hw/display/xlnx_dp: Free FIFOs adding xlnx_dp_finalize()
[qemu/ar7.git] / migration / colo.c
blobde27662cab561605d3d53d3b5659f4e0e78c5a19
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  */
13 #include "qemu/osdep.h"
14 #include "sysemu/sysemu.h"
15 #include "qapi/error.h"
16 #include "qapi/qapi-commands-migration.h"
17 #include "qemu-file-channel.h"
18 #include "migration.h"
19 #include "qemu-file.h"
20 #include "savevm.h"
21 #include "migration/colo.h"
22 #include "block.h"
23 #include "io/channel-buffer.h"
24 #include "trace.h"
25 #include "qemu/error-report.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/rcu.h"
28 #include "migration/failover.h"
29 #include "migration/ram.h"
30 #ifdef CONFIG_REPLICATION
31 #include "replication.h"
32 #endif
33 #include "net/colo-compare.h"
34 #include "net/colo.h"
35 #include "block/block.h"
36 #include "qapi/qapi-events-migration.h"
37 #include "qapi/qmp/qerror.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/runstate.h"
40 #include "net/filter.h"
42 static bool vmstate_loading;
43 static Notifier packets_compare_notifier;
45 /* User need to know colo mode after COLO failover */
46 static COLOMode last_colo_mode;
48 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
50 bool migration_in_colo_state(void)
52     MigrationState *s = migrate_get_current();
54     return (s->state == MIGRATION_STATUS_COLO);
57 bool migration_incoming_in_colo_state(void)
59     MigrationIncomingState *mis = migration_incoming_get_current();
61     return mis && (mis->state == MIGRATION_STATUS_COLO);
64 static bool colo_runstate_is_stopped(void)
66     return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
69 static void secondary_vm_do_failover(void)
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;
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     }
90     migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
91                       MIGRATION_STATUS_COMPLETED);
93     replication_stop_all(true, &local_err);
94     if (local_err) {
95         error_report_err(local_err);
96         local_err = NULL;
97     }
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     }
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     }
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);
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
142 static void primary_vm_do_failover(void)
144 #ifdef CONFIG_REPLICATION
145     MigrationState *s = migrate_get_current();
146     int old_state;
147     Error *local_err = NULL;
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(migrate_get_current());
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     }
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     }
177     replication_stop_all(true, &local_err);
178     if (local_err) {
179         error_report_err(local_err);
180         local_err = NULL;
181     }
183     /* Notify COLO thread that failover work is finished */
184     qemu_sem_post(&s->colo_exit_sem);
185 #else
186     abort();
187 #endif
190 COLOMode get_colo_mode(void)
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     }
201 void colo_do_failover(void)
203     /* Make sure VM stopped while failover happened. */
204     if (!colo_runstate_is_stopped()) {
205         vm_stop_force_state(RUN_STATE_COLO);
206     }
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     }
221 #ifdef CONFIG_REPLICATION
222 void qmp_xen_set_replication(bool enable, bool primary,
223                              bool has_failover, bool failover,
224                              Error **errp)
226     ReplicationMode mode = primary ?
227                            REPLICATION_MODE_PRIMARY :
228                            REPLICATION_MODE_SECONDARY;
230     if (has_failover && enable) {
231         error_setg(errp, "Parameter 'failover' is only for"
232                    " stopping replication");
233         return;
234     }
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     }
246 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
248     Error *err = NULL;
249     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
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     }
260     error_free(err);
261     return s;
264 void qmp_xen_colo_do_checkpoint(Error **errp)
266     Error *err = NULL;
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);
276 #endif
278 COLOStatus *qmp_query_colo_status(Error **errp)
280     COLOStatus *s = g_new0(COLOStatus, 1);
282     s->mode = get_colo_mode();
283     s->last_mode = last_colo_mode;
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     }
300     return s;
303 static void colo_send_message(QEMUFile *f, COLOMessage msg,
304                               Error **errp)
306     int ret;
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);
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));
322 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
323                                     uint64_t value, Error **errp)
325     Error *local_err = NULL;
326     int ret;
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);
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     }
343 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
345     COLOMessage msg;
346     int ret;
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;
362 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
363                                        Error **errp)
365     COLOMessage msg;
366     Error *local_err = NULL;
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     }
379 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
380                                            Error **errp)
382     Error *local_err = NULL;
383     uint64_t value;
384     int ret;
386     colo_receive_check_message(f, expect_msg, &local_err);
387     if (local_err) {
388         error_propagate(errp, local_err);
389         return 0;
390     }
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;
401 static int colo_do_checkpoint_transaction(MigrationState *s,
402                                           QIOChannelBuffer *bioc,
403                                           QEMUFile *fb)
405     Error *local_err = NULL;
406     int ret = -1;
408     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
409                       &local_err);
410     if (local_err) {
411         goto out;
412     }
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;
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     }
439     /* Disable block migration */
440     migrate_set_block_enabled(false, &local_err);
441     if (local_err) {
442         goto out;
443     }
444     qemu_mutex_lock_iothread();
446 #ifdef CONFIG_REPLICATION
447     replication_do_checkpoint_all(&local_err);
448     if (local_err) {
449         qemu_mutex_unlock_iothread();
450         goto out;
451     }
452 #else
453         abort();
454 #endif
456     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
457     if (local_err) {
458         qemu_mutex_unlock_iothread();
459         goto out;
460     }
461     /* Note: device state is saved into buffer */
462     ret = qemu_save_device_state(fb);
464     qemu_mutex_unlock_iothread();
465     if (ret < 0) {
466         goto out;
467     }
468     /*
469      * Only save VM's live state, which not including device state.
