xref: /openbmc/qemu/migration/colo.c (revision ae4c2099)
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 "block/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         local_err = NULL;
97     }
98 
99     /* Notify all filters of all NIC to do checkpoint */
100     colo_notify_filters_event(COLO_EVENT_FAILOVER, &local_err);
101     if (local_err) {
102         error_report_err(local_err);
103     }
104 
105     if (!autostart) {
106         error_report("\"-S\" qemu option will be ignored in secondary side");
107         /* recover runstate to normal migration finish state */
108         autostart = true;
109     }
110     /*
111      * Make sure COLO incoming thread not block in recv or send,
112      * If mis->from_src_file and mis->to_src_file use the same fd,
113      * The second shutdown() will return -1, we ignore this value,
114      * It is harmless.
115      */
116     if (mis->from_src_file) {
117         qemu_file_shutdown(mis->from_src_file);
118     }
119     if (mis->to_src_file) {
120         qemu_file_shutdown(mis->to_src_file);
121     }
122 
123     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
124                                    FAILOVER_STATUS_COMPLETED);
125     if (old_state != FAILOVER_STATUS_ACTIVE) {
126         error_report("Incorrect state (%s) while doing failover for "
127                      "secondary VM", FailoverStatus_str(old_state));
128         return;
129     }
130     /* Notify COLO incoming thread that failover work is finished */
131     qemu_sem_post(&mis->colo_incoming_sem);
132 
133     /* For Secondary VM, jump to incoming co */
134     if (mis->migration_incoming_co) {
135         qemu_coroutine_enter(mis->migration_incoming_co);
136     }
137 #else
138     abort();
139 #endif
140 }
141 
142 static void primary_vm_do_failover(void)
143 {
144 #ifdef CONFIG_REPLICATION
145     MigrationState *s = migrate_get_current();
146     int old_state;
147     Error *local_err = NULL;
148 
149     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
150                       MIGRATION_STATUS_COMPLETED);
151     /*
152      * kick COLO thread which might wait at
153      * qemu_sem_wait(&s->colo_checkpoint_sem).
154      */
155     colo_checkpoint_notify(s);
156 
157     /*
158      * Wake up COLO thread which may blocked in recv() or send(),
159      * The s->rp_state.from_dst_file and s->to_dst_file may use the
160      * same fd, but we still shutdown the fd for twice, it is harmless.
161      */
162     if (s->to_dst_file) {
163         qemu_file_shutdown(s->to_dst_file);
164     }
165     if (s->rp_state.from_dst_file) {
166         qemu_file_shutdown(s->rp_state.from_dst_file);
167     }
168 
169     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
170                                    FAILOVER_STATUS_COMPLETED);
171     if (old_state != FAILOVER_STATUS_ACTIVE) {
172         error_report("Incorrect state (%s) while doing failover for Primary VM",
173                      FailoverStatus_str(old_state));
174         return;
175     }
176 
177     replication_stop_all(true, &local_err);
178     if (local_err) {
179         error_report_err(local_err);
180         local_err = NULL;
181     }
182 
183     /* Notify COLO thread that failover work is finished */
184     qemu_sem_post(&s->colo_exit_sem);
185 #else
186     abort();
187 #endif
188 }
189 
190 COLOMode get_colo_mode(void)
191 {
192     if (migration_in_colo_state()) {
193         return COLO_MODE_PRIMARY;
194     } else if (migration_incoming_in_colo_state()) {
195         return COLO_MODE_SECONDARY;
196     } else {
197         return COLO_MODE_NONE;
198     }
199 }
200 
201 void colo_do_failover(void)
202 {
203     /* Make sure VM stopped while failover happened. */
204     if (!colo_runstate_is_stopped()) {
205         vm_stop_force_state(RUN_STATE_COLO);
206     }
207 
208     switch (get_colo_mode()) {
209     case COLO_MODE_PRIMARY:
210         primary_vm_do_failover();
211         break;
212     case COLO_MODE_SECONDARY:
213         secondary_vm_do_failover();
214         break;
215     default:
216         error_report("colo_do_failover failed because the colo mode"
217                      " could not be obtained");
