2 * COarse-grain LOck-stepping Virtual Machines for Non-stop Service (COLO)
3 * (a.k.a. Fault Tolerance or Continuous Replication)
5 * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
6 * Copyright (c) 2016 FUJITSU LIMITED
7 * Copyright (c) 2016 Intel Corporation
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.
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"
21 #include "migration/colo.h"
23 #include "io/channel-buffer.h"
25 #include "qemu/error-report.h"
26 #include "migration/failover.h"
27 #ifdef CONFIG_REPLICATION
28 #include "replication.h"
30 #include "net/colo-compare.h"
32 #include "block/block.h"
33 #include "qapi/qapi-events-migration.h"
34 #include "qapi/qmp/qerror.h"
35 #include "sysemu/cpus.h"
36 #include "net/filter.h"
38 static bool vmstate_loading
;
39 static Notifier packets_compare_notifier
;
41 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
43 bool migration_in_colo_state(void)
45 MigrationState
*s
= migrate_get_current();
47 return (s
->state
== MIGRATION_STATUS_COLO
);
50 bool migration_incoming_in_colo_state(void)
52 MigrationIncomingState
*mis
= migration_incoming_get_current();
54 return mis
&& (mis
->state
== MIGRATION_STATUS_COLO
);
57 static bool colo_runstate_is_stopped(void)
59 return runstate_check(RUN_STATE_COLO
) || !runstate_is_running();
62 static void secondary_vm_do_failover(void)
64 /* COLO needs enable block-replication */
65 #ifdef CONFIG_REPLICATION
67 MigrationIncomingState
*mis
= migration_incoming_get_current();
68 Error
*local_err
= NULL
;
70 /* Can not do failover during the process of VM's loading VMstate, Or
71 * it will break the secondary VM.
73 if (vmstate_loading
) {
74 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
75 FAILOVER_STATUS_RELAUNCH
);
76 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
77 error_report("Unknown error while do failover for secondary VM,"
78 "old_state: %s", FailoverStatus_str(old_state
));
83 migrate_set_state(&mis
->state
, MIGRATION_STATUS_COLO
,
84 MIGRATION_STATUS_COMPLETED
);
86 replication_stop_all(true, &local_err
);
88 error_report_err(local_err
);
91 /* Notify all filters of all NIC to do checkpoint */
92 colo_notify_filters_event(COLO_EVENT_FAILOVER
, &local_err
);
94 error_report_err(local_err
);
98 error_report("\"-S\" qemu option will be ignored in secondary side");
99 /* recover runstate to normal migration finish state */
103 * Make sure COLO incoming thread not block in recv or send,
104 * If mis->from_src_file and mis->to_src_file use the same fd,
105 * The second shutdown() will return -1, we ignore this value,
108 if (mis
->from_src_file
) {
109 qemu_file_shutdown(mis
->from_src_file
);
111 if (mis
->to_src_file
) {
112 qemu_file_shutdown(mis
->to_src_file
);
115 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
116 FAILOVER_STATUS_COMPLETED
);
117 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
118 error_report("Incorrect state (%s) while doing failover for "
119 "secondary VM", FailoverStatus_str(old_state
));
122 /* Notify COLO incoming thread that failover work is finished */
123 qemu_sem_post(&mis
->colo_incoming_sem
);
124 /* For Secondary VM, jump to incoming co */
125 if (mis
->migration_incoming_co
) {
126 qemu_coroutine_enter(mis
->migration_incoming_co
);
133 static void primary_vm_do_failover(void)
135 #ifdef CONFIG_REPLICATION
136 MigrationState
*s
= migrate_get_current();
138 Error
*local_err
= NULL
;
140 migrate_set_state(&s
->state
, MIGRATION_STATUS_COLO
,
141 MIGRATION_STATUS_COMPLETED
);
143 * kick COLO thread which might wait at
144 * qemu_sem_wait(&s->colo_checkpoint_sem).
146 colo_checkpoint_notify(migrate_get_current());
149 * Wake up COLO thread which may blocked in recv() or send(),
150 * The s->rp_state.from_dst_file and s->to_dst_file may use the
151 * same fd, but we still shutdown the fd for twice, it is harmless.