470      * TODO: We may need a timeout mechanism to prevent COLO process
471      * to be blocked here.
472      */
473     qemu_savevm_live_state(s->to_dst_file);
475     qemu_fflush(fb);
477     /*
478      * We need the size of the VMstate data in Secondary side,
479      * With which we can decide how much data should be read.
480      */
481     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
482                             bioc->usage, &local_err);
483     if (local_err) {
484         goto out;
485     }
487     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
488     qemu_fflush(s->to_dst_file);
489     ret = qemu_file_get_error(s->to_dst_file);
490     if (ret < 0) {
491         goto out;
492     }
494     colo_receive_check_message(s->rp_state.from_dst_file,
495                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
496     if (local_err) {
497         goto out;
498     }
500     qemu_event_reset(&s->colo_checkpoint_event);
501     colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
502     if (local_err) {
503         goto out;
504     }
506     colo_receive_check_message(s->rp_state.from_dst_file,
507                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
508     if (local_err) {
509         goto out;
510     }
512     ret = 0;
514     qemu_mutex_lock_iothread();
515     vm_start();
516     qemu_mutex_unlock_iothread();
517     trace_colo_vm_state_change("stop", "run");
519 out:
520     if (local_err) {
521         error_report_err(local_err);
522     }
523     return ret;
526 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
528     colo_checkpoint_notify(data);
531 static void colo_process_checkpoint(MigrationState *s)
533     QIOChannelBuffer *bioc;
534     QEMUFile *fb = NULL;
535     int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
536     Error *local_err = NULL;
537     int ret;
539     last_colo_mode = get_colo_mode();
540     if (last_colo_mode != COLO_MODE_PRIMARY) {
541         error_report("COLO mode must be COLO_MODE_PRIMARY");
542         return;
543     }
545     failover_init_state();
547     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
548     if (!s->rp_state.from_dst_file) {
549         error_report("Open QEMUFile from_dst_file failed");
550         goto out;
551     }
553     packets_compare_notifier.notify = colo_compare_notify_checkpoint;
554     colo_compare_register_notifier(&packets_compare_notifier);
556     /*
557      * Wait for Secondary finish loading VM states and enter COLO
558      * restore.
559      */
560     colo_receive_check_message(s->rp_state.from_dst_file,
561                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
562     if (local_err) {
563         goto out;
564     }
565     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
566     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
567     object_unref(OBJECT(bioc));
569     qemu_mutex_lock_iothread();
570 #ifdef CONFIG_REPLICATION
571     replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
572     if (local_err) {
573         qemu_mutex_unlock_iothread();
574         goto out;
575     }
576 #else
577         abort();
578 #endif
580     vm_start();
581     qemu_mutex_unlock_iothread();
582     trace_colo_vm_state_change("stop", "run");
584     timer_mod(s->colo_delay_timer,
585             current_time + s->parameters.x_checkpoint_delay);
587     while (s->state == MIGRATION_STATUS_COLO) {
588         if (failover_get_state() != FAILOVER_STATUS_NONE) {
589             error_report("failover request");
590             goto out;
591         }
593         qemu_event_wait(&s->colo_checkpoint_event);
595         if (s->state != MIGRATION_STATUS_COLO) {
596             goto out;
597         }
598         ret = colo_do_checkpoint_transaction(s, bioc, fb);
599         if (ret < 0) {
600             goto out;
601         }
602     }
604 out:
605     /* Throw the unreported error message after exited from loop */
606     if (local_err) {
607         error_report_err(local_err);
608     }
610     if (fb) {
611         qemu_fclose(fb);
612     }
614     /*
615      * There are only two reasons we can get here, some error happened
616      * or the user triggered failover.
