;
832         }
833 
834         vmstate_loading = false;
835         vm_start();
836         trace_colo_vm_state_change("stop", "run");
837         qemu_mutex_unlock_iothread();
838 
839         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
840             failover_set_state(FAILOVER_STATUS_RELAUNCH,
841                             FAILOVER_STATUS_NONE);
842             failover_request_active(NULL);
843             goto out;
844         }
845 
846         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
847                      &local_err);
848         if (local_err) {
849             goto out;
850         }
851     }
852 
853 out:
854     vmstate_loading = false;
855     /* Throw the unreported error message after exited from loop */
856     if (local_err) {
857         error_report_err(local_err);
858     }
859 
860     switch (failover_get_state()) {
861     case FAILOVER_STATUS_NONE:
862         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
863                                   COLO_EXIT_REASON_ERROR);
864         break;
865     case FAILOVER_STATUS_REQUIRE:
866         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
867                                   COLO_EXIT_REASON_REQUEST);
868         break;
869     default:
870         abort();
871     }
872 
873     if (fb) {
874         qemu_fclose(fb);
875     }
876 
877     /* Hope this not to be too long to loop here */
878     qemu_sem_wait(&mis->colo_incoming_sem);
879     qemu_sem_destroy(&mis->colo_incoming_sem);
880     /* Must be called after failover BH is completed */
881     if (mis->to_src_file) {
882         qemu_fclose(mis->to_src_file);
883     }
884     migration_incoming_disable_colo();
885 
886     rcu_unregister_thread();
887     return NULL;
888 }
889