153 if (s
->to_dst_file
) {
154 qemu_file_shutdown(s
->to_dst_file
);
156 if (s
->rp_state
.from_dst_file
) {
157 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
160 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
161 FAILOVER_STATUS_COMPLETED
);
162 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
163 error_report("Incorrect state (%s) while doing failover for Primary VM",
164 FailoverStatus_str(old_state
));
168 replication_stop_all(true, &local_err
);
170 error_report_err(local_err
);
174 /* Notify COLO thread that failover work is finished */
175 qemu_sem_post(&s
->colo_exit_sem
);
181 COLOMode
get_colo_mode(void)
183 if (migration_in_colo_state()) {
184 return COLO_MODE_PRIMARY
;
185 } else if (migration_incoming_in_colo_state()) {
186 return COLO_MODE_SECONDARY
;
188 return COLO_MODE_NONE
;
192 void colo_do_failover(MigrationState
*s
)
194 /* Make sure VM stopped while failover happened. */
195 if (!colo_runstate_is_stopped()) {
196 vm_stop_force_state(RUN_STATE_COLO
);
199 if (get_colo_mode() == COLO_MODE_PRIMARY
) {
200 primary_vm_do_failover();
202 secondary_vm_do_failover();
206 #ifdef CONFIG_REPLICATION
207 void qmp_xen_set_replication(bool enable
, bool primary
,
208 bool has_failover
, bool failover
,
211 ReplicationMode mode
= primary
?
212 REPLICATION_MODE_PRIMARY
:
213 REPLICATION_MODE_SECONDARY
;
215 if (has_failover
&& enable
) {
216 error_setg(errp
, "Parameter 'failover' is only for"
217 " stopping replication");
222 replication_start_all(mode
, errp
);
227 replication_stop_all(failover
, failover
? NULL
: errp
);
231 ReplicationStatus
*qmp_query_xen_replication_status(Error
**errp
)
234 ReplicationStatus
*s
= g_new0(ReplicationStatus
, 1);
236 replication_get_error_all(&err
);
240 s
->desc
= g_strdup(error_get_pretty(err
));
249 void qmp_xen_colo_do_checkpoint(Error
**errp
)
251 replication_do_checkpoint_all(errp
);
255 COLOStatus
*qmp_query_colo_status(Error
**errp
)
257 COLOStatus
*s
= g_new0(COLOStatus
, 1);
259 s
->mode
= get_colo_mode();
261 switch (failover_get_state()) {
262 case FAILOVER_STATUS_NONE
:
263 s
->reason
= COLO_EXIT_REASON_NONE
;
265 case FAILOVER_STATUS_REQUIRE
:
266 s
->reason
= COLO_EXIT_REASON_REQUEST
;
269 s
->reason
= COLO_EXIT_REASON_ERROR
;
275 static void colo_send_message(QEMUFile
*f
, COLOMessage msg
,
280 if (msg
>= COLO_MESSAGE__MAX
) {
281 error_setg(errp
, "%s: Invalid message", __func__
);
284 qemu_put_be32(f
, msg
);
287 ret
= qemu_file_get_error(f
);
289 error_setg_errno(errp
, -ret
, "Can't send COLO message");
291 trace_colo_send_message(COLOMessage_str(msg
));
294 static void colo_send_message_value(QEMUFile
*f
, COLOMessage msg
,
295 uint64_t value
, Error
**errp
)
297 Error
*local_err
= NULL
;
300 colo_send_message(f
, msg
, &local_err
);
302 error_propagate(errp
, local_err
);
305 qemu_put_be64(f
, value
);
308 ret
= qemu_file_get_error(f
);
310 error_setg_errno(errp
, -ret
, "Failed to send value for message:%s",
311 COLOMessage_str(msg
));
315 static COLOMessage
colo_receive_message(QEMUFile
*f
, Error
**errp
)
320 msg
= qemu_get_be32(f
);
321 ret
= qemu_file_get_error(f
);
323 error_setg_errno(errp
, -ret
, "Can't receive COLO message");
326 if (msg
>= COLO_MESSAGE__MAX
) {
327 error_setg(errp
, "%s: Invalid message", __func__
);
330 trace_colo_receive_message(COLOMessage_str(msg
));
334 static void colo_receive_check_message(QEMUFile
*f
, COLOMessage expect_msg
,
338 Error
*local_err
= NULL
;
340 msg
= colo_receive_message(f
, &local_err
);
342 error_propagate(errp
, local_err
);
345 if (msg
!= expect_msg
) {
346 error_setg(errp
, "Unexpected COLO message %d, expected %d",
351 static uint64_t colo_receive_message_value(QEMUFile
*f
, uint32_t expect_msg
,
354 Error
*local_err
= NULL
;
358 colo_receive_check_message(f
, expect_msg
, &local_err
);
360 error_propagate(errp
, local_err
);
364 value
= qemu_get_be64(f
);
365 ret
= qemu_file_get_error(f
);
367 error_setg_errno(errp
, -ret
, "Failed to get value for COLO message: %s",
368 COLOMessage_str(expect_msg
));
373 static int colo_do_checkpoint_transaction(MigrationState
*s
,