218     }
219 }
220 
221 #ifdef CONFIG_REPLICATION
222 void qmp_xen_set_replication(bool enable, bool primary,
223                              bool has_failover, bool failover,
224                              Error **errp)
225 {
226     ReplicationMode mode = primary ?
227                            REPLICATION_MODE_PRIMARY :
228                            REPLICATION_MODE_SECONDARY;
229 
230     if (has_failover && enable) {
231         error_setg(errp, "Parameter 'failover' is only for"
232                    " stopping replication");
233         return;
234     }
235 
236     if (enable) {
237         replication_start_all(mode, errp);
238     } else {
239         if (!has_failover) {
240             failover = NULL;
241         }
242         replication_stop_all(failover, failover ? NULL : errp);
243     }
244 }
245 
246 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
247 {
248     Error *err = NULL;
249     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
250 
251     replication_get_error_all(&err);
252     if (err) {
253         s->error = true;
254         s->has_desc = true;
255         s->desc = g_strdup(error_get_pretty(err));
256     } else {
257         s->error = false;
258     }
259 
260     error_free(err);
261     return s;
262 }
263 
264 void qmp_xen_colo_do_checkpoint(Error **errp)
265 {
266     Error *err = NULL;
267 
268     replication_do_checkpoint_all(&err);
269     if (err) {
270         error_propagate(errp, err);
271         return;
272     }
273     /* Notify all filters of all NIC to do checkpoint */
274     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, errp);
275 }
276 #endif
277 
278 COLOStatus *qmp_query_colo_status(Error **errp)
279 {
280     COLOStatus *s = g_new0(COLOStatus, 1);
281 
282     s->mode = get_colo_mode();
283     s->last_mode = last_colo_mode;
284 
285     switch (failover_get_state()) {
286     case FAILOVER_STATUS_NONE:
287         s->reason = COLO_EXIT_REASON_NONE;
288         break;
289     case FAILOVER_STATUS_COMPLETED:
290         s->reason = COLO_EXIT_REASON_REQUEST;
291         break;
292     default:
293         if (migration_in_colo_state()) {
294             s->reason = COLO_EXIT_REASON_PROCESSING;
295         } else {
296             s->reason = COLO_EXIT_REASON_ERROR;
297         }
298     }
299 
300     return s;
301 }
302 
303 static void colo_send_message(QEMUFile *f, COLOMessage msg,
304                               Error **errp)
305 {
306     int ret;
307 
308     if (msg >= COLO_MESSAGE__MAX) {
309         error_setg(errp, "%s: Invalid message", __func__);
310         return;
311     }
312     qemu_put_be32(f, msg);
313     qemu_fflush(f);
314 
315     ret = qemu_file_get_error(f);
316     if (ret < 0) {
317         error_setg_errno(errp, -ret, "Can't send COLO message");
318     }
319     trace_colo_send_message(COLOMessage_str(msg));
320 }
321 
322 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
323                                     uint64_t value, Error **errp)
324 {
325     Error *local_err = NULL;
326     int ret;
327 
328     colo_send_message(f, msg, &local_err);
329     if (local_err) {
330         error_propagate(errp, local_err);
331         return;
332     }
333     qemu_put_be64(f, value);
334     qemu_fflush(f);
335 
336     ret = qemu_file_get_error(f);
337     if (ret < 0) {
338         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
339                          COLOMessage_str(msg));
340     }
341 }
342 
343 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
344 {
345     COLOMessage msg;
346     int ret;
347 
348     msg = qemu_get_be32(f);
349     ret = qemu_file_get_error(f);
350     if (ret < 0) {
351         error_setg_errno(errp, -ret, "Can't receive COLO message");
352         return msg;
353     }
354     if (msg >= COLO_MESSAGE__MAX) {
355         error_setg(errp, "%s: Invalid message", __func__);
356         return msg;
357     }
358     trace_colo_receive_message(COLOMessage_str(msg));
359     return msg;
360 }
361 
362 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
363                                        Error **errp)
364 {
365     COLOMessage msg;
366     Error *local_err = NULL;
367 
368     msg = colo_receive_message(f, &local_err);
369     if (local_err) {
370         error_propagate(errp, local_err);
371         return;
372     }
373     if (msg != expect_msg) {
374         error_setg(errp, "Unexpected COLO message %d, expected %d",
375                           msg, expect_msg);
376     }
377 }
378 
379 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
380                                            Error **errp)
381 {
382     Error *local_err = NULL;
383     uint64_t value;
384     int ret;
385 
386     colo_receive_check_message(f, expect_msg, &local_err);
387     if (local_err) {
388         error_propagate(errp, local_err);
389         return 0;
390     }
391 
392     value = qemu_get_be64(f);
393     ret = qemu_file_get_error(f);
394     if (ret < 0) {
395         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
396                          COLOMessage_str(expect_msg));
397     }
398     return value;
399 }
400 
401 static int colo_do_checkpoint_transaction(MigrationState *s,
402                                           QIOChannelBuffer *bioc,
403                                           QEMUFile *fb)
404 {
405     Error *local_err = NULL;
406     int ret = -1;
407 
408     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
409                       &local_err);
410     if (local_err) {
411         goto out;
412     }
413 
414     colo_receive_check_message(s->rp_state.from_dst_file,
415                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
416     if (local_err) {
417         goto out;
418     }
419     /* Reset channel-buffer directly */
420     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
421     bioc->usage = 0;
422 
423     qemu_mutex_lock_iothread();
424     if (failover_get_state() != FAILOVER_STATUS_NONE) {
425         qemu_mutex_unlock_iothread();
426         goto out;
427     }
428     vm_stop_force_state(RUN_STATE_COLO);
429     qemu_mutex_unlock_iothread();
430     trace_colo_vm_state_change("run", "stop");
431     /*
432      * Failover request bh could be called after vm_stop_force_state(),
433      * So we need check failover_request_is_active() again.