617      */
618     switch (failover_get_state()) {
619     case FAILOVER_STATUS_COMPLETED:
620         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
621                                   COLO_EXIT_REASON_REQUEST);
622         break;
623     default:
624         qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
625                                   COLO_EXIT_REASON_ERROR);
626     }
628     /* Hope this not to be too long to wait here */
629     qemu_sem_wait(&s->colo_exit_sem);
630     qemu_sem_destroy(&s->colo_exit_sem);
632     /*
633      * It is safe to unregister notifier after failover finished.
634      * Besides, colo_delay_timer and colo_checkpoint_sem can't be
635      * released before unregister notifier, or there will be use-after-free
636      * error.
637      */
638     colo_compare_unregister_notifier(&packets_compare_notifier);
639     timer_free(s->colo_delay_timer);
640     qemu_event_destroy(&s->colo_checkpoint_event);
642     /*
643      * Must be called after failover BH is completed,
644      * Or the failover BH may shutdown the wrong fd that
645      * re-used by other threads after we release here.
646      */
647     if (s->rp_state.from_dst_file) {
648         qemu_fclose(s->rp_state.from_dst_file);
649     }
652 void colo_checkpoint_notify(void *opaque)
654     MigrationState *s = opaque;
655     int64_t next_notify_time;
657     qemu_event_set(&s->colo_checkpoint_event);
658     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
659     next_notify_time = s->colo_checkpoint_time +
660                     s->parameters.x_checkpoint_delay;
661     timer_mod(s->colo_delay_timer, next_notify_time);
664 void migrate_start_colo_process(MigrationState *s)
666     qemu_mutex_unlock_iothread();
667     qemu_event_init(&s->colo_checkpoint_event, false);
668     s->colo_delay_timer =  timer_new_ms(QEMU_CLOCK_HOST,
669                                 colo_checkpoint_notify, s);
671     qemu_sem_init(&s->colo_exit_sem, 0);
672     migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
673                       MIGRATION_STATUS_COLO);
674     colo_process_checkpoint(s);
675     qemu_mutex_lock_iothread();
678 static void colo_incoming_process_checkpoint(MigrationIncomingState *mis,
679                       QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
681     uint64_t total_size;
682     uint64_t value;
683     Error *local_err = NULL;
684     int ret;
686     qemu_mutex_lock_iothread();
687     vm_stop_force_state(RUN_STATE_COLO);
688     trace_colo_vm_state_change("run", "stop");
689     qemu_mutex_unlock_iothread();
691     /* FIXME: This is unnecessary for periodic checkpoint mode */
692     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
693                  &local_err);
694     if (local_err) {
695         error_propagate(errp, local_err);
696         return;
697     }
699     colo_receive_check_message(mis->from_src_file,
700                        COLO_MESSAGE_VMSTATE_SEND, &local_err);
701     if (local_err) {
702         error_propagate(errp, local_err);
703         return;
704     }
706     qemu_mutex_lock_iothread();
707     cpu_synchronize_all_states();
708     ret = qemu_loadvm_state_main(mis->from_src_file, mis);
709     qemu_mutex_unlock_iothread();
711     if (ret < 0) {
712         error_setg(errp, "Load VM's live state (ram) error");
713         return;
714     }
716     value = colo_receive_message_value(mis->from_src_file,
717                              COLO_MESSAGE_VMSTATE_SIZE, &local_err);
718     if (local_err) {
719         error_propagate(errp, local_err);
720         return;
721     }
723     /*
724      * Read VM device state data into channel buffer,
725      * It's better to re-use the memory allocated.
726      * Here we need to handle the channel buffer directly.