374 QIOChannelBuffer
*bioc
,
377 Error
*local_err
= NULL
;
380 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_CHECKPOINT_REQUEST
,
386 colo_receive_check_message(s
->rp_state
.from_dst_file
,
387 COLO_MESSAGE_CHECKPOINT_REPLY
, &local_err
);
391 /* Reset channel-buffer directly */
392 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
395 qemu_mutex_lock_iothread();
396 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
397 qemu_mutex_unlock_iothread();
400 vm_stop_force_state(RUN_STATE_COLO
);
401 qemu_mutex_unlock_iothread();
402 trace_colo_vm_state_change("run", "stop");
404 * Failover request bh could be called after vm_stop_force_state(),
405 * So we need check failover_request_is_active() again.
407 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
411 colo_notify_compares_event(NULL
, COLO_EVENT_CHECKPOINT
, &local_err
);
416 /* Disable block migration */
417 migrate_set_block_enabled(false, &local_err
);
418 qemu_mutex_lock_iothread();
420 #ifdef CONFIG_REPLICATION
421 replication_do_checkpoint_all(&local_err
);
423 qemu_mutex_unlock_iothread();
430 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
432 qemu_mutex_unlock_iothread();
435 /* Note: device state is saved into buffer */
436 ret
= qemu_save_device_state(fb
);
438 qemu_mutex_unlock_iothread();
443 * Only save VM's live state, which not including device state.
444 * TODO: We may need a timeout mechanism to prevent COLO process
445 * to be blocked here.
447 qemu_savevm_live_state(s
->to_dst_file
);
452 * We need the size of the VMstate data in Secondary side,
453 * With which we can decide how much data should be read.
455 colo_send_message_value(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SIZE
,
456 bioc
->usage
, &local_err
);
461 qemu_put_buffer(s
->to_dst_file
, bioc
->data
, bioc
->usage
);
462 qemu_fflush(s
->to_dst_file
);
463 ret
= qemu_file_get_error(s
->to_dst_file
);
468 colo_receive_check_message(s
->rp_state
.from_dst_file
,
469 COLO_MESSAGE_VMSTATE_RECEIVED
, &local_err
);
474 colo_receive_check_message(s
->rp_state
.from_dst_file
,
475 COLO_MESSAGE_VMSTATE_LOADED
, &local_err
);
482 qemu_mutex_lock_iothread();
484 qemu_mutex_unlock_iothread();
485 trace_colo_vm_state_change("stop", "run");
489 error_report_err(local_err
);
494 static void colo_compare_notify_checkpoint(Notifier
*notifier
, void *data
)
496 colo_checkpoint_notify(data
);
499 static void colo_process_checkpoint(MigrationState
*s
)
501 QIOChannelBuffer
*bioc
;
503 int64_t current_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
504 Error
*local_err
= NULL
;
507 failover_init_state();
509 s
->rp_state
.from_dst_file
= qemu_file_get_return_path(s
->to_dst_file
);
510 if (!s
->rp_state
.from_dst_file
) {
511 error_report("Open QEMUFile from_dst_file failed");
515 packets_compare_notifier
.notify
= colo_compare_notify_checkpoint
;
516 colo_compare_register_notifier(&packets_compare_notifier
);
519 * Wait for Secondary finish loading VM states and enter COLO
522 colo_receive_check_message(s
->rp_state
.from_dst_file
,
523 COLO_MESSAGE_CHECKPOINT_READY
, &local_err
);
527 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
528 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
529 object_unref(OBJECT(bioc
));
531 qemu_mutex_lock_iothread();
532 #ifdef CONFIG_REPLICATION
533 replication_start_all(REPLICATION_MODE_PRIMARY
, &local_err
);
535 qemu_mutex_unlock_iothread();
543 qemu_mutex_unlock_iothread();
544 trace_colo_vm_state_change("stop", "run");
546 timer_mod(s
->colo_delay_timer
,
547 current_time
+ s
->parameters
.x_checkpoint_delay
);
549 while (s
->state
== MIGRATION_STATUS_COLO
) {
550 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
551 error_report("failover request");
555 qemu_sem_wait(&s
->colo_checkpoint_sem
);
557 if (s
->state
!= MIGRATION_STATUS_COLO
) {
560 ret
= colo_do_checkpoint_transaction(s
, bioc
, fb
);
567 /* Throw the unreported error message after exited from loop */
569 error_report_err(local_err
);
577 * There are only two reasons we can get here, some error happened
578 * or the user triggered failover.