434      */
435     if (failover_get_state() != FAILOVER_STATUS_NONE) {
436         goto out;
437     }
438     qemu_mutex_lock_iothread();
439 
440 #ifdef CONFIG_REPLICATION
441     replication_do_checkpoint_all(&local_err);
442     if (local_err) {
443         qemu_mutex_unlock_iothread();
444         goto out;
445     }
446 #else
447         abort();
448 #endif
449 
450     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
451     if (local_err) {
452         qemu_mutex_unlock_iothread();
453         goto out;
454     }
455     /* Note: device state is saved into buffer */
456     ret = qemu_save_device_state(fb);
457 
458     qemu_mutex_unlock_iothread();
459     if (ret < 0) {
460         goto out;
461     }
462     /*
463      * Only save VM's live state, which not including device state.
464      * TODO: We may need a timeout mechanism to prevent COLO process
465      * to be blocked here.
466      */
467     qemu_savevm_live_state(s->to_dst_file);
468 
469     qemu_fflush(fb);
470 
471     /*
472      * We need the size of the VMstate data in Secondary side,
473      * With which we can decide how much data should be read.
474      */
475     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
476                             bioc->usage, &local_err);
477     if (local_err) {
478         goto out;
479     }
480 
481     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
482     qemu_fflush(s->to_dst_file);
483     ret = qemu_file_get_error(s->to_dst_file);
484     if (ret < 0) {
485         goto out;
486     }
487 
488     colo_receive_check_message(s->rp_state.from_dst_file,
489                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
490     if (local_err) {
491         goto out;
492     }
493 
494     qemu_event_reset(&s->colo_checkpoint_event);
495     colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
496     if (local_err) {
497         goto out;
498     }
499 
500     colo_receive_check_message(s->rp_state.from_dst_file,
501                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
502     if (local_err) {
503         goto out;
504     }
505 
506     ret = 0;
507 
508     qemu_mutex_lock_iothread();
509     vm_start();
510     qemu_mutex_unlock_iothread();
511     trace_colo_vm_state_change("stop", "run");
512 
513 out:
514     if (local_err) {
515         error_report_err(local_err);
516     }
517     return ret;
518 }
519 
520 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
521 {
522     colo_checkpoint_notify(data);
523 }
524 
525 static void colo_process_checkpoint(MigrationState *s)
526 {
527     QIOChannelBuffer *bioc;
528     QEMUFile *fb = NULL;
529     int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
530     Error *local_err = NULL;
531     int ret;
532 
533     last_colo_mode = get_colo_mode();
534     if (last_colo_mode != COLO_MODE_PRIMARY) {
535         error_report("COLO mode must be COLO_MODE_PRIMARY");
536         return;
537     }
538 
539     failover_init_state();
540 
541     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
542     if (!s->rp_state.from_dst_file) {
543         error_report("Open QEMUFile from_dst_file failed");
544         goto out;
545     }
546 
547     packets_compare_notifier.notify = colo_compare_notify_checkpoint;
548     colo_compare_register_notifier(&packets_compare_notifier);
549 
550     /*
551      * Wait for Secondary finish loading VM states and enter COLO
552      * restore.