727      */
728     if (value > bioc->capacity) {
729         bioc->capacity = value;
730         bioc->data = g_realloc(bioc->data, bioc->capacity);
731     }
732     total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
733     if (total_size != value) {
734         error_setg(errp, "Got %" PRIu64 " VMState data, less than expected"
735                     " %" PRIu64, total_size, value);
736         return;
737     }
738     bioc->usage = total_size;
739     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
741     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
742                  &local_err);
743     if (local_err) {
744         error_propagate(errp, local_err);
745         return;
746     }
748     qemu_mutex_lock_iothread();
749     vmstate_loading = true;
750     colo_flush_ram_cache();
751     ret = qemu_load_device_state(fb);
752     if (ret < 0) {
753         error_setg(errp, "COLO: load device state failed");
754         vmstate_loading = false;
755         qemu_mutex_unlock_iothread();
756         return;
757     }
759 #ifdef CONFIG_REPLICATION
760     replication_get_error_all(&local_err);
761     if (local_err) {
762         error_propagate(errp, local_err);
763         vmstate_loading = false;
764         qemu_mutex_unlock_iothread();
765         return;
766     }
768     /* discard colo disk buffer */
769     replication_do_checkpoint_all(&local_err);
770     if (local_err) {
771         error_propagate(errp, local_err);
772         vmstate_loading = false;
773         qemu_mutex_unlock_iothread();
774         return;
775     }
776 #else
777     abort();
778 #endif
779     /* Notify all filters of all NIC to do checkpoint */
780     colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
782     if (local_err) {
783         error_propagate(errp, local_err);
784         vmstate_loading = false;
785         qemu_mutex_unlock_iothread();
786         return;
787     }
789     vmstate_loading = false;
790     vm_start();
791     trace_colo_vm_state_change("stop", "run");
792     qemu_mutex_unlock_iothread();
794     if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
795         return;
796     }
798     colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
799                  &local_err);
800     error_propagate(errp, local_err);
803 static void colo_wait_handle_message(MigrationIncomingState *mis,
804                 QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
806     COLOMessage msg;
807     Error *local_err = NULL;
809     msg = colo_receive_message(mis->from_src_file, &local_err);
810     if (local_err) {
811         error_propagate(errp, local_err);
812         return;
813     }
815     switch (msg) {
816     case COLO_MESSAGE_CHECKPOINT_REQUEST:
817         colo_incoming_process_checkpoint(mis, fb, bioc, errp);
818         break;
819     default:
820         error_setg(errp, "Got unknown COLO message: %d", msg);
821         break;
822     }
825 void *colo_process_incoming_thread(void *opaque)
827     MigrationIncomingState *mis = opaque;
828     QEMUFile *fb = NULL;
829     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
830     Error *local_err = NULL;
832     rcu_register_thread();
833     qemu_sem_init(&mis->colo_incoming_sem, 0);
835     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
836                       MIGRATION_STATUS_COLO);
838     last_colo_mode = get_colo_mode();
839     if (last_colo_mode != COLO_MODE_SECONDARY) {
840         error_report("COLO mode must be COLO_MODE_SECONDARY");
841         return NULL;
842     }
844     failover_init_state();
846     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
847     if (!mis->to_src_file) {
848         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
849         goto out;
850     }
851     /*
852      * Note: the communication between Primary side and Secondary side
853      * should be sequential, we set the fd to unblocked in migration incoming
854      * coroutine, and here we are in the COLO incoming thread, so it is ok to
855      * set the fd back to blocked.
856      */
857     qemu_file_set_blocking(mis->from_src_file, true);
859     colo_incoming_start_dirty_log();
861     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
862     fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
863     object_unref(OBJECT(bioc));
865     qemu_mutex_lock_iothread();
866 #ifdef CONFIG_REPLICATION
867     replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
868     if (local_err) {
869         qemu_mutex_unlock_iothread();
870         goto out;
871     }
872 #else
873         abort();
874 #endif
875     vm_start();
876     trace_colo_vm_state_change("stop", "run");
877     qemu_mutex_unlock_iothread();
879     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
880                       &local_err);
881     if (local_err) {
882         goto out;
883     }
885     while (mis->state == MIGRATION_STATUS_COLO) {
886         colo_wait_handle_message(mis, fb, bioc, &local_err);
887         if (local_err) {
888             error_report_err(local_err);
889             break;
890         }
892         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
893             failover_set_state(FAILOVER_STATUS_RELAUNCH,
894                             FAILOVER_STATUS_NONE);
895             failover_request_active(NULL);
896             break;
897         }
899         if (failover_get_state() != FAILOVER_STATUS_NONE) {
900             error_report("failover request");
901             break;
902         }
903     }
905 out:
906     /*
907      * There are only two reasons we can get here, some error happened
908      * or the user triggered failover.
909      */
910     switch (failover_get_state()) {
911     case FAILOVER_STATUS_COMPLETED:
912         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
913                                   COLO_EXIT_REASON_REQUEST);
914         break;
915     default:
916         qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
917                                   COLO_EXIT_REASON_ERROR);
918     }
920     if (fb) {
921         qemu_fclose(fb);
922     }
924     /* Hope this not to be too long to loop here */
925     qemu_sem_wait(&mis->colo_incoming_sem);
926     qemu_sem_destroy(&mis->colo_incoming_sem);
927     /* Must be called after failover BH is completed */
928     if (mis->to_src_file) {
929         qemu_fclose(mis->to_src_file);
930         mis->to_src_file = NULL;
931     }
933     rcu_unregister_thread();
934     return NULL;