580 switch (failover_get_state()) {
581 case FAILOVER_STATUS_NONE
:
582 qapi_event_send_colo_exit(COLO_MODE_PRIMARY
,
583 COLO_EXIT_REASON_ERROR
);
585 case FAILOVER_STATUS_REQUIRE
:
586 qapi_event_send_colo_exit(COLO_MODE_PRIMARY
,
587 COLO_EXIT_REASON_REQUEST
);
593 /* Hope this not to be too long to wait here */
594 qemu_sem_wait(&s
->colo_exit_sem
);
595 qemu_sem_destroy(&s
->colo_exit_sem
);
598 * It is safe to unregister notifier after failover finished.
599 * Besides, colo_delay_timer and colo_checkpoint_sem can't be
600 * released befor unregister notifier, or there will be use-after-free
603 colo_compare_unregister_notifier(&packets_compare_notifier
);
604 timer_del(s
->colo_delay_timer
);
605 timer_free(s
->colo_delay_timer
);
606 qemu_sem_destroy(&s
->colo_checkpoint_sem
);
609 * Must be called after failover BH is completed,
610 * Or the failover BH may shutdown the wrong fd that
611 * re-used by other threads after we release here.
613 if (s
->rp_state
.from_dst_file
) {
614 qemu_fclose(s
->rp_state
.from_dst_file
);
618 void colo_checkpoint_notify(void *opaque
)
620 MigrationState
*s
= opaque
;
621 int64_t next_notify_time
;
623 qemu_sem_post(&s
->colo_checkpoint_sem
);
624 s
->colo_checkpoint_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
625 next_notify_time
= s
->colo_checkpoint_time
+
626 s
->parameters
.x_checkpoint_delay
;
627 timer_mod(s
->colo_delay_timer
, next_notify_time
);
630 void migrate_start_colo_process(MigrationState
*s
)
632 qemu_mutex_unlock_iothread();
633 qemu_sem_init(&s
->colo_checkpoint_sem
, 0);
634 s
->colo_delay_timer
= timer_new_ms(QEMU_CLOCK_HOST
,
635 colo_checkpoint_notify
, s
);
637 qemu_sem_init(&s
->colo_exit_sem
, 0);
638 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
639 MIGRATION_STATUS_COLO
);
640 colo_process_checkpoint(s
);
641 qemu_mutex_lock_iothread();
644 static void colo_wait_handle_message(QEMUFile
*f
, int *checkpoint_request
,
648 Error
*local_err
= NULL
;
650 msg
= colo_receive_message(f
, &local_err
);
652 error_propagate(errp
, local_err
);
657 case COLO_MESSAGE_CHECKPOINT_REQUEST
:
658 *checkpoint_request
= 1;
661 *checkpoint_request
= 0;
662 error_setg(errp
, "Got unknown COLO message: %d", msg
);
667 void *colo_process_incoming_thread(void *opaque
)
669 MigrationIncomingState
*mis
= opaque
;
671 QIOChannelBuffer
*bioc
= NULL
; /* Cache incoming device state */
674 Error
*local_err
= NULL
;
677 rcu_register_thread();
678 qemu_sem_init(&mis
->colo_incoming_sem
, 0);
680 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
681 MIGRATION_STATUS_COLO
);
683 failover_init_state();
685 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
686 if (!mis
->to_src_file
) {
687 error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
691 * Note: the communication between Primary side and Secondary side
692 * should be sequential, we set the fd to unblocked in migration incoming
693 * coroutine, and here we are in the COLO incoming thread, so it is ok to
694 * set the fd back to blocked.