553      */
554     colo_receive_check_message(s->rp_state.from_dst_file,
555                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
556     if (local_err) {
557         goto out;
558     }
559     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
560     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
561     object_unref(OBJECT(bioc));
562 
563     qemu_mutex_lock_iothread();
564 #ifdef CONFIG_REPLICATION
565     replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
566     if (local_err) {
567         qemu_mutex_unlock_iothread();
568         goto out;
569     }
570 #else
571         abort();
572 #endif
573 
574     vm_start();
575     qemu_mutex_unlock_iothread();
576     trace_colo_vm_state_change("stop", "run");
577 
578     timer_mod(s->colo_delay_timer,
579             current_time + s->parameters.x_checkpoint_delay);
580 
581     while (s->state == MIGRATION_STATUS_COLO) {
582         if (failover_get_state() != FAILOVER_STATUS_NONE) {
583             error_report("failover request");
584             goto out;
585         }
586 
587         qemu_event_wait(&s->colo_checkpoint_event);
588 
589         if (s->state != MIGRATION_STATUS_COLO) {
590             goto out;
591         }
592         ret = colo_do_checkpoint_transaction(s, bioc, fb);
593         if (ret < 0) {
594             goto out;
595         }
596     }
597 
598 out:
599     /* Throw the unreported error message after exited from loop */
600     if (local_err) {
601         error_report_err(local_err);
602     }
603 
604     if (fb) {
605         qemu_fclose(fb);
606     }
607 
608     /*
609      * There are only two reasons we can get here, some error happened
610      * or the user triggered failover.
611      */
612     switch (failover_get_state()) {
613     case FAILOVER_STATUS_COMPLETED:
614         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
615                                   COLO_EXIT_REASON_REQUEST);
616         break;
617     default:
618         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
619                                   COLO_EXIT_REASON_ERROR);
620     }
621 
622     /* Hope this not to be too long to wait here */
623     qemu_sem_wait(&s->colo_exit_sem);
624     qemu_sem_destroy(&s->colo_exit_sem);
625 
626     /*
627      * It is safe to unregister notifier after failover finished.
628      * Besides, colo_delay_timer and colo_checkpoint_sem can't be
629      * released before unregister notifier, or there will be use-after-free
630      * error.
631      */
632     colo_compare_unregister_notifier(&packets_compare_notifier);
633     timer_free(s->colo_delay_timer);
634     qemu_event_destroy(&s->colo_checkpoint_event);
635 
636     /*
637      * Must be called after failover BH is completed,
638      * Or the failover BH may shutdown the wrong fd that
639      * re-used by other threads after we release here.
640      */
641     if (s->rp_state.from_dst_file) {
642         qemu_fclose(s->rp_state.from_dst_file);
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     migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
667                       MIGRATION_STATUS_COLO);
668     colo_process_checkpoint(s);
669     qemu_mutex_lock_iothread();
670 }
671 
672 static void colo_incoming_process_checkpoint(MigrationIncomingState *mis,
673                       QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
674 {
675     uint64_t total_size;
676     uint64_t value;
677     Error *local_err = NULL;
678     int ret;
679 
680     qemu_mutex_lock_iothread();
681     vm_stop_force_state(RUN_STATE_COLO);
682     trace_colo_vm_state_change("run", "stop");
683     qemu_mutex_unlock_iothread();
684 
685     /* FIXME: This is unnecessary for periodic checkpoint mode */
686     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
687                  &local_err);
688     if (local_err) {
689         error_propagate(errp, local_err);
690         return;
691     }
692 
693     colo_receive_check_message(mis->from_src_file,
694                        COLO_MESSAGE_VMSTATE_SEND, &local_err);
695     if (local_err) {
696         error_propagate(errp, local_err);
697         return;
698     }
699 
700     qemu_mutex_lock_iothread();
701     cpu_synchronize_all_states();
702     ret = qemu_loadvm_state_main(mis->from_src_file, mis);
703     qemu_mutex_unlock_iothread();
704 
705     if (ret < 0) {
706         error_setg(errp, "Load VM's live state (ram) error");
707         return;
708     }
709 
710     value = colo_receive_message_value(mis->from_src_file,
711                              COLO_MESSAGE_VMSTATE_SIZE, &local_err);
712     if (local_err) {
713         error_propagate(errp, local_err);
714         return;
715     }
716 
717     /*
718      * Read VM device state data into channel buffer,
719      * It's better to re-use the memory allocated.
720      * Here we need to handle the channel buffer directly.