696 qemu_file_set_blocking(mis
->from_src_file
, true);
698 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
699 fb
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
700 object_unref(OBJECT(bioc
));
702 qemu_mutex_lock_iothread();
703 #ifdef CONFIG_REPLICATION
704 replication_start_all(REPLICATION_MODE_SECONDARY
, &local_err
);
706 qemu_mutex_unlock_iothread();
713 trace_colo_vm_state_change("stop", "run");
714 qemu_mutex_unlock_iothread();
716 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_READY
,
722 while (mis
->state
== MIGRATION_STATUS_COLO
) {
725 colo_wait_handle_message(mis
->from_src_file
, &request
, &local_err
);
730 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
731 error_report("failover request");
735 qemu_mutex_lock_iothread();
736 vm_stop_force_state(RUN_STATE_COLO
);
737 trace_colo_vm_state_change("run", "stop");
738 qemu_mutex_unlock_iothread();
740 /* FIXME: This is unnecessary for periodic checkpoint mode */
741 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_REPLY
,
747 colo_receive_check_message(mis
->from_src_file
,
748 COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
753 qemu_mutex_lock_iothread();
754 cpu_synchronize_all_pre_loadvm();
755 ret
= qemu_loadvm_state_main(mis
->from_src_file
, mis
);
756 qemu_mutex_unlock_iothread();
759 error_report("Load VM's live state (ram) error");
763 value
= colo_receive_message_value(mis
->from_src_file
,
764 COLO_MESSAGE_VMSTATE_SIZE
, &local_err
);
770 * Read VM device state data into channel buffer,
771 * It's better to re-use the memory allocated.
772 * Here we need to handle the channel buffer directly.
774 if (value
> bioc
->capacity
) {
775 bioc
->capacity
= value
;
776 bioc
->data
= g_realloc(bioc
->data
, bioc
->capacity
);
778 total_size
= qemu_get_buffer(mis
->from_src_file
, bioc
->data
, value
);
779 if (total_size
!= value
) {
780 error_report("Got %" PRIu64
" VMState data, less than expected"
781 " %" PRIu64
, total_size
, value
);
784 bioc
->usage
= total_size
;
785 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
787 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_RECEIVED
,
793 qemu_mutex_lock_iothread();
794 vmstate_loading
= true;
795 ret
= qemu_load_device_state(fb
);
797 error_report("COLO: load device state failed");
798 qemu_mutex_unlock_iothread();
802 #ifdef CONFIG_REPLICATION
803 replication_get_error_all(&local_err
);
805 qemu_mutex_unlock_iothread();
809 /* discard colo disk buffer */
810 replication_do_checkpoint_all(&local_err
);
812 qemu_mutex_unlock_iothread();
818 /* Notify all filters of all NIC to do checkpoint */
819 colo_notify_filters_event(COLO_EVENT_CHECKPOINT
, &local_err
);
822 qemu_mutex_unlock_iothread();
826 vmstate_loading
= false;
828 trace_colo_vm_state_change("stop", "run");
829 qemu_mutex_unlock_iothread();
831 if (failover_get_state() == FAILOVER_STATUS_RELAUNCH
) {
832 failover_set_state(FAILOVER_STATUS_RELAUNCH
,
833 FAILOVER_STATUS_NONE
);
834 failover_request_active(NULL
);
838 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_LOADED
,
846 vmstate_loading
= false;
847 /* Throw the unreported error message after exited from loop */
849 error_report_err(local_err
);
852 switch (failover_get_state()) {
853 case FAILOVER_STATUS_NONE
:
854 qapi_event_send_colo_exit(COLO_MODE_SECONDARY
,
855 COLO_EXIT_REASON_ERROR
);
857 case FAILOVER_STATUS_REQUIRE
:
858 qapi_event_send_colo_exit(COLO_MODE_SECONDARY
,
859 COLO_EXIT_REASON_REQUEST
);
869 /* Hope this not to be too long to loop here */
870 qemu_sem_wait(&mis
->colo_incoming_sem
);
871 qemu_sem_destroy(&mis
->colo_incoming_sem
);
872 /* Must be called after failover BH is completed */
873 if (mis
->to_src_file
) {
874 qemu_fclose(mis
->to_src_file
);
876 migration_incoming_disable_colo();
878 rcu_unregister_thread();