721      */
722     if (value > bioc->capacity) {
723         bioc->capacity = value;
724         bioc->data = g_realloc(bioc->data, bioc->capacity);
725     }
726     total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
727     if (total_size != value) {
728         error_setg(errp, "Got %" PRIu64 " VMState data, less than expected"
729                     " %" PRIu64, total_size, value);
730         return;
731     }
732     bioc->usage = total_size;
733     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
734 
735     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
736                  &local_err);
737     if (local_err) {
738         error_propagate(errp, local_err);
739         return;
740     }
741 
742     qemu_mutex_lock_iothread();
743     vmstate_loading = true;
744     colo_flush_ram_cache();
745     ret = qemu_load_device_state(fb);
746     if (ret < 0) {
747         error_setg(errp, "COLO: load device state failed");
748         vmstate_loading = false;
749         qemu_mutex_unlock_iothread();
750         return;
751     }
752 
753 #ifdef CONFIG_REPLICATION
754     replication_get_error_all(&local_err);
755     if (local_err) {
756         error_propagate(errp, local_err);
757         vmstate_loading = false;
758         qemu_mutex_unlock_iothread();
759         return;
760     }
761 
762     /* discard colo disk buffer */
763     replication_do_checkpoint_all(&local_err);
764     if (local_err) {
765         error_propagate(errp, local_err);
766         vmstate_loading = false;
767         qemu_mutex_unlock_iothread();
768         return;
769     }
770 #else
771     abort();
772 #endif
773     /* Notify all filters of all NIC to do checkpoint */
774     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
775 
776     if (local_err) {
777         error_propagate(errp, local_err);
778         vmstate_loading = false;
779         qemu_mutex_unlock_iothread();
780         return;
781     }
782 
783     vmstate_loading = false;
784     vm_start();
785     trace_colo_vm_state_change("stop", "run");
786     qemu_mutex_unlock_iothread();
787 
788     if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
789         return;
790     }
791 
792     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
793                  &local_err);
794     error_propagate(errp, local_err);
795 }
796 
797 static void colo_wait_handle_message(MigrationIncomingState *mis,
798                 QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
799 {
800     COLOMessage msg;
801     Error *local_err = NULL;
802 
803     msg = colo_receive_message(mis->from_src_file, &local_err);
804     if (local_err) {
805         error_propagate(errp, local_err);
806         return;
807     }
808 
809     switch (msg) {
810     case COLO_MESSAGE_CHECKPOINT_REQUEST:
811         colo_incoming_process_checkpoint(mis, fb, bioc, errp);
812         break;
813     default:
814         error_setg(errp, "Got unknown COLO message: %d", msg);
815         break;
816     }
817 }
818 
819 void *colo_process_incoming_thread(void *opaque)
820 {
821     MigrationIncomingState *mis = opaque;
822     QEMUFile *fb = NULL;
823     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
824     Error *local_err = NULL;
825 
826     rcu_register_thread();
827     qemu_sem_init(&mis->colo_incoming_sem, 0);
828 
829     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
830                       MIGRATION_STATUS_COLO);
831 
832     last_colo_mode = get_colo_mode();
833     if (last_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_fopen_channel_input(QIO_CHANNEL(bioc));
857     object_unref(OBJECT(bioc));
858 
859     qemu_mutex_lock_iothread();
860 #ifdef CONFIG_REPLICATION
861     replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
862     if (local_err) {
863         qemu_mutex_unlock_iothread();
864         goto out;
865     }
866 #else
867         abort();
868 #endif
869     vm_start();
870     trace_colo_vm_state_change("stop", "run");
871     qemu_mutex_unlock_iothread();
872 
873     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
874                       &local_err);
875     if (local_err) {
876         goto out;
877     }
878 
879     while (mis->state == MIGRATION_STATUS_COLO) {
880         colo_wait_handle_message(mis, fb, bioc, &local_err);
881         if (local_err) {
882             error_report_err(local_err);
883             break;
884         }
885 
886         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
887             failover_set_state(FAILOVER_STATUS_RELAUNCH,
888                             FAILOVER_STATUS_NONE);
889             failover_request_active(NULL);
890             break;
891         }
892 
893         if (failover_get_state() != FAILOVER_STATUS_NONE) {
894             error_report("failover request");
895             break;
896         }
897     }
898 
899 out:
900     /*
901      * There are only two reasons we can get here, some error happened
902      * or the user triggered failover.
903      */
904     switch (failover_get_state()) {
905     case FAILOVER_STATUS_COMPLETED:
906         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
907                                   COLO_EXIT_REASON_REQUEST);
908         break;
909     default:
910         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
911                                   COLO_EXIT_REASON_ERROR);
912     }
913 
914     if (fb) {
915         qemu_fclose(fb);
916     }
917 
918     /* Hope this not to be too long to loop here */
919     qemu_sem_wait(&mis->colo_incoming_sem);
920     qemu_sem_destroy(&mis->colo_incoming_sem);
921     /* Must be called after failover BH is completed */
922     if (mis->to_src_file) {
923         qemu_fclose(mis->to_src_file);
924         mis->to_src_file = NULL;
925     }
926 
927     rcu_unregister_thread();
928     return NULL;
929 }
930